Grammatical Theory and Metascience PDF

GRAMMATICAL THEORY AND METASCIENCE
AMSTERDAM STUDIES IN THE THEORY AND

HISTORY OF LINGUISTIC SCIENCE
E.F.K. KOERNER, General Editor
Series IV - CURRENT ISSUES IN LINGUISTIC THEORY
Advisory Editorial Board
Henning Andersen (Copenhagen); Raimo Anttila (Los Angeles)

Tomaz V. Gamkrelidze (Tiflis); Klaus J. Kohler (Kiel)
J. Peter Maher (Hamburg); Ernst Pulgram (Ann Arbor, Mich.)
E. Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Honolulu)
Volume 5
Esa Itkonen
Grammatical Theory and Metascience

A CRITICAL INVESTIGATION
INTO THE METHODOLOGICAL AND PHILOSOPHICAL
FOUNDATIONS OF 'AUTONOMOUS' LINGUISTICS
ESA ITKONEN
University of Helsinki
AMSTERDAM / JOHN BENJAMINS B.V.
1978
©Copyright 1978 - John Benjamins B.V.
ISBN 90 272 0901 4 / 90 272 0906 5
No part of this book may be reproduced in any form, by print, photoprint,
microfilm or any other means, without written permission from the publisher.
PREFACE
By 'metascience' I understand the methodology and/or the philo
sophy of a given science, or of science in general. 'Science' will
be used in the sense of Wissenschaft, i.e., as covering the area of
natural sciences (e.g., physics), human sciences (e.g., sociology),
and formal sciences (e.g., logic). 'Grammatical theory', or more
simply 'grammar', will stand for Saussurean autonomous linguistics,
as distinguished from socio- and psycholinguist!'cs on the one hand,
and from mathematical linguistics on the other.
The present investigation will concern itself with the meta
scientific status of grammatical theory. This does not, however,
mean that I think other forms of linguistics are less important. In
my forthcoming article "Qualitative vs. Quantitative Analysis in
Linguistics" and, in more detail, in my forthcoming monograph Causali
ty in Linguistics , I shall analyse the metascientific status of socio-
and psycholinguistics.
I shall argue here against positivism, or the metascientific
doctrine according to which the model set up by the natural sciences
is directly applicable to all human sciences, including (autonomous)
linguistics. I shall refer to 'hermeneutics' as an alternative, non-
positivistic philosophy of science. However, what I shall say is
fully compatible also with (modern interpretations of) such non-po-
si ti vi s ti c doctrines as phenomenology and marxism. I shall also
argue that grammatical theory is nonempirical. More particularly,
grammatical theory should be regarded, in my opinion, as qualitatively
VI
different not just from the natural sciences, but also from the empi
rical human sciences.
My discussion will to a large extent centre around the status of
transformational grammar (henceforth to be abbreviated as 'TG'). In
the present context I shall not so much criticise TG as a scientific
theory, but rather I shall criticise its metascientific concept of it
self. I have presented my criticism of TG primarily in my article
"The Use and Misuse of the Principle of Axiomatics in Linguistics".
The central issue here concerns the role of normativity in lin
guistic data. I do not think that the importance of this concept
has yet been grasped in current theoretical linguistics. As long as
this continues to be the case, no adequate understanding of the meta
scientific status of linguistics can, in my opinion, be reached.
This book is the second, revised edition of my 1974 dissertation
Linguistics and Metascience. My interest in the topic dates from 1968,
when I could no longer ignore the discrepancies between empirical ex
planation and what was referred to as 'grammatical explanation'.
It was during discussions with Matti Juntunen and Lauri Mehtonen,
in the early seventies, that for the first time I became aware of the
inadequacies of positivism, and realised the need for an alternative
philosophy of science. I owe a great debt of grati tute to Professor
E. F. K. Koerner, who selflessly gave so much of his time to edit
the manuscript.
Helsinki, December 1977 Esa Itkonen

TABLE OF CONTENTS
PREFACE V
TABLE OF CONTENTS VII
1.0. THE IDEA OF 'POSITIVISM' 1

1.1. The Data of Positivist Science; the Definition of
'Empirical ' 2
1.2. Explanation, Prediction, and Testing 4
1.3. Comparison with Peirce's Logic of Science 9
1.4. Theory and Observation 12

1.5. Ontology 16
1.6. Concluding Remarks 18
2.0. THE IDEA OF 'HERMENEUTICS' 20

2.1. Psychology 21
2.2. Sociology 24
2.3. Psychoanalysis and Psychotherapy 30
2.4. Sociology of Knowledge 33
2.5. Philosophy 42
2.6. Logic 48
2.7. Concluding Remarks 54
VIII GRAMMATICAL THEORY AND METASCIENCE
3.0. 20TH-CENTURY LINGUISTIC THEORIES: A BRIEF SURVEY 55
3.1. Saussure 55
3.2. Hjelmslev 59
3.3. Sapir 61
3.4. Bloomfield 68
3.5. Harris 71
3.6. Transformational Grammar '. 75
3.7. Some Recent Developments in Linguistic Theory 87
3.8. Conclusion 89
4.0. PHILOSOPHICAL FOUNDATIONS OF THE CONCEPT OF LANGUAGE 91

4.1. Characterisation of the Traditionist Epistemology 91
4.2. Refutation of the Traditionist Epistemology 94
4.2.1. The Axiomaticity of the Concepts of Person and Thing . . . . 94
4.2.2. Mind, Behavior, and Environment 96
4.2.3. Characterisation of Mental Phenomena; the Notion of

'Pattern' 104
4.2.4. General Characteristics of the Conceptual Distinctions
Employed in the Present Study . 108
4.2.5. The Impossibility of Private Languages 109
4.3. Implications for Linguistic Theory 113
4.3.1. Psycho Unguis tics 113

4.3.2. Theory of Grammar 117
5.0. THE CONCEPT OF LANGUAGE 122
5.1. Ontology: Rules of Language as Constituted by 'Common
Knowledge1 122
5.2. Epistemology: the Distinction between Language and Linguistic
Intuition 131
TABLE OF CONTENTS IX
5.3. Rules of Language and Certainty 141
5.4. Rules of Language and Social Control 151
6.0. THE BASIS OF THE NONEMPIRICAL NATURE OF GRAMMAR 155

6.1. The Difference between Rule-Sentences and Empirical
Hypotheses 156
6.2. Examples of Rules and Rule-Sentences 166
6.3. Two Different Types of Rule-Sentence 168
7.0. THE INELIMINABILITY OF LINGUISTIC NORMATIVITY 175

7.1. A Synchronic Grammar Does not Investigate Spatiotemporal
Utterances, but Correct Sentences 175
7.2. Grammatical Concepts Are not Comparable to Theoretical

Concepts of Natural Science 177
7.3. Rules Are not Regularities of Non-Normative Actions 182
7.4. Grammatical Descriptions Cannot Be Replaced by Psycho-

linguistic and/or Sociolinguistic Descriptions 187
7.5. The Position of Transformational Grammar vis-à-vis

Linguistic Normativity 188
8.0. LANGUAGE AND GRAMMAR 192

8.1. The Basis of the Difference between Natural Science and
Human Science: Observer's Knowledge vs. Agent1 s Knowledge . . . 193
8.2. The Two-Level Nature of the Human Sciences: Atheoretical
vs. Theoretical 198

8.3. The Two-Level Nature of Grammar 208
8.4. The Ontological Reality of Grammatical Descriptions 219
9.0. THE METHODOLOGY OF GRAMMAR 228

9.1. General Remarks 228
9.2. Explanation and Prediction 233
X GRAMMATICAL THEORY AND METASCIENCE
9.3. Testing 245
9.4. Universal Linguistic Theory 263
9.5. Appendix: Examples Taken from the Transformationalist

Literature 264
10.0. GRAMMAR AND LOGIC 276

10.1. The Basis of the Similarity between Generative Grammars
and Systems of Logic 276

10.2. Testing 280
10.3. Explanation 287
11.0. GRAMMAR AND PHILOSOPHY 294
11.1. The Methodology of Classical Philosophy 294
11.2. The Concept of Explication 301

11.3. Grammars as Instances of Explication 307
CONCLUSION 311
NOTES 313
REFERENCES 331
INDEX OF AUTHORS 349
INDEX OF TERMS 352

1.0. THE IDEA OF 'POSITIVISM'
The term ' p o s i t i v i s m ' can be i n t e r p r e t e d i n d i f f e r e n t ways. Ac-
cording to one i n t e r p r e t a t i o n , ' p o s i t i v i s m ' , or more p r e c i s e l y 'neopo-
s i t i v i s m ' , is i d e n t i f i a b l e w i t h the trend i n the philosophy of science
represented during the 1930s by such authors as S c h l i c k , Neurath, Car-
nap, and Reichenbach, and known as ' l o g i c a l p o s i t i v i s m ' or ' l o g i c a l em-
pirism'. This trend was characterised by the methodological requirement
t h a t a l l t h e o r e t i c a l statements, should be wholly reducible to observation
statements; i n a sense, theories would thus be superfluous. This inter-
p r e t a t i o n of ' p o s i t i v i s m ' would exclude from p o s i t i v i s m Popper's ' c r i t i -
cal r a t i o n a l i s m ' as well as Hempel's and Nagel's current conception of
the philosophy of science, which i s a d i r e c t outgrowth from the older
' l o g i c a l empirism'. The two trends last-mentioned d i f f e r i n several r e -
spects, but they are i n agreement on the r e j e c t i o n of the 'strong reduc-
t i o n i s m ' mentioned above: Instead of viewing science as an inventory of
i n d u c t i v e generalisations based upon observation, they emphasise the
r o l e of t h e o r y - c o n s t r u c t i o n .
However, i n one fundamental respect a l l of the types of philosophy
of science mentioned have something i n common: they a l l adhere to the
s o - c a l l e d 'methodological monism', or the conception t h a t a l l empirical
sciences are characterised by common methods of e x p l a n a t i o n , p r e d i c t i o n ,
and t e s t i n g , methods t h a t i n t h e i r most e x p l i c i t form appear w i t h i n the
natural sciences. A l l sciences i n v e s t i g a t i n g human or social phenomena
are presumably to be subsumed under t h i s concept of ' e m p i r i c a l science'.
The word ' p o s i t i v i s m ' has also been used as a common denomination f o r
a l l of those m e t a s c i e n t i f i c schools t h a t subscribe to methodological
monism ( v i z . the u n i f i e d - s c i e n c e i d e a l ) , as i t is defined here ( c f . for
2 GRAMMATICAL THEORY AND METASCIENCE
instance Martindale 1960:56). This usage has been adopted in the subse
quent argument.
1.1. The Data of Positivist Science; the Definition of 'Empirical'
The uniformity of methods of description presupposes the uniformity
of what is to be described: according to positivism, all empirical scien
ces deal with events located in space and time. Knowledge about such
events is obtained through (intersubjective) observation. Primarily, ob
servable events contain only qualities and relations that can be measu
red in terms of length, time, and weight. In addition, also events in
volving such 'secondary' qualities as colour, sound, and smell are allow
ed to be observable, with the understanding that they are to be operati-
onalised, or translated into events involving 'primary', measurable qua
lities only. For Popper for instance the empirical nature of a theory
depends on whether there is a class of 'basic statements' which might
falsify it; basic statements are about observable events, and such an
event is defined as "an event involving position and movements of mac
roscopic physical bodies" (Popper 1965:103). On the other hand, from the
fact that space, time and weight define the notion of observability, it
does not follow, of course, that they establish some kind of universal
conceptual framework to which unobservable microphenomena for example
must also conform (Nagel 1961:170-71, n.12). It is sufficient that
statements about microphenomena imply some basic statements about ob
servable events.
In the light of what precedes, it is understandable that 'observa
bility' has been closely linked to what I would like to call 'empirical-
ness'. In fact, 'empirical' has mostly been defined, in the Poppen an
spirit, as 'falsifiable on the basis of observation'. However, theories
may contain sentences with mixed quantifiers à la '(x)(Ey)(Fxy)' , and
these are not falsifiable, but only testable, i.e. confirmable or dis
confirmable (cf. 1.2. below). Secondly, if observability taken strictly
in the sense of measurability is defined as 'pure' observability, then
we might say that in classical mechanics we have to do with pure obser
vation motivated by theoretical considerations (but not, of course,
THE IDEA OF 'POSITIVISM' 3
with pure observation tout c o u r t ) . In most 'observational' sciences,

however, pure observability is only an ideal which is never achieved:
in such cases, just as in everyday life, observation as such contains
interpretive or 'theoretical' elements.
Consequently, it is best to define 'empirical' as 'testable by (sen
tences referring to) what happens or obtains in space and time', with the
following qualification: That, and only that, which happens or obtains
in space and time can be observed, observation being an act inseparable,
perhaps in more than one way, from theory. From this it follows, among
other things, that i n t u i t i o n , for instance, is not a form of observation.
To call all ways of gaining knowledge by the same name, i.e., 'observa
tion', not only is uninformative but - as we shall see later on - also
conceals important methodological differences.
The whole point of giving a definition of 'empirical' is to provide
a criterion to distinguish (empirical) science from 'metaphysics', phi
losophy and logic being prime examples of 'metaphysics'. The above-
mentioned definition of 'empirical' is clearly able to differentiate
between physics, on the one hand, and philosophy and logic, on the other.
Accordingly the latter two - and all sciences of a similar kind - will
hence be called nonempirical sciences.
Notice that the present notion of 'empirical' is entirely value-free:
it is not implied that logic, because of its nonempirical status, would
be treated as less valuable than for instance physics. On the other hand,
there exists another, value-laden notion of 'empirical', which aims at dis
tinguishing good science from bad science or 'ideology'. I shall have
nothing to say about this latter notion of 'empirical'.
The measurability of observable events is guaranteed by the mathe-
matisation of nature, which largely constitutes the content of Galileo's
'revolution' of science. That is to say, empirical science in the posi-
tivistic sense concentrates upon those (physical) characteristics of
events which can directly be given numerical values, or operationalises
events so that they acquire such characteristics. This idealising proce
dure amounts to a r e - d e f i n i t i o n of events: their similarity or difference
is determined solely on the basis of their measurable properties, and
it becomes meaningless to ask whether two events with the same measu
rable properties are veally similar or interchangeable. Or, as Mittelstrass
(1974b:66) puts it:
Als 'empirisch' oder 'empirisch begründet' kann im Rahmen des Gali-
leischen, für die Methodologie der Physik in Geltung bleibenden Er
fahrungsbegriffes nur noch das Ergebnis einer messenden Pvaxis be
zeichnet werden (cf. 2.5. below).
1.2. Explanation, Prediction, and Testing
According to standard positivism, "science is interested in estab

lishing predictive and explanatory connections between observables"
(Hempel 1965:179). For reasons that will later become apparent, I shall
restrict my discussion to the so-called 'deductive-nomological' (= D-N)
model of explanation (e.g., Hempel 1965:335-38). According to this view,
explanation and prediction are similar, insofar as in both cases it is
necessary to deduce a sentence that refers to a particular event from
a whole that comprises one or more sentences referring to general regu
larities, and one or more sentences referring to particular events, or
so-called 'antecedent conditions'.2 It is also said that regularities
are 'explained' by deducing the universal hypotheses referring to them
from some other, more abstract universal hypotheses. Nevertheless, not
only is the explanation of regularities qualitatively different from the
explanation of events (von Wright 1971:184, n.12), but the former is also
logically secondary with respect to the latter, since particular events
determine which universal hypotheses about general regularities are (as
sumed to be) true. Hempel even claims that (inductive-)statistical ex
planations, too, are meant to explain particular events (Hempel 1965:381).
However, von Wright (1971:13-15) points out that, unlike D-N explanations,
statistical explanations cannot properly be said to explain why an event
occurred, but rather, why its occurrence was to be expected.
Because of the supposed structural symmetry between explanation and
prediction, and of the interdependence of these two notions with testing,
i.e., confirmation or disconfirmation, all these methodologically central
notions can be clarified with the aid of the D-N model. The general form
of this model is as follows:
The standard type of a D-N explanation is a causal explanation,

and in such a case antecedent conditions and explanandum-events are i-
dentifiable as 'causes' and 'effects' respectively. However, there are
non-causal D-N explanations too, that is, explanations based on 'laws of
functional dependence', where two sets of facts determine each other si
multaneously (Hempel 1965:352-53, and Nagel 1961:77-78). In such in
stances it is to some extent a matter of opinion which facts are to be
explained in terms of which others, and the term 'antecedent condition'
is clearly misleading. - von Wright (1971:175, n.35) notes that the ge
neral idea underlying the D-N model, or the "'Popper-Hempel' theory of
explanation", has been "something of a philosophical commonplace ever
since the days of Mill and Jevons".
The simplest case of a D-N explanation is
According to Hempel (1965:275), this sentence expresses an explanation
which "surely is intuitively unobjectionable". Presented in the form
of the D-N model, one obtains as follows:
Fa
Ga
By applying the r u l e of ' u n i v e r s a l i n s t a n t i a t i o n ' to
we get , and from t h i s , together w i t h ' F a ' , we can derive 'Ga'
by Modus Ponens.
'Fa' and 'Ga' r e f e r to observable events, which means t h a t the i n -
dividual-expression ' a ' r e f e r s to a space-time p o i n t or r e g i o n , and the
predicates ' F ' and 'G' r e f e r to measurable p r o p e r t i e s . Moreover, the
events r e f e r r e d to by 'Fa' and 'Ga' ( o r , e q u i v a l e n t l y , the properties
r e f e r r e d to by ' F ' and 'G') must be conceptually independent, t h a t i s ,
'Ga' must not be deducible from 'Fa' alone.
In explanation, we s t a r t from 'Ga' which we know, on the basis of
o b s e r v a t i o n , to be t r u e , and we t r y to f i n d a s u i t a b l e explanans from
which it can be derived. If there is such an explanans, i.e., one which
consists of the true observation-sentence 'Fa' and of the universal hy
pothesis or theory which can, for independent reasons, be
assumed to be true, then we may tentatively consider the event referred
to by 'Ga' as being explained.
In prediction, on the other hand, we start from the explanans, which
we hold to be true, and deduce from it the sentence 'Ga', or predict
that 'Ga' will be true. The truth or falsity of predictions is deter
mined on the basis of observation. Making a prediction, bringing about
the antecedent conditions, and deciding the truth-value of the prediction,
constitute an experiment.
It is customary to discuss only the predictive or experimental as
pect of the testing of theories. However, if testing is equated with
the method of selecting, on objective grounds, one theory from among
others, then the explanatory aspect is just as important.
The predictive componenet of testing is concerned with the question
whether the theory generates only true sentences about observable events
(of the relevant domain). If the prediction 'Ga', which has been made
on the basis of the 'theory' and of the truth of 'Fa',
turns out to be true, then it is said that the observation-report
'Fa&Ga' (predictively) confirms the 'theory'. Of course, no amount of
such reports, e.g. 'Fb&Gb', 'Fc&Gc', etc., can conclusively establish
the truth of That is to say, since 'Fa&Ga' is true,
is true, and if we from this sentence (and from any number of
similar sentences) infer , we are making use of the logi
cally invalid argument form (where 'p' and 'q' stand
for and respectively). From the above it follows
that no theory or universal hypothesis can be conclusively confirmed, or
verified, on the basis of observational evidence.
The weakness of this characterisation of confirmation lies in the
fact that is true not only if 'Fa&Ga' is true, but also if
either '-Fa&Ga' or '-Fa&-Ga' is true, which means that the events re
ferred to by these latter two sentences are also 'confirmatory evidence'
for the truth of . In fact, since is logically
equivalent to '-FavGa', either '- Fa ! or 'Ga' would alone suffice to
make it true and, hence, to 'confirm' '(x) something which
hardly makes sense. To put it in different terms, since '(x) ,
confirmable by 'Fa&Ga', is logically equivalent to '(x)
confirmable by '-Ga&-Fa', it must also be confirmable by the last-men
tioned sentence« This is the basis for the so-called 'paradoxes of
confirmation' (Hempel 1965:3-46), Even though extensively discussed in
the literature on philosophy of science,they are entirely artificial and
result from the gratuitous belief that truth-functional logic, more pre
cisely, universally quantified material implication, provides an adequate
way of expressing empirical hypotheses, or that the truth of an empirical
generalisation is determined exactly in the same way as that of a mate
rial implication.
The predictive component does not yield conclusive confirmations
which means that it cannot conclusively establish that the theory gene
rates only true sentences. But it can conclusively establish that the
theory does not generate only true sentences: all that is needed is one
false prediction. In terms of our example, let us assume that the pre
diction 'Ga' turns out to be false. Since 'Ga' was validly inferred
from &Fa' and now '-Ga' is the case, it follows, by Modus
Tollens, that ' &Fa' must be false too. If 'Fa' is false, the
matter ends there. But let us assume that the antecedent conditions are
as they were supposed to be, namely that 'Fa' is true. We then have the
observation-report 'Fa&-Ga', which is identical to saying that
is false. Now was validly inferred from , and
therefore if the former sentence is false, then the latter sentence is,
again by Modus Tollens, also false. In other words, has
been conclusively dis confirmed, or f a l s i f i e d . 6
Thus it cannot be established that a theory generates only true
sentences. But even if it could, this would not be enough, because the
theory in question might still be ever so narrow or one-sided, and there
fore worthless. Obviously, the predictive component of testing is not
sufficient in itself., but must be supplemented by the explanatory compo
nent, or the requirement that the theory generate all true sentences.
It is immediately evident that a theory may be tested not only on pre
dictive but also on explanatory grounds: there is an event belonging to
the domain supposedly covered by the theory, and we ask whether the
theory explains the event, or generates the sentence (truthfully) refer
ring to it. If it does, it is (explanatorily) confirmed; if it does
not, it is (explanatorily) disconfirmed. More precisely, a theory A is
disconfirmed if there is a set of facts which it is unable to explain,
and there exists an alternative theory Β which is able to explain these
facts, together with all the facts explained by the theory A.
The testing may or may not expand the stock of evidence which the
theory is supposed to account for. Instead of predicting new, unobser
ved facts or of observing new, unpredicted facts, we may ask whether the
theory under scrutiny explains well-confirmed regularities or subtheories,
i.e., whether this theory surpasses others in the generality of its ex
planations :
Theoretical synthesis, with no addition of new evidence, is
classically taken to lend further support (by the very fact of its
being successful at all) to the joint theory than to either theory
taken separately (Harre 1970:170).
However, this principle cannot be taken in an absolute sense, as

Popper (1965:269-73) does, for instance. That is, it is not true that
the more general hypothesis, although unfalsified at the moment, is
always better confirmed, or more probably true, than the less general
one implied by it (Barker 1957:160). After all, there is a countless
number of wel 1-established observational hypotheses which have never been
and will probably never be falsified (even if they have been made more,
precise in the course of time), whereas the various theories that have
been superimposed on them at one time or another have all been falsified
sooner or later.
It must be admitted that, in view of the supposed symmetry between
explanation and prediction, the terminological distinction 'explanatory
testing' - 'predictive testing' is not mandatory. Given an event, it
would be possible to ask, not only whether the theory explains it, but
also whether the theory would have predicted it. Similarly it would be
possible to reformulate predictions as explanations. However, the dis-
tinction I want to make here is not primarily that between prediction

and explanation but - to use neutral terms - that between the descrip
tive capacity of a theory and its object of description. More precisely,
either we first determine what the theory claims, and then check whether
what it claims is in fact the case; or we first determine what is the
case, and then check whether this is what the theory claims. In actual
practice these two points of view always occur together, but it is use
ful to differentiate between them for analytical purposes. It is un
deniable that the former corresponds in a rather natural way to prediction
while the latter corresponds to explanation. - This way of defining the no
tion of testing will facilitate the comparison of natural-science theo
ries with generative grammars and systems of logic (cf. 9.0. and 10.0.
below).
Both in predictive and in explanatory testing i t is required that
the new evidence which is supposed to confirm or to disconfirm the
theory is conceptually independent from previous evidence. This means
that if the theory predicts, with the aid of the anteced
ent conditions Fa and Fb, first the event Ga and then the event Gb,
the latter event (predictively) confirms the theory, over and above the
confirmatory support given by Ga, only if it is logically independent
from Ga. Correspondingly, if the theory explains the events Ga and
Gb, the latter (explanatorily) confirms the theory, over and above what
has been done by Ga, only if it logically independent from Ga. Itshould
be remembered that empirical explanation and prediction, in turn, require
the logical independence between Fa and Ga, on the one hand, and between
Fb and Gb , on the other.
1.3. Comparison with Peirce's Logic of Science
We can sum up the methodological notions introduced so far by

showing their relation to the triad deduction-induction-abduction,
which constitutes the essence of Peiree's logic of scientific method
(cf. Peirce 1958, Bk II, chap.3). Deduction is, first, the method by
means of which a prediction is derived from the theory. Induction is
made use of in determining the outcome of the prediction as well as in
assessing its effect upon the theory. If the event predicted fails to
occur, this means, of course, that the theory in its present form is
falsified. But it also means that something else, some unexpected
event, has occurred. This event demonstrates the need for a new theory
which could explain it (along with all facts previously explained). It
is the function of abduction to suggest such a theory. Hence, deduction
is a necessary component of abduction, because what is abduced is a
theory which explains the unexpected event, i.e., a theory from which
the sentence referring to this event can be deduced. Abduction is sole
ly responsible for the growth of (scientific) knowledge. The value of
the new theory must be ascertained, again, by a combined use of deduc
tion (in the sense of prediction) and induction. And so the scientific
process goes on indefinitely, as illustrated by the following diagram:
In this example the prediction 'Ga' turns out to be false, i.e.,

'Ha' turns out to be true, and this gives rise to a new, more differen
tiated theory.
Explanatory confirmation or falsification comes into play when we
come across an event, for instance one referred to by 'Gd&Hd', and ask
whether or not the theory can explain it (or, alternatively, whether
or not the theory would have predicted it).
Peirce succintly characterises his three basic operations by say-
ing that deduction determines what must be as a matter of logical ne

cessity, induction determines what is as a matter of fact, and abduction
determines what could be as a matter of empirical possibility. His ter
minology is rendered a little opaque by the fact that he uses the term
'abduction' also to refer to the process of eliciting (part of) the
antecedent conditions in a case where the explanandum-event and the re
levant regularities are known.
Predictive confirmation, explanatory confirmation, and abduction
(in the primary sense) can all be represented by the invalid argument
known as the 'fallacy of affirming the consequent1, where 'p' stands
for the theory (plus statements of antecedent conditions) and 'q' stands
for a sentence derived from it:
However, all the three above-mentioned operations are clearly distinct

and necessitate therefore distinct ('pragmatic') interpretations of the
schema in question. In predictive confirmation we start from 'p', and
from it we go forth, or 'progress', to 'q'. When we notice that 'q' is
true, we take this as an indication that 'p' might be true too. On the
other hand, both in explanatory confirmation and in abduction we start
from 'q' and go back, or 'regress', to 'p'. The difference between the
two operations lies in the fact that in the former case 'p' is readily
available (perhaps as one among several alternative theories): we notice
that 'q' can be derived from 'p', and in our opinion this lends support
to 'p'. In abduction by contrast', when we have 'q', 'p' does not yet
exist. The fact referred to by 'q' must be explained, however, and to
this end we invent 'p' which is such that 'q' can be derived from it.
It would be somewhat unnatural to say at this point that 'q' 'confirms'
'p', because the latter has been created expressly to account for the
former.
It is convenient to represent predictive falsification and expla
natory falsification by two different schemas. Although these both
exemplify the valid Modus Tollens argument and are thus logically
12 GRAMMATICAL THEORY AND METASCIEΝCE
e q u i v a l e n t , they are pragmatically q u i t e d i f f e r e n t .
In predictive falsification (cf. A ) , we start from 'p' and derive

'q'. We then notice that 'q' is false, which means that 'p' too is
false. In explanatory falsification (cf. B ) , we start from 'q', which
we know to be true, and then we notice that 'p' (which is available)
entails the negation of 'q' and is thus false. In addition, this latter
figure of thought serves the function of hypothesis elimination, which
is a necessary component of successful abduction: before we arrive at
a satisfactory theory which explains the unexpected event referred to
by 'q', along with all other known facts, we have mentally discarded
several theories because they entail the negation of 'q' or of some
other true sentence.
My notion of 'predictive testing' corresponds to Bocheñski's
(1971:100-04) notion of 'progressive reduction' (which he also calls
'verification'). By contrast his notion 'regressive reduction' (which
he also calIs'explanation') covers solely the area of abduction. As I
have already pointed out, in the philosophy of natural science the
existence of explanatory testing is not clearly distinguished from
predictive testing, on one hand, and from abduction, or the invention
of theories, on the other (see, however, for instance Hempel 1965:29,
η.36). It must be added that positivism has generally paid little
attention to the concept of abduction.
1.4. Theory and Observation
The preceding sketch of the basic methodological notions under

lying (the philosophy of) natural science and, hence, positivist science
is definitely oversimplified, because theories have been represented by
universal implications containing the same concepts, i.e., 'F' and 'G',
which are contained in observational sentences. In reality, the relatio
between theory and observation is of course much more complex. A some-
what more r e a l i s t i c p i c t u r e can be gathered from analysing the i n t e r -

play of t h e o r e t i c a l and observational concepts in D-.N explanations.
Suppose we have to explain the event 01a. (Here ' T ' and '0' stand
f o r ' t h e o r e t i c a l ' and ' o b s e r v a t i o n a l ' , r e s p e c t i v e l y . ) If
is a w e l l - c o n f i r m e d hypothesis, then we might set up the f o l l o w i n g D-N
explanation:
Now, since ' T ' i s a t h e o r e t i c a l concept, we cannot have any d i r e c t (i.e.,

observational) evidence f o r the t r u t h of ' T a ' . Nor do we have any i n d i -
r e c t evidence independent of the t r u t h of i.e., the explanandum
which we are t r y i n g to e x p l a i n . Suppose, however, t h a t there
also e x i s t s a w e l l - c o n f i r m e d hypothesis Suppose f u r t h e r
t h a t we can d i r e c t l y v e r i f y the t r u t h of ' 0 2 a ' . Then we could w i t h some
j u s t i f i c a t i o n set up the f o l l o w i n g t e n t a t i v e i n d u c t i v e argument:
I t is only as a r e s u l t of t h i s i n d u c t i v e step t h a t we are e n t i t l e d to

assume the t r u t h of 'Ta' and to use i t in our D-N e x p l a n a t i o n . - A
s l i g h t l y more complicated (and more r e a l i s t i c ) case is discussed i n
Stegmüller (1974:166-76).
Notice t h a t the requirements previously imposed upon statements of
antecedent conditions are here seen to p o i n t in d i f f e r e n t directions:
Only ' 0 2 a ' r e f e r s to an observable f a c t or event, and only 'Ta' permits
the deduction of the explanandum ' 0 2 a ' . - and
are i d e n t i f i a b l e as maximally simple cases of correspondence rules (cf.
below). Moreover, to be p r e c i s e , the two a's referred to by the expression
'Ta' and 'CLa' (or 'CLa') cannot be s t r i c t l y i d e n t i c a l . Rather, we would
need s o - c a l l e d basis rules to connect the numerical space-time reference
of t h e o r e t i c a l statements l i k e 'Ta' to t h a t of observational ones l i k e
(cf. Stegmüller 1970:317-19).
It is most convenient to characterise the relation of theory and
observation by comparing it against what until recently was the 'ortho
dox' or 'standard' view of the matter (cf. Feigl 1970 and Hempel 1970).
According to this view, a scientific theory is, ideally, an axiomatic
system. Axioms contain the primitive concepts of the theory and impli
citly define the latter in a purely formal or syntactic way. Primitive
concepts are used to define new concepts,, These two types of theoreti
cal concepts acquire empirical meaning through rules of correspondence
which connect them with empirical concepts. The latter, in turn, are
connected through operational definitions with classes of directly ob
servable, measurable events and states. (Sometimes no distinction is
made between rules of correspondence and operational definitions.)
With the aid of standard rules of inference and of suitable statements
of antecedent conditions» it is then possible to derive from this sys
tem of axioms and definitions either true or false statements about par
ticular measurable states and events.
This conception was always meant to be an idealisation, but it has
turned out to be over-idealised. Its revision consists, essentially, in
changing the status of correspondence rules and operational definitions:
from definitions, they are changed into statements which eventually ad
mit of confirmation or falsification. It follows that theoretical con
cepts acquire a genuine meaning and do not just become meaningful through
stipulation. The distinction between theory and observation is also
lessened by noting that theoretical concepts build upon older (and less
abstract) theoretical concepts which largely retain their customary
meanings even after new scientific discoveries have been made.
Theoretical hypotheses are logically prior to observational ones,
because the latter are deduced from the former. As a whole, however,
observation is epistemologically prior to theory, because the latter is
abduced on the basis of the former. The function of definition is simi
lar to that of deduction: these two operations integrate the concepts
and the sentences of the theory, respectively.
Since theoretical hypotheses are tied into a whole by the concepts
they contain, a prediction is always deduced jointly from several hypo

theses (plus statements of antecedent conditions), and therefore it is
impracticable to try to confirm or to falsify one single hypothesis at
a time. If the prediction turns out to be false, one is to a certain
extent free to decide which hypothesis from among the hypotheses consti
tuting the theory has been most directly affected. That is to say, a
theory must be treated as a whole whose parts are linked to each other
by relations that are partly conceptual (or stipulative) and partly em
pirical, but the line between conceptual and empirical may be drawn in
more than one way. Furthermore, because of the great distance between
theory and observation, it is an exaggeration to say that one false pre
diction (even with true statements of antecedent conditions) is able to
disconfirm 'conclusively' the theory: most often a false prediction
effects only a readjustment of the lower-level hypotheses of the theory.
The existence of instances of theoretical concepts is hypothesised
on the basis of observable events which are taken to be caused by them.
(This line of thinking can be continued: unobservable instances of theo
retical concepts can be considered as caused by instances of more theo
retical concepts,) Instances of hypothetical theoretical concepts pro
duce their observable effects only under certain observable conditions,
which means that theoretical concepts are normally of a dispositional
nature. The relation between theoretical and observational concepts
may be either nomological or statistical, but it must above all be a
constant one. In other words, in whatever manner the operational defi
nitions are interpreted (cf. above), it must be possible to formulate
them in terms of space-time coordinates, and this is actually done in
physics for instance. On the other hand, correspondence rules are most
ly formulated in a "comparatively loose and imprecise" way (Nagel 1961:
99). If one ignores the need for operational definitions and concen
trates only on the 'loose and imprecise' character of correspondence
rules, one may with some apparent plausibility try to apply the natural-
science approach, and in particular the 'theoretical concept - observa
tion concept' distinction, to instances where it is not applicable in
fact. This is the case with certain trends in modern theoretical
linguistics for example,as I shall demonstrate later on (7.2. below).
When a (type of) observable event has been e x p l a i n e d , i t i s seen i n
a new and d i f f e r e n t way, namely ' t h r o u g h ' the theory which explains it.
I t has been claimed, in p a r t i c u l a r w i t h i n c r i t i c a l rationalism, that a l l
observation is theory-dependent: f o r i n s t a n c e , the acceptance of 'basic
statements' about observable events is viewed as an a p p l i c a t i o n of the
very same theory which such statements are supposed to confirm or to
falsify (Popper 1965:106). S i m i l a r l y , i t is claimed t h a t from the cur-
rent p o i n t of view, events and f a c t s can only e x i s t w i t h i n one or an-
other s c i e n t i f i c theory (Feyerabend 1968). Although I admit t h a t there
is no hard-and-fast d i s t i n c t i o n between observation and t h e o r y , I cannot
accept the above conclusion of an i n e v i t a b l e l o g i c a l c i r c l e between the
two. The ' s c i e n t i f i s a t i o n ' of one's outlook is not an i r r e v e r s i b l e pro-
cess; r a t h e r , i t is possible to r e - e s t a b l i s h t h a t common-sense or prac-
tical approach of looking at things which precedes science both w i t h i n
c o l l e c t i v e h i s t o r y and w i t h i n i n d i v i d u a l one. Although the p r a c t i c a l
view i s a theory of i t s own, and v a r i e s , to some e x t e n t , both geographi-
c a l l y and d i a c h r o n i c a l l y , i t nevertheless represents the only possibility
f o r c o n s t r u c t i n g an o b j e c t i v e basis on which d i f f e r e n t theories can be
compared and evaluated. A s i m i l a r a t t i t u d e is c h a r a c t e r i s t i c of the
'constructivist' philosophy of the s o - c a l l e d Erlangen school ( c f . , e.g.,
Kambartel & M i t t e l s t r a s s 1973). As we s h a l l see, these questions are
o
in no way limited to natural science only.
1.5. Ontology
The ontology of positivism is still today based on the logical

atomism of the early Wittgenstein; a world constructed along these lines
can be called a Tractatus-viorld (von Wright 1971:44). The thesis of
atomism can be taken as a reformulation of the definition of empiricai-
ness given in terms of spatiotemporal testability (cf. 1.1. above).
Both attempt to draw the line between empirical and nonempirical scien
ce by imposing certain conditions upon the data of purportedly empiri
cal descriptions. The thesis says, essentially, that it must not be the
case that the existence of some states of affairs is logically (or con-
ceptually) necessary to the existence of some other states of affairs.

The distinction at issue is that between empirical science and concep
tual analysis. It is based on the distinction between natural and logi
cal (or conceptual) necessity, which we might illustrate, respectively,
by the sentences "If a piece of metal is heated, it expands"
and "If Bill is a bachelor, he is unmarried" What is at
issue here is 'simply' the difference between what can and what cannot
9
be thought or imagined. The use of a 'psychologistic' concept like
'imaginability' as the criterion of natural vs. conceptual necessity or,
to use other terms, of contingent vs. necessary truth, has sometimes
been objected to, but there is actually no alternative to it (Pap 1958:
216-18): all more formal criteria derive their meaning from the fact
that they rest upon this 'psychologistic', i.e., intuitive, distinction;
or else they have no meaning at all. Now, it seems undeniable that it
is possible to imagine 'q' to be false without being forced to imagine
'p' also to be false; and this shows that the (causal) relation between
the events referred to by 'p' and 'q' is not conceptual. This presup
poses of course a previous acceptance of the customary rules of logic.
Yet it also seems undeniable that if we imagine 's' to be false, there
is no way for us to imagine that 'r' might nevertheless be true. There
fore the relation between (the facts referred to by) 'r' and 's' must be
conceptual. Interestingly enough, there are conceptual relations also
between spatially and temporally distinct actions i or more precisely
between the meanings of such actions (cf. 4.2.2. below).
Conceptual relations cannot exist in space and time, because the
entities between which they hold, i.e., concepts or meanings, do not
exist in space and time. Now, the de facto identity of the thesis of
atomism and the thesis of spatiotemporal testability, as definitions of
empi ri cal ness, can be seen in the fact that while the latter bases the
notion of 'empirical' on what happens or obtains in space and time, the
former excludes from the domain of 'empirical' precisely that which does
not happen or obtain in space and time, i.e., conceptual relations. The
rationale behind the thesis of atomism seems to be the Humean idea that
in space and time anything whatever could occur, even if its occurrence
is highly improbable. Since anything could occur, observably in prin
ciple, in space andtime, empirical sentences and theories, which are
about space and time, are necessarily testable on the basis of obser
vable evidence. The empirical world differs from the 'world of concepts'
precisely because it is not imaginable that in this world anything what
ever could occur. In fact, notions such as 'spatiotemporality', 'occur
rence', and 'observation' constitute a coherent whole which is incompa
tible with any characterisation of the 'world of concepts'.
We noted already that concepts are not spatiotemporal entities.
Nor do they occur or fail to occur in the real world (although it is
true, of course, that given concepts may or may not be consciously held
at a given time and a given place). This is connected with the fact
that relations between concepts hold necessarily, not contingently. It
is a necessary (conceptual) fact that, for instance, 'bachelor' entails
'unmarried', contradicts 'married', is compatible with 'tall' and in
compatible with 'divisible by 3', although usually only entailment and
contradiction are called 'necessary' relations. The necessary character
of conceptual relations is not called into question by the fact that
concepts change in the course of time or, more precisely, that people
replace old concepts by new ones: this only means that new necessary
relations are substituted for old ones. -The distinction between conceptual
analysis and empirical science will be discussed in greater detail in
the next chapter (see 2.0. below).
1.6. Concluding Remarks
In the beginning I defined 'positivism' as that philosophy of

science which claims that all empirical sciences must make use of the
methods of natural science. Above I have characterised these methods so
as to make the basic tenet of positivism more precise. We may ask what
are the sciences which, according to positivism, might be able to acquire
the status of an empirical science. In my view the following disciplines
could well qualify: sociology, economics, linguistics, anthropology,
possibly psychoanalysis and history, and perhaps also classical philo
logy; lastly,if at all, the study of literature, music, and the fine
arts.
As we noted in 1.1., philosophy and logic constitute the opposite

of empirical science. It is a remarkable fact that in the positivist
tradition the relation between empirical science and nonempirical scien
ce has practically never been explicitly discussed. The positivist de
finition of philosophy and logic is a purely negative one: they are
simply those sciences (or 'disciplines') which are not empirical.
I try to put linguistics into its proper place within the system
of sciences. This requires a reinterpretation not only of linguistics,
but of some other sciences as well, as will be seen in the next chapter.
2.0. THE IDEA OF 'HERMENEUTICS'
I shall use the term 'hermeneutics' to denote all those schools of
thought which make an irreducible distinction between observation and
understanding, and claim that the investigation of human phenomena is,
in one way or another, qualitatively different from the investigation of
physical reality. Thus, as a philosophy of science, hermeneutics is
clearly opposed to positivism. Such an interpretation of 'hermeneutics'
appears historically justified. However, as will be seen toward the
end of this chapter, this view will lead to a somewhat modified concep
tion of 'hermeneutic science'.
Hermeneutics constitutes itself, first, as a set of theories of
particular (human) sciences and, second, as a theory of the nature and
the presuppositions of science, or of scientific knowledge, in general.
In the latter instance, hermeneutics overlaps with so-called 'transcen
dental philosophy'. There exists no uniform hermeneutic method, compa
rable for instance to the positivistic notions of explanation ortesting.
This is due to the fact that the subject matter of (the sciences analy
sed by) hermeneutics is not as uniform as the measurable subject matter
of natural science. As a purely informal characterisation, it might be
said that hermeneutics acquires its data through understanding meanings,
intentions, values, norms, or rules, and that the hermeneutic analysis
consists in reflection upon what has been understood. It goes without
saying that, depending on the nature of the investigation, hermeneutic
methods may be combined with more empirically oriented methods.
To my knowledge, Radnitzky (1970) contains the most thorough dis
cussion of the relationship between positivism and hermeneutics (al
though these terms are not used quite in the same way as here). Haber-
THE IDEA OF 'HERMENEUTICS' 21
mas (1970) offers a good exposition of the hermeneutic modes of think

ing. Apel (1973a) emphasises more the purely philosophical aspect of
hermeneutics. Other theoretists will be named below in the course of
discussion.
In this chapter, I shall illustrate the hermeneutic point of view
by starting from experimental psychology and proceeding gradually - in
the direction of decreasing empiricalness - to philosophy and logic.
2.1. Psychology
In the most general terms, hermeneutics says that it makes a dif

ference whether one investigates physical nature or human nature. There
are instances where the truth of this claim is rather obvious (though
far from universally acknowledged). However, I want to begin from an
example where there is apparently no difference between natural science
and human science, and to show that even there the existence of such a
difference can be demonstrated in a precise way. If human beings are
investigated strictly from a physical or biological point of view, then
it is trivially true that the resulting description is pure natural
science. Experimental psychology comes closest to being a science which
while studying humans qua humans, is nevertheless comparable to a natu
ral science. Both in experimental psychology and in natural science
spatiotemporal phenomena give rise to explanatory hypotheses about those
unobservable mechanisms which make the phenomena in question happen.
Therefore it may seem rather natural to assume that one single concept
of science is operative here. This is indeed the prevailing opinion
in modern theoretical linguistics for instance. Thus it is claimed that
there is no difference between a physical concept like 'redness' and a
psychological concept like 'intention': the occurrence of instances of
either concept can and must be scientifically ascertained by means of
objective, measurable criteria (Schnelle 1973:37-39). In what follows,
I shall compare precisely these two concepts.
Applied to the case at hand, my objection to the thesis of a strict
equivalence between experimental psychology and natural science runs as
follows. It is possible, in principle, to investigate, within a given
experimental set-up, the redness or non-redness of all objects of the wor'
without exception. By contrast, investigating, within a given experi
mental set-up, the occurrence or non-occurrence of intentions divides
the objects of the world into two strictly separate classes, namely
those which can, and those which cannot, even in principle, be subjec
ted to experimentation. To illustrate: Suppose that there is a group
A of experimental psychologists and that one half of A, i.e., A-,, intends
to investigate the occurrence or non-occurrence of specific intentions
in a group of test persons. Then the other half, or Α 2 , may justifiably
claim that it is illegitimate to speak about any intentions of A-,, until
their occurrence has been established by means of objective measurements,
and A2 proceeds to make such measurements. But this means - and both A-,
and A2 know that it means - that A2 intends to make the requisite measu
rements. And yet, if A-, and A2 accept the general thesis of positivism,
they must claim that it is illegitimate to speak of knowing any such
intentions of A 2 until they have been measured by A 1 . But measuring
them implies, again, intending to measure them, and these intentions of
A1 ought to be measured first by A 2 ; and so ad i n f i n i t u m . Notice, on
the other hand, that there is no difficulty whatever for A-, to measure
the occurrence or non-occurrence of redness on the faces of the members
of A2 and vice versa. Or, to choose an example from mechanics, there
is no difficulty whatever for A-, and A 2 to use each other as specimens
of freely falling bodies.
The same point can be made even more easily in the following way:
Suppose that a given psychologist intends-, to investigate intentions
of his test persons. Then, in order to justify his undeniable know
ledge of his intention·,, he ought to measure it, which implies intending,,
to measure it, but then he ought to (intend3 to) measure his intention2
first; and so on. By contrast, he could easily investigate the redness
on his own face or use himself as a freely falling body.
To give one more example, suppose that someone is experimentally
investigating the process of understanding in his test persons. It goes
without saying that he understands-, the outcome of his experiment, i.e.,
whether it confirms or falsifies his hypothesis, and more importantly, he
knows that he understands it. But he cannot subject his understanding1
THE IDEA OF H
' ERMENEUTICS' 23
to experimentation; and even i f he c o u l d , i t would be necessary f o r him
to be able non-experimentally to understand 2 the outcome of t h i s e x p e r i -
ment; and so on. I t i s i n f a c t a r a t h e r s e l f - e v i d e n t t r u t h t h a t when I
am conducting experiments, I must know what I am doing; but i t is not
on the basis of experiments (on myself) t h a t Ï have acquired t h i s know-
ledge (of m y s e l f ) . Nor can i t be claimed t h a t , when I see others per-
forming experiments and know what they are doing, I have acquired t h i s
knowledge on the basis of experiments. As Wittgenstein (1967, I I , § 71)
points o u t , an experiment presupposes mutual understanding.
We have seen, f i r s t , t h a t w i t h i n a given experimental situation
there are objects whose i n t e n t i o n s or lack of them cannot be measurably
tested and, second, t h a t these are p r e c i s e l y objects which are known to
have i n t e n t i o n s . By c o n t r a s t , the redness or non-redness of a l l objects,
i n c l u d i n g the measuring instruments and the bodies of the s c i e n t i s t s ,
can be measurably tested in a given experimental situation.
Above, I presented a c l e a r - c u t d i f f e r e n c e between natural science
and experimental psychology. Since the l a t t e r i n v e s t i g a t e s the occur-
rence of instances of concepts ( e . g . , 'intention') on which, as forms
of human a c t i v i t y , a l l sciences - i n c l u d i n g experimental psychology
i t s e l f - are necessarily based, i t i s l o g i c a l l y impossible t h a t i t could
exhaustively describe (the knowledge of) a l l instances of such concepts.
(Instances of ' i n t e n t i o n ' f o r example are necessarily contained in the
very act of d e s c r i p t i o n . ) Attempts at providing an exhaustive e x p e r i -
mental d e s c r i p t i o n i n e v i t a b l y lead to an i n f i n i t e regress, as has been
shown above. The regress can be halted only by g i v i n g up the attempt at
o p e r a t i o n a l i s i n g one's knowledge i n i t s e n t i r e t y and by s t a r t i n g to re
flect on t h a t p a r t of one's knowledge which cannot be operational i sed.
We have seen t h a t even i n experimental psychology there e x i s t s t h i s type
of knowledge, t h a t i s , even there not a l l knowledge about what is to be
i n v e s t i g a t e d , e . g . i n t e n t i o n s , is experimental knowledge. In such a
case, then, experimentation i s not j u s t unnecessary, but also impossible.
By c o n t r a s t , physical science does not i n v e s t i g a t e instances of concepts
on which i t , qua science, i s based. A l l physical p r o p e r t i e s of scien-
t i s t s and of measuring instruments a l i k e are measurable, which means
that, within physics, all (potential) knowledge about what is to be in
vestigated is experimental knowledge.
The unsuccesful attempt to turn all psychological knowledge into
experimental knowledge is clearly analogous to the attempt to formalise
the whole of logic, i.e., to turn all (logical) languages into formalised
languages. Today it is universally agreed that the attempt at total
formalisation is a logical impossibility. In each particular case, the
formalised language must be intuitively understood - otherwise the
formalisation would simply have no point - and this understanding re
mains of course unformalised and can only be expressed through ordinary
language. And if it subsequently becomes formalised, then the new (me-
ta-)formalisation will presuppose an intuitive understanding of its own,
expressible, again, through ordinary language only; and so on. This ge
neral principle has been formulated in the dictum "ordinary language is
always the last metalanguage". The only way to stop the infinite re
gresses of formal languages and of psychological experimentation is
(self-)reflection, ordinary language being just its expression.
At first glance, experimental psychology and logic are rather dis
similar. They agree, however, on using experimentation and formalisation,
respectively, as means of externalisation of knowledge. As we have just
seen, all knowledge cannot be externalised; the non-externalisable part
can only be reflected upon. Externalisation is a more inclusive concept
than operationalisation, and a more restricted one than expression.
As a method, reflection is clearly different from empirical testing
with its predictive and explanatory components (cf. 1.2. above): the
former operates on knowledge whereas the latter operates on events in
space and time. As Habermas (1968:9) has succintly said, lack of re
flection is positivism. Reflection is an objective method and leads to
objective results, but its objectivity is of course different from the
measurable objectivity of natural science.
2.2. Sociology
Just like experimental psychology, main-stream or Durkheimian so

ciology is in important respects similar to natural science: it constructs
(probabilistic) causal models in order to explain actual (social) beha-
viour (cf. Blalock 1964 and Boudon 1974; as for applications to linguis
tics, cf. Itkonen forthcoming a). In this section, however, I shall
concentrate specifically on those aspects that distinguish sociology
from natural science.
The uniformity of the data of natural science is guaranteed by the
fact that only its measurable properties are taken into account. Since
each centimeter or second is identical with each other centimeter or
second, the differences and similarities between (physical) things and
events can be ascertained in a precise and perfectly general way. In
experimental psychology already, it is only seldom that behaviour and
its relevantsurroundings can be given numerical values in an equally re
liable way. Yet, intuitively speaking, it seems to be possible that
different types of behaviour and surroundings relating, e.g., to the
psychology of perception could be assigned to relatively stable and
easily generalisable categories. In sociology, however, even this in
tuitive confidence is lacking to a large extent. Such sociologically
relevant dimensions as welfare or discontentment are not measurable in
the same self-evident way as length or time. Nor has it been possible
to make them measurable through operationalisation, as has been done to
the dimension of colour for example. As a consequence, no feelings or
acts of discontentment are interchangeable, as measurably identical
things or events are, and results obtained in one social context can
only with qualifications, if at all, be generalised to other social con
texts .
Of course, it is possible to give exact space-time coordinates to
particular instances of social behaviour, and in this sense, then, so
cial data are 'measurable'. But it is justifiably felt that, taken by
themselves, space and time are sociologically irrelevant dimensions:
giving a spatio-temporal definition to social behaviour changes it into
physical behaviour and makes the data of sociology simply disappear. On
the other hand, dimensions such as welfare or discontentment are felt,
again justifiably, to be sociologically relevant, whatever their exact
spatio-temporal manifestations turn out to be.
The framework of universal measurability defines the notion of
scientific observation (Beobachtung): space and. time measured i n such a
way is observable. Unable to impose a s i m i l a r , s c i e n t i f i c a l l y relevant
framework on social d a t a , the s o c i o l o g i s t is forced to t r y to interpret
(deuten) or understand (verstehen) them, as best he can, before he can
even t h i n k of s c i e n t i f i c a l l y explaining them. This f a c t is c l e a r l y e v i -
dent i n cases where one sets out to describe a f o r e i g n community (cf.
the discussion of Sapir in 3 . 3 . below). When one describes one's own
community, the element of understanding is apparently lacking, because
one has unconsciously acquired the r e q u i s i t e understanding while grow-
ing up in the community; i n t h i s case, the data are s i t u a t e d on the l e -
vel of one's pre-understanding (Vorverständnis).
Understanding pertains to the meaning (Sinn) of social behaviour or
of i t s r e s u l t s . There is no known way, f o r instance no strategy of
t h e o r e t i c a l concepts, correspondence r u l e s , and operational definitions,
to e l i m i n a t e meaning or to reduce understanding to observation ( c f . a b o v e ;
f o r more d e t a i l s , see 4 . 0 . and 7.0. below). Therefore understanding,
which i t s e l f may be of explanatory character (Weber 1968:546-48), must
remain a precondition f o r s o c i o l o g i c a l explanation:
Soziologie ... soll heissen: eine Wissenschaft, welche soziales

Handeln deutend verstehen und dadurch in seinem Ablauf und seinen
Wirkungen ursächlich erklären will (Weber 1968:542). Diese Mehr
leistung der deutenden gegenüber der beobachtenden Erklärung ist
freilich durch den wesentlich hypothetischeren und fragmentari
scheren Charakter der durch Deutung zu gewinnenden Ergebnisse er
kauft. Aber dennoch: sie ist gerade das dem soziologischen Erken
nen Spezifische (o~p.cit., p. 555).
It is often claimed that understanding is a heuristic tool which

merely facilitates access to the data of sociology. This is an entirely
misleading formulation, however, because understanding provides the only
access to the data of sociology. If this access is blocked there is
nothing to be described; and if the access is distorted the data are
distorted accordingly. It is a remarkable fact that no one has ever
tried to describe a society or, for that matter, a language in purely
observational terms, which is another way of saying that no one has
ever tried to describe a society or a language whichhe has not under
stood (at all). Therefore it is pointless to insist that this could be
done, in principle.
All attempts at reducing understanding to observation are inspired
by the positivistic idea of Einheitswissenschaft. They are characteris
tic of philosophers of sociology, not of practicing sociologists, who
have difficulties enough in trying to explain satisfactorily social data
as such. However, many sociologists think that even if the data of
sociology and physics are qualitatively different, they are nevertheless
governed by the same type of laws. This position is certainly justified
to some extent, but it cannot be accepted without qualifications. Com
pare Runciman (1969:10):
The less extreme positivistic position, which concedes validity
to individual characters and intentions but claims that they are
reducible in terms of general laws of human behaviour, can be
countered simply by pointing out that no such reduction has been
succesfully made. Different people will respond differently to
different situations in the light of knowledge which is now un
foreseeable, and any procedure on the strict model of natural
science is therefore bound to break down.
And:
The historian (and therefore the social scientist) can never be a

thoroughgoing positivist; but he must, once he has realized this,
still try to behave up to a point as though he were (op. cit.., p. 11).
In sociological field work the data depend crucially on how the

people define the sociologist who is investigating them. There is no
effective way of controlling this process of definition because it is
performed by 'Objects' who are, in principle, just as unpredictable or
free in their actions as the sociologist himself. Thus, sociological
field work (on which the subsequent theoretical description is entirely
based) is an interaction, or a process of mutual definition and inter
pretation. The deeper one enters into this interaction, the better and
richer data one obtains but, at the same time, the more one is forced to
give up the scientific, data-controlling attitude (Cicourel 1964, chap.
2). The interactional character of sociological research brings out the
general truth that one cannot understand someone else without the possi
bility of being understood by the latter in return. This possibility
contributes in an essential way to distinguishing understanding from
28
misunderstanding, (It is another,contingent matter that often this possi
bility is not actualised, for instance because of the temporal distance
between the one who understands and the one who is being understood.)
When people are misunderstood, they generally show it. Indeed, it is a
universal phenomenon that people may reject an interpretation, perhaps
a numerical one, which is being imposed upon them - one need only to
think of slaves rejecting the concept of 'slave' - and similar phenomena
may occur, and have occurred in fact, in the context of sociological
interpretations, especially when sociology works in a close cooperation
with the political establishment. In such instances, it is even possible
that the interpretation produces a change in what is being interpreted.
It is well known that in natural science research instruments may affect
the behaviour of research objects; but there this phenomenon is predic
table.
The constant possibility of a change of social behaviour points to
the pervasive historicity of social data: not only types of behaviour and
of surroundings change in the course of time, but also the criteria with
which they are evaluated or 'measured'. Moreover, the variation in time
is compounded by a variation in space.
Now let us, for comparison, consider the situation in physics. It
does not make sense to say that physical objects define or understand
the physicist. Physical objects and the physicist 'interact' only in
the figurative sense of the word; it would be more accurate to say that
only the latter acts, namely by manipulating the former. The degree of
the data-controlling attitude increases with the degree of the physical
'interaction', not vice versa. Physical objects 'reject' a proposed in
terpretation only in the figurative sense of the word. Finally, though
eyery particular conception of natural science is historically given, as
is evident from the discrepancy between the Aristotelian and the Gal i lean
traditions, the current conception consists precisely in describing phy
sical reality in ahistorical terms. Physical reality may have a history,
e.g., it may be true that the universe is constantly expanding, but this
history is expressed in ahi stori cal terms. It might be possible to de
scribe the physical reality differently, but we have decided to consider
in it only that which is measurable by means of units of length, time,

and weight, categories which are explicitly defined as ahistorical enti
ties. The important thing is that this decision has proved to be well-
motivated: it has not led to a conflict with physical facts; rather, it
seems to reflect the very nature of physical data adequately, in a pri
mitive, noncircular sense of 'adequacy'. In sociology, by contrast, a
similar decision, i.e., a decision to adopt a fixed, ahistorical system
of descriptive concepts, leads immediately to a conflict with social
data. This is an undisputable fact. The only thing that can be dispu
ted about, is whether or not this situation must remain the way it is.
In my opinion, no amount of progress can bring about a qualitative change
in this respect. To do justice to the historicity, which sociology shares
with its subject matter, it must work with an essentially open system
of primitive descriptive concepts.
Sociology sets up hypotheses that are confirmed or falsified by
what occurs or obtains in space and time. To that extent, it would be
accurate to call sociology an 'empirical' science. We have seen, on the
other hand, that the data which either confirm or falsify sociological
hypotheses cannot be defined in a uniform, numerical way. Rather, the
data come to be seen as what they are through the process of understand
ing a given culture, or, more generally, a given body of social relation
ships or meanings; and there is no known way to operationalis.e either
this process or its result. Since the knowledge of social data cannot
be operationalised in a uniform way, namely, in terms of "position and
movement of macroscopic physical bodies", it follows that sociological
theories cannot be falsified in such terms, which again means that so
ciology is not an empirical science in the same sense as physics (cf.
1.1.). In other words,sociology contains an irreducible hermeneutic
component.
Where there is a danger of confusion, I shall use the terms 'empi
rical 1 ' and 'empirical2' as equivalent to, respectively, 'empirical in
the physical, measurable sense1 and 'empirical in the social, non-mea
surable sense'.
It is the same fact which separates both experimental psychology
and sociology from natural science: psychologists and sociologists in
vestigate something which they themselves, qua scientists, are part of.
It is impossible for them to reach avantage point from which all psy
chological or social-historical phenomena could be exhaustively described,
Such a vantage point would lie outside of the human nature; but no one
can step outside of himself. By contrast, the physical reality, inclu
ding the bodies of scientists themselves, can be viewed from such a
vantage point.
Within sociology, the fact that scientists investigate what they
are part of, or what is qualitatively similar to them, draws attention
to the historical nature of man and, hence, to problems of interpreta
tion. In the discussion of experimental psychology in 2.1., the role
of interpretation appeared nonexistent. In retrospects it can be seen,
however, that psychological phenomena, like intentions, are identified
as what they are solely on the basis of a common pre-understanding which
psychologists share with their 'data', i.e., test persons. Such a pre-
understanding (or Vorverständnis) is inevitably a culture-dependent or
historical phenomenon.
2.3. Psychoanalysis and Psychotherapy
Sociology investigates phenomena in space and time and offers causal

explanations, for them: it at the same time has to struggle with the prob
lem of interpreting its data. All these elements reappear in the context
of psychoanalysis. However, their constellation is new, and the empha
sis on the hermeneutic component is stronger. - My view of psychoanaly
sis is based primarily on Habermas (1968) and Lorenzer (1973).
The psychoanalyst attempts, essentially, to trace disturbances in
the patient's behaviour back to certain early traumatic experiences, and
to cancel their negative influence. The immediate data which he has to
start from are almost exclusively of verbal character: they consist of
dream-reports and reported free associations« The immediate task is to
interpret this verbal material, or to translate it into ordinary langu
age. Freud already noticed the analogy between this undertaking and
philological work (Habermas 1968:263). Translation between two langu
ages is not possible, unless they possess common meanings. The very
possibility of psychoanalysis rests on the fact that the patient's langu

age is a distorted variant of ordinary language. Its interpretation
cannot succeed, unless the systematicity of the distortion is revealed,
and this presupposes, in turn, the existence of a general psychoanaly
tic theory. There are several such theories, but they all seem to share
the assumption that neurotic disturbances in language and behaviour are
symptoms of, or caused by, an early conflict between one's personal wish
es and properties of one's social environment: the conflict is resolved,
and the consistency between individual and social is reinstated, by re
moving the offending wishes from one's consciousness and, at the same
time, by removing the corresponding expressions from one's language.
This is only an apparent solution, however, because the wishes remain
buried in the unconscious and give rise to neurotic surrogate behaviour.
Rather than interpreting the conflict as existing between two self-
consistent but opposite entities, i.e., the patient and his social en
vironment, it is also possible, and perhaps preferable, to locate the
conflict, or contradiction, in the latter, that is, to claim that one
of these two entities, viz. the social environment, is in fact self-
contradictory. The patient merely internalises the social contradic
tion. This brings him into an insoluble 'can't win' situation, where
neurotic or psychotic behaviour appears as the only 'rational' solution
(Laing & Esterson 1964).
Interpreting the patient's language, or translating it back into
ordinary language, is a typically hermeneutic undertaking, given that
traditional hermeneutics originated in philological analysis and clari
fication of texts (Palmer 1969:75-83). Taken in itself, this phase of
psychoanalytic research has little or no connection with the methodology
of natural science. However, the situation changes when we take into
account that the psychoanalyst is, after all, looking for causal factors,
i.e., those conflicts which have caused, inter a l i a , the disturbances
in the patient's language. To this extent, it seems, then, that the
psychoanalyst is doing empirical science. However, his preoccupation
with causality is of a peculiar nature, because as soon as he discovers
a cause-effect relationship, he tries to abolish it, something which
would be unimaginable w i t h i n natural science. That i s , i t is not j u s t
a question of preventing a cause-effect r e l a t i o n s h i p from being actua-
l i s e d , but of l i t e r a l l y a b o l i s h i n g it.
We saw in 2 . 2 . (above) t h a t the s o c i o l o g i s t must enter i n t o an i n -
t e r a c t i o n w i t h the people he is i n v e s t i g a t i n g . In psychoanalytic r e -
search the degree of i n t e r a c t i o n is considerably higher. The a b o l i t i o n
of the causal f a c t o r s t h a t have been responsible f o r the p a t i e n t ' s neu-
r o t i c behaviour c o n s t i t u t e s the c r i t e r i o n f o r the success of psychoana-
l y t i c work. But the success depends on the a c t i v e cooperation of the
p a t i e n t : the e a r l y c o n f l i c t looses i t s causal g r i p on him, only i f he
understands i t as what i t i s ; t h u s , understanding abolishes causality.
The r e l a t i o n between the psychoanalyst and the p a t i e n t is not j u s t
an i n t e r a c t i o n ; more p r e c i s e l y , i t is a dialogue. In the beginning, the
dialogue may be very d e f e c t i v e , but i t is p r e c i s e l y the task of the
psychoanalytic i n t e r p r e t a t i o n to improve on i t . In the end both p a r t -
ners are supposed to have a normal d i a l o g u e , i . e . , to speak the same,
n o n - d i s t o r t e d language. - Here, I t h i n k , any analogy to natural science
i s too f a r - f e t c h e d to deserve a comment.
Psychoanalysis is a critical science: i t t r i e s to show to the pa-
t i e n t t h a t , in important respects, his (self-)knowledge i s l i m i t e d or
d i s t o r t e d , or simply false. The p a t i e n t achieves, h o p e f u l l y , true know-
ledge when he becomes conscious of those f a c t o r s t h a t have been, from
the o u t s i d e , as i t were, determining his (previous) consciousness. It
has often been noted t h a t psychoanalysis shares t h i s c r i t i c a l attitude
with Marxist sociology. The d i f f e r e n c e between c r i t i c a l science and
prescriptive science i s , a g a i n , one of degree or emphasis ( c f . 2 . 5 . be-
low) .
Insofar as i t can be s a f e l y assumed t h a t the c o n t r a d i c t i o n r e s i d e s ,
p r i m a r i l y , i n the p a t i e n t ' s social environment, the way t h a t psychoana-
l y s i s - o r , r a t h e r , psychotherapy - was characterised in the previous
paragraph as a c r i t i c a l science, i s no longer a p p r o p r i a t e . On t h i s new
assumption, psychotherapy i s c r i t i c a l p r i m a r i l y of e x i s t i n g social con-
d i t i o n s and only secondarily of the p a t i e n t ' s self-knowledge; t h a t i s ,
his knowledge is f a l s e only i n s o f a r as he has taken his social e n v i r o n -
ment as representing an unquestionable and immutable truth.

Habermas (1968, chap. 10) considers psychoanalysis as analogous
to the philosophical critique of 'knowledge. On this interpretation,
the patient and the psychoanalyst fulfil respectively the functions of
knowledge and of critical reflection upon knowledge. This analogy is
illuminating but not very fruitful, because it is not able to represent
the achievement of any actually new knowledge: the truth towards which
the 'critical reflection' is progressing is always known in advance,
i.e., it is that common-sense knowledge which the psychoanalyst already
possesses and which the patient is supposed to (re)gain at the end of
the treatment. However, it would seem to be more fruitful to reverse
the roles in such a way that the psychoanalyst, or his knowledge, comes
to represent the knowledge to be criticised. Of course, the patient
cannot literally take over the function of critical reflection, but in
any case he gives the evidence in the light of which the psychoanalyst
may criticise and improve his own knowledge. That is, from seeing what
kind of social relationships are inadequate to, because destructive of
the human mind, he may draw tentative inferences as to the 'real' natu
re of the human mind and to the type of society adequate to it.
2.4. Sociology of Knowledge
All the human sciences that I have discussed so far share a concern
with particular entities whose location in space and time can, or could,
be specified with a high degree of precision. Even in psychoanalysis,
the traumatic experience that causes neurotic behaviour was itself
caused by an event or a series of events that could be, at least in
principle, spatiotemporally specified; and events of this kind are part
of the subject matter of psychoanalysis. On the other hand, sociology
of knowledge, also called 'phenomenological sociology', is characterised
by the fact that it excludes the consideration of particular space-time
entities. This feature sets sociology of knowledge (as well as in this
respect similar sciences) definitively off from natural science. Metho
dological characteristics that are common to sociology of knowledge as
well as natural science are common to all sciences without exception.
To begin with, I want to explain under which conditions it is rea-
sonable and even necessary to make scientific descriptions without any
regard to particular spatiotemporal entities. Let us consider, for in
stance, the description of such entities as beliefs. It is clear that
a person must possess some internal ,psychologycal mechanisms which en
able him to acquire beliefs or, alternatively, which determine what
kinds of belief he may acquire. It is also clear that certain external,
social factors or mechanisms determine, to a high degree, what kinds of
beliefs he does acquire in fact. Furthermore, it is not only the case
that beliefs are determined by internal and external mechanisms; beliefs
are themselves part of a determining mechanism insofar as they influence
the behaviour of the one who maintains them, or, alternatively, are ma
nifested in and through his behaviour. And yet, after we have described
what determines a belief, and what it determines, there still remains
the task of describing the belief itself. Since psychological and so
cial factors determine only in a rather abstract and unpredictable way
which beliefs one actually comes to uphold, it is necessary to describe
beliefs in their own right, in addition to the fact that it is always
possible and legitimate to do so. This is not to deny that, depending
on the purpose at hand, the description of a belief may occur as only
a part of a larger, e.g., historical description, which must then con
tain also the causes and effects of the belief in. question. (What I have
said here about beliefs, is generally true of attitudes, opinions, ideas,
goals, ideals, etc.)
To give a still clearer example, let us consider, next, such enti
ties as games. All games must be, at least potentially, learned and
played, and since processes of learning and playing necessarily take
place in space and time, it follows that games necessarily have a spa
tiotemporal reference. And yet, although games and (at least poten
tial) processes of learning and playing always occur together, they
are clearly different aspects of one and the same phenomenon. Indeed,
it would be a serious mistake not to distinguish between descriptions
of games and descriptions of how they are learned and/or played. The
former are, directly, descriptions of rules (or norms) and hence, in
directly, of possible correct behaviour. The latter are descriptions
of actual behaviour, whether correct or not.

The distinction between normativity and spatiotemporal ity will
occupy us later at a greater length (cf. 7.0. below). Here I simply
state it as a fact that normativity cannot be reduced to, or exhausti
vely defined in terms of space and time. This is not a repetition of
the thesis that social behaviour cannot be reduced to physical behavi
our, i.e., to mere sounds and movements (cf. 2.2.). It means, rather,
that a norm cannot be adequately defined by means of any number of
factual nonnormative actions, which are of course spatiotemporal, but
not only spatiotemporal, entities.
Games are learned and played under psychological and social con
ditions, and therefore descriptions of the processes of learning and
playing games are psychological and/or sociological (in fact, social-
psychological) descriptions. As such, they are directed towards dis
covering those causal (or perhaps teleological) mechanisms that, together
with (people's subjective conceptions of) the rules themselves, would
explain the game-behaviour; hypotheses about the mechanisms involved
are confirmed or refuted by spatiotemporal actions which may, but need
not, occur under experimental conditions. By contrast, within the de
scription of rules there is no room for mechanisms of any kind. The
notion of (causal) 'mechanism' is inseparably tied to space and time,
that is, to the idea of making something happen in space and time.
12
Rules and norms do not exist in space and time, nor can they be de
fined in spatiotemporal terms. This idea is expressed by calling norm-
descriptions 'conceptual analyses'. It is indeed an analytical truth
that conceptual analyses are incompatible with the (empirical) search
for causal mechanisms. This analytical truth is reflected in the simple
fact that no one would consider for example descriptions of the rules
of chess or poker as causal descriptions. - The same basic distinctions
as here are made by Weber (1968:332-43). Weber (542-48) explicitly re
moves normativity outside of (empirical) sociology.
It can be seen, in retrospect, that the description of beliefs
and ideas, too, has in itself nothing to do with the search for causal
mechanisms. A causal element may be involved both in acquiring a be-
lief or an idea and in maintaining it or acting in accordance with it,
in analogy to the two processes of learning a game and playing it. But
beliefs or ideas in themselves, or knowledge in general, are incompa
tible with causality. This is evident from the simple fact that know
ledge about an event in space and time is not itself in space and time,
whereas causality must obtain between entities in space and time.
In linguistics, for instance, it is customary to speak of 'uncon
scious mechanisms that underlie knowledge' or, simply, of 'underlying
knowledge', which might give rise to the impression that descriptions
of knowledge are somehow imperfect without corresponding causal de
scriptions. However, these expressions refer in fact to the mechanisms
underlying the acquisition and performance of those actions (of sentence
production and recognition) which, or whose results, constitute the
object of the knowledge in question. The only mechanisms that underlie
knowledge, and not for instance actions as the object of knowledge, are
those little-understood mechanisms which produce and maintain (self-)
consciousness in general, and which cannot be limited to individual
psychology (cf. 4.0. below).
A description of a game, of chess for example, cannot be empirical
ly falsified, in that precise sense in which empirical (i.e., empirical1
or empirica 2 ) falsification is defined as falsification on the basis
of particular events or actions in space and time. No particular action
performed in playing chess can invalidate the rules of chess, or the
corresponding description: all such actions are either correct or in
correct, and in both instances they have no impact upon the rules. Con
sequently, according to Popper's 'criterion of demarcation', a game-de
scription is 'metaphysical', since it is not empirical (Popper 1965:40-
42). Of course, a description of chess may be falsified by pointing
out that it misrepresents the rules, or fails to describe them as they
are. But it is impossible to reformulate this as a case of empirical
falsification. It should not be surprising that conceptual analyses
are not empirically falsifiable. In accordance with the use of subin
dices up to now, I shall say that empirical1 theories are testable 1 , em-
piricai ? theories are testable 2 , and conceptual analyses are testable 3 .
37
THE IDEA OF H
' ERMENEUTICS'
One cannot describe a game adequately unless one knows it, or knows
its rules. Therefore it might be said equally well that one describes
a game or the knowledge of a game (for details, see 5.0.). This is the
reason why it is proper to discuss games in the present context. Above,
I have made a clear distinction between describing a game and describing
the process of learning it. This means that I must now make an equally
clear distinction between knowing or mastering a game and learning it.
In practice, the existence of such a distinction is always taken for
granted. For instance, it would be absurd for me to doubt indefinitely
whether or not I have learned the rules of poker. It is no objection
to point to the fact that no one can tell the precise moment from which
on one is no longer learning the game, but masters it. There are count
less distinctions, e.g., those between young and old, or rich andpoor,
whose existence it would be absurd to deny, in spite of the fact that
it is impossible to tell where, precisely, the rich for example ends
and the poor begins. Nor is it an objection to point to the fact that
some people are sometimes forced to start describing a (complex) game
before they master it properly.
Descriptions of games can be Supplemente d, and perhaps even modi
fied, by psychological or sociological descriptions of how they are
learned and played; but they cannot be replaced by the latter. There
exists an important asymmetry between these different types of descrip
tion: it is possible, and quite normal, to describe games in their
own right, without reference to their psychological or sociological as
pects; but it is impossible to describe psychological or sociological
aspects of games without reference to games themselves.
This lengthy exposition was necessary in order to give a general
justification for the kind of science that sociology of knowledge is.
My principal sources in this domain are Mead (1934), Schutz (1962), and
Berger & Luckmann (1966). This type of research was anticipated in
Husserl (1962[1913]:91-96).13
Sociology of knowledge is primarily interested in the analysis
and clarification of common-sense knowledge, not in those conditions
which determine the nature of this kind of knowledge. Thus, unlike
psychoanalysis or Marxist sociology, sociology of knowledge is not a
critical science: it tries to describe knowledge, not to influence it.
However, a critical dimension could be seen as a rather natural exten
sion of the general analytical concern with knowledge.
In the discussion of empirical sociology in 2.2., we noticed that
the description and explanation of social behaviour presupposes fami
liarity with the culture or, more narrowly, with the institutions with
in which the behaviour in question takes place. One cannot even begin
to explain a set of (spatiotemporal) actions until one has understood
them as what they are, that is, until one has succeeded in interpreting
their cultural and institutional background. Now, empirical sociology
and sociology of knowledge investigate the same reality, but they do
so from differing points of view. The latter deliberaty disregards
actual social behaviour, which constitutes the data of the former. In
stead, it is centred on that cultural and institutional background
which makes both the existence and the understanding of social behaviour
possible, given that social is only what is understood as such (cf. 5.1.
below). Sociology of knowledge focusses on the precondition of empiri
cal sociology, i.e., upon what must be known before behaviour can be
seen as social, and therefore it is only natural that sociology of know
ledge is also called 'aprioristic sociology'. Again, it is clear that
there can be no culture without people who 'support' it, but the descrip
tion of a culture can be just as little reduced to the description of
social behaviour as the latter can be reduced to the description of phy
sical behaviour.
The more abstract or general the knowledge described, the more phi -
sophical and the less sociological the description is in character.
Consider, for instance, how Mead (1934:152-64) and Schutz (1962:312-29)
analyse the basic structure of a social interaction between two persons
A and B. A plans and evaluates his actions in terms of what he thinks
that Β expects from him, and vice versa; and A knows that Β plans and
evaluates his own actions in the same way, and that Β also knows that
A knows this, and vice versa. This 'reciprocity of perspectives' con
stitutes the conceptual framework, or the a priori condition, of all
actual, spatiotemporal instances of social interaction, and the descrip
tion of the former differs accordingly from that of the latter.
The analysis of the reciprocity of perspectives is rather philo
sophical in character. It is indeed impossible to empirically falsify
such an analysis. That is to say, a theory of natural science can, in
principle, be falsified by a few events, or even by a single event,
which contradicts it. But we have absolutely no idea of what such an
action would be like which could falsify the 'theory' of the reciproci
ty of perspectives for instance. If someone plans his actions without
adopting in the least the perspective of others, we content ourselves
with stating that his actions are peculiar in this respect. Therefore,
this theory is not empirical, not even in the non-measurable sense of
'empirical'. This result can be generalised so as to apply to all
theories of sociology of knowledge. The most we can say is that these
theories, and generally all conceptual analyses, are false, if they mis
understand or misrepresent the knowledge which is their subject matter.
But this is not a case of empirical falsification. When the number of
those people decreases who possess the body of knowledge to be described,
the description gets a more sociological or, loosely speaking, more 'em
pirical' flavour. In particular, this is true of describing particular
institutions of a given society. However, the points I made while dis
cussing the description of maximally general knowledge remain validhere
as well.
An institution is constituted by a set of norms which define the
appropriate or correct behaviour within it. Norms also define a set of
more or less constant roles. The distinction between norms, on the one
hand, and mere habits or customs, on the other, is only a gradual one;
I shall concentrate exclusively on what are felt to be relatively clear
cases of norms. Depending on the explicitness of the norms involved,
institutions have a more or less pronounced existence. A game with de
finite rules is the limiting case of an institution; in fact, it is a
defining property of institutions, as this term is being used here, that
they possess a recognisable similarity to such a game. Hence, what I
have said about games, applies to institutions as well. There are se-
veral institutions within one and the same society, and therefore it is
normally the case that one person does not know all of them. In this
sense, then, knowledge about institutions is less general and more 'em
pirical' than for instance knowledge about the reciprocity of perspec
tives. However, one is able, in principle, to come to know any insti
tution of one's own society, or even of any other society. Moreover,
we cannot help making a distinction between learning an institution and
mastering it; here the analogy to the games is obvious. When one mas
ters an institution, one knows it. That is, one can never know with
certainty how different people have acted or will act within the limits
of an institution, but one can know its norms with certainty. Descrip
tions of such a knowledge may be false in several different ways, but
they can be empirically falsified just as little as descriptions of the
reciprocity of perspectives, or descriptions of games in general.
Learning to master an institution means learning to act correctly,
and knowing that this is what one is learning. Therefore, the resulting
knowledge can be appropriately called 'agent's knowledge'. On the other
hand, physical reality or, more generally, spatiotemporal reality of any
kind, cannot be learned or mastered in any literal sense; it can be ob
served or hypothesised about, but it cannot be acted out. Therefore the
knowledge about what happens in space and time can be called 'observer's
knowledge'. The basis for the distinction between these two types of
knowledge has been formulated in the classical hermeneutic tradition as
follows:
Die Natur ist uns fremd. Denn sie ist uns nur ein Aussen, kein
Inneres. Die Gesellschaft ist unsere Welt. Das Spiel der Wechsel
wirkungen in ihr erleben wir mit in aller Kraft unseres ganzen We
sens, da wir in uns selber von innen, in lebendigster Unruhe, die
Zustände und Kräfte gewahren,aus denen ihr System sich aufbaut
(Dilthey 1914:36).
Nur was der Geist geschaffen hat, versteht er. Die Natur, der
Gegenstand der Naturwissenschaft, umfasst die unabhängig vom Wir
ken des Geistes hervorgebrachte Wirklichkeit. Alles, dem der
Mensch wirkend sein Gepräge aufgedrückt hat, bildet den Gegenstand
der Geisteswissenschaften (Dilthey 1927:148).
Das Subjekt des Wissens ist hier eins mit seinem Gegenstand, und
dieser ist auf allen Stufen seiner Objektivation derselbe (op. c i t . ,
p. 191).
It is to be noted, however, that traditional hermeneutics fails to

clearly distinguish between knowledge, in particular normative knowledge,
and social behaviour, and, hence, between aprioristic and empirical so
ciology. The relation between knowledge and social behaviour is a dia
lectical one: the two are clearly different, but presuppose each other.
Knowledge is a priori with respect to any particular action: it is part
of the concept of action that an action is understood, at least poten
tially, by the one who performs it, and by others, as what it is; and
this presupposes the existence of a body of knowledge on the basis of
which the action is understood. Similarly, an institution or a norm
is a priori with respect to any particular action subsumable under it.
Durkheim (1938:57) expresses the same idea by saying that a social fact
(fait social) is independent of, rather than a priori with respect to,
any of its individual manifestations. On the other hand, knowledge (of
institutions or norms) cannot be a priori with respect to all social
behaviour.
I view descriptions of any kind of institutional knowledge as qua
litatively similar to sociology of knowledge. This judgement applies
to large areas of cultural anthropology and of jurisprudence for example.
Actual institutional behaviour constitutes data for corresponding empi
rical descriptions.
Sociology of knowledge, as it is known today, makes little or no
use of formal methods. This may be due to the fact that current research
is to a large extent concerned with the preliminary task of interpreta-
t i o n , as it pertains either to the most general features of the social
reality or to specific institutions. It should be evident, however,
that a systematic description of an institution, once it has been inter
preted, or of several institutions at the same time, cannot be carried
out without appropriate means of formalisation. A certain amount of
formalisation appears in fact already in Lévi-Straussian anthropology.
Sociology of knowledge is not necessarily restricted to the ana
lysis of the common knowledge of social reality. It is perfectly pos
sible to subject the common knowledge of physical reality to a similar
anlysis. For instance, Aristotelian physics was based on a method of
analysis which was practically identical with that of sociology of know
ledge (cf. 9.3. below). Therefore it is only logical that Aristotelian
physics should have been called 'phenomenological physics' (Lorenzen
1969b:144).
2.5. Philosophy
There is no clear distinction between sociology of knowledge and

philosophy. It is largely due to an historical accident, which problems
have come to be seen as 'philosophical' ones.15
Analysis of knowledge is what philosophy and sociology of knowledge
are about. Analysis of knowledge means, in turn, analysis of those con
cepts into which knowledge is structured or, equivalently, analysis of
those expressions which are used to express the concepts. Concepts are
tied to norms for their correct understanding and use. It might even
be said that there is an institution connected with every coherent set
of concepts. Such an institution can be experimentally investigated
just as little as any other institution orgame. Rather, the 'institution'
of the use of concepts is the a priori condition for the possibility of
experimentation.
To illustrate: In 2.1. (above) we noted that concepts like 'redness'
and 'intention' are different insofar as there are differences between
the ways in which their instances can be experimentally investigated.
However, as far as the possibility of experimental investigation is con
cerned, the concepts themselves are quite similar: they - or, rather,
their use - cannot be subjected to experimentation, because in this
case it is impossible to uphold the principle that the outcome of a
test must be accepted, whatever it turns out to be. Thus, if a test
person claims that things which we know to be red are not red, or that
stones have such and such intentions, this outcome has no effect upon
our concepts 'redness' and 'intention'; and therefore what we have here
is not a test about these concepts. Rather, it is a test about the
perceptual or cognitive state of the test person. If we were 'testing'
the concept 'redness', for instance, we would accept only such outcomes
where things that are really red are claimed to be red. But this only
means, again, that we are not dealing with genuine tests.

What I have said here about concepts is by the same token true of
all rules or norms: the existence of a rule cannot be established ex
perimentally or by observing actual behaviour. Rather, a rule must be
learned; and once it has been learned, it gives us a criterion with
which we may evaluate actual behaviour as either correct or incorrect.
Of course, learning involves both observing and hypothesising. But when
one has learned a rule, it is known, that means one knows how to act
correctly, and therefore the sentence expressing the rule is not an ex
perimentally testable hypothesis.
Consequently, whatever ontological differences there are between
red things and intentions, and whatever methodological differences there
are, as the result, between corresponding descriptions, experiments with
red objects and those with intentions presuppose in just the same way
the knowledge of the respective concepts. Similarly, when these con
cepts are being operationalised, or given scientific definitions, the
success or failure of operationalisation can only be judged by compa
ring proposed definitions against the pre-experimental knowledge of these
concepts. Giving scientific definitions to everyday concepts does not
eliminate the latter, or the need for them.
The fact that analysis of concepts is methodologically the same,
whether the concepts are about social or physical reality, does not do
away with the general distinction between human science and natural
science, as it was formulated by Dilthey for example. All concepts with
out exception are made and used by man. Therefore the knowledge about
them is agent's knowledge, and they are properly considered as the sub
ject matter of human science. Physical reality, however it is conceptu
alised, is not made by man; here even if the concepts are man-made, the
instances of concepts are not. (As for games, by contrast, man has not
only made the concept 'correct move', but he also makes all correct
moves as well as all incorrect ones.) Therefore the knowledge about
physical reality is observer's knowledge, and physical reality itself
is investigated by natural science. Events in space and time could not
possibly be explained and predicted by mere conceptual analysis.
All concepts are similar in two important respects, First, as we
have just seen, they all belong into a normative context, or an 'insti
tution' of their use, which cannot be experimentally established or
tested, but is simply known, once it has been, learned. Secondly, all
concepts, even philosophical or mathematical ones, are somehow, either
directly or indirectly, manifested in space and time. Therefore the
mere fact that a concept has, for instance by means of a suitable 'func
tion', been put into relation with something in space and time, proves
in itself nothing about the empirical or nonempi ricai nature of the
science which the concept in question is characteristic of (cf. 7.2.
below).
The institution of the use of everyday concepts is identical with
Mittelstrass' (1974b) notion of elementary praxis of distinction and
orientation (Unterscheidungs- und Orientierungspraxis) (for a more exten
sive discussion, see Lorenz 1970, chap.2), Philosophy, in the tradi
tional sense of this word, analyses everyday concepts as well as con
cepts which have grown out of previous analyses of everyday concepts:
Philosophy is the explicitation of our pre-philosophic life in and

with the world. It reveals to us the truth which is s till conceal
ed in it in an implicit way. It brings our original understanding
of the world and of ourselves over into a philosophic comprehend
ing. It brings us to a higher degree of truth, but does not change
our understanding of the world in an essential way,...(Kockelmans
1969:37).
The method of philosophical analysis, or explication, will be described

in 11.0. (below). It is clear that explication is an institution of its
own, that is, an institution which has developed out of simple reflec
tion upon everyday concepts, as it arises for instance in connection
with various disagreements and misunderstandings of daily life. Today,
of course, explication pertains also to scientific concepts. But here
too it remains true that explication does not change in an essential way
our understanding of its object. That is, science so changes in an es
sential way our understanding of (certain aspects of) the world, but ex
plication does not in a similar way change our understanding of science.
- Our knowledge of explication is based on its institutional character.
It is the purpose of explication as practiced in the philosophy of
science not only to clarify, but also to justify the scientific method.
The Erlangen school has made an intriguing attempt at providing a jus
tification (Begründung) for natural science, in particular physics.
This is done by reconstructing, in a stylised form, of course, the ge
nesis of science, that is, by showing how science, as it is today, can
be constructed, step by step, from everyday life. Both science and
everyday life are considered throughout from a realistic point of view,
that is, as combinations of social activity and socially determined
knowledge. The programme of reconstruction can be divided into two,
obviously interconnected parts: first, the reconstruction of scientific
language; second, the reconstruction of scientific measurement. (Notice
that these two a -priori conditions for natural science do not yet exhaust
its methodology.)
As the elementary 'praxis of distinction and orientation', ordinary
language is the inevitable starting point for the reconstruction of
scientific, formal language (cf. 2.6. below). The reconstruction of
scientific language is carried out, in outline, in Kamlah & Lorenzen
(1967), and has been often repeated elsewhere.
The reconstruction of scientific measurement is called 'protophy-
sics'. It is subdivided into the theories of measuring space, time and
mass, or geometry, chronometry, and 'hylometry' (= classical mechanics
without gravitation), which are sciences of increasing complexity. The
objective or empirical-, character of (Galilean) physics is due to the
intersubjective agreement upon those ideal norms, formal isable or explic
able as the axioms of protophysics, which govern actual measurement.
Protophysics has grown out of a practical concern with things and events
(= Herstellung spraxis). It is an a priori science which, instead of
investigating actual physical events, investigates possible physical
events, i.e., the concept 'physical event', as defined by the ideal norms
of measurement (Lorenzen 1969a and Böhme 1976). These norms obviously
constitute an institution of measurement.
Protophysics as conceived by Lorenzen is a descriptive science,
since it describes the ideal norms as they are. However, it has a pre
scriptive function as well for those who have not clearly grasped the
norms in question. The same is true of the philosophy of science in
general. For instance, Hempel's D-N model intends to explicate the no
tion of deterministic explanation as it is; but it may also be seen as
containing an implicit recommendation for how this notion ought to be
applied in practice.
In my opinion, the work of the Erlangen school ought to be seen as
a systematic elaboration of Husserl's thesis that the meaning of modern
science cannot be understood unless the internal connections between
science and everyday life, or Lebenswelt, are brought back to conscious
ness. As for its constructivist method, the Erlangen school is, to be
sure, not indebted to Husserl but, rather, to Dingler (cf. Mittelstrass
1974c). In any case, Lorenz (1970:149-50), for instance, makes use of
the notion Lebenswelt in explicit reference to Husserl. The following
excerpts, which formulate the recurrent theme of Husserl (1954), show
the connection between Husserl and the Erlangenschool:
Die geometrische Methodik der operativen Bes timmung einiger und
schliesslich aller idealen Gestalten aus Grundgestalten, als den
elementaren Bestimmungsmitteln, weist zurück aud die schon in der
vorwissenschaftlich-anschaulichen Umwelt, zuerst ganz primitiv, und
dann kunstmässig geübte Methodik des ausmessenden und überhaupt mes-
senden Bestimmens (Husserl 1954:24).
So macht denn jede gelegentliche (oder auch "philosophische") Rück

besinnung von der kunstmässigen Arbeit auf ihren eigentlichen Sinn
stets bei der idealisierten Natur halt, ohne die Besinnungen radikal
durchzuführen bis zu dem letzlichen Zweck, dem die neue Naturwissen
schaft mit der von ihr unabtrennbaren Geometrie, aus dem verwissen
schaftlichen Leben und seiner Umwelt hervorwachsend, von Anfang an
dienen sollte, einem Zwecke, der doch in diesem Leben selbst liegen
und auf seine Lebenswelt bezogen sein musste (op. cit., p.50).
Since then, this general position has gained ground because of the
popularity of Wittgenstein's later philosophy (cf. below, and 4.0.). How
ever, the idea of Lebensapriori was expressed already by Dilthey (1924:
136) in the following terms:
Die fundamentalen Voraussetzungen der Erkenntnis sind im Leben ge

geben, und das Denken kann nicht hinter sie greifen... sie sind
nicht Hypothesen, sondern... aus dem Leben entspringende Prinzipien
oder Voraussetzungen, welche in die Wissenschaft als die Mittel, an
weiche sie gebunden ist,eingehen.
THE IDEA OF H
' ERMENEUTICS' 47
As part of his c r i t i c i s m of the u n i v e r s a l i s t i c claims of p o s i t i -

vism, Apel (1973b) has shown convincingly t h a t the community of s c i e n -
t i s t s , w i t h i t s own l i n g u i s t i c and methodological r u l e s , is the a priori
c o n d i t i o n f o r the p o s s i b i l i t y of empirical science; and t h i s community
(which has in turn developed out of n o n - s c i e n t i f i c communities) cannot
be i t s e l f i n v e s t i g a t e d by methods of empirical science, but only by
hermeneutic r e f l e c t i o n . These ideas go in part back to Peirce and
Royce. Popper (1965:52 and 55) too notes t h a t the philosophy of empi-
r i c a l science is i t s e l f not an empirical s c i e n c e , b u t i n his termino-
logy a kind of 'metaphysics'. More r e c e n t l y , Popper (1972:162-63) has
i d e n t i f i e d t h i s kind of metaphysics w i t h hermeneutic understanding, or
w i t h the study of a r t (p. 180) or l i t e r a t u r e (p. 185). Such remarks,
coupled w i t h the Popperian thesis of the u l t i m a t e i r r a t i o n a l i t y of
science, hardly provide a s a t i s f a c t o r y c h a r a c t e r i s a t i o n of the p h i l o -
sophy of science. Radnitzky (1976) has t r i e d to show t h a t Popper's no-
t i o n of the Letztbegründung of science is in r e a l i t y the same as Apel's
and, one may add, the same as t h a t of the Erlangen school. In my o p i -
n i o n , what he says is c o r r e c t i n i t s e l f , but represents his own stand-
p o i n t more than Popper's. Whether or not the thesis of the i r r a t i o n a l i t y
of science is meant only f i g u r a t i v e l y , as Radnitzky suggests, i t is r e -
f u t e d , f i r s t , by p o i n t i n g to the f a c t t h a t science grows out of the d i f -
f e r e n t i a t i n g use of language and, second, by noting with Wittgenstein
(1969, § 549), in a way reminiscent of D i l t h e y and Husserl, t h a t
You must bear in mind that the language-game is so to say something

unpredictable. I mean: it is not based on grounds. It is not rea
sonable (or unreasonable).
It is there - like our life.
The question of the l a s t presuppositions of ( s c i e n t i f i c ) knowledge

is c l o s e l y connected w i t h the question of the critique of knowledge.
According to Hegel, i t is possible to strengthen the r e l i a b i l i t y of
one's knowledge by r e t u r n i n g to the p o i n t where t h i s knowledge o r i g i n a -
ted and by s y s t e m a t i c a l l y working one's way back up to the present: the
knowledge one f i n a l l y a t t a i n s at the end of t h i s journey is no longer
the same as i t was at the beginning (Habermas 1968:14-35). This idea
is s t r i k i n g l y s i m i l a r to the c o n s t r u c t i v i s t programme of the Erlangen
school (Kamlah & Lorenzen 1967:22, n.l notwithstanding): For i n s t a n c e ,
protophysics achieves a more accurate understanding of modern physics
p r e c i s e l y by r e t u r n i n g to the logical o r i g i n of physics (which l a r g e l y
coincides w i t h the h i s t o r i c a l one) and by c o n s t r u c t i n g i t s rationally
j u s t i f i a b l e development from there up to i t s present form. On the other
hand, as a form of c r i t i q u e of knowledge, the c o n s t r u c t i v i s t programme
does not make the (transcendental-)hermeneutic r e f l e c t i o n superfluous,
because each method, even the one which t r i e s to unearth a l l presuppo-
s i t i o n s , has i t s own presuppositions, which are i n need of c r i t i c a l re-
f l e c t i o n ( c f . Bubner 1976).
2.6. Logic
Knowledge of logic is acquired through learning the rules for

making valid inferences and, secondarily, for stating necessary truths.
Knowing the rules does not guarantee their correct application in each
particular instance, given human fallibility. Therefore knowledge of
logic also implies the ability to detect contradictions in what is being
said. But since this ability too may be misused, there must be general
rules for attacking and defending the logical truth of assertions.
The fact that logic is deeply rooted in everyday thinking and act
ing, is evident in many ways. In the discussion of psychoanalysis in
2.3. we have already seen, incidentally, that even at the least con
scious or 'logical' levels, the need for eliminating contradictions out
weighs all other psychological needs: the contradiction between indivi
dual and social, or, alternatively, the social contradiction in its in
ternalised form, is removed, and the consistency of one's psyche is
established even at the cost of producing neurotic or psychotic beha
viour. The same concern with consistency reappears in connection with
logicians' system-construction. Moreover, even the most elementary forms
of decision-making contain logical inferences. For instance: "If I want
to achieve A, I have to do B. I want to achieve A. Therefore I have to
do B." Finally, experimental research shows that where people are appa
rently making fallacious inferences, they are often in fact inferring
THE IDEA OF 'HERMENEUTICS 49
correctly, but are using premises which are different from those intend
ed, that is, they are not peerforming the task they are supposed to (Hen
ie 1962). And when it is shown test persons that the (perhaps only
apparent) fallacies which they have committed are in contradiction with
other, de facto valid inferences, it is always the fallacious inferences,
never the valid ones, which are subsequently modified (Wason 1964),
Hence, in addition to the passive ability of delecting contradictions,
one also possess the active ability of replacing logical invalidity
by logical validity.
These few remarks may suffice to refute the old myth of the illogi
cal character of ordinary thinking. Of course, logic proper represents
only one part of ordinary thinking. But this does not mean that the
domain of ordinary thinking which goes beyond logic would have to be
contrary to logic.
Logic is not something 'eternal', something which is only contingent
ly related to man. On the contrary, it is based on, or grows out of
everyday life, precisely like philosophy and science in general (cf.
2.5.). Since logic is ultimately a social phenomenon, an adequate de
finition of logic ought to present logic in use. This requirement would
seem to be intuitively obvious. More importantly, it is also conceptu
ally necessary, because - as will be shown in 4.0. (below) - the exis
tence of all languages, including logical ones, is necessarily based on
their intersubjective use. Therefore, the definition of logic must be
a pragmatic one. In particular, it must be able to reconstruct the
situation of the use of logic, i.e., the characteristic interaction of
those who are in this situation. As we have seen in the beginning of
this section, this interaction must consist of acts of attacking and
defending logical truth, to be performed in accordance with definite
rules. It is desirable, moreover, that these acts could be presented
as a natural extension of acts of attacking and defending empirical
truth.
A dialogical or g a m e - t h e o r e t i c a l conception of logic, which satis
fies the afore-mentioned requirements, has been developed by Lorenzen
and Lorenz since the late fifties (see, e.g., Kamlah & Lorenzen 1967,
chaps. 5-6, Lorenzen 1969a, and Lorenz 1973). In addition to its prag
matic nature, this conception has still other advantages over more tra
ditional conceptions. First, both the junctors and the quantifiers are
introduced in the same, dialogical way. Second, although in truth-
functional logic the truth-value of a complex formula can be determined
only if the truth-values of its elementary sentences are already known,
game-theoretical logic does not have to accept this clearly unrealistic
restriction; rather, the truth or falsity of elementary sentences is
established only in the course of the dialogue-game.
The game is a succession of attacks and defenses between two dispu
tants, i.e., the 'proponent' and the Opponent'. The junctors and, or,
if - then, and not are introduced as follows: If the proponent asserts
the sentence 'A&B' , the opponent attacks it by questioning either 'A'
or 'B'; if the proponent can defend both 'A' and 'B', he has won; other
wise he has lost. If 'AvB' is asserted, the opponent attacks it by
questioning it as a whole, and the proponent wins if he is able to de
fend either 'A' or 'B'. If is asserted, the opponent attacks
it by asserting 'A', the proponent attacks by questioning 'A', but if
the opponent is able to defend 'A', the proponent wins only if he is
able to defend 'B'. Finally, if '-A' is asserted, the opponent attacks
by asserting 'A', and the proponent attacks in his turn by questioning
'A'. The opponent wins or loses depending on whether he can or cannot
defend 'A'.
The rules for the use of quantifiers, which are superimposed upon
the above-mentioned rules, are introduced as follows: If the proponent
asserts a universally quantified sentence '(x)Fx', the opponent attacks
by picking out an object a which he thinks is a counter-example. If
the proponent is able to defend 'Fa', he has won (which does not mean,
of course, that he has proven the truth of '(x)Fx'). On the other hand,
the opponent attacks an existentially quantified sentence '(Ex)Gx' by
simply questioning it, and then the proponent must pick out a suitable
object b and defend 'Gb'. If he is unable to do so, he has lost (which
does not mean, of course, that the opponent has proven the falsity of
'(Ex)Gx').
The game starts from the entire sentence and proceeds gradually to
the elementary sentences of the type 'Fa'. In order to win, one seeks
an object which could falsify a universally quantified sentence, or ve
rify an existentially quantified sentence. If one has been able to find
or, perhaps, to produce such an object, one has won the game. If this
aspect of dialogue-games is emphasised, they could be called 'games of
seeking and finding'. The dialogical introduction of the predicates,
junctors, and quantifiers serves at the same time the programme of
re constructing the scientific language (cf. 2.5.).
A complex sentence is empirically true if the proponent always
wins, that is, if he is able to defend it against any opponent, by de
fending those constituent, elementary sentences which he has asserted
in the performance of any game connected with the complex sentence. If
any opponent is able to win, the sentence is empirically false.
A complex sentence is logically true , if the proponent has to de
fend an elementary sentence which has been previously defended by the
opponent. In such a case, the opponent has been forced to defend and
to attack one and the same sentence, which means that denying the
complex sentence has led to a contradiction. The contradiction
means that one and the same sentence has been attacked and de
fended, whatever this sentence is, and therefore it is permissible to
consider merely sentence-formulae, instead of sentences. Finally, logi
cal implication is defined in such a way that a formula 'B' is said to
be implied by the formulae 'A1'...'An ', if 'A1 '... 'An' are asserted
by the opponent, and the proponent is able to defend 'B' so as to force
the opponent into a contradiction. - The dialogical approach has also
been extended to modal logic (Lorenzen 1969a, chap.6).
In sum, the truth, whether empirical or logical, is here defined
as the defensi bili ty of a sentence or a formula against any opposition,
which is equivalent to the existence of a winning strategy connected
with the sentence or the formula.
Since the mid-sixties, Hintikka has also been developing a game-
theoretical conception of logic (cf. Hintikka 1973). He admits the for
mal similarities between Lorenzen's and Lorenz' conception and his own.
However, in his opinion, the 'absolutely cruciar difference between the
two consists in the fact that Lorenzen's and Lorenz' games are purely
formal, 'indoor' games, whereas his own 'outdoor' games are played in
constant reference to the external reality (Hintikka 1973:80-81; also
pp. 108-09). This is a curious misunderstanding. As we have seen, in
the di alogical conception empirical truth is ascertained, ultimately,
by finding out whether or not certain objects possess certain properties
and relations. For instance, the opponent wins the game connected with
the sentence "For all x, if χ is an atheist, χ is stupid or wicked",
because he is able to find an object, i.e., Bertrand Russell, who, al
though an atheist, is not wicked (nor stupid)(Lorenzen 1969a: 26-27).
Similarly, the proponent wins the game connected with the sentence "In
all Bavarian lakes there are fishes", only if he can find an object
with the property 'fish' in Tegernsee, which is a Bavarian lake suggest
ed by the opponent (op. cit., p.31). More generally, Kamlah & Lorenzen
(1967) devote an entire chapter to the question of how the truth-value
of an elementary sentence can be determined:
Wer über Einzelgegenstände - über Personen, Dinge, Ereignisse -

wahre Aussagen machen will, der muss sich in jedem Fall und in der
jeweils geeigneten Weise Zugang zu diesen Gegenständen verschaffen,
sei es durch Beobachtungen, durch Experimente«, durch Befragung von
Zeugen, Interpretation von Texten und so fort (op. c i t . , ρ.124).
Consequently, the f o l l o w i n g c h a r a c t e r i s a t i o n of H i n t i k k a ' s (1973:81)
'games of seeking and f i n d i n g ' applies with equal force to d i a l o g i c a l
games:
They are not 'indoor games'; they are 'played' in the wide world
among whatever objects our statements speak about. An essential
part of ail these games consists in trying to find individuals
which satisfy certain requirements.
The dialogical conception of logic is a stylised reconstruction of

that institution which is constituted by the rules of making valid in
ferences and, secondarily, of formulating necessarily true sentences.
There is no method by which the adequacy of this reconstruction could
be definitively established. However, it is possible to present intui
tives conceptual, and experimental, evidence for it (cf. the beginning
of this section).
Logical knowledge is acquired through participation in the above-

mentioned institution. In addition to reconstructing the situation of
the intersubjective use of logic, there are other ways of presenting
logical knowledge in a systematic form, for instance the axiomatic
method. This method will be described in 10.0. (below), where I discuss
axiomatisations of deontic propositional logic in relation to generative
grammars. It is important to notice that, unlike the dialogical method,
the axiomatic method does not even raise the question as to the origin,
or the basis , of logical knowledge.
To a still higher degree than philosophy of science, logic is a
prescriptive science. It starts by formalising those rules of inference
which are used in ordinary language, but it is also constantly engaged
in devising more precise ways of speaking and in extending this preci
sion to new domains (for a further discussion, see Itkonen 1976a).
The general idea of logic which I have outlined here is closely
similar to that contained in Wittgensteinian philosophy, In Winch's
(1958:100) formulation, "criteria of logic are not a direct gift of
God, but arise out of, and are only intelligible in the context of,
ways of living or modes of social life". Here again, however, Husserl
was the first to clearly conceive and formulate the issue (cf. 2.5.),
as can be seen from the following lengthy quotation:
Wir können dafür auch sagen: die vermeintlich völlig eigenständige
Logik, welche die modernen Logistiker - sogar unter dem Titel einer
wahrhaft wissenschaftlichen Philosophie - glauben ausbilden zu kön
nen, nämlich als die universale apriorische Fundamentalwissenschaft
für alle objektiven Wissenschaften, ist nichts anderes als eine
Naivität. Ihre Evidenz entbehrt der wissenschaftlichen Begründung
aus dem universalen lebensweltlichen Apriori, das sie beständig,
in Form wissenschaftlich nie universal formulierter, nie auf wesens
wissenschaftliche Allgemeinheit gebrachter Selbstverständlichkeiten,
immerzu voraussetzt. Erst wenn einmal diese radikale Grundwissen
schaft da ist, kann jene Logik selbst zur Wissenschaft werden. Vor
her schwebt sie grundlos in der Luft und ist, wie bisher, so sehr
naiv, dass sie nicht einmal der Aufgabe inne geworden ist, welche
jeder objektiven Logik, jeder apriorischen Wissenschaft gewöhnlich
en Sinnes anhaftet: nämlich zu erforschen, wie sie selbst zu be
gründen sei, also nicht mehr 'logisch', sondern durch Rückleitung
auf das universale vor-logische Apriori, aus dem alles Logische,
der Gesamtbau einer objektiven Theorie, nach allen ihren methodo
logischen Formen, seinen rechtmässigen Sinn ausweist, durch wel-
chen also alle Logik selbst erst zu normieren ist (Husserl 1954:
144) .
Of course, it is neither possible nor meaningful to discuss the

problem of the basis, or the justification, of formal methods in all
those contexts where such methods are used. But to ignore even the
existence of this problem, is indeed 'naive'. It is a curious fact,
but a fact nevertheless, that formalists are often the last to think
about the meaning and the genuine applicablity of their formalisations.
Positivism and formalism are, essentially, simplistic philosophies.
They are based on the a priori assumption that all entities that can
be ('scientifically') spoken about, must be of one and the same type.
However, truth is more complex.
2.7. Concluding Remarks
My general thesis is that synchronic autonomous linguistics, or

'grammar', is methodologically similar to such hermeneutic sciences as
logic and philosophy. This claim will be justified in detail in 10.0.
and 11.0. (below). However, both logic and philosophy contain a pre
scriptive component, which seems to be absent from grammar as well as
from other forms of linguistics (yet cf. 6.3.). Consequently, of all
the sciences discussed in this chapter it is sociology of knowledge
that comes closest to grammar. The two can be characterised as de
scriptive normative sciences: they investigate institutions, rather than
actual (institutional) behaviour, from the descriptive point of view.
Psychology and sociology have, of course, their linguistic counterparts
in psycholinguistics and sociolinguistics (yet cf. n.13).
My main disagreement with traditional hermeneutics concerns the
status of logic: it is generally thought that both natural science and
formal logic are equally opposed to the concept of hermeneutic science.
For me, however, formal logic is a full-fledged hermeneutic science.
3.0. 20TH-CENTURY LINGUISTIC THEORIES: A BRIEF SURVEY
The m e t a s c i e n t i f i c self-understanding of modern l i n g u i s t i c s is a l -
most e x c l u s i v e l y p o s i t i v i s t i c . In what f o l l o w s , I s h a l l t r y to document
this claim. As a r u l e , l i n g u i s t s have made no d e t a i l e d statements of
their metascientific positions. Therefore d i f f e r e n t schools of l i n g u i s -
t i c s w i l l be c h a r a c t e r i s e d , from the m e t a s c i e n t i f i c p o i n t of view, only
on the basis of t h e i r g e n e r a l , i n e x p l i c i t a t t i t u d e towards the question
of whether l i n g u i s t i c s c o n s t i t u t e s a natural science among o t h e r s , or
whether i t has, i n common w i t h other human sciences, c e r t a i n characte-
r i s t i c s which d i s t i n g u i s h i t from t y p i c a l natural sciences.
3.1. Saussure
Saussure's philosophy of science as expressed in the Cours de

linguistique générale (CLG in the following) is not easy to expound.
He emphasises throughout the conventional nature of language, which
could be taken to imply a difference between linguistics and standard
natural sciences. However, he also points out that, as a sub-area of
semiotics {sémiologie), linguistics is a part of social psychology,
and hence ultimately of general psychology (CLG 33).20 Now it is clear
that the unique character of conventional (or normative) data is bound
to disappear if they are introduced together with all other types of
psychological data, namely as data to be investigated by the predomi
nantly experimental methods of general psychology. Saussure seems to
realise this because he emphasises that language must be clearly dis
tinguished from faculté de langage: The former is a social entity (cf.
below), while the latter is a quasi-natural entity, and they must be
investigated, respectively, by linguistics and psychology, conceived
as autonomous sciences (CLG 25-27, 34).
It is questionable whether Saussure quite succeeds in his attempt

to clearly delimit linguistics against psychology. On the other hand,
he apparently believes that he has discovered genuine methodological
differences between linguistics and certain other human sciences (in
cluding even jurisprudence and political history). However, these
alleged differences are fictitious. It could be pointed out first that
language is clearly not the only object which can be studied both syn-
chronically and diachronically (CLG 114), - this is so, even if for the
sake of argument we retain Saussure's unsatisfactory view that diachrony
deals exclusively with particular elements; and secondly, on the assump
tion that 'panchronic laws' of linguistics are comparable to regulari
ties in nature, it does not make sense to claim that such laws exist,
but are independent of any concrete facts (CLG 134-35).
Of even greater interest is the need of clarification of Saussure's
account of the social nature of language (langue). To begin with, he
characterises langue, which he views as the only genuine subject matter
of linguistics, as a social institution:
La langue est la partie sociale du langage, extérieure à l'indi
vidu, qui à lui seul ne peut ni la créer ni la modifier; elle
n'existe qu'en vertu d'une sorte de contrat passé entre les membres
de la communauté (CLG 31). Nous venons de voir que la langue est
une institution sociale (CLG 33).
Langue d i f f e r s from other social i n s t i t u t i o n s p r e c i s e l y i n t h a t i t is

a system of communication, a system of signs. By v i r t u e of t h i s , it
c o n s t i t u t e s i t s e l f as the central p a r t ofsemiotics or "une science
qui étudie la vie des signes au sein de l a vie s o c i a l e " (CLG 33). It
is important to note t h a t , although langue is conceived of as a social
phenomenon, the only access to i t is provided by the individual consci-
ousness of the speakers, namely, by what we would today c a l l ' t h e native
speaker's l i n g u i s t i c intuition':
La synchronie ne connaît qu'une perspective, celle des sujets par

lants, et toute sa méthode consiste à recuillir leur témoignage;
pour savoir dans quelle mesure une chose est une réalité, il faudra
et il suffira de rechercher dans quelle mesure elle existe pour
la conscience des sujets (CLG 128).
57
20-TH CENTURY LINGUISTIC THEORIES
With his langue - parole dichotomy Saussure formulates, but does
not s o l v e , the problem of the social vs. i n d i v i d u a l , or o b j e c t i v e vs.
s u b j e c t i v e modes of existence of language: "Par la p a r o l e , on désigne
l'acte de l ' i n d i v i d u r é a l i s a n t sa f a c u l t é au moyen de la convention
s o c i a l e , qui est la 1 angue" (Saussure 1957:10). The question as to the
precise nature of langue and, by i m p l i c a t i o n , of l i n g u i s t i c s may be
reduced to the f o l l o w i n g questions: How does a convention or r u l e
e x i s t (= o n t o l o g y ) , and how is i t known (=epistemology)? I s h a l l pre-
sent my answer in 5.0. (below). Modern t h e o r e t i c a l l i n g u i s t i c s mis-
understands or ignores these questions, equating ' s u b j e c t i v e ' w i t h i n -
t u i t i o n and ' o b j e c t i v e ' w i t h observation, and f a i l i n g to see t h a t , taken
i n i t s e l f , observation too is a wholly i n d i v i d u a l and s u b j e c t iv e process.
Within a conceptual framework which defines the notions 'subjective'
and ' o b j e c t i v e ' in the way i n d i c a t e d , the notion ' s o c i a l ' must remain
u t t e r l y incomprehensible. I t is no accident, then, t h a t w i t h i n TG, f o r
i n s t a n c e , the s o c i a l and normative nature of language simply disappears
from the view ( c f . 7.5. below).
What Saussure has to say, i n g e n e r a l , about language being a fait
social is in i t s e l f c o r r e c t but not very r e v e a l i n g . When he attempts
to answer the question of the nature of the (synchronic) ' l a w s ' of
language, however, his account becomes confused. Given t h a t langue is
a social i n s t i t u t i o n , i t would be natural to t h i n k t h a t ' l a w s ' of langu-
age are analogous to constituents of i n s t i t u t i o n s , i . e . , r u l e s . Saussure
c o r r e c t l y notes t h a t " t o u t e l o i s o c i a l e a deux caractères fondamentaux:
e l l e est impévative e t e l l e est générale; e l l e s'impose, e t e l l e s ' é t e n d
à tous les cas, dans certaines l i m i t e s de temps e t de l i e u , bien enten-
du" (CLG 130); notice t h a t loi sociale is here i d e n t i c a l w i t h 'rule',
and not w i t h ' r e g u l a r i t y ' , in the sense to be explained i n 6 . 0 . However,
he continues by endeavouring to d i s t i n g u i s h l i n g u i s t i c rules from social
ones, on the basis of the alleged f a c t t h a t although the former are ge-
n e r a l , they are not imperative. Contrary to what one might expect, t h i s
statement does notaim at the equation o f l i n g u i s t i c rules w i t h natural
r e g u l a r i t i e s , as Saussure admits t h a t "sans doute [la l o i synchronique]
s'impose aux i n d i v i d u s par la c o n t r a i n t e de l'usage c o l l e c t i f " (CLG 131;
see also p.104). What he means by the non-imperative character of
l i n g u i s t i c rules i s , r a t h e r , that language does not possess any force
that could guarantee the maintenance of a given r u l e . But this is of
course a universal truth about all social r u l e s , not j u s t l i n g u i s t i c
ones, as was in f a c t implicitly admitted by Saussure himself when he
stated that social rules are valid within certain limits of time and
place (cf. the quotation from CLG 130 above).
When we elucidate Saussure's account of this point, i t is apparent
that in his theory l i n g u i s t i c rules are de facto analogous to social
r u l e s ; this is in perfect agreement with his view that language is a
social i n s t i t u t i o n ; and social rules are to be distinguished from re-
g u l a r i t i e s in nature, since the l a t t e r are not ' i m p e r a t i v e ' , i . e . , nor-
mative. I t may be added that on this i s s u e , crucial from the methodo-
logical standpoint, Saussure seems to have been misled by the errors
in his own analysis referred to above. At l e a s t , he has not drawn any
e x p l i c i t methodological consequences from the difference between rules
and r e g u l a r i t i e s .
As can be seen from the preceding argument, Saussure's methodolo-
gical statements which go beyond the description of language as a sys-
tem of signs are mere hints and suggestions. Therefore I cannot but
agree with Koerner's (1973:53) following assessment:
...the social nature of language, langue as a fait social, etc.

do not constitute, anywhere in the whole of the Cours, an integral
part of Saussure's theory...
Since Saussure wished to make linguistics a science in its own

right and with a frame of reference of its own, sociological ex
planations of linguistic behaviour were of only secondary or even
tertiary importance to him (op. c i t . , p. 58).
Saussure was interested, above all, in delineating the subject

matter of linguistics as an autonomous science. Within synchrony, this
subject matter coincided for him with that of traditional grammar. As
his famous comparison of language with the game of chess demonstrates,
linguistics was to describe a system of entities whose relations to
each other were determined by conventions or rules. Hence, language
was defined as a game or i n s t i t u t i o n , but - as we have seen - Saussure
59
was s a t i s f i e d w i t h t h i s r e s u l t and d i d not s e r i o u s l y i n q u i r e i n t o the
o n t o l o g i c a ! and epistemological nature of the 'language-game' sodefined.
Nor did he i n q u i r e i n t o i t s psychological and s o c i o l o g i c a l substrata,
although he was f u l l y aware of t h e i r existence. He c l e a r l y considered
a l l these questions as l y i n g outside of l i n g u i s t i c s proper.
3.2. Hjelmslev
Above ( 3 . 1 . ) , Saussure was located somewhere between p o s i t i v i s m

and hermeneutics. Hjelmslev, the founder of ' g l o s s e m a t i c s ' , takes i n
turn a p o s i t i o n at the i n t e r s e c t i o n of Saussure and p o s i t i v i s m . He ex-
p l i c i t l y takes the Saussurean conception of language as his s t a r t i n g -
p o i n t , and elaboratesand r e f i n e s several Saussurean themes (suchas 'ex-
pression-content' and 'form-substance' d i s t i n c t i o n s , the i n t e r p l a y of
syntagmatic and paradigmatic r e l a t i o n s , and language as a system of va-
l u e s ) , which have been a l l but ignored by other schools of s t r u c t u r a l
linguistics. S i m i l a r l y , he places natural languages w i t h i n the context
of general s e m i o t i c s , which extends "from the study of l i t e r a t u r e , art,
and music, and general h i s t o r y , a l l the way to l o g i s t i c s and mathema-
21
tics"(PTL 108). Games, understood as " a b s t r a c t transformation systems",
are the l i m i t i n g case of semiotic systems. The t e s t f o r deciding whether
or not a given system i s a ' s e m i o t i c ' consists i n the discovery of
whether the two planes of expression(-form) and content(-form) can or
cannot be demonstrated as having the same s t r u c t u r e throughout, w i t h a
one-to-one r e l a t i o n between the e n t i t i e s of the one plane and the e n t i -
t i e s of the other (PTL 112). In the case of a genuine sign-system the
two s t r u c t u r e s are not isomorphic. I f , f o r i n s t a n c e , we consider the
expression-form /man/, and the corresponding content-form 'man', i t is
clear t h a t they are both f u r t h e r analysable, i . e . , the former is ana-
lysable f i r s t i n t o phonemes, and then i n t o d i s t i n c t i v e f e a t u r e s , whereas
the l a t t e r i s analysable i n t o meaning components (or 'sematic m a r k e r s ' ) .
However, no one-to-one r e l a t i o n e x i s t s between the unanalysable e n t i t i e s
of the two planes.
I t would appear t h a t the preceding c h a r a c t e r i s a t i o n of the subject
matter of l i n g u i s t i c s suggests a n o n - p o s i t i v i s t i c i n t e r p r e t a t i o n of
glossematics. Nevertheless, Hjelmslev's sketchy account of the metho
dology to be applied in the description of natural languages (and other
semiotic systems) thus characterised hardly supports this conclusion.
First of all, Hjelmslev contrasts his own approach to language with
what he regards as the standard methodology of humanities, or human
sciences. According to this conception, human, as opposed to natural
phenomena, are non-recurrent, and consequently cannot be subjected to
exact and generalising treatment; this implies that the only possible
method utilisable by human sciences is mere description of particular
facts and events, that is to say, a method which is nearer to poetry
than to exact science (PTL 8-10). It is precisely my thesis that a
dichotomy exists between natural sciences and human sciences, and hence
between positivism and hermeneutics, The above-mentioned form of such
a dichotomy, however, is quite obviously false. For example, when we
study a human institution, we have to make reference to (possible) par
ticular actions that are in conformity with the rules of this institu
tion, that is, to particular correct actions; naturally enough, this
does not imply that we are as a result merely giving a ('poetic') de
scription of these particular actions for their own sake. In other
words, a given analysis of rule-governed intentional behaviour is always
meant to be general in the sense that, as it stands, it automatically
applies to, and defines, an indefinite number of correct actions. While
the existence of a rule guarantees at least the possibility of the re
currence of correct actions instantiating this rule, it is true, on the
other hand, that an action that does not fall within the category of
rule-governed behaviour possesses no similar guarantee of recurrence.
But from this it does not follow, of course, that such an action is non
recurrent , and that consequently its description possesses no generality.
Actions are identified by the intentions 'behind' them, and there are no
logical obstacles, although there may sometimes be considerable factual
obstacles, which prevent the recurrence of actions with identical inten
tions. Moreover, even if a given action is de facto non-recurrent, the
method of analysing it, viz. the so-called 'practical syllogism' is of
course meant to be generally applicable (cf. von Wright 1971; Stoutland
1976).
20-TH CENTURY LINGUISTIC THEORIES 61
In any event, Hjelmslev, with a view to avoiding the alleged weak

nesses of standard human sciences, undertakes the construction of a uni
versal linguistic theory on the basis of the model of the hypothetico-
deductive theories of more advanced (i.e., natural) sciences. He further
distinguishes between the arbitrary and the appropriate aspects of his
theory. The former expresses the fact that a fully explicit scientific
theory can be viewed as an uninterpreted deductive system which is in
dependent of experience, whereas the latter expresses the obvious fact
that a theory must be able to account for empirical data. The theory
has to fulfil the following 'empirical principle': "The description
shall be free of contradiction (self-consistent), exhaustive, and as
simple as possible" (PTL 11). Different self-consistent and exhaustive
descriptions derived from the same theory are evaluated on the basis of
their simplicity, whereas different theories are evaluated on the basis
of the extent to which they approximate to the ideal formulated in the
'empirical principle' (in whatever way this is to be decided in practice).
By virtue of its calculative character, the universal theory "must be of
use for describing and predicting not only any possible text composed in
a certain language, but ... any possible text composed in any language
whatsoever" (PTL 17). 22
Hjelmslev's notions of explanation and testing remain vague, as is
to be expected, given that at the time he wrote his Prolegomena, precise
explications of these notions were only in the process of emergence. In
any event, a universal linguistic theory of the glossematic type would
apparently be a kind of general typology of natural languages: these
would be described and classified on the basis of the various (combina
tions of) dependences which they contain at the hierarchically-ordered
levels of the expression-plane and the content-plane, both in the para
digmatic and in the syntagmatic dimensions. Uldall (1957) has tried to
apply greater precision to this 'algebra of language', but his account
of the 'glossematic form' is no more than an application of the familiar
calculi of classes and first-order functions.
3.3. Sapir
Unlike other representatives of American structural (including trans-
formational) linguistics, Sapir is a methodologist who has integrated
linguistics into the wider framework of the social and human sciences,
as distinct from the natural sciences. Sapir's conception of society,
and of the role played in it by language, bears a striking similarity
to that of Mead, Schutz, and Winch, and I would not hesitate to call
him a de facto hermeneutician (as this word is understood here). Of ne
cessity, the foregoing calls into question TG's reinterpretation of his
tory, according to which Sapir (in much the same way as Humboldt and
23
Descartes) was f i r s t and foremost a forerunner of TG. TG a r r i v e s at
t h i s i n t e r p r e t a t i o n of Sapir by concentration upon a s i n g l e aspect of
his work, i . e . , phonology. I t is true t h a t both S a p i r ' s phonology and
the TG phonology are d i f f e r e n t f r o m , because more a b s t r a c t than, the
post-Bloomfieldian 'taxonomie' phonology. This merely negative charac-
t e r i s t i c , which also applies to Trubetzkoy ( c f . his concept of 'archi-
pnonerne') and to Hjelmslev ( c f . his concept of 'phoneme form') for in-
stance, does not by any means j u s t i f y the claim of a strong s i m i l a r i t y
between S a p i r ' s and TG's p o s i t i o n s . And as regards the general nature
of language, S a p i r ' s p o s i t i o n is not j u s t d i f f e r e n t from, but d i r e c t l y
opposite to t h a t of TG, as w i l l be seen below.
From the outset S a p i r , l i k e Saussure, draws a c l e a r l i n e between
l i n g u i s t i c and natural phenomena: "Language is p r i m a r i l y a c u l t u r a l
and social product and must be understood as such" (Sapir 1949f:166).
More p r e c i s e l y , language is comparable to a s o c i a l i n s t i t u t i o n ; insti-
t u t i o n a l behaviour is of course subject to psychological description
and e x p l a n a t i o n , but the primary task is to describe i t qua i n s t i t u t i o -
n a l , t h a t i s , to describe the i n s t i t u t i o n w i t h i n which i t is performed:
We can profitably discuss the intention, the form, and the histo
ry of speech, precisely as we discuss the nature of any other phase
of human culture - say art or religion - as an institutional or
cultural entity, leaving the organic and psychological mechanisms
back of it as something to be taken for granted (Sapir no date: 11) .
As institutional behaviour, speaking is characterised by.its inherent

meaningful ness, Sapir maintains:
Speech, like all elements of culture, demands conceptual selection,

inhibition of the randomness of instinctive behaviour. That its
'idea' is never realized as such in practice, its carriers being

instinctively animated organisms, is of course true of each and
every aspect of culture (op. c i t . , p. 46. n.2).
This quotation expresses succinctly the notion that any adequate

theory of language, as a part of a theory of cultural behaviour, must
be able to account both for the similarities and the differences between
intended action and performed action, for instance, between intended
sentence and uttered sentence. As the object of grammatical description,
the 'idea' of speech has priority over actual speech. This is justa re
formulation of Saussure's claim of the priority of 'langue' over 'paro-
25
le' as the subject matter of linguistics. In other words, in describing
language it is necessary to go beyond 'sense data', or the purely physi
cal side of language (Sapir 1949b:45). The contention that "no entity
in human experience can be adequately defined as the mechanical sum or
product of its physical properties" (Sapir 1949c:46), does in fact
neatly sum up two constant themes in Sapir's work: First, under whatever
aspect it is viewed, human experience is always organised into 'con
figurations', i.e., structures; secondly, intentional or meaningful en
tities are not reducible to physical ones, and therefore - since meaning
cannot be observed, but only understood - a distinction has to be made
between observation and understanding. (This is of course the axiom of
hermeneutics.) From the nature of language, as characterised above,Sapir
(1949f:166) draws the conclusion that
better than any other social science, linguistics shows by its data
and methods, necessarily more easily defined than the data and me
thods of any other type of discipline dealing with socialized be
haviour, the possibility of a truly scientific study of society
which does not ape the methods nor attempt to adopt unrevised the
concepts of the natural sciences.
Observation shows us movements and sounds connected with human bo

dies, but it cannot in itself tell us what the people in question are
doing or whether, in fact, they are doing anything. To decide this
question, we must be able to arrive at the meaning of this (observable)
behaviour, but we can only achieve this through understanding. Since,
as Mead among others has repeatedly emphasised, meaning exists only in
a social context, understanding is a socially acquired process. Sapir
(1949i:546-47) thus comes to the conclusion, later to be defended in
detail by Wittgenstein and Winch, that all intentional behaviour must
be of a social nature:
It is impossible to say what an individual is doing unless we have

tacitly accepted the essentially arbitrary modes of interpretation
that social tradition is constantly suggesting to us from the very
moment of our birth. Let anyone who doubts this try the experiment
of making the painstaking report of the actions of a group of nati
ves en gaged in some form of activity, say religious, to which he
has not the cultural key. If he i s askillful writer, he may succeed
in giving a picturesque account of what he sees and hears, or thinks
he sees and hears, but the chances of his being able to give a re
lation of what happens in terms that would be intelligible and
acceptable to the natives themselves are practically nil.
This account is d i r e c t l y r e l e v a n t to our discussion ( i n 2 . 2 . above)

of how the data of sociology are acquired i n the f i r s t place. Under-
standing is a two-way process: i t is not only the case t h a t the s c i e n -
t i s t must understand the behaviour of the people described but - p r e c i s e -
l y to check whether or not he has in f a c t understood them - the people
described must also be able to understand and accept at l e a s t the non-
t h e o r e t i c a l p a r t of the d e s c r i p t i o n . To use Apel's (1973b) terminology,
the objects of a s o c i o l o g i c a l or s o c i a l - a n t h r o p o l o g i c a l investigation
must be p o t e n t i a l ' c o - s u b j e c t s ' of the social scientist.
The i n d i v i d u a l gradually acquires the key to an a l i e n c u l t u r e by
s t a r t i n g to i m i t a t e o v e r t behaviour, and, as Sapir (1949d:105-06) put
it, " i n the process of f a l l i n g in w i t h the ways of society one i n effect
acquiesces in the meanings t h a t inhere i n these ways".
From the c i t a t i o n s given so f a r , i t is clear t h a t Sapir i s a c t u a l l y
o u t l i n i n g a f u l l y - f l e d g e d philosophy of human sciences. But he is not
content w i t h the mere announcement o f methodological g u i d e l i n e s , he also
gives exemplary hermeneutic analyses of several s p e c i f i c a l l y human phe-
nomena which are simply incompatible w i t h any kind of p o s i t i v i s t i c de-
scription. A case i n p o i n t is the analysis presented i n his "The Me<an-
27
ing of R e l i g i o n " . Where are the r u l e s of correspondence and the ope-
r a t i o n a l d e f i n i t i o n s which would connect in a f i x e d way the concepts,
e . g . , ' s p i r i t u a l s e r e n i t y ' , t h a t appear i n S a p i r ' s account w i t h p a r t i -
cular events in space and time? Since such concepts must be defined
afresh for every culture, the conceptual differences between different
cultures should themselves be defined by means of culture-independent
rules of correspondence, if this positivistic programme of interpreting
Sapir's description in terms of space and time is to be carried out. I
claim that there is no way to operational ise a concept like 'spiritual
serenity in the Ojibwa culture', still less a concept like 'Ojibwa
culture' itself. So long as no correspondence rules are proposed, even
tentatively, to prove the opposite, I am entitled to disregard all ob
jections to my claim, or more generally to the claim that cultural con
cepts are qualitatively different from concepts such as 'atom' and can
not be treated according to the canons of natural science. Consequently,
in "The Meaning of Religion" Sapir is not making an empirical1 descrip
tion. Nor - as far as I can see - is he making an empi rical2 descrip
tion: because he is not speaking about, or even clearly implying, any
specific types of action, the occurrence, or lack of occurrence, of par
ticular actions cannot falsify his description in any straightforward
way. I think most positivists would agree with me that the description
in question is nonempirical. From this they would infer that it is non-
scientific as well. I would disagree with such a conclusion. Sapir's
description is clearly testable: it would be false precisely in case re
ligion is not what he claims it to be. There are objective methods for
deciding this question, but these methods are not as simple as the con
firmation and the falsification employed by empirical science. Nor is
there any reason why they should be as simple, apart from the aprioris-
tic requirement of the 'unity of science' (which, of course, is no ge
nuine reason). Consequently, I shall say that descriptions of cultural
or institutional phenomena are testable 3 . If they can be taken to pro
vide explanations, then they are to be characterised as explanatory3.
Sapir's analysis of various Amerindian religions is remarkably si
milar to Winch's (1964) interpretation of Zande magic. Both authors
reject the Western, excessively technological standards as a vantage
point from which to make intelligible and to evaluate (i.e., criticise)
the religious institutions of primitive societies. Rather, these in-
s t i t u t i o n s have to be understood w i t h i n t h e i r own conceptual universes,
even to the extent t h a t the p o s s i b i l i t y of a communication between d i f -
f e r e n t universes comes to be recognised as a genuine, i f not unsolvable
problem. Winch also acknowledges the importance of language i n forming
the Weltanschauung of a given s o c i e t y , and in f a c t , he u n w i t t i n g l y gives
a s u c c i n t formulation to the 'Sapir-Whorf hypothesis' by quoting W i t t -
g e n s t e i n ' s remark from the Tractatus ( 5 . 6 2 ) : "The l i m i t s of my language
mean the l i m i t s of my w o r l d . "
Furthermore, i t is not too much to say t h a t Sapir a n t i c i p a t e s some
of the most important i n s i g h t s of Lévi-Straussian s t r u c t u r a l i s m . In
discussing the p a t t e r n i n g of social u n i t s , Sapir (Ί949g:340-41) notes
t h a t i n p r i m i t i v e s o c i e t i e s f u n c t i o n a l groups are generally subsidiary
to k i n s h i p , t e r r i t o r i a l , and status groups. I t has turned out t h a t i n
e n t i r e l y unrelated s o c i e t i e s the organisations of such non-functional
groups e x h i b i t formal or s t r u c t u r a l s i m i l a r i t i e s to a s t r i k i n g degree;
t h i s n e a r - i d e n t i t y of form obviously poses a problem. Sapir prepares
the way f o r an answer by r e l a t i n g the above-mentioned formal s i m i l a r i
t i e s , and analogous s i m i l a r i t i e s i n other f i e l d s as w e l l , to "a c e r t a i n
innate s t r i v i n g f o r formal e l a b o r a t i o n and expression and to an uncon
scious p a t t e r n i n g of sets of r e l a t e d elements of experience" (Sapir
1949e:156). When Sapir advocates "a social psychology of form which has
hardly been more than adumbrated", he seems to be g i v i n g an accurate
c h a r a c t e r i s a t i o n of the basic o b j e c t i v e s of Lévi-Straussian s t r u c t u r a -
lism:
The true unraveling of the basic and largely unconscious concepts

or images that underlie social forms has hardly been begun. Hence
the anthropologist is in the curious position of dealing with im
pressive masses of material and with a great number of striking
homologies, not necessarily due to historical contact, that he is
quite certain have far-reaching significance, but the nature of
whose significance he is not prepared to state. Interpretative
anthropology is under a cloud, but the data of primitive society
need interpretation none the less. The historical explanations
now in vogue, often exceedingly dubious at best, are little more
than a clearing of the ground toward a social interpretation; they
are not the interpretation itself (Sapir 1949g:338-39).
At the most general level, Sapir criticises existent human scien-

67
ces by reason of their predilection to specialise in one single, dis
connected aspect of human or social behaviour, with man being ignored
as the ultimate, integral object of inquiry. Instead, he envisages "an
inclusive science of man, one that does the best it can to harbor the
value judgements of experiencing human beings within its own catholic
'universe of discourse'" (Sapir 1949j: 584). Similar attempts have been
made before. For instance, Dilthey (1924:139-240) outlined a 'descrip
tive and dissecting (zergliedernd) psychology', which, in contrast to
empirical, explanatory psychology, would consider man as the phocal
point of those multiple contexts of meaning which constitute his social-
28
historical environment as he himself experiences it. Sapir mentions
cultural anthropology and psychiatry as possible candidates for evolu
tion into such an 'inclusive science of man', but adds that to date
psychiatry has disqualified itself, by reason of its having, quite un
realistically, ignored the effects of the larger context of society
on the individual psyche. Thus Sapir (1949j: 588) notes:
For all practical purposes a too low income is at least as signi
ficant a datum in the causation of mental ill-health as a buried
Oedipus complex or sex trauma. Why should not the psychiatrist
be frank enough to call attention to the great evils of unemploy
ment or of lack of economic security?
Sapir concludes by making the proposal that psychiatry should re

commend a more equal distribution of wealth. As a psychiatrically-mind-
ed social scientist, Sapir wishes to grasp man in his totality, and in
the process he breaks the barriers of value-free science: not content
with mere description of social reality, he wishes also to change it.
Sapir's psychiatry is a critical science, with the emphasis laid on
the critique of the social environment (cf. 2.3. above). Or, if we con
sider his reference to cultural anthropology, we might say that he is
striving for a sociology of knowledge furnished with the critical di
mension (cf. 2.4.).
Sapir's theory of the human sciences could be criticised because
of its very generality. For instance, he fails to distinguish clearly
between normative and spatiotemporal aspects of institutions. Detailed
comparisons with the methods of natural science are also lacking. Just
as in the case of Saussure, however, anyone voicing such a criticism
would show a lack of historical perspective.
3.4. Bloomfield
Bloomfield has endorsed the positivistic philosophy of science in

a rather extreme form, as a combination of A. P. Weiss's 'physicalist'
psychology, and Carnap's and Neurath's 'logical empirism'. As a behaviou
rist he endeavours to account for meaning by resorting to the stimulus-
response model: meaning is the speaker's stimulus (i.e., physically or phy
siologically definable features of his total situation) and the hearer's
response, whereas an utterance is the speaker's (substitute) response
and the hearer's (substitute) stimulus (Bloomfield 1935:22-27). Interes
tingly enough, Bloomfield admits somewhat later (p. 145) that, from the
standpoint of linguistics, it is a necessary assumption that each lin
guistic form has a constant and definite meaning. However, as far as
their physical features are concerned, the speaker's situations may
differ in innumerable, uncontrollable ways, and may eventually involve
practically anything in the world. As a consequence. Bloomfield is
forced into the paradoxical conclusion that the necessary assumption
mentioned above cannot be confirmed in a single case:
Actually, however, our knowledge of the world in which we live is

so imperfect that we can rarely make accurate statements about the
meaning of a speech-form .... It is true that we are concerned
not so much with each individual as with the whole community. We
do not inquire into the minute nervous processes of a person who
utters, say, the word 'apple', but content ourselves rather with
determining that, by and large, for all the members of the commu
nity, the word 'apple' means a certain kind of fruit. However, as
soon as we try to deal accurately with this matter, we find that
the agreement of the community is far from perfect, and that every
person uses speech-forms in a unique way (Bloomfield 1935:74-75;
emphasis added),
And later on (p. 158) :
We assume that each linguistic form has a constant and definite

meaning, different from the meaning of any other linguistic form
in the same language ... We have seen that this assumption cannot
be verified ...
Bloomfield has borrowed his notion of 'accuracy' from the me-

69
thodology of natural science. That is, his approach is precisely the
opposite of the correct one: instead of trying to define meaning in
terms of physical features, he ought to have accepted meaning as given,
and to have used it as the criterion for decision as to which physical
features are relevant or distinctive.
The errors in his reasoning are instructive in many ways, not only
because they illustrate the inadequacy of the behaviouristic approach.
Rather, they should lead to question the appropriateness of the whole
notion of empirical confirmation in this context, and, by the same token
to the realisation that only the latter member of the dichotomy 'obser
vation - understanding' is adequate for the study of linguistic data.
Neither 'taxonomie' linguistics nor TG has ever been able to grasp the
nature and the implications of the above dichotomy in any systematic
way. The only area in which this dichotomy has received due attention
is that of the study of sounds, since the linguist cannot help drawing
a distinction between (physical) phonetics and (mental) phonology.
Similarly, when Bloomfield states his basic principle that "in
every speech-community some utterances are alike in form and meaning",
he admits that it is simply on the basis of "our everyday knowledge"
that we are capable of making the judgements in question (Bloomfield
1935:77-78). It is clear that everyday knowledge of language is iden
tical with the native speaker's linguistic intuition. However, intui
tion is not reducibleto observation, and linguistic schools which strive
to dispense with the distinction between intuitive and observational
knowledge are attempting to force linguistic data into the straitjacket
of methodological monism.
In his explicitly metascientific statements, Bloomfield subscribes
to 'physicalism' (which is a form of the 'strong reductionism' noted on
p.l):
... all scientifically meaningful statements must be translatable

into physical terms - that is, into statements about movements
which can be observed and described in coordinates of space and
time (Bloomfield 1936:90).
This standpoint is diametrically opposed to that held by Sapir.

Furthermore, Bloomfield is particularly interested in that part of the
physi cal i s ti c programme that is concerned with the translation of men
tal istic terminology into physical terms. It is evident that if such
entities as ideas, concepts, and thoughts need to be eliminated in
favour of something more tangible, then the most likely substitutes are
the words and sentences which have been thought to express these enti
ties. (Actually, this last sentence may serve to show how difficult
it is to rid oneself of mentalistic terminology.) Accordingly, in
Carnap (1937:233-37 and 284-92), sentences that speak about mental en
tities are treated as 'quasi-syntactical', that is, as sentences which
seem to be about the world, but are in fact about language. Equivalent-
ly, within the framework of Weissian psychology, mental entities are
reduced to language, viz. a 'noise' which the speaker makes, and which
"acts with a trigger-effect upon the nervous systems of his speech-
fellows" (Bloomfield 1936:93). In conformity with the physicalistic
programme, Bloomfield does in fact try to translate the 'material mode'
of sentences that contain expressions for mental entities into the 'for
mal mode' of sentences that contain expressions for suitable linguistic
entities; for example, he replaces 'idea' by 'definition' and 'what can
be conceived' by 'what the definition says'. In work of this kind, he
even perceives one of the major tasks of linguistics. (Fortunately, he
did not take his own advice quite literally.)
Positivist as he was, Bloomfield never identified the linguist's
task with describing a given corpus of utterances» On the contrary, he
was fully aware of the 'creative' nature of language:
... it is obvious that most speech-forms are regular, in the sense

that the speaker who knows the constituents and the grammatical
pattern, can utter them without ever having heard them; moreover,
the observer cannot hope to list them, since the possibilities of
combination are practically infinite (Bloomfield 1935:275).
Today when reading books on linguistics, one easily gets the impres
sion that (aside from Humboldt) TG has discovered the creativity of
language. To Bloomfield, however, this fact was Obvious' (cf. the
quotation above) and in no need of elaboration. TG's emphasis on crea
tivity rests on its use of recursive rules (which were not available in
Bloomfield's time). However, the emphasis on recursivity rests on mis-
understanding, because only the use of a quite restricted amount of re-

cursivity can be justified in natural language grammars (cf. Itkonen
1976a). This also means that Bloomfield is correct in emphasising, as
against TG, that the possibilities of combination are practically , but
not theoretically, infinite,
3.5. Harris
Harris develops the Bloomfieldian conception of l i n g u i s t i c s , but

f o r the most p a r t omits e x p l i c i t reference to the philosophy of science
t h a t underlies i t . In view of the general disparagement of 'taxonomic'
l i n g u i s t i c s by TG, i t is i n t e r e s t i n g to note t h a t , at the level of ge-
neral methodology, H a r r i s ' s view of grammar as a s c i e n t i f i c theory i s
c l o s e l y s i m i l a r to t h a t of Chomsky. As w i l l be gathered from a quota-
t i o n on p. 78, Chomsky considers a grammar as a theory t h a t predicts
utterances and explains them by s t a t i n g the s t r u c t u r e s of these p r e d i c t -
ed utterances on each l i n g u i s t i c l e v e l , and by showing t h a t these s t r u c -
tures conform to the rules of the grammar.
In H a r r i s ' s o p i n i o n , a grammar is concerned w i t h " r e g u l a r i t i e s in
selected aspects of human behaviour" (Harris 1961:22), more p r e c i s e l y
with " r e g u l a r i t i e s i n the d i s t r i b u t i o n a l r e l a t i o n s among c e r t a i n features
of speech" (op. cit., p.5). These can only be described by concentrating
on a ' c o r p u s ' , but "the i n t e r e s t i n our analysis of the corpus derives
p r i m a r i l y from the f a c t t h a t i t can serve as a predictive sample of the
language" (p.244; emphasis added). In other words, "when a l i n g u i s t
o f f e r s his r e s u l t s as a system representing the language as a whole, he
i s predicting t h a t the elements set up f o r his corpus w i l l s a t i s f y all
other b i t s of t a l k i n g in t h a t language" ( p . 1 7 ; eraphasis added). 29 Hence,
a grammar makes p r e d i c t i o n s and i s e i t h e r confirmed or disconfirmed (and
i f the l a t t e r , then modified) depending upon whether or not i t s predic-
t i o n s are matched by l i n g u i s t i c data. Consequently, H a r r i s - t y p e grammars
are t e s t a b l e p r e c i s e l y i n the same way as Chomsky-type grammars. Harris
i s i n f a c t constantly concerned w i t h the d i s c o n f i r m a b i l i t y of his d e s c r i p -
t i o n s , as i s c l e a r l y evident from the f o l l o w i n g passage:
It is ... often convenient to make the division into morphemic seg

ments first in the case of those utterances and parts of utterances
in which the difference in adequacy among various alternative seg
mentations is extreme. The less obvious choices of segmentation
can then be decided with the help of the classes of morphemic seg
ments which have already been set up. Even then, new data may lead
us to rescind some of our previous segmentations in favor of alter
native ones which pattern better with the new data (Harris : 163).30
Harris (p.369) arrives at (and/or describes) linguistic structures on the

basis of segmentation and classification: "As a result of these operations
we not only obtain initial elements, but are also able to define new
sets of elements as classes or combinations (sequences, etc.) of old
ones".
Utterances have hierarchical structures consisting of different
levels in such a way that "each element is identified relatively to the
other elements at its level, and in terms of particular elements at a
lower level" (p.370); "each stretch of speech in the corpus in now com
pletely and compactly identifiable in terms of the elements at any one
of the levels" (p.364). Since the Harris-type grammar is a predictive
theory of the linguistic structure, it 'explains' the predicted utter
ances in a manner quite similar to the Chomsky-type grammar, namely
by stating their structures at each linguistic level, and showing that
these structures conform to the rules of the grammar. To be sure, in
the Harris-type grammar, rules of grammar are mostly not explicitly
formulated, but are rather exhibited by the structures in the represen
tative sample, as they have been discovered on the basis of the standard
operations of segmentation and classification.
From the above, it follows that from TG's point of view, 'taxonomie'
grammars and transformational grammars must have the same metascientific
status, i.e., they are both to be seen as testable, explanatory theories.
(Notice that we are discussing grammars of particular languages, not
universal linguistic theories; for the latter, see 9.4.). However,
since - as we shall see later on - formal and systematic descriptions
of linguistic data do not provide genuine explanations (cf. 9.2.), in
this respect Harris's position actually seems to be more correct than
Chomsky's, given that the former plays down the 'explanatory' aspect of
grammar. In any event, it is definitely incorrect to say that post-
Bloomfieldian American linguistics was merely concerned with 'observa
tional adequacy', i.e., with presenting the observed primary data cor-
73
rectly. (Yet this distortion, as formulated in Chomsky (1964d:28-30
and 97-98) and elsewhere is today almost universally accepted as true.)
Harris (1957:142) explicitly points out that he is concerned with ma
king g e n e r a l i s a t i o n s ; this should in fact be evident to anyone, given
that the Harris-type grammar is a -predictive theory, as we have seen
above. Moreover, even when dealing with his data, Harris by no means
restricts his attention to merely observing whatever occurs in speech.
In morphemic analysis, for instance, he selects a definite frame of sub
stitution which serves as the basis of the (substitution) test by means
of which morphemes and morpheme sequences are grouped into (substitution)
classes (Harris 1957:143-44). It is clear that substitution frames do
not occur in speech qua substitution frames. The considerations which
lead to the selection of one substitution frame over others refer to the
over-all simplicity of the description; that is, they are of a purely
theoretical nature and may in no way be simply derived from observation:
The criterion which decides for -ing, and against un-, as the rele
vant environment in determining substitution classes [for verbs]
is therefore a criterion of usefulness throughout the grammar, a
configurational consideration (Harris:143, n.6)
And more explicitly:

This means that we change over from correlating each morpheme with
all its environments, to correlating selected environments (frames)
with all the morphemes that enter them. The variables are now the
positions, as is shown by the fact that the criterion for class
membership is substitution. The element which occurs in a given
class position may be a morpheme which occurs also in various other
class positions. We merely select positions in which many morphemes
occur, and in terms of which we get the most convenient total de-
scription (p.150; emphasis added,).
In this connection, mention should be made of one additional respect

in which the contrast between Harris and Chomsky is much less real than
it would appear at first sight. The former expressly attempts to con
struct discovery procedures for grammars, whereas for the latter, who
has adopted the methodological maxim that the context of discovery and.
the context of justification need to be strictly separated, the interest
in discovery procedures is one of the major flaws of 'taxonomie' lin-
guistics. 31 Instead, Chomsky recommends replacing discovery procedures
by 'evaluation procedures' which are used to determine which of two al
ternative grammars is the better one.
This seemingly clear contrast of opinions calls for two comments.
First, the principle of separating the contexts of discovery and justi
fication, is not unknown to Harris, but he interprets it in his own way,
viz. by (implicity) distinguishing between methods which might be called
'discovery in practice' and 'discovery in principle': The latter is the
theoretically justifiable reconstruction of a discovery procedure (one
might even say: the context of justification for discovery procedures),
whereas the former represents the way in which linguists actually pro
ceed. This distinction is clarified by the following passage, which
is characteristic of pre-transformational American linguistics:
In determining the morphemes of a particular language, linguists

use, in addition to distributional criteria, also (in varying de
grees) criteria of meaning difference. In exact descriptive lin
guistic work, however, such considerations of meaning can only be
used heuristicaliy, as a source of hints, and the determining cri
teria will always have to be stated in distributional terms (Harris
1961:365).
Secondly, it is generally agreed outside standard positivism that

all aspects of scientific discovery cannot be disregarded, because con
centration on the -post hoc justification (or 'rational reconstruction')
of well-established theories gives a wholly distorted, static picture of
science. A philosophy of science which neglects the question of the
growth of science, the central question for every practising scientist,
is inadequate almost by definition. Notice that even TG's 'evaluation
procedure' tries to capture one aspect of the growth of the linguist's
knowledge, viz. the selection of the better of two proposed grammars.
The interpenetration of the contexts of discovery and justification
can be illustrated more precisely as follows. Suppose that a transfor
mationalist is deliberating on whether the verbs of a given language
should be described by means of a single v-category, supplemented by
strict subcategorisation rules to be found in the dictionary, or with
three distinct categories V a, F b , Vc, along with corresponding subindices
assigned to every verb, or by a common category W for both verbs and
adjectives, supplemented by strict subcategorisation rules. It then
would become evident that the category, and ultimately the grammar,
which he discovers is the one which he can best justify on the basis of
considerations referring to the over-all simplicity of the description.
But this standard procedure of a transformationalist grammarian is pre
cisely the same as the one which Harris explicitly considers as his own
(cf. above). Hence, discovery is always implicit in justification, and
any absolute distinction between the two is just an artifact.
It may be added that the replacement of discovery procedures by less
demanding evaluation procedures has in practice been largely symbolic:
it may be true that it is impossible to devise exact and rigorous disco
very procedures for grammars; but no exact and rigorous evaluation pro
cedures for grammars have been devised, either. Nor should this be sur
prising, considering that procedures for purely formally determining
the superiority of one theory over another have so far not been devised
in any science, whether empirical or not.
To sum up: Harrisian grammars are explanatory and testable theories
with the same right as those advocated by TG. In keeping with the re
quirement of 'strong reductionism', however, Harris tries to make his
approach as observational as possible and to minimise the role of theo
ry. Like Bloomfield, Harris thus represents the strict positivistic
standpoint that characterised much of the philosophy of science in the
1930s. It would be unfair to criticise them for subscribing to the pre
dominant methodological trend of their own time. To be sure, alternative
methodologies were available already then.
3.6. Transformational Grammar
When TG is compared with earlier schools of linguistics, its most

original contribution is seen to lie in the domain of syntax. To
some the contrast seems pronounced enough to justify the talk of a
'Chomskyan revolution' in linguistics; in phonology, whether synchronic
or diachronic, and in semantics such claims are certainly less justified.
It has also been claimed that Chomsky revolutionised linguistics at
the metascientific level as well. It is with this claim that I shall
come to terms in the present section.
It is generally thought that TG's originality in metascience rests
on two closely connected factors, namely its theoveticol and m e n t a l i s t i o
character. First, let us consider more closely the t h e o r e t i c a l charac
ter of TG. From the beginning it has been claimed that TG, as no school
of linguistics before it, has been able to turn grammars into genuine
theories and hence to lift linguistics onto the level of exact natural
sciences. (That is, the adherents of TG have never bothered to make
the distinction between natural, i.e., empirical1, sciences and empiri
cal, i.e., empirical2, social sciences.) Lees (1957:377), reviewing Syn
tactic Structures, noted:
... Chomsky's book on syntactic structures is one of the first se
rious attempts on the part of a linguist to construct within the
tradition of scientific theory-construction a comprehensive theory
of language which may be understood in the same sense that a chemi
cal , biological theory is ordinarily understood by experts in those
fields.
It may seem strange to some linguists that a grammar can be consi

dered to be a theory of a particular language, and not just a re
ordering or abbreviation of a text. But when we consider the ge
nerality which must be required of a grammar ... we see that it is
analogous ... to a scientific theory embodying proposed laws of
nature (p.380).
The other most important result of Chomsky's theory of language is

his very strict axiomatization of linguistic theory. He has chosen
to take seriously the requirement that a grammar be not merely an
arbitrary reorganization of some corpus, but (in a specifiable
sense) a simplest machine which will generate all and only the gram
matical sentences of a language (p,381).32
As it stands, the claim of the uniquely theoretical character of

TG is clearly false. As we have seen in 3.5. (above), because of its
predictive character, a Harri sian grammar is not "just a reordering or
abbreviation of a text" or "an arbitrary (!) reorganization of some
corpus", as Lees would have it. Similarly, Harrisian 'substitution
classes', elicited on the basis of selected 'substitution frames', are
clearly theoretical concepts. Furthermore, both Harris and Chomsky
agree that the theoretical concepts must be constructed in accordance
with considerations referring to the over-all simplicity of the grammar:
Notice that simplicity is a systematic measure; the only ultimate
criterion in evaluation is the simplicity of the whole system. In
discussing particular cases, we can only indicate how one or an

other decision will affect the over-all complexity. Such valida
tion can only be tentative, since by simplifying one part of the
grammar we may complicate other parts (Chomsky 1957:55-56).
This is also Harris' position (cf. p. 73).
Below this level of abstraction, however, there is of course a

genuine difference between Harris and Chomsky, or between 'taxonomie'
linguistics and TG. Harris tries to minimise the number of theoreti
cal concepts and to keep them as concrete as possible, whereas Chomsky
accepts no similar constraints. This difference in methodology is inti
mately related to a difference in what is regarded as legitimate data
for grammatical descriptions: because Chomsky allows the free use of in
tuitive criteria, he is able to perceive in linguistic data a great num
ber of relations (e.g., relations between different sentence-types) that
are more abstract than anything accepted by 'taxonomists' as being ob
jectively given. The description of a wider range of phenomena quite
naturally necessitates the use of a greater number of theoretical con
cepts. However, no clear answer is given to the question of how, pre
cisely, the theoretical concepts are to be interpreted, i.e., whether
in terms of intuition or of observation.
The theories of 'taxonomie' linguistics are much less abstract than
those of TG. Bach (1965) sees these two positions in linguistics as
representative of two clearly distinct scientific traditions, namely
the 'Baconian' and the 'Keplerian'. In his opinion these two traditions
are exemplified, in the modern philosophy of science, by the logical
empirism of the 1930s and for instance by Popper (1965) and Hempel (1952),
respectively (Bach 1965:114 and 118, n.10). This means that at the le
vel of general methodology the 'revolution' brought about by TG amounts
to bringing the metascientific assumptions of linguistics up to date.
TG regards American 'taxonomie' linguistics as 'positivistic' be
cause of its allegiance to logical empirism (see, e.g., Chomsky 1968:2;
Postal 1968a:231-32). I have argued that the current Popper-Hempel-
Nagel conception of the philosophy of science, to which TG subscribes,
is merely an improvement of logical empirism and deserves, by virtue of
its adherence to methodological monism, to be equally subsumed under the
notion of 'positivism' (cf. p.l). Hence, TG explicitly espouses the
positivistic philosophy of science, 'positivism' being understood, here
and elsewhere, in the sense of 1.0. (above).
The supposed analogy between grammars and theories of natural scien
ce has been construed by Chomsky as follows:
A grammar of a particular language can be considered, in what seems
to me a perfectly good sense, to be a complete scientific theory
of a particular subject matter, and if given a precise enough form,
a formalized theory. Any interesting scientific theory will seek to
relate observable events by formulating general laws in terms of
hypothetical constructs, and providing a demonstration that certain
observable events follow as consequences of these laws. In a parti
cular grammar, the observable events are that such and such is an
utterance and the demonstration that this observable event is a con
sequence of the theory consists in stating the structure of this
'predicted utterance on each linguistic level, and showing that this
structure conforms to the grammatical rules or the laws of the the
ory (Chomsky 1975 [1955]:77; emphasis added).
Given the assumed symmetry between prediction and explanation (cf.

1.2.), the demonstration that a predicted utterance is a consequence
of the grammar amounts to an explanation of this utterance. - In Syn
tactic Structures (p.49) Chomsky repeats the above statement, and express
ly equates the grammatical concepts with theoretical concepts of physics.
A nearly identical passage recurs in Chomsky (1964a:223). The same insis
tence upon the strict similarity between linguistics and physics charac
terises Chomsky's most recent statements: "... our scientist S ... stu
dies language exactly as he studies physics, taking humans to be 'natu
ral objects'" (Chomsky 1976:183).
As an explanatory and predictive theory, a grammar must of course
be testable. The predictions which are used in testing grammars are
more precisely characterised by Bach (1964:176) as follows:
A grammar for a specific language forms the basis for a series of
predictions of the following sort: If Τ is a terminal string in
phonetic transcription derivable from a grammar G of a language L,
then Τ is in L. That is, the grammar generates only sentences of
the language and no nonsentences. Further, if T' is in L, then T'
is derivable from G; i.e., all the sentences of the language are
derivable from the grammar. Finally, the grammar must assign a
set of markers (structural descriptions) to each generated string.
The markers form the basis for other predictions about relations
between sentences, ambiguity, and so on.
From the preceding quotations it is clear that TG intends grammars

to be formalised empirical theories or, briefly, axiomatic theories (see
also Chomsky 1964b:576). This calls for a few comments. A generative
grammar is definable as an axiomatic system (see, e.g., Wall 1972:197-
212), but it cannot be an axiomatic theory (of a language L ) , for the
simple reason that, unlike the sentences constituting an axiomatic theo
ry, the rewriting rules constituting a generative grammar, e.g., S→NP
+ VP, do not admit of truth-value. Similarly, aside from the last lines
of derivations, the 'theorems' of a generative grammar are not sentences,
but just strings of symbols. Of course, what transformationalists have
always meant - although they have not been able to express it clearly -
is that the me tagrammar of a grammar is a genuine theory, i.e., a theory
which claims, truly or not, that such and such rewriting rules express
generalisations about L and generate all and only correct sentences of
L or, more realistically, generate relatively many correct and relative
ly few incorrect sentences of L. Thus Lieb's (1974) sharp rejection of
TG for instance is unfounded, as it stands, because it rests solely on
the trivial fact that expressions like S →NP + VP are not sentences.
In view of the fact that TG has brought to bear on linguistics the
'Kepierian' point of view, it is curious to note how few references can
actually be found in transformationalist literature to modern represen
tatives of the 'Keplerian' philosophy of science. There are passing
references to Popper, Hempel, and Nagel for instance in Chomsky (1964d:
98), Postal (1968a:295, n.7), and Katz & Bever (1974:57, n.26), but the
applicability of the methods of natural science to linguistic data is
always simply taken for granted. It is only outside of TG circles that
the relation of linguistics and philosophy of science has become the
object of serious study (cf. 3.7. below). The incessant use of such
terms as 'explanation', 'prediction', and 'empirical testability' does
nothing to lend methodological rigour to TG descriptions, because their
meanings either are never made clear or, if made clear, turn out to be
misconceived.
For instance, consider TG's notion of 'empirical'. In Chomsky &
Katz (1974) and Chomsky (1976) the following type of statement occurs
a dozen times: "the defining characteristic of empirical science: non-
trivial theories are underdetermined by data" (Chomsky & Katz 1974:350);
"nontrivial empirical theory is underdetermined by evidence" (Chomsky
1976:209). 'Underdetermined by data1 is identical with 'falsifiable',
which in turn is considered identical with 'noncircular' and'nontauto-
logical' (Chomsky 1976:173-74). It is by virtue of this characterisa
tion that "linguistics is an empirical science, not a branch of logic
and mathematics" (Chomsky & Katz 1974:354).
This notion of 'empirical' is plainly wrong, apparently because
Chomsky and Katz have less than adequate knowledge of such nonempi rical
sciences as formal logic. They seem to be led by the principle 'logic
is tautological'. However, such a formulation is quite unacceptable.
Deontic logicians, for instance, attempt to construct systems or theo
ries which would generate all intuitively valid or tautological formu
lae and no intuitively invalid or nontautological formulae. Such theo
ries can be, and are, falsified by contrary evidences which means that
they are underdetermined by data, or 'empirical' in Chomsky's and Katz's
sense (cf. Itkonen 1975b, and 10.0. below). Considering that linguistics
is concerned with correctness, as logic is with validity or tautologi-
calness, it would be just as meaningful to say that 'linguistics is cor
rect' as to say that 'logic is tautological'.
Consequently, if one wishes seriously to inquire into the empirical-
ness or nonempiricalness of linguisties, then use must be made of the de
finition of 'empirical' given in 1.1. (above). 'Falsifiability' is not
enough; what is needed is spatiotemporal falsifiabili ty (or more preci
sely, testability). For a grammar to be an empirical theory, it is re
quired, first of all, that it deals with spatiotemporal data. It is also
required that the basis for 'grammatical predictions' is conceptually
independent from what is predicted or, equivalently, that in 'grammat
ical explanations' the explanandum-event (or -fact) is conceptually inde
pendent from the antecedent conditions. In 9.0. (below) I hope to show
that, against the nearly universal belief, these requirements are not,
and cannot be, met by any type of grammar. This result presupposes of
course an explanation of the precise nature of the subject matter of
grammar.
TG's position on the last-mentioned issue seems clear enough: A
grammar is comparable to a theory of natural science precisely because
it deals only with spatiotemporal, observable phenomena. However, the
situation is in fact more confused. Chomsky's Syntactic Structures,
wherein linguistics is explicitly compared to physics, also contains
the following passage:
... we assume -intuitive knowledge of the grammatical sentences of

English and ask what sort of grammar will be able to do the job of
producing these in some effective and illuminating way. We thus
face a familiar task of explication of some intuitive concept - in
this case, the concept 'grammatical in English', and more generally,
the concept 'grammatical' (Chomsky 1957:13; emphasis added).
Here we have a genuine confusion which has v i t i a t e d TG's conception

of s c i e n t i f i c methodology from i t s i n c e p t i o n . TG has never been able to
grasp the d i s t i n c t i o n between e x p l i c a t i o n , i . e . , analysis of concepts
(or of conceptual knowledge), on the one hand, and empirical explanation
and p r e d i c t i o n of spatiotemporal data, on the other. Given t h a t the
former is the method used by philosophers and l o g i c i a n s while the l a t t e r
is the method used by empirical s c i e n t i s t s , i t follows t h a t TG has never
f e l t i t necessary to d i s t i n g u i s h between philosophy and l o g i c , on the
one hand, and empirical science, on the other. Therefore the TG-type
'methodology' contains i n d i s t i n c t l y each and every form of scientific
a c t i v i t y and, as such, i t i s not only u n i n f o r m a t i v e , but also i r r e p a r a b -
ly self-contradictory.
The conception of grammar t h a t I have been discussing so f a r r e -
presents the standpoint of 'autonomous l i n g u i s t i c s ' , not too d i f f e r e n t from
Saussure's p o s i t i o n . That i s , whether the subject matter of grammar i s
conceived as a set of observable events or as a body of conceptual know-
ledge o r , i n c o n s i s t e n t l y , both, i t is a s e l f - c o n t a i n e d , purely linguis-
t i c realm of phenomena. With the subsequent psychologisation of l i n g u i s -
t i c s by TG, e x p l i c i t l y formulated i n Aspects, the s i t u a t i o n becomes,
however, even more confused. This is also where the m e n t a l i s t i c charac-
t e r of TG comes i n t o p i c t u r e .
By its mentalistic subject matter TG understands two different

things, without making, however, almost any distinction between them:
on the one hand, conscious intuitive knowledge of language; on the other,
mechanisms of sentence perception and production that lie beyond the le
vel of consciousness. The confusion between these two domains has been
produced inevitably, first, by calling both of them 'knowledge' and,
secondly, by postulating somewhere between them a nonexistent middle
term called the 'ideal speaker's intuition', which seems to possess cha
racteristics of both. A confusion of the kind discussed here occurs in
Chomsky (1965:19) for instance, where the expressions 'tacit knowledge'
and 'the linguistic intuition of the native speaker' are used to refer
to one and the same thing.
In other words, we have here a confusion between the following two
types of entities: on the one hand, the concept 'correct sentence of a
language L', which is the object of conscious knowledge; on the other,
utterances of a language L, which are manifestations of unconscious 'know
ledge'. In the former case, 'knowledge' equals 'consciousness', while
in the latter, 'knowledge' is a hypothetical dispositional concept. The
elimination of any distinction between conscious and unconscious know
ledge is officially sanctioned by Chomsky's decision to let 'cognize'
stand for 'knowing' in both senses: "For psychology, the important
notion will be 'cognize', not 'know'" (Chomsky 1976:165). Denying the
relevance of consciousness agrees with Chomsky's conviction that linguists
study humans as "natural objects" (op. cit p.183), given that conscious
ness is the only thing which distinguishes humans from inanimate things.
Notice that Chomsky is here forced to make, in the classical positivist
tradition, an absolute break between' linguists and their objects of in
vestigation, since he surely would not deny that linguists ought to have
at least some conscious idea of what they are doing. The inconsistency
of this position hardly needs to be pointed out: Chomsky, for one, has
investigateci nothing but his own conscious knowledge of English, for in
stance his knowledge of the sentences "John is easy to please" and "John
is eager to please". But then it is simply false to say that he has been
investigating a natural object, because natural objects have no conscious
ness. 34
The nature of the TG-type mental ism becomes evident from a conside
ration of the 'mentalistic' (or 'rationalist') explanations that grammars
are thought to provide. The following passage provides a good starting
point:
The important point about mentalism here is that Hockett, like

Bloomfield, can apparently not accept that language is an abstract
object represented in a physical system, an object determining in
part the behaviour of the system (Postal 1968a:295).
The behaviour of a physical system is determined by causal f a c t o r s . Katz

(1966:182) i n f a c t argues t h a t the mental e n t i t i e s postulated by TG are
"causally e f f e c t i v e y e t o b s e r v a t i o n a l l y inaccessible phenomena", i n every
respect comparable to unobservable e n t i t i e s postulated by the p h y s i c i s t .
Elsewhere ( i n Katz 1967:75) he pleads s u c c i n t l y f o r a "causal conception
of mentalism". Now 'causal e x p l a n a t i o n ' is synonymous w i t h 'mechanistic
e x p l a n a t i o n ' (von Wright 1971:2). I t f o l l o w s , perhaps s u r p r i s i n g l y , that
TG-type m e n t a l i s t i c explanations are in f a c t mechanistic explanations.
Their sole requirement is t h a t they possess a c e r t a i n degree of a b s t r a c t -
ness. Thus, Postal (1968a:295, n.7) notes t h a t i t was the work of people
l i k e Carnap, Hempel, and S c h e f f l e r , i . e . , work i n s p i r e d by natural science,
which "completely destroyed the philosophic basis of antimental ism". This
remark shows t h a t , as properties of TG, 'theoretical' and ' m e n t a l i s t i c '
are literally identical. So much is evident also from a passage where
Chomsky i d e n t i f i e s "the general antipathy to theory" w i t h "the so-called
34a
' a n t i - m e n t a l ism' "(Chomsky 1964:70, n.8).
S i m i l a r l y , Lenneberg, who is one of the f i r m e s t supporters of TG
w i t h i n the f i e l d s of psychology and physiology, emphasises the import-
ance of mechanistic explanations i n a way which may seem paradoxical at
f i r s t glance:
The great achievement of contemporary psychology was the replacement

of mentalistic explanations by mechanistic ones and the simultaneous
insistence upon empirical testability of hypothesized laws (Lenne
berg 1964:600) .
However, he is using the term 'mentalism' in a sense in which most

people would probably use it and which is directly opposite to TG's sense
of 'mentalism'. In TG's sense, then, mentalism or rationalism in linguis-
tics means simply acceptance of the actual canons of natural science, in
stead of some simplified versions thereof:
A theory is empirical if it is about the empirical world, and as
such, comfirmable or disconfirmable on the basis of observation and
experimentation. Thus, Chomsky's rationalism is every bit as em
pirical as Bloomfield's empiricism (Katz &Bever 1974:75, n.25).
A m e n t a l i s t i c grammar is no longer a purely l i n g u i s t i c grammar in

the Saussure-Sapir-Bloomfield t r a d i t i o n . That i s , instead of describing
j u s t utterances or sentences or l i n g u i s t i c knowledge, i t t r i e s to uncover
the -psychological mechanisms determining l i n g u i s t i c behaviour:
In a good sense, the grammar proposed by the linguist is an expla

natory theory; it suggests an explanation for the fact that ... a
speaker of the language in question will perceive, interpret, form,
or use an utterance in certain ways and not in others (Chomsky 1968;
23) .
Perceiving, interpreting, forming, and using utterances are process

es that go on in space and time. Such processes are investigated in
terms of 'external observation' (Chomsky 1972:14-15, and 1973:111) and
experimentation, i.e., in terms of the methodology outlined in 1.0.
(above).
Thus we see that the TG-type mental ism, or 'rationalism', is a form
of positivistic psychology. Because it tries to convert all knowledge
into external, experimental and observational knowledge, and cannot even
conceive of a different type of knowledge (cf. Chomsky 1973:111), it is
theoretically untenable for reasons indicated in 2.1. This may not be
regarded as a serious defect because it is common to positivistic psy
chology in general and does not affect in any way the results of actual
experimental research (although it does affect the philosophical inter
pretation of such results). What is serious, however, is that TG wants
to be at the same time, and with the same data and methods, both a gram
matical and a psychological theory. On the one hand, TG is a grammat
ical, or linguistic, theory which claims to be based on observation and
experimentation: Intuitions take somehow the role of observable data
(Dougherty 1974:126 and 133); experiments in turn are supposedly simi
lar to those used in astronomy, i.e., they are "questions put to the
nature", and the "linguist testing his grammatical formulations poses

questions which can only be answered by appealing to observations"
(Dougherty 1974:130-31). On the other hand, TG is a psychological theo
ry which stubbornly refuses to make genuine experiments, although it
could do so, or at least refuses to let experimental results affect non-
experimental grammatical descriptions, i.e., descriptions resulting from
35
Dougherty-type pseudo-experiments. It follows that grammatical non-
experimental descriptions serve, circularly, as the only basis for con
firmation or falsification of claims about their own psychological real
ity, i.e., about how language is represented in the human mind, although
the only (independent) evidence that could possibly confirm or falsify
such claims is provided by 'external evidence', including experimental-
psychological research on linguistic behaviour.
Today this kind of criticism has become quite common. Its possible
impact on the work within TG will be discussed in the light of the work
by Fodor & Bever & Garrett (1974)(see 8.4. below).
Because the methodology of naturai science is the one that TG espous
es, TG is by definition unable to account for the specifically human cha
racter of language. In Chomsky (1966a) and elsewhere Chomsky is opposed
to mechanistic explanation of linguistic behaviour, but this is incon
sistent because - as we have seen - TG has no other type of explanation
to offer in its stead; at most, it is just a question of increasing the
complexity of proposed mechanistic explanations. In this context Chomsky
insists on the creative aspect of language, which he defines as 'rule-
governed creativity' (Chomsky 1964d:22) or as freedom within innately
given limits:
Language is a process of free creation; its laws and principles are
fixed, but the manner in which the principles of generation are
used is free and infinitely varied (Chomsky 1973:182).
The interesting thing is that, due to his general posi ti vi sti c out
look, Chomsky simply does not possess the methodological concepts to
capture the creative, non-mechanical character of human behaviour, for
instance, language use. The first step in this direction would be the
realisation that human sciences differ systematically from natural seien-
ces, for instance in the ways indicated in 2.0. (above). On another

count, Chomsky (1973:175) clearly thinks that the "restrictive attribu
tes of mind" which constrain human creativity and freedom (of language
and thought) can be investigated empirically, or in accordance with the
standard methodology of natural science. Experimental study of the li
mits of thought, however, requires a knowledge of both sides of these
limits, i.e., also of what cannot be thought. But this is obviously
impossible (cf. Itkonen 1974:291-94).
The non-creative or non-innovative character of recursivity (cf.
n.36) is in conflict with any reasonable interpretation of that funda
mental tenet of TG which says that a (recursive) grammar is able to ge
nerate an infinite number of completely novel sentences. To non-trans
formationalists, this notion of complete novelty has always seemed ex-
37
tremely puzzling. Therefore the following clarification offered by
Postal (1968b:267) is certainly welcome:
The purpose of this book has been to provide the basis for an ex
planation of an almost miraculous and easily overlooked fact: Any
speaker of a human language, like English, French, or Chinese, can
produce and understand utterances which are completely novel to
him. ... As an illustration of this novelty, you will observe that
the sentences on this page are completely new to you; that is, you
have never seen exactly these sentences before. Perhaps the easiest
way to convince yourself that normal use of language involves com
pletely novel expressions is to try to find in a book or a news
paper some sentences which you can reasonably claim to have expe
rienced before in their entirety. A search of this sort will re
veal an interesting fact: Even in a long book it is unlikely that
you can find a repetition of the same sentence.
In other words, a sentence A is completely novel i n r e l a t i o n to a

sentence B, i f only i t is not exactly identical with, or a repetition
o f , B. This is a good i l l u s t r a t i o n of the i r r e s p o n s i b l e way i n which
TG has often been g i v i n g new meanings to w e l l - e s t a b l i s h e d words, w h i l e
denying the existence of such a p r a c t i c e . Notice t h a t , according to TG's
usage, even a system c o n s i s t i n g of the two rules X→XO and X→0 i s able
to generate an i n f i n i t e number of expressions, i . e . , s t r i n g of zeros,
which are completely novel w i t h respect t o , because not exactly i d e n t i c a l
w i t h , each o t h e r .
To sum up, i n s o f a r as terms l i k e 'mentalism' or ' c r e a t i v i t y ' are
used i n connection w i t h TG, they do n o t h i n g , i n s p i t e of t h e i r conno-

t a t i o n s , to a l t e r the thoroughly p o s i t i v i s t i c character of TG.
3.7. Some Recent Developments in Linguistic Theory

As f a r as I know. Saumjan is the f i r s t to have provided a clear
formulation of the underlying p o s i t i v i s t i c assumptions of TG, and to
have attempted to raise l i n g u i s t i c s from a ' d e s c r i p t i v e ' science to an
'explanatory' one by bringing i t squarely i n t o the hypothetico-deductive
tradition. (The same attempt was made, w i t h varying degrees of success
and e x p l i c i t n e s s , by B l o o m f i e l d , Hjelmslev, Chomsky, and o t h e r s . ) Saum-
jan is also the f i r s t to have e x p l i c i t l y i d e n t i f i e d grammatical explana-
tions w i t h D-N explanations (Saumjan 1971; the Russian o r i g i n a l in 1965).
Although I think i t an easy matter to demonstrate t h a t Saumjan's account
is untenable, i t does possess the m e r i t of s t a t i n g the p o s i t i v i s t i c po-
s i t i o n in a consistent and e x p l i c i t way, so t h a t e i t h e r agreement or
c r i t i c i s m becomes p o s s i b l e . Saumjan's own theory of l i n g u i s t i c d e s c r i p -
t i o n s , i t must be s a i d , contains several f a l l a c i e s . F i r s t and foremost,
his notion of 'dynamic', as opposed to ' s t a t i c ' , synchrony is w i t h o u t
substance. More p r e c i s e l y , the idea t h a t generative grammars (as opposed
to 'taxonomie' grammars) in some sense reveal the 'dynamic inner f u n c t i o n -
ing of language' i s based upon the f a l l a c y of a s c r i b i n g the properties
of a way of describing to the thing described. I t should be c l e a r , how-
ever, t h a t , to o f f e r a simple example, the immediate-constituent analy-
sis of a sentence remains the same, i r r e s p e c t i v e of whether i t is pre-
sented in a ' s t a t i c ' , H a r r i s - t y p e grammar, or in the phrase-structure
component of a 'dynamic', TG-type grammar. Moreover, Saumjan's view
(which i s , i n c i d e n t a l l y , repeated i n Sampson 1975) t h a t a l l factually
occurring utterances are c o r r e c t , does not admit of any sensible i n t e r -
pretation.
Botha (1968) has independently given a c o n s i s t e n t l y positivistic
i n t e r p r e t a t i o n of TG, which i s i n a l l relevant respects s i m i l a r to t h a t
presented by Saumjan. In his l a t e r works Botha has grown i n c r e a s i n g l y
critical of TG (Botha 1971 and 1973). He notes, i n p a r t i c u l a r , t h a t TG
does not f u l l y s a t i s f y the requirements of empirical explanation or of
empirical t e s t a b i l i t y . In his opinion TG ought to be developed so as
to satisfy them, but he does not make any concrete proposals about how
this could be achieved.
TG is subjected to a similar criticism in Derwing (1973) and in
Sampson (1975). They note that TG, as it is today, is not a genuine
empirical science, and propose changes and/or reinterpretations which
should transform it into such a science (for discussion, see 7.0. below).
It is not always clear whether Botha, Derwing, and Sampson mean their
results to be applicable only to TG, or to grammatical description in
general.
It is possible to agree with the criticism offered by the above-
mentioned authors, while arriving at a fundamentally different conclusion
In a series of publications I have defended the thesis that theories of
autonomous linguistics are not, and cannot be, empirical theories (I tko-
nen 1969, 1970, 1972a, 1972b, 1974, 1975a, 1976b). As a result, the
methodology and the philosophy of (autonomous) linguistics need to be
rethought in their entirety. In the present discussion I hope to give as
full a justification as possible of this view and explore as many of its
implications as I can.
Apel (1973c) has subjected TG to a careful philosophical scrutiny,
and comes up with the conclusion that, because of the impossibility of
the "Subjekt-Objekt-Trennung" (cf. 2.1. - 2. above) TG is, to a cer
tain degree, not an empirical, explanatory science, but rather a re
construction of a given rule-system, or of a competence pertaining to
such a system (cf. also Habermas 1971b:171-75). 'To a certain degree'
is, of course, problematic. Apel (1973c:283) thinks that formal univer
sals, which he erroneously identifies with transformations tout court,
are genuine laws of nature, but he is apparently unaware that such formal
universals as are known today (supposing that there are any) are arrived
at purely on the basis of grammatical analysis and hence cannot possibly
be given a deterministic or nomic status. Furthermore what little ex
perimental research has been done, shows that transformations do not have
psychological reality, i.e., deterministic power (cf. 8.4. below). On the
other hand, it would be pointlessfor Apel to speak about non-discovered
(and probably non-existent) formal universals. Perhaps worse, Apel takes
over from TG the confusion between rules of language and rules of grammar,
While this confusion permits TG to see rules of language as non-normative
(cf. 7.5. below), it permits Apel (1973c:281 and 284) to see rules of
grammar - other than 'formal universals' - as normative. From this it
follows, implausibly, that grammars can be considered as somehow social
and normative entities.
In keeping with Apel's general philosophical line, Andresen (1974)
offers a detailed analysis of the concept 'explanation' in TG, and arrives
at the conclusion that TG explanations do not meet the criteria of D-N
explanation. Weydt (1975) achieves the same result. In Dretske (1974)
and Hutchinson (1974) a clear distinction is drawn, with differing value
judgements, between grammatical explanations and psychological or menta
listic explanations, the former being denied an empirical status. (In
fact, puzzlement about their status is expressed already in Botha l968:
109.) Ringen (1975:36) argues that "if grammars are to be compared with
scientific theories at all, they should be compared with axiomatic theo
ries in nonempirical sciences like logic and mathematics and not with
theories of physics and chemistry". Lass (1976) comes up with the con
clusion that phonological theories, instead of being empirical in Popper's
sense, are forms of 'rational metaphysics'. Similarly, according to Kac
(forthcoming), the normativity of linguistic data necessitates the non-
empirical character of grammatical descriptions; he proposes a type of
description which, unlike TG descriptions, would directly account for
linguistic normativity.
The issue of the psychological reality of grammatical descriptions
has been seriously taken up by scholars like Andersen, Antti la, Campbell,
Hsieh, Linell, Skousen,Steinberg and Krohn, who all represent, in dif
fering ways, what has come to be called 'concrete phonology'.
3.8. Conclusion
Saussure, Hjelmslev, Sapir, Bloomfield, Harris, and Saumjan consider

language as an autonomous, self-contained object of study. Saussure and
Sapir realised that in whatever way, precisely, the mode of existence of
language is to be defined, it is not individual, but social. More pre
cisely, language does not exist at the level of social spatiotemporality,
but at that of (social) institutions. In Sapir's opinion, language as

an institutional or cultural entity needs methods of description which
are not those of empirical sociology or psychology, let alone of pure
natural science. Saussure's position is less clear. Even Sapir deve
lops, however, no explict methodology. Bloomfield, Harris, and Saumjan
regard language as an observable, physical entity, and therefore they
subscribe to positivism. Hjelmslev does likewise, for somewhat unclear
reasons. TG regards language not as an autonomous entity, but as a psy
chological one. It subscribes to the methodology of natural science;
yet, at the same time it concentrates upon intuitive knowledge and dis
misses the use of experiments.
In sum: Saussure's methodological position cannot be pinned down.
Hjelmslev represents inexplicit positivism. Sapir represents inexplicit
hermeneutics. Bloomfield, Harris, and Saumjan represent explicit and
consistent positivism. TG represents explicit and inconsistent positi
vism.
4.0. PHILOSOPHICAL FOUNDATIONS OF THE CONCEPT OF
LANGUAGE
With regard to its relation to knowledge, language may be consider

ed either as merely its expression or - more plausibly, I think - as
virtually identical with it. In both instances it is clear that the
question concerning the nature of language cannot be treated apart from
the question concerning the nature of knowledge. In other words, lin
guistics cannot be discussed in entire abstraction from general episte-
mology. In the present chapter an attempt will be made to substantiate
this claim.
4.1. Characterisation of the Traditionist Epistemology
The history of epistemology has been dominated by one powerful

tradition, which has been represented by mind-matter dualists like
Descartes and Locke, subjective idealists like Berkeley, empiricists
like Hume, and phenomenalists like Russell and Ayer. The basic tenet
of this line of thought (which, in agreement with Saunders & Henze 1968,
I will term 'traditionism') is that subjective experiential data are
the primary source of knowledge. This egocentric position entails that
public things and qualities are thought to be in some way constructed
out of subjective experiences. Furthermore, the knowledge of other
minds is supposed to be gained inductively , on the basis of the so-
called argument from analogy: When I perceive that bodies (constructed
out of my sense-impressions and) resembling mine behave under similar
circumstances in the same way as my body does, I may infer with a high
degree of probability that these bodies are possessed by minds which
think and feel in ways similar to mine. Hence my point of view is gi
ven as the inescapable Cartesian foundation of everything else, where-
as in contradistinction to the existence of my mind, the existence of
other minds is a contingent fact. In addition, the existence of other
minds (which is only probable) is contingently connected with the be
haviour 'pointing towards' it, which means more specifically that psy
chological or mental concepts such as pain, knowledge, and intention
possess no necessary or conceptual link with their manifestations. A
similar conclusion can also be drawn from the consideration of my mind,
as opposed to other minds: In relation to me, the existence of psycho
logical concepts is necessarily given, but since it appears self-evi
dent that I know what I think or feel without having to observe my own
behaviour, the relation between psychological concepts and behaviour is,
again, contingent.
According to the traditionist conception, what is immediately gi
ven is my experience, especially sense-impressions and perceptions.
That is to say, the immediately given and indubitable aspect of other
human beings (if that is what they are) is their physical appearance,
i.e., the movements performed, and the sounds emitted by their bodies.
On the basis of this observational evidence, I make inductive inferen
ces concerning their possible mental states and processes (which are
instances of corresponding psychological concepts); however, these in
ferences are at best probable inferences. Mind and the related psycho
logical concepts represent something that I cannot literally observe,
but only understand. Mind in its various ramifications is something
'behind' the external, observable behaviour. Since mind is a deriva
tive and more or less unreliable notion, understanding is a derivative
and unreliable way of gaining knowledge. Consequently, as far as other
human beings are concerned, I ought to try to replace 'mind' by 'beha
viour' , and understanding by observation. This means that no differen
ce in principle should exist between the natural sciences and the human
sciences; in both cases, I, as a scientist, deal with observable events
(obviously in conformity with the hypothetico-deductive method), and
all apparent differences between the two types of science are merely
attributable to the failure of the human sciences to imitate the exam
ple of physics with a sufficient degree of fidelity. This is the only
FOUNDATIONS OF THE CONCEPT OF LANGUAGE 93
reason why the human sciences have not yet been able to rise from the
merely descriptive level to the explanatory one. At most, understand
ing can be used as a heuristic device in setting up positivistic causal
explanations of observable events belonging to the domain of human be
haviour.
When the traditionist thesis is briefly expressed in the above
manner, it probably seems nothing more to the average scientist than
harmless philosophical speculation which has no connection with his
work. However, upon closer inspection it has already become apparent
that this thesis has methodological implications that have been extre
mely effective and decidedly harmful in the history of scientific re
search influenced by positivismo First of all, human sciences have
been assimilated to natural sciences; secondly, both within human and
within natural sciences the status of the scientist, i.e., the one who
observes the behaviour of things and (other) human beings, has remained
unaccounted for. The former deficiency has been especially criticised
by hermeneutics, while the latter has been criticised by all those
schools of philosophy and metascience (including hermeneutics, of course)
which emphasise the nature of science as a historically given, social
enterprise. Recently, Apel (1973b) has shown in detail that traditio-
nism in the form of 'methodical solipsism' provides still today the
epistemological foundations of the positivistic philosophy of science.
The scientist is not in the world, in which are his objects of re
search, including all other human beings; rather, he is on the limit
of the world. Or, as the early Wittgenstein put it:
Dass die Welt meine Welt ist, das zeigt sich darin, dass die Grenz
en der Sprache (der Sprache, die allein ich verstehe) die Grenzen
meiner Welt bedeuten ... Ich bin meine Welt ... Das Subjekt gehört
nicht zur Welt, sondern es ist eine Grenze der Welt ... Hier sieht
man, dass der Solipsismus, streng durchgeführt, mit dem reinen Rea
lismus zusammenfällt. Das Ich des Solipsismus schrumpft zum aus
dehnungslosen Punkt zusammen, und es bleibt die ihmkoordinierte Rea
lität (Wittgenstein 1969a: 5.62, 5.621, 5.63, 5.632, 5.64)
It is paradoxical, or one of the ironies of the history of ideas,

that positivism, generally regarded as the philosophy of 'hard facts',
rests on such an exotic epistemological foundation.

The traditionist thesis may be reformulated in linguistic terms
as follows: Since the knowledge of the objective or public world is
somehow derived from subjective experiences, ordinary intersubjective
languages which refer to public things and qualities must (or could)
have been preceded by subjective or private languages referring to sub
jective experiences only. Hence 'private language' could be taken to
mean a language logically prior to any intersubjective language. How
ever, private languages can just as well be considered from a more ge
neral perspective, and be defined as languages logically independent
of any intersubjective language; that is, each word of a private langu
age must be conceptually independent of public, objectively given phe
nomena.
During the last 25 years, the notion of private language has been
the object of intense philosophical discussion. This notion was initi
ally developed by the later Wittgenstein, who used it as an explication
of his own earlier views, as expounded in the Tractatus . By his de
monstration of the impossibility of private languages, he expressly re
futed those mostly implicit and unexpressed presuppositions which have
since the time of the Tractatus underlain the epistemology of positivism.
4.2. Refutation of the Traditionist Epistemology
4.2.1. The Axiomaticity of the Concepts of Person and Thing

I believe it to be generally felt that the traditionist account
of knowledge is in some way unnatural. No one seriously believes that
the existence of other persons is less than certain. More precisely,
I cannot seriously believe that I am the only person who exists with
certainty. Moreover, not only the notion of person, but also the no
tion of thing implied by traditionism is unsatisfactory: no one has
ever succeeded in showing how, precisely, public things could be con
structed out of subjective experiences. The existing tentative formu
lations can be demonstrated as making circulary use of that very notion
of thing which they should define. It is not surprising, therefore,
that several thinkers, including Marx and Engels, Dilthey, Mead,
Husserl, and Heidegger, have come to reject the traditionist epistemolo-

38
gy.
In what follows, I shall apply the Wittgensteinian approach to
prove the primarily social nature of knowledge and language. This
approach ought to be seen as part of a more general trend in (social)
philosophy (cf. n.38). We may start by concentrating upon the internal
inconsistencies of the analogy argument, which may be represented more
explicitly in the form of the following (invalid) inference:
(i) If I have the experience A, I am usually in the situation

Β and behave in the way C.
(ii) (I perceive that) a body resembling mine is in the situa
tion Β and behaves in the way C.
Therefore it is probable that the body has the experience A and

thus possesses a mind similar to mine; i.e., I have here to do
with another person.
The premises which should make the conclusion probable are supposed
to represent that primary state of knowledge in which I know nothing
but myself and my private experiences: this is the whole point of tra-
39
ditionism. It is the conclusion that introduces the concept of some
one else's experience, and hence of someone else. However, it is quite
easy to see that there can be no concept of 'I' without the correlative
concept of 'he' (and 'you' and 'we'), just as, for example, there can
be no concept of 'wrong' without the correlative concept of 'right'.
Consequently, in contradistinction to the basic assumption of traditio-
nism, the concept of person (i.e., I and others) must be assumed from
the outset, a priori , and all purely egocentric accounts of knowledge
are inherently inconsistent. Furthermore, since the world which I
perceive is necessarily intersubjective, the things in it are public
things, not constructions out of my private sense-impressions. There
fore knowledge is social at its origin; and the social origin of know
ledge guarantees the social control of knowledge (cf. 5.3. below).
This is not to deny the reality of sense-impressions but only to main

tain that they are dependent upon public things and qualities; as a
consequence, they cannot be 'subjective' or 'private' in any absolute
sense. The concepts of thing and person are axiomatic, since, instead
of being something which we could gradually construct out of our expe
riences, they determine the form of our experiences. The same is true,
incidentally, of rules of inference which are meant to be applied in
the analogy argument, or any other argument: such rules must be objec
tively given, but it is obviously impossible to construct them for ex
ample out of my subjective feelings of certainty.
The thesis of intersubjectivity may be formulated more precisely
as follows: First, 'I' does not exist in a vacuum, but rather in combi
nations like 'I do' and 'I think'. Second, 'I do' and 'I think' refer
to at least potentially self-conscious processes, and not to some in
stinct-like unconscious processes, which means that they entail, and not
only are entailed by, 'I know I do' and 'I know I think'. Third, the
concept of 'I' is - as we saw above - interdependent with the concept
of 'others'. From these three points taken together, it follows that
a person A cannot know that he is doing or thinking X, and thus cannot
do or think X, unless he is able to know what it is for some other per
son Β to do or think X. The same is true, in turn, of B's knowledge of
his own actions and thoughts with respect of A's (possible) actions and
thoughts. Hence, it can be shown on purely conceptual grounds that A
must be able, in principle, to identify B's various mental states or
processes, and vice versa. This means that mental states and processes
exist only at the level of common knowledge i.e., of a common ability
to identify them, wherever they occur.41
4.2.2. Mind, Behaviour, and Environment
The existence of the above-mentioned common ability was establish

ed as the result of a kind of 'transcendental deduction'. The question
still remains open how this ability is put into practice. It is only
trivial to state that we come to know what others are doing or thinking
by understanding their behaviour. (Behaviour may be either observed or
understood. But notice that we neither understand observable, or phy

sical, behaviour nor observe understandable behaviour. Rather, we un
derstand understandable behaviour and observe observable behaviour; cf.
immediately below). Hence, as a first approximation we might say that
identification of mental states and processes rests on public behaviour
al criteria. However, Ï want to argue, with Wittgenstein, that this is
not just a contingent but a necessary truth, which means that there is
no sensible way to separate a mental state or process from its (under
standable) criteria. Probably no one would deny that an action qua
action must necessarily possess some public aspect. It is easier to
think that thoughts and feelings might be only contingently related to
any public criteria. To give an extreme example, it seems at least lo
gically possible that even if A must be able, in principle, to identify
B's thoughts and feelings, he might do this on the basis of telepathy.
However, although this could be sometimes the case it could not be the
case always. Thoughts and feelings cannot be consistently conceived of,
if they are deprived of any relation to public behaviour, that is, of
any possibility of even indirect expression or manifestation. Without
this possibility, it becomes impossible to tell one thought or feeling
from another. Therefore the relationship of psychological concepts to
their public criteria is not just contingent. Moreover, A can, as a
matter of conceptual possibility, understand B's behaviour only if it
does not vary at random, but exhibits a certain uniformity. Consequ
ently, all actions, thoughts, and feelings, including those of any pur
ported Cartesian ego, are conceptually dependent upon the constancy of
their public, i.e., intersubjectively understandable, manifestations:
"An 'inner process' stands in need of outward criteria" (Wittgenstein
1958: §580). 42
After discussing the relationship between mental phenomena and
their behavioural criteria, let us consider their relation to the si
tuational context of the latter. The thesis of atomism says that there
can be no conceptual relations between distinct physical states of
affairs, e.g., between an event and its physical context, which means
that any event might conceivably occur in any environment (cf. 1.5.
above). By contrast, there are conceptual relations between actions,

considered either as such or as criteria for the occurrence of thoughts
and feelings, and their (social) environments. In other words, it is
simply not true that any action might conceivably occur in any environ
ment. To give an example, let us suppose that in a context A there is
an act of obeying, i.e., B. Is the existence of Β independent of A?
Certainly not. It is a conceptual truth that one can obey only when
commanded to do something. Therefore it is a conceptually necessary
condition to the existence of Β that there has been C, i.e., an act of
commanding performed by someone else. C may be simply identified with
A. 4 3
To continue, let us take A as our point of departure. Does the
existence of A necessarily require the existence of something else?
Here the situation is slightly more complicated than in the previous
instance. Of course, if someone gives a command, it is not conceptu
ally necessary that he will be obeyed. But it is clear that there is
a conceptually necessary relation between the types, i.e., concepts,
of which A and Β are tokens: if the concept of obeying (and that of
disobeying) were nonexistent, the concept of commanding would, of ne
cessity, be nonexistent, too, with the consequence that acts of com
manding would be conceptually impossible. That is to say, the exist
ence of A does not require the existence of B, but it does require the
existence of the concept exemplified by B. Moreover, in order to exist,
A in fact requires the existence, not only of the concepts of command
ing, obeying, and disobeying, but also of a definite state of affairs.
Evidently, this state of affairs is not simply the action B, but the
exclusive disjunction of B, i.e., the act of obeying, and its contra
ry C, i.e., the act of disobeying. If A is followed neither by Β nor
by C, then, appearances notwithstanding, A does not exist. For instance,
if in the presence of a corpse I say 'Open the window!', and nothing
happens, I have been neither obeyed nor disobeyed, but then I have not
given a command either. This makes clear that 'C' does not simply
equal '-B', because what the corpse 'does' is both '-B' and '-C'; this
is why 'B' and 'C' are contraries, and not contradictories.
Notice that it would be quite wrong to say that in order to exist, a
natural event like rain requires the existence of the concept which it
exemplifies, let alone the existence of some related concepts. Rain
would exist even if man had not conceptualised it, and indeed even if
man did not exist. But acts of commanding would not exist, if they
as well as acts of obeying and disobeying had not been conceptualised
by man.
The above discussion of the conceptual relation between commanding
on one hand and either obeying or disobeying on the other (i.e., 'A=D'
where 'D' = ' B ^ C ' ) , may be profitably compared with our previous dis
cussion (in 1.5. above) of the empirical relation between the heating
of a piece of metal and its expanding ('E F'). These two discussions
can be summed up by noting that the falsity of 'D' entails the falsity
of 'A', whereas the falsity of 'F' does not entail the falsity of 'E'.
There is only one possible objection: We could make the relation of 'E'
to 'F' a conceptual one by d e f i n i t i o n , i.e., by taking it to be a defi
ning property of metal that it expands when heated. As a consequence,
'E' would entail 'F', because '-F' would entail '-E': if a piece of what
we thought was metal does not expand after we have heated it, i.e., if
'F' is false, then eo ipso what we heated was not metal, i.e., 'E' is
false, contrary to what we first thought. However, even upon this in
terpretation the relation between 'E' and 'F' remains completely differ
ent from that between 'A' and 'D'. Natural events are what they are,
regardless of how we define o r interpret them; and they would continue
to exist even if we did not define or interpret them at all. By con
trast, being interpreted is essential to being an action (cf. the prin
ciple "'I do' entails 'I know I do'"). Consequently, if upon the prevail
ing interpretation the relation between two actions, or action complex
es, is conceptual then its conceptuality is not just due to some defi
nitional trick but really 'is in the world', that is, in the social
world of actions and institutions.
What was said here about actions such as commanding and obeying
holds true more generally of all well-established forms of social in
teraction: there can be no buying without selling, and so on. Now, it
might be somewhat misleading to regard conceptual relations as holding

just between separate actions, as I did above. It would perhaps be
more accurate to say that they constitute (part of) those customs or
institutions (of interpretation) which, in the first place, give mean
ing to actions, or make them what they are. Such customs or institu
tions specify not only types of action, but also types of persons who
are to perform the actions in question, and types of (spatial and/or
temporal) settings in which they are to be performed. One needs only
to think of what is required for an act of marrying a woman to be per
formed. Thus the (social) environment to which an action stands in a
conceptual relation may consist of different kinds of components.
It may not always be clear what is the exact environment of an
action, and - due to the conceptual relation between action and envi
ronment - the nature of the action itself remains unclear in such a
case. However, just as there are conceptual limits upon which actions
can be performed in given environments, there are also conceptual li
mits upon how an environment can be interpreted. It may on occasion
be unclear whether a person has been commanded to do something so that
he can either obey or disobey, or whether the environment is such as
to allow a wedding to take place. But at least it is certain that a
man standing alone and without any equipment of communication in the
middle of nowhere can then and there neither give a command nor marry
anyone. - In sum, as far as mental phenomena are concerned, they ne
cessarily require public criteria not only of the behavioural, but
also of the situational type. It is literally an optical illusion
which makes us think that actions could be conceived atomistically
i.e., in isolation: they only seem to occur independently or their
environments.
Wittgenstein characterised the constancy of mentally relevant
behaviour and its relation to social environment as follows: 46
'Grief' describes a pattern which recurs, with different varia

tions, in the weave of our life. If a man's bodily expression
of sorrow and joy alternated, say with the ticking of a clock,
here we should not have the characteristic formation of the
pattern of sorrow or of the pattern of joy (Wittgenstein 1958:
II, i).
An expectation is embedded in a situation, from which it arises.

The expectation of an explosion may, for example, arise from a
situation in which an explosion is to be expected (I, §581).
Could someone have a feeling of ardent love or hope for the space
of one second - no matter what preceded or followed this second?
- What is happening now has significance - in these surroundings.
The surroundings give it its importance and the word 'hope' refers
to a phenomenon of human life. (A smiling mouth smites only in a
human face.)(§583).
Now suppose I sit in my room and hope that N.N. will come and
bring me some money, and suppose one minute of this state could
be isolated, cut out of its context: would what happened in it
then not be hope? - Think, for example, of the words which you
perhaps utter in this space of time. They are no longer part of
this language. And in different surroundings the institution of
money doesn't exist either (§584).
An intention is embedded in its situation, in human customs and

institutions. If the technique of the game of chess did not exist,
I could not intend to play a game of chess. In so far as I do in
tend the construction of a sentence in advance, that is made pos
sible by the fact that I can speak the language in question (§337).
In order to get clear about the meaning of the word 'think' we

watch ourselves while we think; what we observe will be what the
word means I - But this concept is not used like that. (It would
be as if without knowing how to play chess, I were to try and
make out what the word 'mate' meant by close observation of the
last move of some game of chess .)(§316).
And if things were quite different from what they actually are -
if there were for instance no characteristic expression of pain,
of fear, of joy; if rule became exception and exception rule; or
if both became phenomena of roughly equal frequency - this would
make our normal language-games lose their point (§142) .
Behaviour which is or can be understood is meaningful behaviour.

Conceptual relations between actions and their environments are indeed
comparable to meaning-relations , as these exist in language. Whether
an instance of human behaviour has meaning or not, can only be decided
by considering the wider aspects of the social or cultural context of
the behaviour in question. Actions can be understood (and, in the first
place, understood as actions), only if one has the 'cultural key' to
them, as Sapir would have said. In the following remark von Wright
asserts the analogy between l i n g u i s t i c and n o n - l i n g u i s t i c actions as

well as the culture-dependent nature of action in general:
Intentional behaviour, one could say, resembles the use of langu

age. It is a gesture whereby I mean something. Just as the use
and the understanding of language presuppose a language community,
the understanding of action presupposes a community of institutions
and practices and technological equipment into which one has been
introduced by learning and training. One could perhaps call it a
life-community. We cannot understand or teleologically explain be
haviour which is completely alien to us (von Wright 1971:114-15;
cf. also 3.3. above).
From the fact that the environment of an action partly determines

its meaning, it follows that if an action is transferred into a differ
ent environment, it may no more be the same action (or, perhaps, any
action at all). Conversely, if a given environment in which certain
actions are performed as a matter of course is deprived of these ac
tions, it ceases to be the environment which it was before.
Because meaning cannot be measured in terms of length, time, and
weight, it transcends the conceptual universe of the positivist. This
explains why positivists have taken it for granted that the relations
between actions and their environments satisfy the thesis of atomism
(cf. 1.5. above). It is only on this (false) assumption that one can
try to force human intentional behaviour into the model of (causal)
D-N explanations (cf. 1.2. above).
So far we have come to see that our knowledge of mental facts is
inseparable from our knowledge about behavioural and situational facts.
But from this it by no means follows that we should be able to give any
nearly exhaustive lists of public, behavioural or situational criteria
to be used in identifying various actions, thoughts, or feelings. This
apparently paradoxical fact is due to the 'creativity', or more gene
rally the unpredictability, which is typical of human actions: There is
obviously no limit to the number of the ways in which, for instance, one
might try to make someone else angry, or might manifest one's anger.
However, by virtue of the mutual ability of understanding (whose exist
ence has been proved above), we can be sure that as new criteria of men
tal phenomena appear, they will be understood by others.
The unpredictability of criteria is coupled with that of the envi

ronments in which they appear. In other words, what is the environment
of an action, often cannot be known in advance, but only after the action
has been done, because it is only at that moment that one can see how
the agent had interpreted, or even had decided to interpret, the envi
ronment of his action. Moreover, a new way of interpreting an environ
ment may amount to creating a new type of environment, not to speak of
less contemplative ways of creating new environments (which in turn may
require, or make possible, new types of mental phenomena or at least
new criteria of old types). It is for these reasons that historians
generally cannot predict actions, but can try to explain them only
after they have been produced. - It should be remembered, however, that
there are conceptual limits upon how a given environment can be inter
preted.
Mental concepts are sometimes compared to theoretical concepts of
natural science (cf. 1.4. above). It is apparent from the previous
paragraph why this analogy is defective. The exact nature of those ob
servable events to which a theoretical concept is linked by a chain of
correspondence rules and operational definitions is in principle open,
due to the openness of the nature of the circumstances in which they
may occur. But once the circumstances are known, the nature of the ob
servable events in question must be predictable , at least within a de
finite range of statistical variation. If the predictions are not
born out, the concept has to be modified; o r , to put it in other terms,
the correspondence rules and/or the operational definitions have been
shown to be false. By contrast, public criteria of (an instance of) a
mental concept are essentially u n p r e d i c t a b l e ; and the occurrence of an
unpredicted criterion by no means calls for a modification of the cor
responding concept. This difference between the two types of concepts
is intimately connected with the fact that the mental concepts in
question are of an atheovetical character. Theoretical concepts of
psychology act more like theoretical concepts of physics. Atheoretic-
al concepts are not hypothetical (we know what 'anger' means) even if
we often do not know whether they are applicable in definite situations.
By contrast, theoretical concepts are necessarily hypothetical.

Wittgenstein's method of eliciting criteria of mental phenomena,
for example those of imagination, consists in 'looking at', or remind
ing himself of the ways in which corresponding words, e.g., 'imagina
tion', are used. Such an analysis easily reveals the infinite variety
of, e.g., the criteria for imagining. Wittgenstein's method is empha
tically not that of introspection of one's consciousness while one is
imagining. It is impossible to identify any single and coherent 'con
tent of consciousness' which is always present when one is imagining
(or hoping, intending, calculating, and so on) and which could thus
constitute the meaning of 'imagination'. Similarly, it is hopeless
to try to determine the meaning of 'imagination' somehow in abstrac
to. Although it can hardly be said that the meaning of a word liter
ally is its use, it is most certainly shown by its use.
From the above, it follows that, although public criteria are the
logical basis of mental phenomena, all lists of such criteria are bound
to be incomplete. This implies that no single criterion is a necessary
condition, and no amount of criteria constitute a sufficient condition,
for the occurrence of a given mental phenomenon. It is further quite
clear that when we make use of such criteria, we will sometimes be
wrong. Thus we may on occasion be led to think that a person is think
ing or doing something, when he is not, and the reverse may happen too.
Nevertheless it would not make sense to suppose that we could be wrong
all the time, that is, that no one would ever be thinking or doing
what others would understand him to be thinking or doing. This would
be just as senseless as assuming that everybody could make nothing but
false moves in every game, although one in fact arrives at such a con
clusion by the apparently innocuous inference from 'x sometimes happens'
to 'x might always happen' (cf. Wittgenstein 1958:§344-45).
4.2.3. Characterisation of Mental Phenomena: the Notion of

'Pattern'
The realisation that experiences qua possible objects of (communic

able) knowledge are logically dependent upon, or inseparable from inter-
subjectively understandable behaviour in characteristic surroundings
does not of course mean reducing them to behaviour; nor does it mean
claiming that experiences independent from public criteria do not
exist, but only that nothing can be said about them. Wittgenstein
justly notes that purely subjective, and ex definitione incommunic
able, experiences have no place in the public, intersubjectively
understandable language in which we talk about our thoughts and feel
ings for instance; and such an language, or an artificial but still
intersubjectively understandable language based on such a language,
is the only logically possible language (cf. 4.2.5. below). Con
sequently, the question about the existence of something about which
nothing can be said in any language, turns out to be a spurious one.
On the other hand, with regard to the status of communicable experi
ences, Wittgenstein goes out of his way to repudiate the charges of
simple-minded behaviourism: "And now it looks as if we had denied men
tal processes. And naturally we don't want to deny them" (Wittgenstein
1958: I, §308).
Nonetheless, Wittgenstein admittedly remains vague in regard to
the positive definition of the nature of mental phenomena and of the
method of gaining knowledge about them, i.e., understanding them.
The following remark seems to indicate the direction in which one
should look for the answer:
When it looks as if there were no room for such a form between

other ones you have to look for it in another dimension. If there
is no room here there is room in another dimension (Wittgenstein
1958: II, p.200),49
I try to make the meaning of this somewhat cryptic statement clear

by applying it to the analysis of actions. The reason why actions can
be taken as representative of mental phenomena is that intentions are
an inseparable part of them. Now the positivistically-minded usually
claim that actions consist of nothing but observable parts, and hence
simply fall within the domain of observation. But it is the characte
ristic pattern of bits of observable behaviour, embedded in characte
ristic environments, which reveals the presence of an intention over
and above the purely physical, observable side of an action. It might
be said that behaviour of this kind is (understood as) 'directed' (Tay

lor 1964:55). More precisely, given the conceptual relation between
action and social environment, the pattern consists equally of bits of
observable behaviour and also of observable environment.
Although (observable) bits of behaviour and of environment are ob
served, the pattern qua pattern is understood. Behaviour is observed
or understood depending on whether it is considered in itself or as
contained in a characteristic pattern: We observe a sound when we sim
ply hear it as it is. We understand it, e.g., as an English word, when
we consider the present situation as a speech situation (which means
establishing the conceptual links between an action and its environment),
and when we further embed the present speech situation into the larger
context of English language. One and the same behaviour may be embed
ded into different patterns, and thus be understood in different ways,
but - as we have seen - there are conceptual limits upon which patterns
may be taken to exist objectively.
To use Wittgenstein's expression, we might now say that patterns
exist in a different dimension from their (atomistically conceived)
parts. The first dimension is understood, the second is observed.
Since a pattern is more than its parts, we have the (justified) feel
ing that understanding is more than the mere observation. Therefore,
when we say that mental phenomena, e.g., intentions, are something
'behind' observable behaviour, we are in a sense right. But we are
wrong insofar as we try to locate mental phenomena in the same (spa
tial) dimension as observable behaviour, with only the difference that
the former are supposedly behind the latter
It seems to me that Mead has in mind precisely the same distinc
tion which I have been trying to make here, when he explains why his
'social behaviourism' need not deny the existence of mental phenomena:
... the existence as such of mind or consciousness, in some sense
or other, must be admitted - the denial of it leads inevitably to
obvious absurdities ... we may deny [the] existence [of mind] as
a psychical entity without denying its existence in some other
sense at all; ... Mental behaviour is not reducible to non-mental
behaviour. But mental behaviour or phenomena can be explained in
terms of non-mental behaviour or phenomena, as arising out of,

and as resulting from complications in, the latter (Mead 1934:
10-11; emphasis added).51
Saunders & Henze (1968:18) call Wittgenstein a 'nonreductionist

behaviourist', and, in light of the above quotation, this term applies
to Mead as well. Throughout the present investigation I am trying to
defend the standpoint of nonreductionist behaviourism. I shall have
to deal mostly with the three dimensions of observable behaviour, ac
tions, and norms. Although actions are patterns of observable beha
viour, they are not reducible to the latter. Similarly, although
norms are patterns of actions, they are not reducible to the latter
(cf. 5.1. below).
A sentence referring to an action neither entails nor is entail
ed by any definite set of sentences referring merely to movements and/
or sounds. One and the same action(-type) may be performed in an in
definite number of different ways, so that knowing that a certain ac
tion has been performed does not entail knowing which particular move
ments have been made or which particular sounds have been emitted in
performing this action. By contrast, the mere fact that certain move
ments are made or certain sounds are emitted, does not entail that any
thing is being done intentionally (cf. Taylor 1964:54-57; von Wright
1971:111-13). This is Brentano's well-known 'thesis of intentionality'.
Its latter part shows, once again, why we are justified in thinking
that understanding (a pattern) is 'more' than observing (its atomis-
tically-conceived parts). 52
Apart from the inherent inconsistency of the analogy argument,
the traditionist position vis-à-vis actions is just as unnatural as
any behaviourist view contradicting the thesis of intentionality. It
is certainly not the case that, from the fact that a human-looking body
emits the sounds [aIm bo:d], I draw the inductive inference that pro
bably a person is saying 'I am bored'. More generally, it is not the
case that I believe, on the basis of past experience, that all these
bodies similar in appearance to my body are equipped with minds; rather
my attitude towards these persons (because that is what they are) is
an attitude towards beings equipped with minds (Wittgenstein 1958: II,
iv; 1969b: §404). Moreover, it would not make sense to ask whether
this attitude, which guides all my actions, can be justified by offer
ing reasons for it. The reasons for our actions will come to an end
sooner or later, and then we will just act; but to act without reasons
does not mean acting unreasonably (Wittgenstein 1958: I, §211 and 289;
1969b: §110 and 1.48). On the contrary, if justification by experience
did not come to an end sooner or later, it would not be justification
(Wittgenstein 1958: I, §485). Or to put it otherwise, doubt which has
no end is not even doubt (Wittgenstein 1958: II, v; Wittgenstein 1969b:
§115 and 150). It is the way people think and live, their form of life,
that shows what they accept as the ultimate justification (Wittgenstein
1958: I, §325, and II, xi, p.226; 1969b: §148 and 204). This idea is
of course identical with Husserl's conception of the role of the Le~
bensapriori (cf. 2.5. above). To say that the concept of person is
axiomatic is just another way of saying that it is part of our form of
1ife.
4.2.4. General Characteristics of the Conceptual Distinctions

Employed in the Present Study
Together with the axiomanaticity of thing and person, the refutation
of traditionism also establishes the axiornaticity, or irreducibility,
of the interconnected distinctions 'observation - understanding',
'event - action', 'physical - mental'. The conceptual apparatus which
I am introducing in this study contains many other dichotomies of the
same type. We have already come across the distinction between learn
ing a game and. mastering it in 2.4. (above), and we shall make use, inter
alia, of distinctions between 'rule' and 'regularity', 'correct' and
'incorrect', 'atheoretical' and 'theoretical'. The use of such dis
tinctions is sometimes objected to because of their relative nature.
I state here once and for all that in my opinion this 'objection' has
no convincing force in its favour. All distinctions which either in
volve social life (e.g., 'physical - mental') or obtain in it (e.g.,
'correct - incorrect') are relative. This is a trivial truth which
need not be repeated once it has been stated. However, the important
thing is that even if each of the distinctions concerned forms a con-
tinuum, the end points of such a continuum are absolutely different

(in the relevant respect) and, moreover, they both represent extreme
ly large numbers of important cases. For instance, there is an infi
nite number of physical entities and mental entities as well as correct
entities and incorrect entities.
In this context two opposite mistakes are often made. Let us take
as an example the 'correct - incorrect' distinction. On the one hand,
from the fact that some cases are unclear, it is inferred that all cases
are unclear; this is the standpoint of the current empiricist trend in
socio- and psycholinguistics (cf. 5.4. and 7.4. below). On the other
hand, presumably because of their untidiness, the factually existing
unclear cases are taken to be purely apparent, so that clear cases are
what exists in 'reality'; this is the 'classical' standpoint of TG:
Chomsky ... views each and every string of the language as belong
ing to one or the other of the two categories 'grammatical' or 'un-
grammatical'. For him, the middle range of 'undecidable cases'
reflects not some inherent gradient in the phenomena which a de
scriptively adequate rule must represent but simply incomplete
knowledge on the part of the linguist (Katz & Bever 1974:11).
The fallaciousness of both of these lines of thought should be

evident. Take the distinction between young and old: It would be
equally absurd to claim that since some people are neither young nor
old, all people are neither young nor old, and that in reality there
are only young people and old people. I hope to avoid both of these
fallacies. All distinctions concerned are relative, but at the same
time they have huge numbers of absolutely clear cases in their favour.
Both aspects of these distinctions may be legitimately explored. I for
one feel that, in the instances to be explored in this study, the end
points of a continuum are more important that its middle section.
4.2.5. The Impossibility of Private Languages

The refutation of traditionism, as presented above, can be trans
lated without difficulty into a refutation of the possibility of pri
vate languages. We may first consider that type of private language
which would most fully satisfy the requirements of traditionism, that
is, which would be logically prior to any intersubjective language.
Given that, as a matter of logic, there can be no 'I without 'he',

'you', and 'we', I could not say that I, in the logically primary sense,
privately invent a rule of language and then begin to follow it, since
at that point I could not even have the concept of myself.
Second, we may consider the case in which a person already lives
in the public world and speaks an intersubjective language, but then
tries to invent a private language, viz. a language which would be lo
gically independent of any intersubjective language. Each word of such
a language would have to refer to some purely subjective experiences
which would be logically independent of publicly identifiable phenome
na. However, we have seen that even if there exist subjective experi
ences of this kind, nothing can be said about them, which for all, both
practical and philosophical, purposes amounts to saying that they do
not exist. From the fact that nothing can be said about these experi
ences, it follows, more particularly, that one could not without con
tradiction try to discuss the question of whether or not there can be
a private language referring to them. - Notice that this instance is
different from the one, to be discussed in 5.3. below, where it is
admitted that an alien logic might exist even if we cannot know it.
Here we are discussing a private language which we ourselves suppo
sedly ought to be able to invent and to use. Yet we cannot know or
imagine such a language. Any such language which we try to imagine is
bound to contain concepts and, hence, words of some public language.
This is why we cannot make even the first step towards constructing a
private language (cf. n.48).
Third, let us consider a possible objection which says that we
have inferred too much from the refutation of the analogy argument:
even if one needs the concept of other persons, one does not actually
need other persons. Suppose that there is a man who is completely un
acquainted with social life, and particularly with social rules, but
whose world-conception is nevertheless roughly similar to ours, once
the social aspect is eliminated; he sees trees and stones, and does
not have to struggle with the problem of how to construct them out of
his sense-impressions. Is it possible for this man to invent a (pri-
vate) language for himself, and to use it to refer to (observable)

things and qualities?
To begin with, it can be plausibly argued that this question is
unacceptable, since it rests upon unacceptable assumptions. Presumably,
the man would have to acquire the concept of himself, that is, the con
cept of his own person, by contrasting himself with the surrounding
(non-social) nature. But it is not at all clear that the resulting con
cept of person would in any way resemble our concept of person. Conse
quently, it is not at all clear that the man could be said to be per
ceiving things in the same way as we perceive them, or that he could
be said to be doing something (such as inventing a rule and following
it) in the same sense as we are doing it. However, the decisive argu
ment is that if the man invents a rule and tries to follow it (suppos
ing, for the sake of argument, that he can do such things), then he has
no independent checks on whether or not he is following the rule cor
rectly. He uses a word as he remembers having used it before, but if
he wants to check whether he is using it correctly, all he can do is
to compare his present use with exactly the same memory which already
constitutes the basis for it (cf. Kenny 1975:192-93). Perhaps he in
fact uses the word correctly, as some outsider might be able to as
certain; but he has no way of knowing it himself. Everything that
would seem correct to him, would be correct, which would of course de
prive the word 'correct' of its meaning. To think that one is follow
ing the rule correctly is entirely different from following the rule
correctly (Wittgenstein 1958: §202 and 258). Consequently, there can
53
be no private rules. Rules just like concepts exist on the level
of common knowledge (cf. 4.2.1. above), and common knowledge is in
separable from social control of what is known.
It still may be objected that even if there is a group of per
sons, there might yet be no genuine social control, supposing that
each member of the group is continuously using words incorrectly.
Now, if they are making mistakes at random, they have no public
language; and obviously they have no private languages either. Thus
we cannot speak here of making mistakes. On the other hand, if they
all make mistakes in the same way, it is misleading to speak of making

mistakes at all.
The necessarily social nature of rules illustrates the methodical
aspect of the notion 'methodical solipsism'. The positivistic philo
sophy of science implies, not only that one man can observe 'from the
outside' the rest of the world, including all other human beings, but
also that one man can invent and apply the rules of scientific research.
The latter point indicates that some of the differences between stan
dard positivism and Popperism may be more apparent than real.
The three preceding arguments (which are, in a sense, of decreas
ing strength) apply to three different types of private languages.
More precisely, these types are ordered in such a way that the second
argument, and the third, also apply to the first type, and the third
argument also applies to the second type.
The refutation of traditionism, viz. private languages, makes
it clear, first of all, that such psychological concepts as knowledge
are inseparable from outward criteria, and secondly, that rules must
be intersubjective or social. When these two results are combined
and applied to the question relating to the nature of language, it
follows that language, as well as the knowledge of it, is inseparable
from the use of language which conforms to social rules. Wittgenstein
(1958: I, §508-10) has illustrated this point by the following example.
If I imagine that 'a b c d' means 'The weather is fine', does it follow
that 'b' now has the meaning 'weather'? No, just as it does not follow
that there is a book on the table, if I imagine that there is. Of course,
the reason is that no actual use or practice exists in support of the
claim that 'b' has this or that meaning. Meaning exists only in the
context of use; and since no language without meaning exists, there is
no language without the use of language. Given that, as a matter of
logic, (regular) use must be public, not private, it follows that a
theory of language which does not explicitly account for the social
rules constituting the use of language is inadequate on both factual
and logical grounds.
Wittgenstein is generally credited with philosophically justifying

this social and functional conception of language. However, Mauthner
(1923) anticipated several of Wittgenstein's basic insights. In parti
cular, he affirmed the impossibility of private languages and, together
with it, the social nature of knowledge:
Die Sprachbewegungen des unter sprachlosen Mitmenschen allein re

denden Individuums wären aber gar nicht Sprache. Ein einzig sprech
ender Mensch unter sprachlosen Volksgenossen is t ebensowenig vor
stellbar wie ein redender Gott, der den Menschen die Sprache erst
schenkte. Oder er wäre wie der Teilnehmer an einem ausgedehnten
Telephonnetze, das keinen zweiten Teilnehmer hätte. Seine Zweck
bewegungen wären nicht Sprache. Sprache werden diese Bewegungen
erst durch ihre über das Individuum und über die Wirklichkeit hin
ausgehende Eigentümlichkeit, da sie bei einer Gruppe von Menschen
die gleichen, da sie dadurch verständlich, da sie nützlich sind.
Als sozialer Faktor erst wird die Sprache, die vor Erfindung der
Buchdruckerkunst noch nicht einmal in einem Wörterbuch beisammen
war, etwas Wirkliches. Eine soziale Wirklichkeit ist sie; abge
sehen davon, ist sie nur eine Abstraktion von bestimmten Bewegung
en (Mauthner 1923:17-18).
Wenn Begriff und Wort, wenn Denken und Sprache ein und dasselbe
ist, wenn ferner die Sprache sich historisch und im Gebrauche des
Individuums nicht anders als sozial bilden konnte, so muss auch
das Erkennen der Wirklichkeit eine gemeinsame Tätigkeit der Mensch
en sein {op.cit. , p.30). 56
4.3. Implications for Linguistic Theory

In this section, as elsewhere, I take TG as a representative example
of current theoretical linguistics. The relevance of the preceding phi
losophical discussion to an evaluation of this particular theory of
grammar lies in the fact that, in accordance with its general posi ti-
vistic assumptions (cf. 3.6. above), TG maintains a conception of langu
age which is demonstrably equivalent to the private-language conception.
4.3.1 . Psycholinguistics
First of all, it is of importance to point out that the question

of whether knowledge is primarily subjective or intersubjective is more
fundamental than the question of whether, supposing (falsely) that know
ledge is primarily subjective, its acquisition requires a more intricate
innate apparatus (= 'rationalism') or a less intricate one (='empiricism'
Consequently, in this more fundamental context, the well-advertised

distinction between rationalism and empiricism disappears (see, e.g.,
Chomsky 1965:47-59).
The same happens to another well-known distinction, viz. that
between mentalism and behaviourism. We have already seen that TG,
while failing to differentiate clearly between grammar and psycho-
linguistics, purports to explain observable events, and similarly to
test its explanatory hypotheses against new observational evidence.
This means that TG, including TG-inspired 'mentalistic' psychology,
views itself as a genuine hypothetico-deductive, i.e., positivistic,
science. The question that may be asked is, where lies the difference
between TG and behaviouristic psychology, which also claims to be a
hypothetico-deductive science dealing with observable events. It lies
merely in the fact that TG postulates more complex systems of theore
tical concepts, that is, an innate linguistic 'theory' and an acquired
'competence' of a particular language, with a view to accounting for
observable linguistic behaviour. It goes without saying that the dif
ference between TG-type psychology and behaviourism is a difference
not in kind, but in degree. In fact, TG-type mentalistic explanations
have turned out to be mechanistic explanations (cf. 3.6. above).
The Wittgensteinian psychology, by contrast, establishes the ir
reducible, qualitative difference between observation and understand
ing, and hence between positivistic and nonpositivistic (i.e., herme-
neutic) explanations of human behaviour. In view of this, one must
be careful not to be misled when Chinara & Fodor (1966) attack Witt
genstein's (nonpositivistic) 'logical behaviourism' in the name of
(positivistic) 'mentalism'. However, the situation is considerably
clarified by the following passage, which illustrates the use by TG
of the 'analogy argument', and hence its particular variant of method
ical solipsism. Chinara & Fodor (1966:414) reject the thesis of the
logical dependence of psychological concepts upon public criteria, and
continue:
The belief that other people feel pains is not gratuitous even on
the view that there are no criteria of pains. On the contrary, it
provides the only plausible explanation of the facts I know about

the way that they behave in and vis-à-vis the sorts of situations
I find painful.
Here we have the lone observer who believes that other people feel
and think - in fact, that there ave other people - because on this be
lief he can best explain to himself the movements and sounds made or
emitted by familiar-looking bodies around him. It can be said that
the TG-inspired psychologists and philosophers represent the methodic
al solipsism in its purest form.
Certain recent developments within TG might seem to call the pre
ceding account into question. Partly in connection with the attempt
to integrate the theory of speech acts into TG, Fodor, Bever and
Garret (1974:1-21) now explicitly recognise the difference between
action and event in a way which amounts to a de facto acceptance of
the thesis of intentionality. However, this apparent departure from
the positivistic standpoint does not change TG's general methodological
position, as I will try to demonstrate in what follows.
Fodor et al. start from the correct observation that human actions,
and in particular acts of speaking, differ significantly from events
investigated by physics in that their occurrence or non-occurrence can
not be predicted in terms of their environment (cf. 4.2.2. above).
However, as they see it, this difference is not as great as one might
think. They view a human being as a mechanism with simple external
states and complex internal states; that is, the internal states of
the mechanism determine its output behaviour to a much higher degree
than its external states do. Therefore, the output behaviour cannot
be predicted merely on the basis of information about the external
states or the environment. However, insofar as the internal states
become known, the behaviour becomes predictable, and in the unlikely,
but theoretically conceivable case where the internal states are known
entirely, the behaviour is entirely predictable. Consequently, the
difference between action and event is, according to Fodor et al.
after all but a matter of degree.
The same conclusion may be reached in the following way. Fodor
et al. (1974:3-4) differentiate between events related to the human

body, for instance pure reflexes, and actions similarly related, by
assuming that these two types of phenomena are underlain by different
types of 'schemas'. And as might be expected, the schemas underlying
actions are assumed to be characteristically more complex than those
underlying events. In fact, Fodor et al. go so far as to identify or
individuate different actions by reference to the schemas which sup
posedly underlie them. For instance, different acts, or act-tokens,
of writing one's name are said to belong to one and the same act-type,
and to differ from any other apparently quite similar act-tokens, be
cause one and the same schema underlies the former act-tokens and other
schemas underlie the latter act-tokens. But this formulation clearly
means putting the cart before the horse. We cannot use something which
we do not know, to identify something which we do know. For instance,
the characteristic property of snow is its whiteness, and not the fact
that we possess - as we certainly do - some psychological mechanisms
which enable us to perceive its whiteness.
The confusion here involved can be cleared up as follows. On the
one hand, I consciously identify acts of writing one's name as what they
are, on the basis of certain public or social criteria. These criteria
are rather self-evident; ordinary people can discuss them and agree on
them quite easily. On the other hand, I can make the additional hypo
thesis that the conscious social criteria which we all use in identify
ing acts of writing one's name are underlain by unconscious psycholo
gical mechanisms of a certain type, to be discovered only by means of
psychological experimentation.
The situation is the same in linguistics. We identify the diffe
rence between 'John drinks bourbon' and 'Does John drink bourbon?' by
means of qui te obvious public criteria, but we do not know those parti
cular mechanisms which make us so identify it. Recent developments in
psycholinguistics have made it overwhelmingly clear that the analysis
of our consciousness, or of the object of our consciousness,58 must be
sharply distinguished from the analysis of the psychological mechanisms
presumably underlying those actions which, or whose results, we are
conscious of. It is simply wrong to assume that linguistic grammars

are as such psychologically real, or are as such identical with mental
grammars. The dichotomy between these two types of grammar is just a
special case of the general dichotomy between conscious and unconscious
means of identifying (results of) actions.
Consciousness is a specifically human phenomenon. The hypotheti-
co-deductive method is unable to differentiate between conscious and
unconscious, or nonconscious, phenomena (cf. the Katz & Bever quotation
p. 8 4 ) . It is perfectly legitimate to offer psychological explanations
modelled upon the hypothetico-deductive method. What must be regarded
as illegitimate, however, is to think - as Fodor et al. do - that appli
cations of this method could bring out the special nature of human beha
viour. In particular, it is impossible that the scientist could hypo
theti co-deductively describe his own behaviour in its entirety, for
"Even ideally refined behaviourism can ... merely explain the behaviour
of the observed, not of the observing scientist" (Schutz 1962:54; cf.
also 2.1. above).
In light of the Katz & Bever quotation referred to above, TG psy
chology, as a relatively sophisticated variant of posi ti vi sti c psycho
logy, might well be taken to represent a form of 'refined behaviourism'.
Since according to this view the scientist can hypothetico-deductively
describe everybody else except himself, we see that TG remains as com
mitted to methodical solipsism as ever.
4.3.2. Theory of Grammar
The distinction between competence and performance has been inter

preted in Chomsky's 'standard theory' of TG in such a way that (know
ledge of) language is primary while (knowledge of) its use is secondary.
This conception of language is at the same time non-functional, viz.
formal, and individualistic, or in other words 'Cartesian'. It has been
expounded in detail by Moravcsik (1967), who claims that such communica
tive skills as announcing or requesting which imply a reference to the
(intentional) situation of language-use do not belong to the primary
linguistics skills, i.e., those which are part of the competence. More-
over, if it is permissible to speak of the use of language at all in

connection with the competence, it can supposedly only be said that one
uses language to articulate one's physical and social environment, and
to 'freely express' one's beliefs and thoughts. Humboldt seems to have
held similar views with respect to the primary characteristics of langu
age, and on several occasions Hjelmselv voices the idea that 'system
is primary with respect to process'.60
Such a formal (and individualistic) conception is demonstrative
ly untenable. A language into which the roles of speaker and hearer
have not been built in as essential constituents of the concept of
speech situation (which in turn is inseparable from the concept of
language itself), viz. a language in which the intersubjective or social
aspect of rules is treated as theoretically secondary, is equivalent
to a private language. Such a language bears no resemblance to natural
languages, and consequently a linguistic theory which either implicitly
or explicitly endorses this private-language conception is inadequate
on factual grounds. This is the case of the Chomsky-type TG. However,
since private languages are not only factually 'queer' but also logic
ally impossible, the Chomsky-type TG is, in this respect, not only
factually inadequate but also logically inconsistent. This incon
sistency is well illustrated by Moravcsik's (1967) claim that since
language use is secondary, hearers could understand sentences without
knowing the conditions for their correct use. What this implies is
in fact that all speakers could be misusing language all the time,
and could apparently also refuse to use language at all, but would
still remain speakers of this language ( but what language?). This
is the fallacy of making the inference from 'x sometimes happens' to
'x might always happen'.
It may be added that Fodor et al. (1974:167-70) try to base the
semantic component of their grammar-conception on the Wittgensteinian
idea that the meaning of a word is its role in a language-game.
This view has no real connection with their theory, however. Rules
are necessarily social entities (cf. 4.2.5. above), but the authors
do not discuss the social aspect of language at all.
As is well known, TG intends to describe the ideal speaker-hear

er's competence within a totally homogeneous speech-community. From
the primacy of the individualistic or private aspect, it follows for
TG that the similarity (or, in the above idealisation, the identity)
of the linguistic rules which differentmembers of the same community
follow is a contingent fact. To be sure, it is explained by a refe
rence to the similar psychological and physiological make-up of the
speakers of a given language, but this, again, is a contingent fact.
The refutation of private languages shows, however, that the near-
identity of the rules in a speech-community is due to their intersub-
jectivity, which in turn is a necessary characteristic of rules.
It is important to realise that there is no boundary between innate
and social aspects of language: The disposition to socialisation is
itself innate; furthermore, as Leontiev (1971) has noted, the social
setting does not merely act as a trigger that actualises innate struc
tures but rather plays an active role in the formation of specifical
ly human, including linguistic, abilities.
From the above, it is evident that for both philosophical and
methodological reasons TG has been unable to draw the distinction
between particular uses of language and the general principles of
language use, of which the latter obviously are ingredients of the
linguistic competence and thus constitute an essential subject matter
for any theory of language.
It has been shown (in 4.2.5. above) that language is logically
interdependent with public use of language. Use of language consists
in actions conforming to intersubjective or socially valid rules.
Intentions are conceptually necessary parts of actions, including
speech acts. Thus every adequate theory of language must take speech
intentions into account. This idea has found a clear formulation in
the theories of language of Grice (1957, 1968), Strawson (1971), and
Searle (1969); (cf. also Itkonen 1972c). It should further be noted
that even within its own framework TG cannot consistently deny the re
levance of speech intentions, since even if - as TG claims - language
were used only to freely express one's beliefs and thoughts, it would
still remain true that the free expression of one's beliefs and thoughts
is an action which, as a matter of logic, requires the presence of the

corresponding intention.
According to this functional or pragmatic conception, the speech
act is the primary unit of language. Within the speech act, one may
go on to distinguish between the level of intersubjective interaction
between speaker and hearer, viz. the level of 'metacommunication', and
the level of a reference to extralinguistic reality, viz. the level of
communication (cf. Habermas 1971a). Here language serves as a proto
type of science: even natural sciences that are concerned with observ
able, ultimately non-human events (= 'communication') must presuppose
the social rules of scientific activity (='metacommunication'). The
traditional concept of 'sentence' proves to be a unit secondarily ab
stracted from the (primary) speech act.
Considerations that bear varying degrees of similarity to those
presented above have led an increasing number of linguists to replace
the formal Chomskyan notion of competence by a notion that involves a
pragmatic or communicative component (cf. in particular Hymes 1974).
It is also interesting to note that the same functional conception can
be applied even in the study of logic (cf. 2.6. above). To my know
ledge the most satisfactory treatment of the pragmatic basis of langu
age is Schneider (1975), which combines the Wittgensteinian approach
with the constructivist approach of the Erlangen school.
Sentences and types of speech acts are equally normative entities:
the former are concepts exemplified by utterances which in turn are
results of act-tokens exemplifying the latter. Speech act grammars
analyse the concept 'correct (type of) speech act' just as sentence
grammars analyse the concept 'correct sentence'. Consequently, there
are no methodological differences between the two types of grammar.
Since sentence grammars are much more well-established than speech
act grammars, it is understandable that I shall concentrate on the
former, in spite of the fact that they have just been seen to be lo
gically secondary with respect to the latter. It. is only when we go
beyond the analysis of the concept of speech act, and start investi
gating how definite groups of people actually speak under definite
circumstances, that we arrive at a methodologically different type of

linguistic description. This is precisely the difference between gram
mar and sociolinguistics or, more generally, between aprioristic and
empirical human sciences (cf. 2.4.-6. and 2.1.-2. above).
5.0. THE CONCEPT OF LANGUAGE
The term 'rule' may be understood in two different ways. On the
one hand, rules are norms which govern intentional social behaviour and
are in turn manifested by this same (rule-following) behaviour. On the
other hand, rules are sentences referring to, or establishing, rules in
the former sense of 'rule'. I distinguish between these two types of
entities by calling them 'rules' and 'rule-sentences', respectively.
It is for instance a rule of English that the definite article precedes
(and does not follow) the noun. This rule is (truthfully) described
by the rule-sentence "In English the definite article precedes the
noun." 65
In this chapter I shall examine how rules exist, and how they are
known. The status of rule-sentences will be analysed in 6.0. (below).
5.1. Ontology: Rules of Language as Constituted by 'Common

Knowledge '
An analysis of rules presupposes an analysis of actions. On this

point we can refer to the results of 4.2.2-3. (above). It is possible
to abstract from every action the intentional element which, properly
speaking, constitutes an action qua action. (This 'intentional element'
is to be understood, not as some psychical substance, but as a 'pattern'.)
What is left, in such a case, is a physical, observable event. There
is no action without the corresponding physical event, which might be
called the 'substratum' of the action. Intentions, which are necessary
constituents of actions, must be at least potentially conscious: to
do something, one must be able to know, at least under some description,
what one is doing. Thus knowledge is, in principle, inseparable from
THE CONCEPT OF LANGUAGE 123
action. We have seen in 4.0. (above) that knowledge is necessarily

social. From this it follows, first, that to be able to do an action
x, a person A must know the action-concept 'x', that is, he must be
able to identify instances of 'x' done by himself or by others. Second,
A must know that others know 'x' and, third, A must know that others
know that he knows 'x'. When this is the case, we say that 'x' is
an object of common knowledge.
In other words, if the following three conditions are true of an
entity χ and of practically any two persons A and Β of a community,
then χ is said to be an object of common knowledge:
1) A knows1 x
2) A knows2 that Βknows1x.
3) A knows3 that Β knows2 that A knows1 x.
If only the first level of this three-level knowledge would hold
of (practically) every member of a community, we would have a set of
solipsistic consciousnesses or, supposing in particular χ to be a rule
of language, a set of private languages. For conceptual reasons, this
is impossible, as was shown in the preceding chapter. If only the
first two levels were to hold of any two members of a community, we
would have a set of cases of Apel's (1973b) 'methodical solipsism'.
Every member of the community would only have a unidirectional access
to every other member: although he could (try to) control others, he
could not conceive of the possibility that he might in turn be con
trolled by them. Hence it is only at the third level that genuine
s o c i a l i n t e r a c t i o n (which is a logical precondition to the existence
of consciousness and language) becomes possible. Still higher levels
of common knowledge (or belief) exist, but they do not change the basic
pattern of social interaction which can be regarded as established al
ready at the third level.
In light of what precedes, the question concerning the mode of
existence of actions must receive a somewhat differentiated answer.
At first glance, actions seem to exist just in space and time. How
ever, the concepts which they exemplify exist at the level of common
knowledge. The same is true of concepts exemplified by events, or of
concepts in general, but it is the characteristic property of actions

that they, in contradistinction to events, are inseparable of the way
in which they are conceptualised (cf. p. 99). This is why there can
be conceptual relations between actions, although there cannot be such
relations between purely spatiotemporal entities.
We must distinguish between normative and non-normative actions;
they might be illustrated, respectively, by speech acts and acts of
opening a window. Every respect in which an action is or could have
been incorrect reveals the existence of a rule; any speech act fol
lows or fails to follow a large number of different types of rules,
from pragmatic to phonological. A normative action qualifies either
as correct or as incorrect. This distinction may occasionally re
semble those between successful and unsuccessful, rational and ir
rational, or usual and unusual actions, yet all those distinctions are in
principle distinct. To the extent that the 'correct - incorrect' dis
tinction cannot be kept apart from the others, the existence of the
corresponding rule is uncertain, a phenomenon which is only to be ex
pected sometimes, given the general relativity of the distinctions in
volved (cf. 4.2.4. above).
To me there seems to be a good sense in which even non-normative
actions can be said to be implicitly normative. All actions are inse
parable from their interpretations; but interpretation, or use of con
cepts, is a clear-cut institution which consists of rules of its own
(cf. p. 42); thus there exists the constant possibility of correct vs.
incorrect interpretation. To illustrate: The action of opening a win
dow certainly does not seem to be normative in the least. But insofar
as it is an action, the one who does it must be able to identify it as
what it is: it is certainly a reasonable requirement that people must
be able to know what they are doing, if they are really doing something,
and not just undergoing certain uncontrolled processes. Therefore, the
one who is opening a window must be able to identify his action; and
this implies at least the possibility ofmisinterpretation. In this
sense, then, all actions are implicitly normative. Moreover, their
normativity becomes explicit when the interpretation in question is

given a linguistic expression: A man who is opening a window, and who
says "I am closing the window", or "I am painting the wall", is, on
the face of it at least, vising words incorrectly.
Because common knowledge is about what everybody knows that every
body knows, or ought to know, it necessarily contains social control
of what is or ought to be known. We have seen in 4.2.5. (above) that
there is a difference between being correct and seeming correct, or,
more generally, between reality and appearance, and that this diffe
rence can be established only by checks independent of each other,
i.e., checks to be provided by a social community.
What is true of actions is, at a higher level of complexity, also
true of rules. Just as abstraction of the intentional element produces
the (physical) substratum of an action, so abstraction of the normative
element produces the substratum of a rule, this substratum consisting
in a regularity of (non-normative) actions; just as the existence of
an action is, as it were, spread over both space-time and common know
ledge, so is the existence of a rule: it is commonly known to be what
it is, and it is manifested by a regularity of spatiotemporal actions,
and is, again, commonly known to be so manifested. I would like to
emphasise that in my opinion rules are commonly known as rules , not as
construed out of something more elementary, to which they could again
be reduced. That is, I consider normativity as not reducible to non
normativity. Actions are 'patterns' of observables, and rules are
'patterns' of actions, but the dimensions of intentionality and nor
mativity, constituted respectively by these two types of 'patterns'.,
are not reducible to ontologically simpler forms.
Since rules exist as objects of common knowledge, they cease to
exist, once they cease to be its objects. As I understand it, there
cannot be rules which no one knows. Empirical sociology is mainly con
cerned with (statistical) uniformities in social behaviour that are not
known to people themselves. In such a case, according to my terminology,
we have to do not with rules but with (causally effective) regularities.
On the other hand, regularities of physical events may also be objects
of common knowledge, but whether they are or not, has no influence upon
their factual existence. To put it more simply, there can be no norma
tive entities where there are no people, whereas there certainly can
be, and are, physical entities where there are no people.
Since I regard a language as a system of rules, I have given here
my account of what it means to say that a language exists. Objections
to the effect that there are no (genuine) rules of language, will be
discussed in 5.3. and 6.1-2.
However, I must add that at the level of common knowledge langu
age does not exist as a system, but rather as a set of rules. It is
the grammarian's task to work out the system in question. In other
words, atheoretical knowledge of language as a set of rules shades off
into theoretical knowledge of language as a system of rules. This is
why, after a system has been proposed, we sometimes have the feeling
that we knew it all along (cf. 11.1. below). This is also the sense
in which it might be said that the grammarian reveals the language as
it 'really' is.
In language, the inseparability of doing and knowing that one
does, is actualised as the inseparability of speaking and knowing that
one speaks; one's knowledge of speaking, or of object language, becomes
expressed as metalanguage. In other words, natural language is charac
terised by the fact that it functions both as object language and meta
language. This linguistic reflexivity is not a logical flaw, but just
the expression of the inherent reflexivity of consciousness in general.
The strict separation of object language and metalanguage, as it has
been carried out in formal logic, is analogous to the research situa
tion in natural science, where knowledge pertains to something which
has no knowledge or is treated as having none (cf. 2.1. above). In
aphasia the loss of metalinguistic capacities means that one's langu
age withdraws from under social control insofar as one is no longer
able to perceive differences between one's own language and the langu
age of others (cf. Schlieben-Lange 1975b). At first glance, this
might appear to be the beginning of a development toward a private
language. Yet what we have here, is in reality no longer a genuine
language, precisely because it lacks (the means of expressing) the re

flective linguistic consciousness. There is no language without con
sciousness, and no consciousness without social control of it. It is
also interesting to note that the social contradiction, the internali-
sation of which is taken to cause neurotic behaviour (cf. 2.3. above),
may characteristically have the form of a contradiction between object
language and metalanguage: the prospective patient continuously receiv
es messages and comments on these messages in such a manner that their
meanings systematically contradict each other (Schlieben-Lange 1975b).
Moreover, the neurosis so created is linguistically manifested preci
sely as the loss of metalinguistic capacities.
Just as knowledge about norms, all scientific knowledge is common
knowledge, except when it is just being newly produced. Of course, it
is not knowledge in the sense that we know which theories are true, but
in the more modest sense that we know which theories have been proposed
and what is, approximately, their degree of uncertainty.
I have presented here one particular way of refuting the so-called
methodological individualism, which has been, and still is, an import
ant trend in the philosophy of sociology (cf. Nagel 1961:535-46; for
an overview, see Lukes 1968). According to this view, institutional
and normative phenomena must be reduced to actions performed and,
possibly, to beliefs held by individual persons.
It has become evident by now why, in my opinion, the program of
methodological individualism cannot be carried out: since a rule exists
at the level of common knowledge, it cannot be analysed as a set of
particular beliefs held by individual persons. As an object of common
knowledge, a rule is the focal point of an extremely complex configura
tion of interdependent knowledge-relationships. Analysing this con
figuration as a mechanical sum of its parts would destroy the very no
tion of common knowledge and replace it by disconnected instances of
subjective; Cartesian knowledge. Methodological individualism is in
fact merely an application of Cartesian or traditionist epistemology.
to sociology, and the falsity of the latter entails the falsity of the
former. (It may be added that methodological individualism has not

even been aware of the fact that actions already constitute counter
examples to its idea of a homogeneous space-time ontology, viz. an
ontology which admits only of the existence of distinct spatiotempo
rally definable occurrences.)
Accepting the autonomous existence of normali vity means accepting
an ontology which is much more complex than any variant of the positi-
vistic space-time ontology. This is as it should be, however, because
it is precisely the most obvious flaw of positivism that it is utterly
uncapable of accounting, first, for the existence of the subject matter
of sciences like philosophy and logic, and second, for the existence
of science in general. Now, an ontology which contains both spatio-
temporality and normativity, as well as the continuum between them,
would make the existence of every type of science understandable. A
philosophy of science incorporating such an ontology might conceivably
construct a general framework into which it can rather naturally place
all sciences, including itself. A good philosophy does not make itself
an exception, and must therefore be reflexive. It may also be noticed
that even if the ontology here outlined is rather complex, it is hardly
more complex than the ontology required by the 'possible worlds' seman
tics (cf., e.g., Quine 1969:153), which positivistically-minded philo
sophers may without any scruples make use of in connection with their
logical work.
I have presented here my conception of the mode of existence, not
just of actions and norms, but of all social phenomena, from institu
tions to tools. Social reality is meaningful, and meaning exists only
as an object of common knowledge; there can be no private meanings. When
science investigatessocial reality, the causes which it may discover
are unknown before their discovery, and hence cannot, at least not qua
causes, be ob.iects of common knowledge. In my terminology, then, struc
tural properties of a society for instance are not social phenomena,
as long as they remain unknown.
Opponents of methodological individualism have always expressed
views more or less similar to what I have been saying here. In particular,
it is customary to describe the equivalent of common knowledge in terms

of spatial metaphors emphasising the interdependence of the instances
of knowledge constituting this notion. Consider, for example, how
Durkheim (1938 [1895]: 11-12) characterises his 'faits sociaux':
Ainsi ce n'est pas leur généralité qui peut servir à caractériser

les phénomènes sociologiques. Une pensée qui se retrouve dans
toutes les consciences particulières, un mouvement que répètent
tous les individus ne sont pour cela des faits sociaux.
And:
... la société n'est pas une simple somme d'individus, mais le

système formé par leur association représente une réalité specifi
que qui a ses caractères propres. Sans doute, il ne peut rien se
produire de collectif si des consciences particulières ne sont
pas données; mais cette condition nécessaire n'est pas suffisante.
I1 faut encore que ces consciences soient a s s o c i é e s , combinées,
et combinées d'une certaine manière; c'est de cette combinaison
que résulte la vie sociale et, par suite, c'est cette combinaison
qui l'explique. En s ' a g r é g e a n t , ense p é n é t r a n t , en se fusionnant,
les ames individuelles donnent naissance à un être, psychique si
l'on veut, mais qui constitue une individualité psychique d'un genre
nouveau ( o p . c i t . , p.127; emphasis added). Voilà dans quel sens et
pour quelles raisons on peut et on doit parler d'une conscience
collective distincte des consciences individuelles (ibid., n.l;
emphasis added) .
However, Durkheim fails to distinguish between institutions and

their causes, as well as between learning an institution and knowing it.
It is at least partly because of this lack of differentiation that he
arrives at his controversial view that 'social facts are things', things,
to be sure, which are not reducible to ontologically simpler phenomena
(cf. the quotations above and also the clarifications he offers in his
Preface of 1901, p.xi).
Mauthner (1923:19) describes the existence of language in even more
graphical terms: "Wo ist also das Abstraktum 'Sprache' Wirklichkeit? In
der Luft. Im Volke, zwischen den Menschen."
It seems to me that Durkheim's term 'conscience collective' and
Hegel's (and, following him, Dilthey's) term 'objektiver Geist' can be
given a clear and sensible meaning, if they are reinterpreted as
equivalent to common knowledge, as here defined. Dilthey (1927:208)

defines his conception, and delimits it against Hegel's, as follows:
Ich verstehe unter ihm [sc. dem objektiven Geist] die mannigfa
chen Formen, in denen die zwischen den Individuen bestehende Ge
meinsamkeit sich in der Sinneswelt objektiviert hat.
Earlier he had stated (p.150):
Indem so der objektive Geist losgelöst wird von der einseitigen

Begründung in der allgemeinen, das Wesen des Weltgeistes ausspre
chenden Vernunft, losgelöst auch von der ideellen Konstruktion,
wird ein neuev Begriff desselben möglich: in ihm sind Sprache,
Sitte, jede Art von Lebensform ebensogut umfasst wie Familie,
bürgerliche Gesellschaft, Staat und Recht. Und nun fällt auch
das, was Hegel als den absoluten Geist vom objektiven unterschied:
Kunst und Religion und Philosophie unter diesen Begriff.
Hence, the 'objective mind' is based on the fact that existing

ways' of thinking and feeling are necessarily common to all.
In conclusion, I want to point out some rather obvious difficult
ies connected with defining the social mode of existence as being an
object of common knowledge. Like all social concepts, common knowledge
is a relative or gradual one, and therefore the existence based on it
is also relative or gradual. (However, there are absolutely clear
cases; cf. 5.3. below). Moreover, even if a norm exists only if it is
known to exist, it is far from clear to what extent it is required, in
addition, that it be manifested in actual behaviour. Norms and ideals
are often worth describing even after they have ceased to become mani
fest in the behaviour of any group of people. It is precisely the
task of cultural history to show the significance of norms and ideals
which are no longer followed or maintained. It would be rash to say,
unconditionally, that such norms and ideals do not exist any more (as
I would say of norms and ideals which have been completely forgotten).
Evidently, one should distinguish here between different types or de
grees of existence; by contrast, in the realm of physical reality some
thing either exists or does not exist. Even more importantly, there
may be norms or ideals which have, as yet, never become manifest in
the behaviour of any group of people, but which we think should do so,
and are therefore definitely worth describing (cf. 6.3. below). Again,
it would be rash to say that such norms and ideals do not exist.
There are also instances where (potential) human products must be

said to exist, even if they are not (yet) objects of any actual know
ledge. When something is known, and when there is a known method to
draw conclusions from it, it must be said that all conclusions exist,
and not just those which someone has actually happened to draw, or
which are known to have been drawn. Similar arguments apply to other
forms of unactualised but actual i sable common knowledge.
Finally, I would like to point out that the notion of common know
ledge must not be confused with Popper's notion of 'third world'. The
latter contains all possible objects of knowledge or thought, of which
those of common knowledge, even in its relativised versions, are a va-
nishingly small portion. The 'third world' is clearly Platonistic, but
Popper insists that it is nevertheless created by man (e.g., Popper
1972:179, n.8). However, it would seem that if the possibility of every
thing that can ever be thought is objectively given, then it must pre
exist all thought.
5.2. Epistemology: the Distinction between Language and
Linguistic Intuition
Certainly no one would deny that the physical reality investigated
by natural science exists objectively . On the other hand, it is only
in a subjective act of observation that one can acquaint himself with
this objective reality. That is to say, one cannot have someone else's
experiences: If I close my eyes, I do not see anything, even if you
keep your eyes open. Consequently, in natural science we are bound to
make the distinction between objective reality and subjective acts of
knowledge pertaining to it. (Of course, I do not mean that the acts in
question are subjective in any absolute sense. They operate with soci
ally-established concepts, and may themselves become an object of in
vestigation, as in psychology of perception).
In linguistics, the corresponding subjective act is called 'lin
guistic intuition'. (I temporarily distinguish between subjective
linguistic knowledge and linguistic intuitions as acts of actualising
this knowledge.) What does linguistic intuition pertain to? The posi-
ti vi sti c answer is that it pertains to spatiotemporal phenomena, i.e.,

actual utterances and reactions to utterances. However, this would be
a misleading answer, as I hope to show in 7.0. It is possible, to be
sure, to observe uttered sounds and written marks as purely physical,
spatiotemporal phenomena. But if they are understood as linguistic
entities, the primary component of what is understood in this (intui
tive) manner consists of the rules which, precisely, constitute that
which is being understood, as a specimen of language. (This view is
argued in detail in Friedman 1975; cf. 7.3. below).
It is undisputable that a grammar investigates correct sentences.
The concepts 'correct sentence' and 'sentence exemplified by spatio-
temporal utterances' are, however, entirely different. On the one
hand, there is an indefinite number of correct sentences which have
never been exemplified in space and time, but which constitute, never
theless, a legitimate subject matter of grammar, just like there is in
chess an indefinite number of correct (combinations of) moves which have
never been actually made, but which must be accounted for by any ade
quate description of chess. On the other hand, there is an indefinite
number of spatiotemporal utterances which do not exemplify correct,
but incorrect sentences. Consequently, even in cases where linguistic
intuition seems to pertain to actual utterances, it pertains primarily
to rules, as defined in 5.1. (above), since utterances are what they are
solely by virtue of the fact that they exemplifyeither correct or in
correct sentences, and what is a correct sentence in a given language
is determined by the rules of this language. The fact that linguistic
intuition pertains to rules of language is even more evident in cases
where one neither hears uttered sentences nor sees written sentences,
but just thinks of, or reflects upon language.
To sum up, rules, and not spatiotemporal occurrences, are the ob
jective reality to which subjective linguistic intuition pertains to.
For instance, the rule expressed by "In English the definite article
precedes the noun" exists objectively, that is at the level of common
knowledge, and my subjective intuition pertains to this rule. This
conception follows logically from the autonomy of the normative di
mension (cf. 5.1.).
I surmise that most linguists find it difficult to accept the dis

tinction between language and linguistic intuition in the form present
ed here. To make it more acceptable, I shall offer analogies from lo
gic and philosophy. This is in keeping with my general claim that
sciences like grammar, logic, and philosophy are qualitatively similar
(cf. 10.0. and 11.0.).
Logical intuition is obviously something which one can have only
for oneself, that is, it is something subjective. On the other hand
logical rules of inference and logical truths like the following are
72
certainly something objective.
(It should be noted that logical truths, just as well as conceptual

truths, are rule-governed in the sense that it is incorrect not to
accept them.) Hence, we have here the same situation as in connec
tion with language: subjective intuition pertains to objective rules.
Logic is a social phenomenon, and its objective existence means exi
stence at the level of common knowledge (cf. 2.6. above). It is, I
think, because Husserl and Russell, on the one hand, were convinced
of the objectivity of logic and, on the other, regarded physical rea
lity as objective par excellence, that they were led to claim, in cer
tain phases of their careers, that logic is somehow 'eternal' and in
dependent of man.
Next, let us consider a very simple example of philosophical ana
lysis: the concept 'father' is identical with 'male parent'. This is
an objective fact. That is, the corresponding sentence certainly de
scribes an objective conceptual relation, which is normative in the
sense that it would be incorrect not to accept it. And yet everyone
can understand this relation only subjectively, on the basis of what
might be called 'conceptual intuition'. (Of course, intuition pertain
ing to conceptual relations is practically identical with linguistic
intuition pertaining to sentences expressing such relations.) There
could be no conceptual, necessary relation between 'father' and 'male

parent', or between any other concepts, if the existence of concepts
as objective, normative entities were denied, that is, if it were claim
ed that there are no objective concepts 'father' and 'male parent', but
only separate subjective acts of conceptual intuition about 'father'
and 'male parent', acts to which their objects would be inherent. On
this interpretation, we would have to say, either that the necessary
relations hold between separate acts, or that there are no necessary
relations. Both alternatives,however, would be equally unacceptable.
Moreover, it is simply wrong to claim that a sentence like "A father
is a male parent" is about an indefinite number of intuitive, subjec
tive acts. This can be seen from the fact that, on such an interpre
tation, it would be impossible to find out the truth or falsity of any
such sentence; or at least, their truth or falsity could only be of
statistical nature. All these views are clearly false, and therefore
the view from which they were derived, namely the thesis of the non-
objective character of concepts, is false, too. This means at the
same time that conceptual relations, and more generally all facts
residing at the level of common knowledge, are not psychological facts.73
Precisely the same argument applies to logical relations obtaining
between formulae. Consider the equivalence '[p&(qVr)]≡(p&q)V(p&r)]'.74
If one does not allow for the objective existence of the formulae, i.e.,
the formula-types, 75 'p&(qVr)' and '(p&q)V(p&r)', it follows that the
equivalence in question should obtain, not between two formulae (to
which subjective intuitions happen to pertain just now), but between
two intuitions (which pertain to the formulae in question). For seve
ral reasons, such an argument is nonsensical, one reason being that it
simply does not make sense to say that the one intuition is a conjunc
tion and the other a disjunction.
The same argument applies also to the issue of linguistic rules
vs. linguistic intuition. Its relevance to linguistic methodology will
be seen when we come to discuss whether linguistic descriptions satisfy
the general requirement upon empirical explanation and testing, viz.
that the antecedent conditions be conceptually independent of the fact
explained or predicted, or that an empirical hypothesis be testable

against new, conceptually independent evidence (cf. 9.0. below).
In grammar, logic, and philosophy, then, temporally definite,
subjective acts pertain to something objective, namely rules exist
ing at the level of common knowledge, whose temporal limits are quite
indefinite. We might say, then, that common knowledge constitutes a
rule, whereas intuition pertains to it. More precisely, to return to
our definition of 'common knowledge', A's three-level knowledge about
x is simply identical with his (subjective) intuition about x. His
intuition about x is his contribution to common knowledge about χ
and, eo ipso, constitutes the latter in part. Since χ is, in turn,
literally created by common knowledge about x, it follows that A's
intuition both pertains to χ and, in part, creates it. - It should
be remembered, however, that common knowledge, and hence the exist
ence of x, is, strictly speaking, independent of (the intuition of)
any particular person.
The objects of linguistic, logical, and conceptual (or 'philoso
phical') intuition differ from those of observation in that the former
are dependent on, while the latter are independent of man. If someone
should feel that the object of intuition, as here defined, cannot be
compared at all with the object of observation, because the latter is,
presumably, 'present' to the agent of the subjective act in quite an
other way than the former, it may be useful to compare intuition, more
specifically, with the memory of (once observed) physical facts. Both
in intuition and in remembering the object of the respective act is
equally 'absent'. (Of course, this is not to deny the general diffe
rence between social and physical data.)
As we have seen already, the similarity between grammar, logic,
and philosophy rests on the fact that in each case we have to do with
an institution, viz. an institution of correct speaking, correct in
ferring, or correct use of concepts (cf. 2.5-6. above). This simila
rity can be now specified more narrowly by pointing out that in each
case we also have to do with intuition pertaining to language, i.e.,
either language as such (as in grammar), or language qua expression
of conceptual relations (as in philosophy), or language, whether natu

ral or formal, qua expression of junctors and operators relevant to
truth-value (as in logic). The basic similarity between all these
types of (objects of) intuition explains why it is so easy to move
from logic and philosophy to grammar, and vice versa, as witnessed by
a book like Davidson & Harman (1972) for instance.
Within TG, the distinction between 'grammatical' and 'acceptable'
corresponds, in a way, to that between language and linguistic intui
tion. To use 'correct' as a superordinate term, it might be said that
'grammatical', as a concept belonging to competence, represents what
is really correct, whereas 'acceptable', as a concept belonging to
performance, represents what is apparently correct. TG defines a
language as the set of sentences generated by the optimal grammar;
such sentences are indeed tautologically 'grammatical'. However, this
is a very odd notion of language. From the unlimited use of recursi-
vity in grammar, it follows that, in this interpretation, speakers are
able to understand only a tiny fraction of their native language. Con
fronted with this result, TG, instead of modifying its notion of 'langu
age', modifies its notion of 'speaker': language, as here defined, is
said to be spoken and understood by an 'ideal speaker'. But this non
entity is simply identical with an axiomatic system (cf. Itkonen 1976a:
208-15).
My notion of language is entirely different from TG's insofar as
it is a social one (cf. 4.0. above). Language is a set of rules exist
ing at the level of common knowledge, and grammar is a (theoretical)
description of these rules, or of this knowledge. My correct— incor
rect distinction obtains at the level of language, not of intuition,
and is in this respect similar to TG's grammatical — ungrammatical dis
tinction; on the other hand, I have no counterpart to the acceptable —
unacceptable distinction. What is correct or incorrect, is commonly
known. The cases where such a knowledge does not exist with any great
amount of agreement are explained by reference to a decrease in social
control (cf. 5.4. below).
THE CONCEPT OF LANGUAGE 13 7
The distinction between language and linguistic intuition, or
between objective and subjective, corresponds to the distinction be
tween reality and appearence, It is equally necessary to maintain this
distinction whether it be in natural science, linguistics, logic, or in
philosophy; we would otherwise be forced to accept the traditionist e-
pistemology which fails to distinguish between, for instance, being
correct and seeming correct, and must therefore be rejected. In most
instances objective and subjective fall together; this fact is express
ed by saying that we come to know in a subjective act what is objecti
vely given. When a group of people is looking at a tree, in most cases
they all agree, without even a shade of doubt, that what they are look
ing at is a tree. Similarly, when a group of English speakers are re
flecting upon whether the English article precedes or follows the noun,
or whether 'cat' means an animal or a number, in most instances they
all agree unanimously that the article precedes the noun, and that
'cat' means an animal. The same applies to the truth of the formula
'[(pVq)&-p] q' and of the sentence "A father is a male parent".
Here disagreement is empirically possible; it is perfectly possi
ble that someone would claim, against the opinion of others, that what
they are looking at is not a tree, but, say, a butterfly, or that 'cat'
means a number. It is precisely this constant possibility of disagree
ment which reveals the existence of a distinction between reality and
appearance, or objective and subjective, even if they de facto fall to
gether in most cases. In instances like those described above, possi
ble disagreement is deemed pathological; it does nothing to relativise
the distinction between objective and subjective. It is conceptually
necessary that disagreement of this kind is the exception, and not the
rule. Our concepts and our consciousness thereof are based on the pos
sibility of social control and would disappear without it (cf. 4.2.5.
above). Disagreement presupposes agreement: "In order to make a mis
take, a man must already judge in conformity with mankind" (Wittgen
stein 1969b: §156).
It would be equally misleading to say that everything is real or
that everything is only apparent. Even when the two fail together -
as they almost always do - they must be kept apart. A grammar inves
tigates a language, or what is correct. Psycholinguistics investiga
tes, i n t e r alia, linguistic intuition, or what seems correct. (Notice
that even if a grammar does not investigate intuition as such, it must
nevertheless rely on it, because it is only through the subjective act
of intuition that one gets hold of language as objectively given.) The
same distinction obtains between logic and psychology of logic: the
former investigates formulae which ave (both intuitively and formally)
valid, whereas the latter investigates formulae which seem valid. The
same distinction recurs, again, between physics and psychology of per
ception: the former investigates what happens as a matter of fact,
whereas the latter investigates what seems, or is perceived, to happen.
One may also notice that reality and appearance are relative to the
point of view: what is apparent from the point of view of grammar, lo
gic, or physics, is real from the point of view of psychology: for
psychology, intuition and perception are real, and what is apparent,
are the various subjective acts by means of which intuition and per
ception are (perhaps mistakenly) identified. Wittgenstein formulates
the decisive importance of the respective points of view in the follow
ing terms:
Is an experiment in which we observe the acceleration of a freely

falling body a physical experiment, or is it a psychological one
showing what people see in such circumstances? - Can't it be
either? Doesn't it depend on its surroundings: on what we do with
it, say about it (Wittgenstein 1967: I, §160).
Labov for example finds it somehow theoretically unacceptable that

subjective intuition is employed to acquire knowledge about language as
a social entity; he labels this view as 'Saussure's paradox' (Labov
1972:185-87). However, there is nothing paradoxical in this procedure.
Quite the contrary, it is conceptually impossible that rules qua rules
could be reflected upon in any other way. Hence, Saussure was perfect
ly right to distinguish between langue, parole, and faculté de langage
and to move the latter two realms of phenomena, to be investigated, re
spectively, by empirical sociology and psychology, outside of unguis-
tics proper, i.e., of grammar, The point is precisely, that even if

language is asocial entity, it is also a normative one, and, as such,
it cannot be the subject matter of a science subsumable under the
general category of empirical sociology; rather, it must be investi
gated by a science falling under the general category of aprioristic
sociology (cf. 2.2. and 2.4. above). What Saussure might be criti
cised for, is that he did not maintain the distinctions involved with
sufficient consequence. For instance, he apparently found it diffi
cult to accept the ontology of autonomous normativity, which led him
to say, occasionally, that langue, though a 'fait social', is of psy
chic nature and exists in the brain (Saussure 1962:32), something
which a 'methodological holist' like Durkheim would never have said.
It is no accident, then, that Durkheim anticipates the subjective
— objective distinction as it has been defined here:
Les moralistes ne sont pas encore parvenus à cette conception très

simple que, comme notre représentation des choses sensibles vient
de ces choses mêmes et les exprime plus ou moins exactement, not
re représentation de la morale vient du spectacle même des règles
qui fonctionnent sous nos yeux et les figure schématiquement: que,
par conséquent, ce sont ces règles et non la vue sommaire que nous
en avons, qui forment la matière de la science, de même que la phy
sique a pour objet les corps tels qu'ils existent, non l'idée que
s'en fait le vulgaire. Il en resuite qu'on prend pour base de la
morale ce qui n'en est que le sommet, à savoir la manière dont elle
se prolonge dans les consciences individuelles et y retentit (Durk-
heim 1938:30-31).
Dilthey (1927:84) emphasises that what is attained in subjective

acts trying to get hold of common knowledge, or objective mind', does
not belong to the category of psychological knowledge:
Das Verstehen dieses [objektiven] Geistes ist nicht psychologische

Erkenntnis. Es ist der Rückgang auf ein geistiges Gebilde von
einer ihm eigenen Struktur und Gesetzmässigkeit.
For instance, to know why a given person uttered a given sentence

at a given moment is a case of psychological knowledge. On the other
hand, to know the (conventional, standard) meaning of the sentence, or
of the speech act, in question is a case of non-psychological know
ledge, i.e., (subjective) knowledge getting hold of (objective) con-
ceptual r e l a t i o n s e x i s t i n g at the level of common knowledge. 75a

Collingwood (1946:218) shows t h a t the same s u b j e c t i v e - o b j e c t i v e
d i s t i n c t i o n applies also to h i s t o r i c a l facts:
For the historian, the activities whose history he is studying

are not spectacles to be watched, but experiences to be lived
through in his own mind; they are objective, or known to him,
only because they are also subjective, or activities of his own.
History exists, necessarily as part of the present, only so long

as it is known to exist: "The historical process is itself a process
of thought, and it exists only in so far as the minds which are parts
of it know themselves for parts of it" (Collingwood 1946:226).
For Collingwood, to be sure, common knowledge, or 'corporate mind'
( o p . c i t . , p.219), is only part of the objective historical facts which
are subjectively known through a process of 're-enactment', or simply
of understanding. Taken as a totality, however, history must be seen
as a type of common knowledge, because it exists only insofar as it is
(commonly) known to exist. This is the most obvious interpretation of
the thesis (which I take to be established beyond doubt) that in social
matters the past is linked to the present in part conceptually, and not
just by a contingent, causal tie. There are less obvious, but still -
I think - correct interpretations of this thesis which all have to do
with social knowledge and meaning: When for example an institution is
being changed, it happens with the (unconscious) collective intention
that this same institution ought to be,and eventually will be, differ
ent in certain respects. It is this (near-)identity of the meaning of
the institution in the process of change, and the knowledge of it,
which ties the different phases of the process of change together by
a semi-conceptual tie.
An individual speaker possesses a dual role as the agent of (sub
jective) linguistic intuition and as a 'supporter' of the common know
ledge constituting (objective) language. To the extent that his par
ticular intuitions are mistaken, he fails to support common knowledge.
He is able, nevertheless, to regain this role, in each particular case,
insofar as he remains responsive to social control. A man's intuitions
disappear with him, but common knowledge is independent of any single

man.
In the preceding argument, I have used the term 'intuition' to
refer to the act of actual ising one's subjective knowledge about some
thing objectively given. From now on, in conformity with the general
usage, I shall not distinguish between subjective linguistic knowledge
and its actualisations. Both types of phenomena will be covered by
the term 'linguistic intuition'.
5.3. Rules of Language and Certainty
In analytical philosophy it is customarily thought that logical

formulae and conceptual sentences, for example "[(pvq)&-p] q" and "A
father is a male parent", are necessarily true, and that their truth
is known with absolute certainty. On the other hand, it is thought
that no sentence which is, in one way or another, about space and time
can be necessarily true or known with certainty to be true. Such sen
tences are conceived as empirical, or testable (in principle) on the
basis of what happens in space and time (cf. 1.1. and 1.5. above).
Since the existence of a given language seems to be an empirical, spa
tially and temporally delimited fact, and since the use of language
seems to be just a process which goes on in space and time, it is main
tained, in particular, that no sentences about language can be known
with certainty to' be true (or false).
In my view such a general conception is entirely false. In 6.0.
(below) I shall demonstrate its falsity particularly in relation to
sentences about rules of language. In this section I intend to show,
with reference to Wittgenstein's philosophy, that there is no basis
for dividing sentences into two such strictly separate classes.
To begin with, mathematical and logical truths are open to sub
jective mistakes just like truths of other types:
I cannot be making a mistake about 12 χ 12 being 144. And now one

cannot contrast mathematical certainty with the relative uncer
tainty of empirical propositions. For the mathematical proposi
tion has been obtained by a series of actions that are in no way
different from the actions of the rest of our lives, and are in
the same degree liable to forgetfulness, oversight and illusion
(Wittgenstein 1969b: §651).
If the proposition '12 χ 12 = 144' is exempt from doubt, then so

too must non-mathematical propositions be (op. cit., §653).
The question 'But mightn't you be in the grip of a delusion now

and perhaps later find this out?' - might also be raised as an ob
jection to any proposition of the multiplication tables (§658).
The special character of mathematical and logical truths supposed

ly resides in the fact that it is impossible to doubt them. However,
other types of truths, notably those about language, share the same
character:
We say: if a child has mastered language - and hence its applica

tion - it must know the meaning of words. It must, for example,
be able to attach the name of its colour to a white, black, red
or blue object without the occurrence of any doubt.
And indeed no one misses doubt here; no one is surprised that we

do not merely surmise the meaning of our words (Wittgenstein 1969b:
§§522-23).
'I don't know if this is a hand.' But do you know what the word
'hand' means? And don't say 'I know what it means for me now'.
And isn't it an empirical fact - that this word is used like this?
And here the strange thing is that when I am quite certain of how
the words are used, have no doubt about it, I can still give no
grounds for my way of going on. If I tried I could give a thou
sand, but none as certain as the very thing they were supposed to
be ground for (§§306-07; cf. also, in particular, §§369-70).
Wittgenstein (§306) fails, however, to make an important distinc

tion. It is certainly a contingent (rather than empirical) fact that the
rule governing the use of 'hand' is what it is, since it is easy to
imagine this rule being replaced by another. More precisely, what we
have here is the contingent existence of a norm. It is in fact a mis
leading shorthand expression to say that "This word is used like this".
What is meant, is: "This word is used like this, if it is used correct
ly", which is equivalent to "This word ought to be used like this". In
the same way, all grammatical statements about language are of course
intended to be statements about correct language. Now, this 'ought'-
element is not empirical, nor can it be so reinterpreted (cf. 7.0. be
low). Wittgenstein realises this in part when he asks: "Is it that
rule and empirical proposition merge into one another?" (op. cit., §309),
although I would place contingent (rather than empirical) and normati

ve elements here at different levels, rather than let them merge into
one another.
In addition, the presence of the normative element explains the
fact which seems to puzzle Wittgenstein, namely that he can give no
grounds for his certain knowledge about how a word is (to be) used.
The normative force of a genuine rule resides precisely in the fact
that one knows what one ought to do. Hence, the force of a rule re
sides in the rule itself, so to say, not somewhere outside of it. It
is futile to argue, on the basis of (non-normative) physiological,
psychological, social or economic facts, that there is a rule of a
certain kind, if its existence is not evident without such arguments.
Conversely, if people are convinced that there is a rule of a certain
kind, it is futile to argue, on the basis of physiological etc. facts,
78
that it does not exist.' Moreover, although all rules have different
kinds of non-normative substrata, it is impossible to redefine them
solely in terms of the latter (cf. again 7.0. below).
From the above, it follows that there can be no qualitative dif
ferences between logi co-mathematical certainty and linguistic certain
ty, the latter being restricted, of course, only to the simplest type
79
of statements about language. What needs to be pointed out in par
ticular, is that in Wittgenstein's opinion linguistic certainty per
tains, not only to semantic aspects - although those are naturally
the most interesting ones from the philosophical point of view - but
also to phonological aspects of language (cf. also my standard example
from morpho-syntax concerning the place of the English definite artic-
le):
We know, with the same certainty with which we believe any mathe
matical proposition, how the letters A and Β are pronounced, what
the colour of human blood is called, that other human beings have
blood and call it 'blood' (Wittgenstein 1969b: §340).
Every language-game is based on words and objects being recogniz

ed again. We learn with the same inexorability that this is a
chair as that 2 x 2 = 4 (op.cit., §455).
It is hardly surprising that linguistic knowledge should turn out

to be at least as certain as logi co-mathematical knowledge. When ex
ploring the limits of possible doubt, Descartes goes so far as to doubt
the most elementary arithmetical truths. However, he does not even come
to think of the possibility of doubting the meaning of those very words
with which he expresses his doubt. It is easy to show that such a doubt
would no longer be rational (cf. Kenny 1975:205). Therefore those lin
guists who, in the name of 'absolute falsiticationism', claim that no
truths about language can be known with certainty, are expressing a
doubt more radical than the Cartesian one, and are, hence, refuting
themselves by reductio ad absurdum (cf. the results of 4.0. above).
The preceding argument may give rise to the following objection:
How can the claim that we know the meanings of words with certainty
be reconciled with the fact that philosophers apparently for ever
dispute about, and thus obviously do not know, the meanings of words
like 'time', 'knowledge', or 'freedom'?
I think we must distinguish here between two types of knowledge,
viz. 'atheoretical' and 'theoretical', to be introduced properly in
8.0. (below). Philosophers know with absolute certainty what parti-
cular kind of uncertainty is connected with a given word; for in
stance, they know for sure that the one connected with (the meaning
of) 'time' differs from the one connected with 'freedom'. To put it
in different terms, they know, atheoretically, many different types
of (correct) uses of 'time' etc., but what they do not know, is the
theoretical framework or system which could put each of these uses
into its proper place. It has been suggested, notably by Wittgenstein,
that there can be in fact no such systems. However, because this claim
is about theory, it is itself of theoretical nature, and therefore its
truth can be known just as little as the truth of theoretical claims
in general.
It follows quite inescapably, then, that we do know, atheoreti-
cally, the meanings of the words of our language (as well as many
other atheoretical facts about language). Chomsky for one is opposed
to this view, claiming that it can be experimentally determined "what
words really mean" (Chomsky 1972:23). Suppose that one of Chomsky's

experiments would reveal that 'dog' really means 'cat'. Of course he
would disregard such an experiment, for the sensible reason that its
result conflicts with what he (intuitively or pre-experimentally) knows
to be the case. But then, clearly, he would not be conducting experi
ments at all but merely imitating, for whatever reason, the terminolo
gy of natural science. To see this, we only need to consider an ana
logous case in physics: Our experiments reveal the existence of wave-
particles, and we accept this result although it patently conflicts
with our intuitive conception of the nature of the universe. But this
is precisely why we have, in physics, to do with genuine experiments.
Notice that Chomsky has abandoned here one of his earlier positions,
according to which "there is no way to avoid the traditional assump
tion that the speaker-hearer's linguistic intuition is the ultimate
standard that determines the accuracy of any proposed ... operatio
nal test" (Chomsky 1965:21).
Wittgenstein is not content with eliminating the traditional di
viding-line between logico-mathematical and linguistic truths. He al
so points out that there are sentences about the external, physical
world whose truth can be doubted just as little as that of the most
secure logico-mathematical or linguistic sentences. Such apparent
ly empirical sentences characteristically possess the property that,
as long as we do not give up the law of contradiction (and stop think
ing rationally), their falsity would necessitate the falsity of prac
tically everything else; and this consequence is unacceptable because
we cannot consistently (i.e., again, without giving up the law of con
tradiction) think that literally everything we (supposedly) know is
false. Wittgenstein (1969b: §52) says:
... 'At this distance from the sun there is a planet' and 'Here
is a hand' (namely my own hand). The second can't be called a
hypothesis. But there isn't a sharp boundary line between them.
For it is not true that a mistake merely gets more and more im
probable as we pass from the planet to my own hand. No: at some
point it has ceased to be conceivable.
This is already suggested by the following: if it were not so, it

would also be conceivable that we should be wrong in evevy state
ment about physical objects; that any we ever make are mistaken.
So is the hypothesis possible, that all the things around us don't

exist? Would that not be like the hypothesis of our having mis
calculated in all our calculations? (op. cit. , §54-55).
Hypotheses or uncertain claims about physical, social, or logical

matters all presuppose indubitable pre-experimental knowledge of the
relevant domains:
I want to say: propositions of the form of empirical propositions,
and not only propositions of logic, form the foundation of all
operating with thoughts (with language)(op. c i t . , §401).
If I say 'we assume that the earth has existed for many years
past' (or something similar), then of course it sounds strange
that we should assume such a thing. But in the entire system
of our language-games it belongs to the foundations. The as
sumption, one might say, forms the basis of action, and there
fore, naturally, of thought (§411).
Seemingly empirical sentences of this type cannot be genuinely

empirical, because we cannot even (consistently) think of what might
count here as disconfirmatory evidence. For a theory or hypothesis
to be empirical "criteria of refutation have to be laid down before
hand: it must be agreed which observable situations, if actually ob
served, mean that the theory is refuted" (Popper 1963:38; cf. 1.1.
above). In the case at hand, this requirement cannot be fulfilled:
If someone doubted whether the earth had existed a hundred years
ago, I should not understand, for this reason: I would not know
what such a person would still allow to be counted as evidence
and what not (Wittgenstein 1969b: §231).
It strikes me as if someone who doubts the existence of the earth

at that time is impugning the nature of all historical evidence
(op. oit., §188).
Taken by themselves, sentences about rules of language do not

seem to possess the property that their falsity would entail the fal
sity of practically everything else. This is due to the fact that
the existence of particular languages is contingent in a way in which
the existence of our world-picture or of our logic is not. We have

seen, however, that the falsity of sentences claiming certain know
ledge of rules of language would indeed necessitate the falsity of the
sentences claiming any kind of certain knowledge; and this is clearly
an unacceptable consequence. As for the contingent vs. necessary cha
racter of existence, the difference between language and world-
picture, or logic, is not as great as it might seem at first glance.
It is an obvious fact that different cultures may possess world-pic
tures which differ in quite fundamental respects, and that even the
scientific world-picture may change considerably in the course of
time (cf. Wittgenstein 1969b: §94-97). Similarly, even if we cannot
imagine a logic fundamentally different from ours, we can imagine
the possibility of such a logic; in this special sense, then, even
the existence of our logic and mathematics is 'contingent':
Now can I prophesy that men will never throw over the present
arithmetical propositions, never say that now at last they know
how the matter stands? Yet would that justify a doubt on our
part? (op. cit., §652; cf. also Stroud 1966).
Similarly, the fact that English may (in fact, will) change in
the future, does not in the least call into question my knowledge of
English such as it is today.
Because mistakes and delusions are always empirically possible,
we reach a conclusion similar to the one we reached in connection with
the identification of mental phenomena (in 4.2.2. above): it is con
ceptually necessary that people, when dealing with the stock of every
day knowledge, are usually right in what they say, think or do, and
there are objective criteria for deciding this question; but in each
particular case it is possible that an individual person is making a
mistake:
But since a language-game is something that consists in the re

current procedures of the game in time, it seems impossible to
say in any individual case that such-and-such must be beyond doubt
if there is to be a language-game - though it is right enough to
say that as a rule some empirical judgement or other must be be
yond doubt (Wittgenstein 1969b: §519).
At the level of language, what is known with certainty, are rules

of language as defined in 5.1.-2. (above). Truths like "In English the
definite article precedes the noun", "[(pvq)&-p] q", and " 2 x 2 = 4"
are exempted from doubt and used as criteria to find out whether or not
someone understands English, logic, or arithmetic. If someone doubts
the truth of such and similar sentences, this tells us nothing about
them, but quite a lot about the one who is doubting their (necessary)
truth. Their special character is not due to the feeling of certainty,
which one is likely to experience when thinking about them; such a feel
ing may be ill-founded, as the constant possibility of mistake or de
lusion demonstrates. Rather, this character is due to the objective
fact that there is no way to falsify them, a fact, to be sure, which
arouses, or is accompanied by, feelings of certainty.
From the fact that logical truths are no more certain than some
truths about the external world, it does not follow that logic is an
empirical science (Wittgenstein 1969b: §98). The same argument applies
to grammar. Notice in particular that among the sentences which speak
about the external, physical world, the unfalsifiable ones are clearly
the exception; they pertain to the most fundamental properties of our
world-picture. As regards sentences about language, the situation is
entirely different: Rules constitute a language, and are known with
81
certainty; (atheoretical ) rule-sentences, each of which describes a
single rule, are accordingly known with certainty to be true (or false),
which means that they are unfalsifiable; consequently, a representati
ve number of sentences about language, namely those which describe lan
guage most directly, are similar to logical sentences and different
from the vast majority of sentences about the external world. It must
be added, to anticipate what I am going to say below (in 8.3.), that
a theoretical grammar which tries to systematise for example 50,000
rules at the same time, is definitely not known to be either true or
false. As has been noted before, this is a universal property of all
theories. I consider a theory whose truth is not uncertain, or open
to doubt, as a contradiction in terms.
The precise meaning of a sentence, and hence the precise degree

of certainty attached to it, depends on the context in which it is
uttered. Even the most secure sentences may appear doubtful when they
are imagined as being uttered in exceptional circumstances; for instan
ce, " 2 x 2 = 4" is false, when it is agreed that '2' means '3', and
it is meaningless when it is agreed that '2' means 'cat'. Similar
even if less drastic considerations have been used in current linguist
ic discussion to support the view that linguistic facts, the meaning
of words and the correctness of sentences, for instance, can never be
known with certainty. It is in this vein that Bever (1970:346) urges
us to "give up the belief in an 'absolute' intuition about sentences".
I regard this position as definitely false. It is a psychological
fact that contexts can be created where people come to doubt their
senses and even their reason. But this fact does not relativise in
the least the objective existence of physical reality and of logic,
or our (common) knowledge thereof. Analogously, there may be contexts
in which the correctness and meaningful ness of even the most ordinary
sentences can be made appear doubtful, but this does not alter the
objective fact that they are correct and meaningful. The assumption
underlying this claim is that there is something like the 'standard'
or 'normal context' in which sentences about different realms of phe
nomena are, and ought to be , considered (cf. Wittgenstein 1969b: §27).
In linguistic matters, this normal context is the one which has serv
ed as the basis for grammar-writing at least during the last two
millenia. During that time, this context, in which basic metalin
guistic questions like "What does X mean?" or "Is Y correct?" are,
and are known to be, asked and answered in a standard fashion, has
proved its worth, primarily in language-teaching.
First, the existence of a well-established context in which rules
of language are known with certainty, is an undisputable fact. Second
ly, it is a justifiable fact, because grammar cannot be replaced by
psycho- or sociolinguistics, which is another way of saying that lin
guistic normativity cannot be done away with (cf. 7.0. below). It
goes without saying that, as a social phenomenon, this normal context
cannot be defined in precise terms (cf. 4.2.4. above). Yet we cannot

help viewing language within it. It is precisely within this context
that we are convinced, for instance, of the correctness of the senten
ces "John is easy to please" and "John is eager to please". The firm
ness of our conviction is evident from the fact that if a psycholin-
guistic experiment produces evidence to the contrary, we conclude that
the experiment - and not the sentences - is incorrect (cf. the discus
sion of Chomsky above).
One final remark. Rules of language, as defined here, are certain
ly known or knowable. But let us assume, for the sake of argument,
that there is an English-speaking person who is not able to grasp the
truth of "In English the definite article precedes the noun". Does
this fact relativise the truth of this sentence, or the existence of
the corresponding rule? - Not at all. We can safely assume that there
is an English-speaking person unable to grasp the truth of "A father
is a male parent", without having to draw the conclusion that this
sentence is not conceptually, but only probably true. The mere fact
that we, as linguists, know the truth of rule-sentences, would already
suffice to distinguish them from empirical hypotheses the truth of
which no one can know (cf. 6.1. below).
Coseriu is one of the few linguists who would agree with what I
have been saying here:
Gelegentlich wird behauptet, das Sprechen sei eine 'unbewusste'

Tätigkeit oder die Sprecher 'seien sich' der Normen der Sprache,
die sie sprechen, 'nicht bewusst'; das ist jedoch ein unglücklicher
Gedanke, der abzulehnen ist (Coseriu 1974:49).
Denn das sprachliche Wissen - das Sprechen- und Verstehenkönnen -

ist kein theoretisches Wissen, das heisst, es lässt sich nicht
oder zumindestens nicht in allen seinen Teilen motivieren. Dennoch
ist in jedem Sprecher, der seine Sprache spricht, ein klares und
sicheres Wissen (op. cit.,p.50).
And finally:
In Wahrheit aber sind sich die Sprecher des Systems und der soge
nannten 'Gesetze der Sprache' voll bewusst. Sie wissen nicht nur,
was sie sagen, sondern auch, wie es gesagt wird (und wie es nicht
gesagt wird); auf eine andere Weise könnten sie zumindest nicht
sprechen. Andererseits geht es sicher nicht darum, das Sprach

instrument zu 'verstehen' (was Sache des Sprachwissenschaftlers
ist), sondern darum, es anwenden zu können, die Norm erhalten
(nachschaffen) und in Uebereinstimmung mit dem System Sprache
schaffen zu können (op. cit., p.51).
5.4. Rules of Language and Social Control
Up to now I have been speaking exclusively about linguistic know

ledge which is both general and certain. Knowledge of this type (which
must of course be of normative, conceptual nature) pertains to rules
of language. On the other hand, it is clear that all linguistic know
ledge is not of this type and thus does not pertain to rules of lang
uage. Uncertain linguistic knowledge may be either a) atheoretical or
b) theoretical.
There are two principal types of uncertain atheoretical knowledge
of language. Each linguistic change is characterised by the fact that
the rules undergoing the change hold only approximately: when a (rule-
governed) entity A is changing into, or is being replaced by, a (rule-
governed) entity B, there is a period during which it is impossible to
say that either A or Β is definitely correct or definitely incorrect.
I would say that in such cases the social control of rules has decreas-
ed. Where this happens, statistical description of factual occur
rences, that is empirical description, is in order even at the level
of grammar. 83
Retrospectively it can be said that in the case of genuine rules
of language, discussed in the preceding sections and in 6.1.-2. (below),
social control is absolute. When this is, and is (commonly) known to
be, the case, consideration of factual occurrences is just as unneces
sary as it would be in description of any well-established game or in
stitution, and - it should be noted - such a description is of a con
ceptual , and not an empirical, nature.
The existence of linguistic change and of the concomitant lack of
social control brings about a qualification of what has been said in
this chapter so far, but it does not constitute an objection to it.
First, it is obviously not the case that all rules of a language are
changing at the same time. Here as elsewhere, uncertainty presupposes

certainty; rules which hold only approximately presuppose the existence
of rules which hold absolutely. Grammatical description traditionally
has concentrated on the latter kind of rules. Secondly, and more import
antly, the uncertainty typical of linguistic change always obtains with
in the limits of absolute c e r t a i n t y : the uncertainty about the correct
ness vs. incorrectness of the two entities A and B, of which the one is
in the process of supplanting the other, is to be contrasted with the
absolute certainty with which we know that all other (possible) entities
D, C, etc. are incorrect. This is why linguistic change, or linguistic
variation, relativises only to a quite limited extent the general the
sis about absolute atheoretical knowledge of language.
The second instance where social control of atheoretical linguist
ic knowledge, or of rules of language, decreases, is represented by ex
traordinary use of language. It is indeed a self-evident truth that,
where something exceptional is being done, rules must prove insufficent.
Therefore it is only to be expected that social control will decline,
and intuitions will diverge, to the extent that sentence-types not
rooted in actual use of language are being taken into consideration
(cf. 4.2.5. above).
In current theoretical linguistics this problem has acquired a
central importance. Transformationalists originally based their ana
lysis on such unequivocally correct sentences as "The man hit the
ball" and "John found the boy studying in the library". Little by
little, however, they have been driven to analysing more and more
complex or outlandish sentences, of the type "John reminds me of
himself" or "What's this outmoded tack doing being taken on a
complex social issue like ureic deriboflavinization?", with the re
sult that it is increasingly difficult, and often impossible, to
reach a general agreement concerning correctness, synonymy, ambiguity,
etc. of sentences under discussion (cf. Botha 1973:178-85 for elabo
ration). In fact, far-reaching theoretical claims have been (alle
gedly) refuted simply because of the unreliability of the intuitive
data on which they are based. This situation, which is rightly con-
sidered intolerable (Labov 1972:191-202), can be avoided easily enough,

however. Where linguistic knowledge is unreliable, that is, where
there are no absolutely binding rules of language, it is natural to
take recourse to observation of actual speech and/or to psycholinguis-
tic experimentation. These alternatives have been espoused, for exam
ple, by Labov and Bever respectively. It must be emphasised, however,
that the unreliability of some intuitions by no means entails the un
reliability of all intuitions. We have already seen that this falla
cious inference has been made at least by Bever when he urges the re
jection of an "absolute intuition about sentences" (cf. 5.3. above).
- More generally, it is hardly a healthy development that sentences
like "John reminds me of himself" become increasingly an object of
linguistic analysis. In cases like this, "language goes on holiday"
(Wittgenstein 1958: I, §38). Since there are no reliable criteria
(which could only be supplied by actual practice, actual use of langu
age), these sentences "hang in the air" (op. cit., I, §380), and any
thing whatever can be said about them.
Linguistic change, or more generally linguistic variation, and
extraordinary use of language are, then, the two cases where atheore-
tical linguistic knowledge is less than certain, or where the social
control of such knowledge is less than absolute. The possibility of
spontaneous change is a necessary precondition for the continuous
functioning of language, and distinguishes natural language from such
artificial normative systems as formal logic or the game of chess.
Moreover, linguistic change represents the exact point at which ling
uistic normativity and linguistic spatiotemporality contact each other,
or merge into each other. For my general conception of science, such
a point is of absolutely crucial importance since it provides the na
tural link between the empirical sciences (cf. 1.0, and 2.1.-2. above)
and conceptual analyses in the widest sense (cf. 2.4.-6.).
On the other hand, theoretical linguistic knowledge is always un
certain, which means that grammars and any other kinds of theories ex
pressing such knowledge are always liable to falsification (i.e., fal
sification 3 ). Falsifiability is indeed a defining property of theories.
It seems to me that Russell & Whitehead's axiomatisation of proposi-

tional logic might be mentioned as an example of a scientific descrip
tion which starts out as a theory and ends up as an 'instrument of
language' (cf. Wittgenstein 1967: II, §29); that is, it is no longer
subjected to doubt or revision, but is used, instead, to express and
explore doubt about other matters (cf. also 10.0. below). It goes
without saying that there are intermediate cases between clear cases
of what is atheoretical and what is theoretical.
It makes sense to ask for additional criteria of knowledge only
when knowledge is less than certain. This means that at the (atheore
tical) level of the data of grammar, such criteria, to be provided by
psycho- and/or sociolinguistic research, can be made use of only when
investigating entities exhibiting (diachronic/social/geographic) varia
tion or representative of extraordinary use of language. At the (the
oretical) level of grammar, on the other hand, it is always possible
to make use of psycho- and/or sociolinguistic evidence in order to
decide which one among - using purely grammatical criteria - equally
good grammars is the best one. In addition, it is possible (though
not necessary) to modify a proposed grammar on the basis of psycho-
and/or sociolinguistic information. Here we have a working inter
action between empirical theories and conceptual analyses.
6.0. THE BASIS OF THE NONEMPIRICAL NATURE OF GRAMMAR
A grammar is a theory of a language. A theory is empirical if,

and only if, it is testable on the basis of space-time entities, i.e.,
events or actions. There are two ways in which a theory may be non-
empirical: Either it does not deal with space-time entities at all, in
which case it cannot of course be falsified spatiotemporally. Or it
deals, at least apparently, with space-time entities, but is formulat
ed in such a way as not to be falsifiable by them.
Philosophical and logical theories are nonempirical in the former
sense. Within theoretical sociology, moreover, it is generally agreed
that rules are not spatiotemporal, but conceptual entities. Compare
the following statements:
Zunächst kann die 'Norm': die Spielregel also, als solche zum
Gegenstand rein gedanklicher Erörterung gemacht werden (Weber
1968:337; emphasis added).
. . . the sociological concern with rules, as also with the roles

which they govern, is a concern with forms of behaviour which
can be assessed as correct or incorrect performances (Ryan 1970:
137; emphasis added). Thus the regular causal sequence has
[in connection with rules] an 'inside' to it, namely the concep-
t u a l , rule-governed sequence (op. cit., p. 140; emphasis added).
To know a rule is to know which entities exemplify, or fail to

exemplify, the concept 'correct entity E', whether or not they occur in
space and time. Hence knowledge of a rule is knowledge of a (certain
kind of) concept; it is not knowledge of what occurs or will occur in
space and time. This property of (the knowledge of) rules has its lin
guistic counterpart in the fact that a sentence describing a rule, viz.
a rule-sentence, can be neither confirmed nor falsified by what occurs
or will occur in space and time.
From t h i s c h a r a c t e r i s a t i o n of the standard type of r u l e i t follows

t h a t f o r the same reason as philosophical and l o g i c a l theories ( c f . above)
theories which i n v e s t i g a t e only rules could not possibly count as empi¬
rical .
In 5.0. (above) I made i t appear at l e a s t p l a u s i b l e t h a t rules of
language are s i m i l a r to genuine r u l e s . In the present chapter I w i l l
attempt to show that rule-sentences describing (non-variable) rules of
language and those describing rules of game d i f f e r equally from empirical
hypotheses in t h a t they cannot be f a l s i f i e d on the basis of space-time
entities. In f a c t , I w i l l demonstrate t h a t they are u n f a l s i f i a b l e tout
court, which means t h a t they a r e , i n t h i s technical sense, 'necessarily
t r u e ' or ' n e c e s s a r i l y f a l s e ' . This r e s u l t should not be s u r p r i s i n g ,
given t h a t (non-variable) rules are known with certainty ( c f . 5 . 3 . above)
The nonempirical nature of (synchronic) grammars follows now conclusive-
l y from the nonempirical nature of rule-sentences, if (and only i f ) it
can be shown t h a t grammars indeed i n v e s t i g a t e o n l y , or mainly, rules f i -
guring as denotata of nonempirical rule-sentences. This l a t t e r point,
viz. the i n e l i m i n a b i l i t y o f l i n g u i s t i c n o r m a t i v i t y qua o b j e c t o f gram-
matical d e s c r i p t i o n s , w i l l be proved i n 7.0. (below).
6.1. The Difference Between Rule-Sentences and Empirical

Hypotheses
Our discussion rests on the fundamental distinction between regu-

larities in nature and rules of human behaviour.85 It is an axiom of the
philosophy of natural science that a universal hypothesis referring to
a presumed regularity is falsified, if there occur counter-instances to
it (cf. Popper 1965:41, and Hempel 1965:39-40); for example the hypothe
sis "All pieces of metal expand when heated" is falsified, if we find a
piece of metal that does not expand when heated. On the other hand, a
sentence referring to a rule is not falsified simply because there occur
(what looks like) counter-instances to it. Consider the sentence "In
poker a full house beats a flush". This sentence is not falsified by the
fact that in one particular game a player with a flush takes in the pot
even though someone else is holding a full house. Such a performance is
THE NONEMPIRICAL NATURE OF GRAMMAR 157
incorrect, whereas the rule-sentence is about correct performances o n l y .

I t i s a remarkable f a c t t h a t behaviour v i o l a t i n g a r u l e does not f a l s i f y
the corresponding rule-sentence. The reason is t h a t what one does has
no d i r e c t r e l a t i o n (although i t c e r t a i n l y has some r e l a t i o n ) to what one
ought to do; and a rule-sentence is p r e c i s e l y about what one ought to do.
In other words, a rule-sentence is about possible ( c o r r e c t ) actions which
ought to be done, and not about f a c t u a l a c t i o n s , whether c o r r e c t or i n -
c o r r e c t , which are done. (Of course, a r u l e would cease to e x i s t - i n
any strong sense of ' e x i s t e n c e ' - i f c o r r e c t actions conforming to it
were no longer done as a matter of f a c t ; but t h i s is a d i f f e r e n t question.)
Now since counter-instances are simply i r r e l e v a n t , we cannot even s p e c i -
fy the circumstances in which our rule-sentence could be taken to be f a l -
sified. But t h i s means that rule-sentences do not s a t i s f y the most basic
requirement imposed on empirical hypotheses and t h e o r i e s , according to
which "criteria of refutation have to be l a i d down beforehand: i t must
be agreed which observable s i t u a t i o n s , i f a c t u a l l y observed, mean t h a t
the theory i t r e f u t e d " (Popper 1963:38).
This d i f f e r e n c e between empirical hypotheses and rule-sentences has
been c l e a r l y recognised by Ryan (who, to be s u r e , f a i l s to d i s t i n g u i s h
between rules and rule-sentences):
A causal generalization has only one task to fulfil, namely tell

ing us what will and will not happen under particular conditions,
irregularities are thus falsifying counter-examples to the causal
law. But rules are not falsifiable in any simple way - except of
course that it may be false to say that there is a rule - and
breaches of a rule are errors on the part of those whose beha
viour is governed by it (Ryan 1970:141).
By now, I t h i n k I have established beyond reasonable doubt t h a t

rule-sentences describing rules of game ( e . g . , rules of poker) and em-
p i r i c a l hypotheses describing (presumed) r e g u l a r i t i e s in nature differ
from each other i n t h a t the l a t t e r a r e , and the former are n o t , falsi-
f i a b l e on the basis of particular observable space-time events or situ
ations. In what f o l l o w s I am going to show t h a t rule-sentences describ-
ing rules of language are i n every r e l e v a n t respect s i m i l a r to r u l e -
sentences r e f e r r i n g to rules of game and a r e , hence, d i f f e r e n t from em-
pirical hypotheses.
Consider our previous rule-sentence " I n English the d e f i n i t e ar-
t i c l e precedes the noun". This sentence is not f a l s i f i e d even i f we
should come across an utterance l i k e " G i r l the came i n " : such an u t t e r -
ance is incorrect whereas our rule-sentence is about c o r r e c t utterances
(and sentences) o n l y , On the other hand, our rule-sentence cannot of
course be f a l s i f i e d by an utterance l i k e "The g i r l came i n " : such an
utterance is c o r r e c t , or such as the rule-sentence says i t ought to be.
Consequently, the rule-sentence can be f a l s i f i e d n e i t h e r by i n c o r r e c t
utterances nor by c o r r e c t u t t e r a n c e s , which means i n e f f e c t t h a t it
cannot be spatiotemporally f a l s i f i e d at a l l . 'Empirical' equals 'spa-
tiotemporally falsifiable'.86 Since our rule-sentence is spatiotem-
p o r a l l y u n f a l s i f i a b l e , i t must be nonempirical. Moreover, according to
one common d e f i n i t i o n , 'necessarily t r u e ' equals ' u n i v e r s a l and u n f a l s i -
fiable'.87 Since our rule-sentence is both universal (cf. n.86) and
unfalsifiable (cf. immediately above), it follows, then, that it is
also necessarily true in the present, technical sense. - Before we
accept this conclusion, I shall dispose of one rather obvious objection
that may be brought up.
One might admit that in the present state of English our rule-sen
tence about the definite article, and all similar rule-sentences (cf.
6.2.), are in fact unfalsifiable. But - one might argue - it is clearly
possible to imagine a state of English in which phrases like "girl the"
would be correct. In this imaginary English, then, our rule-sentence
would apparently be false, which would mean that it is not unfalsifi
able, or necessarily true, after all.
There are several reasons why this argument cannot be upheld.
First; the rule-sentence is here supposed to be falsifiable by the
existence of a rule differing from the one which exists in fact. It
will be recalled that an empirical hypothesis is falsifiable only by
spatiotemporal events or facts. But it is not a spatiotemporal fact
that a language contains this or that rule. This can be seen most di
rectly from the definition of rule, which says that rules exist only
at the level of common knowledge (cf. 5.1. above). Moreover, I shall
show in 7.0. (below) that there is no possible way to redefine rules in

spatiotemporal terms.
Second; the argument against my thesis, viz. against the nonempi¬
rical nature of rule-sentences, is valid only if it is able to establish
a complete analogy between rule-sentences and empirical hypotheses.
There can be no such analogy, however, because it would mean that hy
potheses of physics could be falsified only by the change of those
physical regularities to which they were supposed to refer. But this
is of course not how hypotheses of physics are universally understood
as being falsified or falsifiable. Besides, the whole notion of a
change of (physical) regularities, or 'laws of nature', is very suspi
cious.
Third; consideration of changes which a state of language might
conceivably undergo conflicts with the fundamental attitude of syn
chronic -grammatical description. Analogously, a description of the
rules of poker must disregard the possibility that they will perhaps
be changed in the future. A description of the present rules of poker
cannot be falsified by any future changes, since it is the express pur
pose of such a description to describe the rules before any change has
occurred. Therefore, a synchronic rule-sentence which could be falsi
fied only diachronically, cannot be falsified at all. The sentence
"In English the definite article precedes the noun", which is intended
to refer only to English as it is today , and which is known with abso
lute certainty to refer truthfully to it, cannot be falsified by refer
ence to some other state of English, past or future, which it is defi
nitely not intended to refer to. Notice also that if English were to
contain a rule according to which "girl the" could be correct, then we
would know this rule with the same certainty as we know the actual rule
oo
today. - We have seen that a similar situation exists even in logic

and mathematics: it is possible that the future rules of inferring and
calculating will be different from the present ones, but this is no
reason to doubt the latter today (cf. p.147)
The sentences which we have been examining here might be reformu
lated as universal implications e.g. in the following way: "For all x,
if χ is a piece of metal and is being heated, then χ will expand".
"For allxand for all y, if χ is a full house and y is a flush in poker,

then χ beats y", "For all χ and for all y, if χ is a (correct) English
definite article and y is a noun correlated with x, then χ precedes y".
In the first implication the relation between the denotata of the ante
cedent and of the consequent is an empirical one, as is shown by the
fact that the antecedent may conceivably be true even if the consequent
is false (cf. 1.5. and 4.2.2. above). On the other hand, in the latter
two implications the relation between the denotata of the antecedent
and of the consequent is a conceptual one: it is impossible for the
antecedent to be true if the consequent is false, because it is a de
fining property of any full house and any flush, as correctly under
stood, that the former beats the latter, just as it is a defining pro
perty of any correct definite article in English that it precedes the
noun correlated with it. If in a particular game a full house does not
beat a flush, the game is not being played correctly. If in a particu
lar utterance of an English sentence a definite article does not pre
cede the noun correlated with it, English is not being spoken correct-
ly.89 - We see that rules are in conflict with the thesis of atomism:
they determine properties which are internal, or conceptually tied, to

entities under discussion, e.g., definite articles or full houses. This
only repeats the result we have reached above.
The rule about the place of the definite article in English obvious
ly has no exceptions, and is also known to have none. But - it might
be asked - is it not true that most rules of language do have excepti
ons, and do these not constitute falsificatory evidence against the cor
responding rule-sentences? Answering this question gives me an oppor
tunity to specify my notions of rule and rule-sentence more precisely.
As an example of a rule of language, the rule about the English
definite article is unnecessarily abstract or general. For my purposes,
it is sufficient to divide this rule up into a set of lower-level rules
which determine for each particular noun of English that, when the noun
is correlated with the definite article, the article precedes, and does
not follow, it. The corresponding rule-sentences are of the type "In
English the definite article precedes the word man", "In English the de-
finite article precedes the word woman", etc. It is clear that such
sentences refer to genuine rules, and not to particular spatiotemporal
events, because they refer to word-types (or word-concepts), and not
to any particular one from among those potentially infinitely many
spatiotemporal word-tokens which may exemplify any given word type. 90
My rules and rule-sentences are primarily meant to exist at this
level of concreteness . I take it for granted that ordinary people,
not to speak of professional linguists, are able to know with certain
ty also rules of a more abstract type. But I have no need for such
rules, because I am interested only in proving that in language there
are genuine rules; and the most concrete kinds of rules are best suit
ed for my purpose. From this requirement of concreteness it follows,
among other things, that at the linguistic level examined here there
are absolutely no exceptions because the (traditional) 'exceptions'
ave themselves rules. Consider the rule according to which the end
ing of the English plural is -s (and not -t, -rk, -buruburu, etc.).
I break down this rule into a set of rules which state the correct
plural forms of each particular English noun. According to one of
these rules, the correct plural for boy is boys', according to another,
the correct plural for man is men. To repeat, I am sure that in addi
tion to knowing the rules which determine the correct plural form for
each noun taken separately, ordinary speakers are also able to know
the generalised rule according to which -s is the correct ending for
almost all nouns. I am also sure that ordinary speakers can be guided
into knowing even more general types of rule.
I am, however, not concerned here with establishing the boundary
line between atheoretical and theoretical, or what can, and what cannot
be known. This is obviously an empirical question, which can be an
swered only on the basis of psychological experimentation (cf. n.90);
and the answer must certainly be formulated in statistical terms. Here
as elsewhere, I am solely concerned with conceptual questions. (This
should be kept in mind throughout my argument. The methodology of an
empirical science - not to speak of the methodology of grammar - is
not itself an empirical science.) Accordingly I am concerned here with
e s t a b l i s h i n g the existence of those absolutely clear cases t h a t may be

taken as the two extremes of the continuum leading from a t h e o r e t i c a l
to t h e o r e t i c a l : At the one end, we have rule-sentences l i k e " I n English
the d e f i n i t e a r t i c l e precedes the word man" or " I n English the p l u r a l
of boy is boys"; at the other end, we have grammatical hypotheses like
the ' s u b j e c t r a i s i n g t r a n s f o r m a t i o n ' or the 'A-over-A p r i n c i p l e ' (cf.
9.5. below). There can be, and a r e , f a l s i f i c a t o r y counter-instances to
such hypotheses, f o r the obvious reason t h a t n e i t h e r t h e i r scope nor
t h e i r t r u t h i s known w i t h c e r t a i n t y ( c f . 5 . 3 . - 4 . above). On the other
hand, we have seen t h a t there can be no counter-instances (or 'excep-
tions') to rule-sentences, the term 'rule-sentence 1 being used here to
r e f e r to rules of the most r e s t r i c t e d and concrete type ( f o r more exam-
ples, c f . 6 . 2 . below). Moreover, the counter-instances to (theoretical)
grammatical hypotheses are rules - and not events or occurrences -
91
which the hypotheses f a i l i n one way or another to account f o r .
A rule-sentence l i k e " I n English the d e f i n i t e a r t i c l e precedes the
word man" cannot be f a l s i f i e d because i t is known to be t r u e . On the
other hand, a rule-sentence l i k e " I n English the d e f i n i t e a r t i c l e f o l l o w s
the word man" i s known to be f a l s e , but i t can be said to be falsified
j u s t as l i t t l e as f o r instance the sentence "A f a t h e r is a female parent"
can be.
U n t i l now we have been discussing the concept of f a l s i f i c a t i o n . Next,
l e t us consider the concept of c o n f i r m a t i o n . I t is an axiomatic t r u t h of
the philosophy of natural science t h a t empirical universal hypotheses
can only be confirmed to a higher or lower degree, which means t h a t they
may eventually be assumed to be t r u e ; but they can never be conclusively
confirmed, or v e r i f i e d , i . e . , known to be t r u e . By c o n t r a s t , true r u l e -
sentences are of course known to be t r u e : t h i s is p r e c i s e l y the reason
why they are u n f a l s i f i a b l e . To deny t h i s is to claim t h a t even when
examining English as we know i t , we can only 'observe' p a r t i c u l a r cor-
r e c t occurrences "the man", "the man", "the man" e t c . ' c o n f i r m i n g ' the
'hypothesis' " I n English the d e f i n i t e a r t i c l e precedes the word man",
but t h a t we can never know f o r sure whether t h i s 'hypothesis' i s t r u e .
But such a claim i s surely absurd. Since our ( t r u e ) rule-sentence can-
not be falsified by incorrect utterances (nor, of course, by correct ut

terances), it cannot be verified by correct utterances, either.
According to another axiom concerning the nature of empirical hypo
theses, an existential hypothesis like "There are unicorns" can be veri
fied by finding a confirmatory instance, but cannot be falsified, i.e.,
conclusively disconfirmed: if we do not find any unicorns, this does not
prove that there are none (although it makes this highly probable), Now
if we make an analogous existential statement concerning English, e.g.,
"In English there is a preposition b l i p which is fully synonymous with
92
on", it is clear that we know its falsity with absolute certainty. To
deny this is to argue, absurdly, that although we do not personally know
of any such English preposition, this does not prove that it does not
exist in English as we know it. And even if we did make an utterance
like "This book is blip the table", which apparently contains a confir
matory instance verifying our 'hypothesis', we would simply brush aside
this example as irrelevant, because it is - once again - incorrect
English. This means that, analogously to the case of falsification of
(atheoretical) true universal statements about English, we cannot even
specify the conditions under which we would be willing to consider our
(atheoretical) false existential statement about English as verified.
And since this (false) statement could not possibly be verified, it can
not properly be called falsified either, just as our unfalsifiable rule-
sentence concerning the place of the definite article could not proper
ly be called verified (cf. above).
A doubt concerning the truth of "In English the definite article
precedes the word man" and the falsity of "In English there is a prepo
sition blip which is fully synonymous with on" is fundamentally differ
ent from a doubt concerning the truth of an empirical universal hypothe
sis and the falsity of an empirical existential hypothesis. In these
latter instances, as we have seen, a doubt is a conceptual necessity,
whereas in the former cases a doubt is excluded. Therefore, if we have
to find in the context of natural science an equivalent to a doubt of
the former type, we should compare it, rather, to a doubt concerning the
outcome of particular experiments or the observational data in general:
I see that this piece of metal expands when heated but I doubt it; and
no amount of evidence can convince me that my doubt is baseless. This
example makes it clear that doubts of this kind are essentially irra
tional. If they were taken seriously, empirical (or grammatical) re
search could not even begin. It would be just as irrational for me to
doubt, when playing poker, whether a full house r e a l l y beats a flush,
in spite of the fact that I have always been playing, and seen every
body else play, according to this rule and that, if I ask others, they
tell me that this is indeed the correct way of playing.
It is important to notice that in my opinion the paradigmatic case
of language description is the one where the linguist is describing his
native language. It is easy to see that this is the ideal to which ling
uists describing foreign languages have always attempted to approximate
in their efforts to better understand the languages under study. Whether
this has been acknowledged or not, the primary object of linguistic de
scriptions has always been language as it is known by those who fluent
ly speak it, and the use of informants for instance is nothing but a
short-cut to that intuition which, due either to the lack of time or
to self-imposed methodological restrictions (as in the time of American
'taxonomic' linguistics), one was not in a position to acquire. However,
native speakers have no uniquely priviledged status. New languages can
be learnt, and when one speaks a foreign language reasonably well, one
is entitled to consider describing it with roughly the same competence
as native speakers are.
We may conclude that in matters of confirmation and falsification
empirical hypotheses and (linguistic) rule-sentences behave in funda
mentally different ways, due to the fundamental difference between re
gularities and rules. This difference could be summed up in the follow
ing way: Rules are known to exist, and they determine which occurrences
(i.e., normative actions) are, and are known to be, correct, whereas oc
currences (i.e., events and non-normative actions) determine which re
gularities are assumed to exist. Or, from the perspective of linguistic
expression, rules determine which rule-sentences are known to be true,
whereas occurrences determine which empirical universal hypotheses are
assumed to be true.
I think the facts which I have been discussing here are quite un
ambiguous; I also think that the conclusions which I have drawn from
these facts are beyond dispute. It might seem surprising then that TG
and most other schools of linguistics either have never discussed these
relatively simple questions at all or have at most made summary claims
which are tantamount to a denial of the conclusions drawn here. However,
there is a logical explanation for this seemingly peculiar attitude.
The difference between rules (of language) and regularities, which was
demonstrated above, is based on the normative nature of rules. Now,
probably because of the general intellectual climate of our century,
it is generally thought that linguistics is an empirical science in
the positivistic sense, i.e., comparable to any standard natural scien
ce (cf. 3.0. above). But there is, and could be, nothing normative
about natural events, e.g., the movements of planets or gas molecules.
And this non-normativity of events exemplifying regularities in nature
is conceptually linked to the fact that regularities can never be known
with certainty, which means that hypotheses referring to them can never
be known with certainty to be true. Consequently TG, as well as most
other schools of linguistics, must claim that in language there are no
rules, but only regularities, and that sentences referring to these
'regularities' are genuine universal hypotheses, i.e., can never be
known for sure to be true. We have just seen that this attitude is
contrary to the facts.
As a representative of positivistic linguistics, Harris (1961:254)
was logical enough to claim, i n t e r alia, that "given the present Eng
lish system in which / η/ does not occur initially, the possibility
that someone will pronounce an English utterance containing initial / η / ,
e.g., in / n e n / is very remote", with the clear implication that if such
an utterance should occur, it would have to be accounted for in the pho
nological description of English. Here Harris is straightforwardly
adopting the position that in language, just as in nature, all actual
occurrences are equally relevant. But we have seen that the fact that
someone utters a form like "girl the", for example, has no relevance to
the description of English, because such a form is incorrect, and there-
fore i r r e l e v a n t . I t is clear t h a t on t h i s issue TG would not wish to

share H a r r i s ' s p o s i t i o n , but the i n t e r e s t i n g thing is t h a t i t has in
f a c t no right to exclude i t s e l f from sharing i t . To repeat, TG claims
to be a natural science dealing w i t h observable events: but natural
events cannot be e i t h e r c o r r e c t or i n c o r r e c t s whereas utterances and
sentences must be; and the notion of correctness is c o r r e l a t i v e w i t h
the notion of r u l e .
The 'ordinary-language philosophers' have f u l l y recognised the
nonempirical nature of statements about language ( c f . Cavell 1971a and
93
Hare 1971). Since they are concerned w i t h the use (and hence the
meaning) of p h i l o s o p h i c a l l y i n t e r e s t i n g words, the t r u t h of the s t a t e
ments they make is of course f a r from e v i d e n t . This means t h a t , i n
stead of being comparable to rule-sentences, such statements represent
a c e r t a i n type of t h e o r e t i c a l hypothesis about (what i s meant i n ) lan
guage ( c f . 8 . 3 . below).
6 . 2 . Examples of Rules and Rule-Sentences
I t i s an essential requirement f o r rule-sentences t h a t they be

absolutely t r i v i a l : sentences about language whose t r u t h or f a l s i t y
i s not known immediately and beyond the p o s s i b i l i t y of doubt are ex
definitione not rule-sentences. Therefore there can be no s c i e n t i f i c
i n t e r e s t i n s t a t i n g rule-sentences of a given language. However, there
is a considerable m e t a s c i e n t i f i c i n t e r e s t i n merely r e a l i s i n g t h a t the
re ave such sentences, i n view of the f a c t t h a t , because of t h e i r non-
empirical n a t u r e , they f l a t l y c o n t r a d i c t the p e r s i s t e n t claim t h a t
linguistics ( i n the sense of 'grammar') i s an empirical science. That
i s , they prove t h a t n o n - t r i v i a l grammatical t h e o r i e s , whose t r u t h or
f a l s i t y i s not known, are not empirical t h e o r i e s , but hypothetical con
ceptual d e s c r i p t i o n s , given t h a t both ( a t h e o r e t i c a l ) rule-sentences and
(theoretical) grammars speak - i n d i f f e r e n t ways - about the same
(normative) reality.
When I d i r e c t a t t e n t i o n to r u l e s , I am not i n t r o d u c i n g any new
entities. A l l t h a t I o f f e r i s reinterpretation of some well-known
facts. This can be made more precise as f o l l o w s . Each science must
have its own set of basic statements, i.e., statements dealing with the
simplest aspect of that region of reality with which the science in
question is concerned. The basic statements of natural science are
about particular spatiotemporal occurrences (cf. above). What do the
basic statements of grammar look like? Bach (1974:61-63) and Leech
(1974:84-90) give examples of, respectively, morpho-syntactic and se
mantic basic statements. For instance:
"The past tense of play is played. The past tense of sing is not
singed hut sang."
"Sing is a verb, It is by virtue of this property that it can

occur in contexts like these:
me a song.'
To is more fun that to be silent .

but not like these:
The best is Home on the range.
John is er that Mary.
The extremely man told me a long story."
"Sing is a verb of the class that can occur in the progressive as
pect {He is singing a song) , as opposed to verbs like know (' He is
knowing the answer). "
"Sing can occur in the passive: This song was first sung by Rudy
Vallee. ( This man is resembled by his mother.)"
"Sing can stand with two NP's in its predicate: She sang me a song
(compare She saw me a sailboat)."
"I am an orphan is synonymous with I am a child and have no father

or mother. "
"I am an orphan entails I have no father."
"This orphan has a father is a contradiction."
To this list we can add the following phonological rules for in
stance: "The words pill and pan (kill and can, etc.) begin similarly"
and "The words pill and kill (pan and can, etc.) begin differently".
Notice that those 'basic statements of grammar', i.e., rule-senten
ces, which contain theoretical terms like 'progressive aspect' or 'syno-
nymous', are not immediately understandable to a layman. To t h i s ex-

t e n t , then, they do not j u s t express p r e s c i e n t i f i c or a t h e o r e t i c a l eve-
ryday knowledge. However, i t is obvious that they could be r e f o r m u l a t -
ed i n purely a t h e o r e t i c a l terms. Besides, the meaning of those theore-
t i c a l terms t h a t occur i n the rule-sentences here considered can be e a s i -
ly taught to everybody.
False rule-sentences most often look simply r i d i c u l o u s , f o r instance:
"The past tense counterpart of He is singing is He singebat". But t h i s
is p r e c i s e l y the proof of the absolute c e r t a i n t y w i t h which we know t h e i r
falsity.
There are cases d i f f e r e n t from those enumerated above, where the
social control of rules is less than absolute ( c f . 5.4. above). For i n -
stance, i s i t ( d e f i n i t e l y ) c o r r e c t or i n c o r r e c t to say "You d o n ' t know
nothing"? However, such l e s s - t h a n - c l e a r cases do not r e l a t i v i s e cases
which are c l e a r . In a d d i t i o n , even the uncertainty about the former
type of examples obtains w i t h i n the l i m i t s of absolute c e r t a i n t y . We
know, f o r i n s t a n c e , t h a t only sentences l i k e "You know n o t h i n g " , "You
d o n ' t know a n y t h i n g " , and "You d o n ' t know nothing" are here possible
candidates f o r correctness. Nothing would be easier than to imagine
non-existent expressions which might be perceived as t r y i n g to say the
same t h i n g , but which would be (known to be) d e f i n i t e l y i n c o r r e c t , e . g . ,
"You knewing anything n o t " .
6 . 3 . Two D i f fe r e n t Types of Rule-Sentence
I t i s often held t h a t rule-sentences (generally r e f e r r e d to as

'rules') lack t r u t h - v a l u e . Thus, according to the p o s i t i o n of e a r l i e r
p o s i t i v i s m and, more generally, of the ' i n s t r u m e n t a l i s t ' school of the
philosophy of science, universal ( t h e o r e t i c a l ) hypotheses are i n t e r -
preted as ' r u l e s ' , or recommendations, f o r making p r e d i c t i o n s about
observable events, which is then taken to imply t h a t universal hypothe-
ses are n e i t h e r true nor f a l s e . And l o g i c a l t r u t h s are even more f r e -
quently i n t e r p r e t e d as rules f o r making c o r r e c t inferences. Such a
view is understandable i n so f a r i t concentrates upon the p r e s c r i p t i v e
or imperative f u n c t i o n of rule-sentences, because imperatives are ob-
v i o u s l y n e i t h e r true nor f a l s e , but rather appropriate or inappropriate
and, i f the former, then e i t h e r obeyed or disobeyed. "Close the door"

would be inappropriate i f u t t e r e d in the middle of a desert or i n a
room w i t h a closed door. I t would be a p p r o p r i a t e , i f i t were u t t e r e d
in a room w i t h an open door and addressed to a person who is able to
close the door. I t might j u s t be possible to i n t e r p r e t the sentence
" I n English the d e f i n i t e a r t i c l e precedes the word man" as meaning
"Place the d e f i n i t e a r t i c l e in f r o n t of the word man, when speaking
E n g l i s h ! " . But i t i s q u i t e unarguable t h a t under normal circumstances
the sentence i n question is ascribed a d e s c r i p t i v e f u n c t i o n , i . e . , it
i s taken to be a d e s c r i p t i o n , and a true d e s c r i p t i o n , of a r u l e of Eng-
l i s h , j u s t as i t s negation " I n English the d e f i n i t e a r t i c l e does not
precede the word man" is taken to be false. Consequently I accept it
as a f a c t about language t h a t i t s rule-sentences have t r u t h - v a l u e . The
t r u t h or f a l s i t y of a rule-sentence is determined by the existence or
the non-existence of the r u l e which i t purports to r e f e r t o .
Logical t r u t h s l i k e ' p v - p ' d i f f e r from rule-sentences in t h a t they
are not m e t a l i n g u i s t i c , t h a t i s , they do not speak about (object) lang-
uage, l i k e the l a t t e r . (We saw i n 5 . 2 . above t h a t metalanguage is simp-
l y the expression of one's r e f l e x i v e consciousness as i t p e r t a i n s , spe-
c i f i c a l l y , to acts of speaking.) Yet, l o g i c a l t r u t h s might be said to
be based on natural language i n s o f a r as they a r e , o r i g i n a l l y , f o r m a l i -
sations of c e r t a i n types of necessary t r u t h s expressible in natural
language. In t h i s sense, t h e n , natural language, as well as the t h i n k -
ing and a c t i n g i n t o which i t i s embedded, c o n s t i t u t e s the Sinnes funda
ment of l o g i c ( c f . 2.6. above). Replacing the symbols f o r the t r u t h -
values, i . e . , 'T' and ' F ' , by a r b i t r a r y symbols l i k e '1' and ' 0 ' may
be a convenient way of b r i n g i n g out the purely mechanical, manipulative
aspect of formal l o g i c , but i t should not make one f o r g e t the meaning
of l o g i c . That ' p v - p ' has on a l l i n t e r p r e t a t i o n s the value '1', means
n o t h i n g , u n t i l one learns t h a t ' 1 ' stands f o r ' t r u e ' , and t h a t t h i s is
a s p e c i a l way of saying t h a t (any sentence exemplifying) the formula
' p v - p ' i s necessarily t r u e .
Consequently, i t i s p e r f e c t l y proper to speak of l o g i c a l truths.
Of course, rules of inference like
which show how one is to get from given premises to the conclusion,
cannot be called true or false, but only valid or invalid. However,
they can be easily transformed into formulae admitting of truth-va
lue, e.g., "[(pvq)&-p] q". This is why validity is generally used
as a superordinate term applying both to rules of inference and to
formili ae.
In this chapter I have limited the notion of rule-sentence in
such a way that (atheoretical) rule-sentences and (theoretical) gram
mars not purporting to refer to existing rules are decreed to lack
truth-value (cf. the discussion of the impossibility of diachronic
falsification of rule-sentences in 6.1. above). I now have to reveal
the presupposition on account of which it makes sense to treat lin
guistic descriptions in this way. It is a self-evident truth that a
natural language fulfils a multiplicity of different functions. It
appears to be generally felt that language fulfils these functions
well enough, one reason for this being that mostly they do not exist
prior to and independently of language, but are rather made possible
by language and develop (spontaneously) together with it. Be it as
it may, the fact is that it is only seldom that there arises a need to
change, i.e., to improve, language in such a way that it better suits
those functions which it is supposed or required to fulfil. And when
such a need does arise, it mostly manifests itself in the creation of
technical sublanguages or formal languages which stand clearly apart
from natural languages and, hence, are not part of the (traditional)
subject matter of linguistics. It is important to realise that it is
only in the context of an activity with no built-in interest in con
scious change and improvement that, strictly speaking, it makes sense
to establish the synchronic - diachronic dichotomy. Our everyday
speech is generally thought to be such an activity, that is, the state
of language preceding a linguistic change has no conscious or inten
tional (i.e., conceptual) relation to the state of language that will
emerge a f t e r the change; consequently, i t makes sense to e s t a b l i s h syn-

chronic studies which have, at l e a s t in p r i n c i p l e , no r e l a t i o n to the
h i s t o r y of language. 94 And, in the synchronic c o n t e x t , only sentences
referring to existing rules are t r u e .
However, there are also other types of 'games', i . e . , rule-govern-
ed a c t i v i t i e s , than the one exemplified by speaking. For instance, the
research conducted by a natural s c i e n t i s t , which follows i t s own proce-
dural r u l e s , contains a b u i l t - i n i n t e r e s t in improving i t s e l f , i.e., in
i n v e n t i n g new rules capable of f a c i l i t a t i n g the f u l f i l m e n t of the over-
a l l goal of the a c t i v i t y in q u e s t i o n ; and such new, e f f e c t i v e rules of
course replace the o l d e r , less e f f e c t i v e ones. Now, i t i s obvious t h a t
my notion of rule-sentence which was so devised as to apply to games
analogous to natural languages does not apply to games analogous to the
p r a c t i c e of the natural sciences. According to the previously given
d e f i n i t i o n of rule-sentence, a sentence recommending or imagining a so
f a r non-existent r u l e is n e i t h e r true nor f a l s e . I f such a sentence
is i n t e r p r e t e d as p u r p o r t i n g (and f a i l i n g ) to r e f e r to some e x i s t i n g
r u l e , then i t is simply f a l s e . We have, however, j u s t seen t h a t in
some contexts a sentence which does not r e f e r to any (as y e t ) existing
rule may precisely for this reason be the ' t r u e ' one:95 such a sentence
recommends a new r u l e which i s , according to some pre-established stan-
dard, more c o r r e c t t h a t the e x i s t i n g r u l e s , and t h e r e f o r e the actions
conforming to the new r u l e would be more c o r r e c t than the c u r r e n t l y
performed actions which conform to e x i s t i n g rules and are of course
c o r r e c t i n r e l a t i o n to these. I f , w i t h respect to the p r a c t i c e of
science, only sentences describing what is done (and not what ought to
be done) were t r u e , science could never have developed out of magic,
in the f i r s t p l a c e , and we would, f o r i n s t a n c e , s t i l l believe t h a t the
earth is f l a t . (Notice t h a t actions are c o r r e c t in r e l a t i o n to r u -
l e s , whereas rules are c o r r e c t in r e l a t i o n to the c l e a r l y identifiable
o v e r - a l l f u n c t i o n or goal of the t o t a l a c t i v i t y , supposing t h a t i t has
one. Insofar as speaking has no such f u n c t i o n , i t is not reasonable
to ask whether rules of language are c o r r e c t or i n c o r r e c t . )
What I said about the p r a c t i c e of a natural s c i e n t i s t applies to
the p r a c t i c e of a social s c i e n t i s t in a two fold way, Here, t o o , there
are pre-established standards which determine what is to be considered

as true or f a l s e w i t h i n the social t h e o r y , but the o r i g i n of these stan-
dards is not the same as w i t h i n the natural sciences, as Marcuse (1964:
6) has pointed out:
'Truth' and 'falsehood' of needs designate objective conditions to

the extent to which the universal satisfaction of vital needs, and
beyond it, the progressive alleviation of toil and poverty, are u-
niversally valid standards. But as historical standards, they do
not only vary according to area and stage of development, they al
so can be defined only in (greater or lesser) contradiction to the
prevailing ones.
Consequently the prevailing standards, as opposed to the 'real' or

universally valid standards, are precisely those of a false conscious
ness, or 'ideology' in the pejorative sense.
To put it more precisely, here the standards of truth and false
hood are not primarily set up at the level of science but at the level
of the social reality which is the object of science, and social scien
tists merely take over and define these standards. The basis for this
is that, unless social scientists, under the influence of positivism,
are willing to ignore the social character of their subject matter and
hence to accept a distorted view of social sciences, they have to admit
that they cannot trascend the social reality in which they live and
which at the same time is the object of their study; because they are
inside , they cannot bring from outside any standards of measurement
and truth with which the social reality could be confronted (cf. 2.2.
above).
Since the social reality is permeated with conflicting, all-em
bracing standards of truth and falsehood, which coincide, roughly, with
conflicting social classes, social scientists cannot avoid accepting,
either implicitly or explicitly, one or another of those standards. And
if, as is done within 'critical' or 'emancipatory' sociology for in
stance, they accept the above-mentioned standards of the satisfaction
of vital needs and of the alleviation of toil and poverty, they reveal
a built-in interest in developing and improving not only (the rules of)
their own science but also its object. From the above it follows that
if one accepts the standards of critical sociology, which are, on the
global scale, certainly the most universal ones, then - just as in

the case of sentences describing the scientific practice - only those
sentences pertaining to social reality which describe something that
does not exist (i.e., alternative social realities) but which ought to
exist are 'true'. Sociological practice consists in defining the not-
yet-existing truth on the basis of, and in contradi ci tion to the exist
ing falsehoods and in trying to change the latter into the former; com
pare Marcuse's (1964 : xii) observation that the "values attached to the
alternatives do become facts when they are translated into reality by
historical practice. The theoretical concepts terminate with social
change" (cf. also the Sapir-quotation p.67)
Whether or not there will be a change as envisaged, constitutes
the criterion for the success of the theory. However, it does not seem
adequate to say that it constitutes as well the criterion for the truth
of the latter, because a theory may correctly identify the truth, while
simultaneously analysing and predicting its own, at least temporary,
lack of success.
By contrast, posi ti vi sti c (including Popperian) sociology rejects
the above-mentioned standards of truth and falsehood. This leaves two
possibilities open: either it tries to prevent a change as envisaged
above from coming about; or it forbears either to bring the change about
or to prevent it from coming about (for these notions, cf. von Wright
1968:38-39). The latter alternative corresponds to the position held
by traditional 'value-free' science. Now, the social change at issue
is a qualitative change; it concerns not just particular actions, but
social regularities (rather than rules) governing particular actions.
No similar change is possible within natural science, because the change
of physical regularities, if possible at all, lies at any rate outside
the human sphere of influence. This non-parallelism between social
science and natural science might be objected to on the grounds that
the variant social regularities which are quite obviously open to the
human influence are in reality just manifestations of some invariant
social regularities comparable to genuine laws of nature. This objec
tion is without substance, however, because sociology has never been
able to uncover such invariant regularities (cf. 2.2. above).

Because the subject matter of natural science lies outside the
human sphere of influence, one can try neither to bring about nor to
prevent a change of the laws of nature; nor can one forbear to do
either. We have seen, by contrast, that these possibilities of action
are necessarily open to the social scientist. He cannot avoid being
confronted with the normative, or indeed e t h i c a l , question as to what
he ought to do about the subject matter of his own science, i.e.,
whether to try to bring about a certain type of change, or to try to
prevent it from coming about, or to forbear to do either. What is im
portant here, is to see that even forbearance is an answer to a value
question. Therefore, contrary to what seems to be the case at first
glance, there can be no value-free social science.
Earlier prescriptive grammars often tried to prevent linguistic
changes from coming about. Today it is generally thought that any sci
entific form of linguistics must adopt the attitude of forbearance to
wards linguistic behaviour. However, this view is in no way inevitable,
and it has been effectively challenged by Hymes (1974, chap. 10) and by
other representatives of what can only be called 'emancipatory linguist
ics' .
Critical sociology is comparable to philosophy of science, logic,
and mathematics insofar as all these sciences are concerned with invent
ing new and better, or more effective, forms of activity within the do
main of their data. I have indicated before that there is no hard-and-
fast division between prescriptive science and critical science. The
aporopriateness of the one label or the other depends on whether the new
forms of activity merely differ from old ones or are in contradiction to
them. As a descriptive science, (the traditional form of) linguistics
differs from critical sociology as well as from (the traditional form of)
logic.
7.0. THE ΙΝELIMIΝ Α Β I L Ι Τ Υ OF LINGUISTIC NORMATIVITY
There are two main objections against my p o s i t i o n : E i t h e r the ana

logy between games and synchronic states of natural 'languages i s d e f e c t i
ve; or both game-descriptions and synchronic-grammatical d e s c r i p t i o n s are
empirical. In 5 . 0 . - 6 . 0 . (above) I refuted the former o b j e c t i o n by show
ing t h a t language contains genuine r u l e s . In the present chapter I i n
tend to' r e f u t e the l a t t e r o b j e c t i o n . P a r t i c u l a r actions are c o r r e c t i f ,
and only i f , they exemplify any of the concepts ' c o r r e c t action A ' , ' c o r
r e c t action B ' , e t c . , which are (known to be) determined by the r e l e v a n t
rules. Since ' r u l e ' and ' c o r r e c t n e s s ' are interdependent, an attempt
to redefine the one i n purely empirical terms implies an attempt to r e
define also the other i n s i m i l a r terms. Such a wholesale attempt to get
r i d o f n o r m a t i v i t y i n general i s involved i n the claim t h a t all descrip
t i o n s of r u l e s , whether rules o f game or rules of language, are e m p i r i c a l .
Prima facie already, t h i s claim i s rather i m p l a u s i b l e . Since - as we
have seen 6 . 1 . - there i s a genuine and i r r e d u c i b l e d i f f e r e n c e between
rule-sentences and empirical hypotheses, any r e d e f i n i t i o n of n o r m a t i v i t y
i n purportedly empirical terms must contain a f a l l a c y of one type or an
other. That t h i s i s indeed the case, w i l l be seen below. I shall dis
cuss f o u r representative types o f c r i t i c i s m s of the notion o f l i n g u i s t i c
normativity. In conclusion, I s h a l l examine TG's p o s i t i o n on t h i s ques
tion.
7.1. A Synchronic Grammar Does not Investigate Spatiotemporal

Utterances, but Correct Sentences
The most s t r a i g h t f o r w a r d way to equate grammars with empirical the

ories i s to say t h a t they i n v e s t i g a t e only spatiotemporal l i n g u i s t i c oc-
currences, i.e., utterances and, possibly, reactions to utterances.

This line of thinking seems to be congenial to American structural
linguistics as represented by Bloomfield and Harris (cf. 3.4.-5. above).
The difficulty with this position is that, even if we grant for the
sake of argument that grammars investigate utterances, it is neverthe
less clear that they do not investigate everything that ever is utter
ed, but only correct utterances. So the problem is how to pick out
those (correct) utterances which are de facto investigated by grammars,
while not making use of the normative, nonempirical concept of 'correct
ness'. That is, the problem here is how to redefine the concept of
'correctness' in purely spatiotemporal terms, i.e., how to operationali-
se (our knowledge of) this concept. In what follows I shall examine
the solution to this problem offered in Sampson (1975).
Sampson (1975:62) undertakes to give a spatiotemporal redefinition
of 'correctness' by stating, as a first approximation, that everything
that a fluent speaker of English utters, is correct and must be descri
bed by a grammar of English. However, it is obviously false to try to
define 'correctness' in terms of 'fluency'. The correct procedure is
precisely the opposite one: whether or not a person is (considered) a
fluent speaker of English, depends on whether or not he speaks English
correctly. Thus 'correctness' remains, so far, a primitive notion.
Furthermore, Sampson (p.66) is forced to admit that even fluent
speakers make incorrect utterances, and that the speech situations con
tain (practically) no spatiotemporal criteria on the basis of which in
correct utterances could be differentiated from correct ones. But this
means precisely that there is no way to operationalise the concepts
'correct utterance (or sentence)' and 'incorrect utterance (or sentence)'
The non-operationalisable character of linguistic knowledge, or of
the concepts functioning as objects of such knowledge, is a fact which
has not totally escaped linguists' attention. When discussing the con
cept of 'synonymy', Chomsky (1969) correctly notes that it cannot pos
sibly be defined in operational terms pertaining to the speech situati
ons, and he concludes (p.65):
THE INELIMINABILITY OF NORMATIVITY 177
What a person does or is likely to do and what he knows, may be

related, in some way that cannot, for the moment, be made precise;
the relation is, however, surely in part factual and not a strict
ly conceptual one.
This is only a more general formulation of the familiar truth that

"the speaker-hearer's linguistic intuition is the ultimate standard that
determines the accuracy of any proposed ... operational test" (Chomsky
1965:21).
Now, Sampson (1975) tries to escape from this dilemma by referring
to the general fact that people make mistakes even when there are no ob
servable facts in terms of which the mistakes might be explained. So it
is not surprising that people also make mistakes in their speech (p.66-
67); this more particular fact should apparently be somehow derivable
from the former one.
This is a rather weak argument. What Sampson does, is in fact to
state my general thesis: (knowledge of) normativity in general is irre
ducible. It is surely awkward to claim that linguistic normativity can
be defined in spatiotemporal terms because it is a special case of gene
ral normativity which cannot be so defined. In fact, the reason why
(linguistic) knowledge cannot be operationalised in terms of (linguistic)
behaviour is simply the general creativity or unpredictability of human
actions (cf. 4.2.2. above).
I have refuted here the straightforward argument to the effect that
all the linguist has to do is to observe linguistic occurrences and de-
no
scribe them. Insofar as Labov's plea for the primacy of the observa
tion of actual speech can be understood as a variant of this argument,
it has also been refuted here. In the two following sections I shall
examine slightly more elaborate attempts to dispose of linguistic norma
tivity.
7.2. Grammatical Concepts Are not Comparable to Theoretical

Concepts of Natural Science
A natural science makes use of a strategy of theoretical concepts

and correspondence rules: To explain the behaviour of observable enti
ties, the existence of unobservable entities, e.g., electrons, that
exemplify the corresponding theoretical concept, e.g., 'electron', is

postulated, and the unobservables are partially interpreted in terms
of the observables by means of correspondence rules and operational
definitions indicating a constant relation between the occurrences of
both types of entities. Consequently a description of natural science
could be reduced to the following basic components: observable space-
time, theoretical concepts, and rules of correspondence (cf. 1.4. abo
ve). If this descriptive apparatus can be applied without distortion
to linguistic data, then linguistics is of course a genuine natural
science.
In the previous section we have seen in a preliminary fashion that
the strategy of theoretical concepts and correspondence rules is in
fact impracticable within linguistics. The attempt at operationalising
a concept like '(in)correct utterance (or sentence)' means giving it
the status of a theoretical concept and trying to find out suitable ru
les of correspondence. The failure of operationalisation makes the ana
logy between grammar and natural science collapse. The reason of this
failure was seen to lie in the elementary fact that human behaviour
differs in fundamental ways from the 'behaviour' of physical things.
Here I shall examine somewhat more closely what is involved in the at
tempt at reinterpreting grammatical descriptions in this way as descrip
tions of theoretical natural science.
Chomsky has been claiming from the beginning that grammatical con
cepts like 'phoneme' and 'phrase' are comparable to theoretical concepts
of physics like 'electron', and that grammatical rules are comparable
to the laws of a physical theory (cf. p. 78). This is TG's official
position up to the present day (see further references in 3.6. above).
The same position has been further elaborated in Kasher (1972).
Kasher interprets 'sentence' as a theoretical concept and defines
it as a string of letters plus its 'verbal context', i.e., the relevant
syntactic, semantic, and pragmatic structures as conceived by the re
spective linguistic theories. 'Sentence' is thus equivalent to the full
grammatical description. 'Utterance', on the other hand, is an observa
tional concept consisting of an 'inscription' and a set of indexes rela-
ted to the speech situation/ Kasher then defines an 'instance func

tion' correlating utterances with sentences. He further distinguishes
between (correct) utterances and 'defective utterances' just as well
as between (correct) sentences and 'doubtful sentences'. It might seem,
then, that the descriptive apparatus of natural science has been suc
cessfully introduced into grammar: we seem to have to do here only with
observable space-time (= 'utterance'), theory (= 'sentence'), and rules
of correspondence (= 'instance functions'). However, such a view is
totally mistaken.
In 7.1. (above) we saw that Sampson (1975) at least tries, even if
unsuccessfully, to solve the problem of normativity in positivistic
terms. By contrast, Kasher (1972) prefers simply to ignore it. But of
course the problem remains precisely the same as before: How is one to
distinguish correct utterances from 'defective' ones? And once again,
the only possible answer is that this happens on the basis of one's nor
mative linguistic knowledge. The fact that Kasher ignores the exist
ence of the rules of language which determine that a given utterance
exemplifies a correct sentence, makes it evident that he fails to see
sentences as atheoretical concepts. But certainly nothing is gained
by eliminating the distinction between (atheoretical ) sentences and
their theoretical descriptions. Since 'verbal context', as part of
'sentence', is explicitly defined by reference to existing grammatical
theories, and since 'utterance' is defined as an instance of 'sentence'
as so defined, it follows, absurdly, that neither sentences nor utter
ances can be said to exist in a speech community with no grammatical
tradition. Kasher fails to explain how initial linguistic data, i.e.,
correct or incorrect sentences and utterances as they are known by the
linguist, are to be adjusted to his 'theoretical - observational' di
chotomy, but he seems to assume that linguistic normativity will some
how disappear in this process. There is of course no basis for such
an assumption.
Secondly, and as a corollary of what precedes, it is clear that
instance functions are in no way comparable to correspondence rules.
Kasher's 'verbal contexts' are in reality theoretical grammatical de-
scriptions of atheoretical sentences, i.e., atheoretical concepts , or

of the rules determining such concepts; and his 'utterances' are in
reality exemplifications of sentences as here defined. Therefore
grammatical descriptions deal with space-time just as little as any
other conceptual analyses do. This is directly born out by the fact
that Kasher speaks nothing about what really goes on in space and
time, i.e., which utterances r e a l l y occur in which spatially and tem
porally defined circumstances. Without explicitly recognising the
existence of atheoretical sentences, he is merely pointing out which
utterances would exemplify atheoretical sentences described by his
'verbal contexts', if such utterances happen, under whatever circum
stances, to occur in space and time. The reason why Kasher knows so
little about actual space-time, i.e., actual linguistic occurrences,
and is thus unable to formulate genuine rules of correspondence, is
simply that, in contradistinction to physical behaviour, human beha
viour is characteristically creative or unpredictable. What we
do know, however, are the rules, and hence the sentences, of our lang
uage. We know which sentences are correct, irrespective of whether
they have ever been uttered, or alternatively which utterances would
be correct if they were to occur. It is this normative knowledge
which constitutes the object of grammatical descriptions, even though
positivistic linguists are unable to perceive this fact.
Far from being comparable to correspondence rules, Kasher-type
instance functions formulate an abstract relation between theory and
space-time, a relation which holds not only in grammar, but also in
logic, mathematics, and philosophy of science. For illustration, let
us examine inferences. In connection with their everyday decision
making different people may on different occasions make the following
inference: "If I want to buy this thing, I must get the money for it.
I want to buy this thing. Therefore I must get the money for it".
All such inferences made on different occasions represent of course
as many inference-tokens exemplifying the corresponding inference-
type quoted above. Within propositional logic, for instance, this
inference-type may in turn be given a theoretical description charac
terising it as a case of Modus Ponens and involving such theoretical
concepts as ' m a t e r i a l i m p l i c a t i o n ' and ' t r u t h f u n c t i o n ' . Using the

Kasher-type d e s c r i p t i v e apparatus we might say now t h a t i n f e r e n c e - t o -
kens are ' u t t e r a n c e s ' and t h e i r t h e o r e t i c a l t r u t h - f u n c t i o n a l descrip-
t i o n s are 'sentences'. Obviously nothing would be easier than to f o r -
mulate an 'instance f u n c t i o n ' between utterances and sentences so de-
f i n e d , although t h i s would require i g n o r i n g the c r u c i a l r o l e of i n f e -
rence-types corresponding to the atheoveticdl concept of sentence (cf.
above). But surely such a r e i n t e r p r e t a t i o n of l o g i c would not t r a n s -
form l o g i c i n t o a natural science. S i m i l a r l y any p a r t i c u l a r acts of
m u l t i p l i c a t i o n , or ' m u l t i p l i c a t i o n - t o k e n s ' , exemplifying a given 'mul-
t i p l i c a t i o n - t y p e ' might be c o r r e l a t e d w i t h theoretical-mathemati cal
d e s c r i p t i o n s of the l a t t e r . Or r e c u r r e n t acts of e x p l a i n i n g a given
phenomenon might be c o r r e l a t e d , via the concept exemplified by these
a c t s , w i t h Hempel's t h e o r e t i c a l d e s c r i p t i o n of (D-N) explanations.
From t h i s i t does not f o l l o w of course t h a t mathematics and p h i l o -
sophy of science would be natural sciences, or even empirical s c i e n -
ces. - To sum up, i t is a rather obvious f a c t t h a t i n a l l normative,
nonempirical sciences l i k e grammar, l o g i c , or philosophy, theoretical
concepts are in one way or another r e l a t e d to what people do.
From the f a c t t h a t instance functions are not comparable to cor-
respondence rules i t also f o l l o w s t h a t , contrary to Chomsky's o p i n i o n ,
grammatical concepts and rules are not comparable, r e s p e c t i v e l y , to
t h e o r e t i c a l concepts and hypotheses of natural science; or i f they a r e ,
so are the concepts and statements of other normative sciences t o o .
For instance (the ordinary language i n t e r p r e t a t i o n of) the 'subject
raising' t r a n s f o r m a t i o n , which is needed to systematise the i n t u i t i v e
concept ' c o r r e c t sentence ( i n L ) ' , i s a t h e o r e t i c a l empirical hypothe-
sis to the same extent as (the ordinary language i n t e r p r e t a t i o n o f )
the axiom of p r o p o s i t i o n a l deontic l o g i c ' - ( 0 p & 0 - p ) ' , which is needed
to systematise the i n t u i t i v e concept ' v a l i d (deontic) f o r m u l a ' . Simi-
l a r l y , e i t h e r both of the concepts 'noun phrase' and ' t r u t h f u n c t i o n '
are empirical t h e o r e t i c a l concepts, or n e i t h e r of them i s ; and since
' t r u t h f u n c t i o n ' is not e m p i r i c a l , n e i t h e r is 'noun p h r a s e ' .
It may be claimed that the particular events correlated with
grammatical theories do not concern the uttering of sentences, but
102
the learning of language. Such a proposal makes no difference,
however. Not only rules of correct speaking, but also those of
correct inferring and calculating must of course be learned. But
this obvious fact certainly does not make logic or mathematics any
more empirical; and the same is true of grammar as well. Describing
a game is different from describing how it is learned (cf. 2.4. above).
7.3 Rules Ave not Regularities of Non-Normative Actions

In the two previous sections I discussed attempts at showing that
the subject matter of grammar is purely spatiotemporal, and that gram
mar is accordingly a pure natural science. This standpoint was refuted
in some detail, but prima facie already, it is quite implausible in
view of the fact that actions, as distinguished from events, are not
purely spatiotemporal, and that utterances are undeniably results of
linguistic actions. This non-material nature of language has been
once again proved in Friedman (1975), where it is shown, in particular,
t h a t words cannot be described in the 'material object language' em
ployed by (theoretical) physics, since rules of use are an inseparable
part of words, and those cannot be described in such a language. I
also agree with Friedman's over-all assessment of why linguists have
been so eager to view their subject matter in material, or purely spa
tiotemporal terms:
The initial tendency to view words and sentences as exclusively

concrete occurrences or alternatively as classes of these, arises
I think out of two confusions. The first, as in synecdoche, is
to consider the salient features of an object as representative
of its totality. In this way the evident concreteness of the
sound of words leads one to ignore the extent to which use, how
ever intangible, is necessary to word-hood, The second error is
linked to the former: it is the conflation of exclusively physic
al systems (such as the solar system) with systems (such as lan
guage) which also contain material elements (Friedman 1975:94).
It may be added that this attitude of "materialist imperialism'

(as Friedman calls it) is mainly due to the influence of positivism.
It is the same attitude which underlies the tendency to eliminate the

distinction between (physical) event and (human) action.
So it is clear that the subject matter of grammar cannot be de
fined in terms of physical s pace-time. But it still may be argued,
with a higher degree of initial plausibility, that it can be defined
in terms of social space-time containing an irreducible meaningful or
conceptual component. Such a non-normative redefinition of rules, in
cluding rules of language, has been undertaken in Lewis (1969). Lewis'
analysis is further developed in Bennet (1973) to provide a non-norma
tive, or 'nominalist', account of speech acts.
Lewis bases his analysis on the concept of 'common knowledge',
roughly in the sense in which it was defined here (see 5.1. above).
He gives a recursive definition which generates an infinite number of
higher-order beliefs (or 'expectations') concerning what everyone· be
lieves (or 'has reason to believe'). I am not concerned here with the
factual correctness of this definition. Nevertheless I would like to
point out that the definition seems to be defective insofar as it achie
ves recursivity only by inadvertently using the word 'to indicate' in
two fundamentally different senses.
The definition of 'indicate' is as follows: "A indicates to Β that
C" = "If Β has reason to believe that A, B would thereby have reason
to believe that C". Now, in the sentence "That the streets are wet in
dicates to everyone that it has been raining", which is modelled upon
Lewis' third premise on p.52, 'indicates' stands for a relation between
matters of fact which holds contingently in the external world. This
can be seen more clearly by translating the sentence into its explicit
form: "If everyone has reason to believe that streets are wet, then
everyone would thereby have reason to believe that it has been raining".
By contrast, in a sentence like "That the streets are wet indicates to
everyone that everyone has reason to believe that streets are wet",
which is modelled upon Lewis' second premise, 'indicates' stands for
a conceptual relation, i.e., a relation which is a constitutive part
of our very concepts of knowledge and belief. The explicit form of
the sentence is: "If everyone has reason to believe that streets are
wet, then everyone would thereby have reason to believe that everyone
has reason to believe that streets are wet". This sentence is a neces
sary truth reminiscent of the theorem of epistemic logic
In the former sentence 'thereby' stands for a contingent, inductive
connection whereas in the latter it stands for a conceptual, deductive
connection. It is the latter type of sentence which is employed to
constitute the recursion step of Lewis' definition. Now, if it is ad
mitted that 'indicate' stands here for two different things, then the
recursive generation of higher-order beliefs cannot even get started
(cf. Lewis 1969:52-60).
Whether or not Lewis' conception of common knowledge is acceptable
in the technical detail, he utilises it to define rules, or 'conventi
ons', in such a way as not to make use of any inherently normative
concepts. The most important part of his definition is the following
(p.78):
A regularity R in the behaviour of members of a population Ρ when

they are agents in a recurrent situation S is a convention if and
only if it is true that, and it is common knowledge in Ρ that, in
almost any instance of S among members of P,
(1) almost everyone conforms to R;
(2) almost everyone expects almost everyone else to
conform to R; . . .
A regularity R is known on the basis of the memory about past ac

tions conforming to it, and this memory serves as the basis for the
expectation about similar actions: "Given a regularity in past cases,
we may reasonably extrapolate it into the (near) future" (Lewis 1969:
41). Lewis does not explicitly define the concept of 'correct action',
but it is clear that correct actions are actions conforming to a regu
larity of the relevant type. And, as we have seen, such a regularity
exists at the level of common knowledge about what is either remember
ed or expected. There are no explicitly normative elements in this
definition. Has Lewis genuinely succeeded, then, in eliminating the
problem of normativity?
First of all, it is clear that the regularity of actions necessa
rily correlated with any given rule contains both correct and incor
rect actions. Therefore, if Lewis' definition is to be upheld, 'me-
mory' and 'expectation' must be defined in such a selective way as to
mean correct actions only. With respect to memory it is not at all
clear whether this can be done in any natural way: when one thinks of
past cases of a regularity, one can certainly remember also incorrect
ones. It seems more natural to say that the actions which one expects
to come upon in connection with a regularity are precisely correct ac
tions; but even this cannot be asserted with absolute certainty. Let us
assume, however, that correct actions can be satisfactorily defined in
terms of what is (commonly) remembered and expected as a matter of fact.
Now it is a self-evident fact that memory and expectation are just as
subject to human fallibility as any other forms of human activity: it
is true of them, too, that what one does has no direct relation to what
one ought to do. In other words, we cannot help distinguishing between
correct and incorrect memories as well as between correct and incorrect
expectations. Hence there seems to be no way to eliminate the concept
of correctness or normativity.
This last observation is my basic objection against Lewis' (and
Bennett's) analysis. There are some complementary points as well.
Lewis' factual, non-normative approach is evident from the 'almost'-
qualifications figuring in his definition of 'convention'. But it is
clear that any such statistical account of rules is quite unable to
express the absolute 'either - or' character of all genuine rules (cf.
5.2.-3. above). Moreover, Lewis fails to see the implicit normativity
of his notion 'recurrent situation S'. Just as we need rules to iden
tify and to perform correct actions, we need rules to identify situati
ons as recurrent. (Such rules for the use of concepts have been dis
cussed p.42 and 100.) It is perhaps the basic insight of Winch (1958)
that we need definite criteria, whose use is governed by rules, to
identify entities as same or different, and that as regards social
entities, such criteria are internal to them.
At any rate, Lewis' concept of rule has the merit that he does
not try to define rules purely as sets of actions, or purely in terms
of social space-time. That is to say, even if he tries to reduce ru
les to regularities of actions, he does not try - as far as I can
see - to reduce in the same way that common knowledge which neces-
sarily pertains to such regularities. However, as an analysis of rules
of language, Lewis' account is rather unsatisfactory. He admits that
there must be such rules. However, he refuses to name any and doubts
whether any can be genuinely known (Lewis 1969:60-68). This common
view, which in my opinion is based on the confusion between rules of
language and rules of grammar, as well as on the general positivistic
attitude concerning language, has been refuted in detail in 5.0. and
6.0. (above). 1 0 3
In this chapter I have been discussing the tendency to deny the
autonomous character of normativity, which leads to various attempts
at reducing normativity as far as possible to the categories of spa
ce and time. In my opinion this very general tendency is a manifes
tation of the empiricist theory of knowledge, which maximises the role
of environment and minimises the role of mind in the acquisition of
knowledge. It is a self-evident truth that rules of language, logic,
and mathematics are all learned, little by little, in definite spatio-
temporal situations, and that, once they are mastered, their mastery
becomes manifest in similar situations. But there is certainly more
to learning and mastering rules than those definite situations in
which they are learned and applied. This additional element is the
active and creative contribution of the human mind. It is a general
truth that human creativity in its various forms cannot be captured
by means of correspondence rules or of any similar method. There is
no reason why the creativity related to rules should be different.
(This does not mean, of course, that the learning and the application
of rules could not, and should not, be investigated empirically.) It
is generally agreed that rules of logic and mathematics cannot, at
least in any direct sense, be reduced to the situations in which they
are learned or applied. I do not see why rules of language should be
treated any differently. However, Lewis (1969:100) draws an absolute
distinction between rules of language on one hand and rules of logic
and mathematics on the other, by claiming that the latter "may have
nothing to do with the conduct of human agents, except that human
agents might benefit by taking account of them". The discussion in
2.6. (above) was directed against precisely this conception of logic
7.4. Grammatical Descriptions Cannot Be Replaced by

Psycholinguists c and/or Socio Unguistic Descriptions
Sampson and Kasher clearly view grammars as empirical theories.

Lewis seems to think likewise (even though the role of common know
ledge makes this position problematic). On the other hand, certain
linguists - e.g., Botha, Derwing, and Hutchinson - who also ad
here to the posti vistic conception of science, have come to the con
clusion that grammars, at least those of the TG variety, are not em
pirical theories (cf. 3.7. above). Since, in their view, "the
subject matter of science is the class of public, not private, events"
(Derwing 1973:248), they are led to argue for a new conception of
grammar. According to it, the existing methods of grammatical descrip
tion must be modified in such a way that grammars become directly open
to experimental falsification, given that the only alternative to ex
perimentation is thought to be speculation (Derwing 1973:306-08). The
experiments envisaged here would be mainly of a psycholinguistic nature,
with the consequence that the distinction between grammar and psycho-
linguistics would be effectively eliminated.
I largely agree with the criticism which the above-mentioned au
thors have levelled against TG's metascientific self-understanding. I
should like to mention in particular Botha's by now classical refutation
of TG's claim to psychologic reality, or of the view of TG as a form
of 'mentalistic linguistics' (Botha 1971:167-70), as well as Derwing's
(1973:270-96) demonstration of the lack of empirical testability in TG.
However, these authors fail to connect the nonempi rical nature of (TG-
type) grammars with the normativity of linguistic data. They also fail
to see that the fact of normativity opens up the possibility to subsume
grammar under the general concept of human science (cf. 2.0. above). It
is this lack of perspective, viz. the view that positivism is the only
possible philosophy of science, which makes linguists like Chomsky,
Sampson, and Kasher consider grammars simply as empirical theories, and
makes linguists like Botha, Derwing, and Hutchinson attempt to turn
grammars into empirical theories, once they have realised that grammars,
as they exist today, are nonempirical.
However, experimentation and speculation do not exhaust the possi
bilities of intellectual activity. Speculation may not be scientific,
but conceptual analysis or explication is a valid scientific alternative
to experimentation; and explication deals neither with public events
nor with private ones. In order to prove that grammar, or grammatical
explication (cf. ll.0. below), cannot be replaced by experimental psy
cholinguistics, I simply refer to the obvious truth that description of
games cannot be replaced by description of game-performances. The re
levance of this truth to linguistics resides in turn in the analogy be
tween rules of language and rules of game, which I have proved in 5.0.
and 6.0. Experimentation with language presupposes knowledge of langu
age, and grammatical descriptions concentrate upon this knowledge.
The same argument which has been used here against the primacy of
psycholinguistics, can also be used against the primacy of socioling-
uistics (cf. 7.1.). 1 0 5
7.5. The Position of Transformational Grammar vis-à-vis
Linguistic Normativity
TG tries to dispense with linguistic normativity roughly in the
same - unsuccessful - way as Sampson has recently proposed. In
7.1.-2. (above) it became evident that, as against Chomsky's opinion,
grammars do not deal just with observable events, and rules of grammar
are not comparable to theoretical hypotheses of natural science. On the
other hand, TG's position is further complicated, and is indeed made in
consistent, by the fact that it also wants to account for the existence
of linguistic intuition. In what follows, I shall examine, against this
background, how TG comes to terms with the rule - regularity distinction.
Given that TG deals with intuition, i.e., rules of language as
they are intuitively known, but aspires to be a natural science, it is
not surprising that its official position on the rule - regularity
dichotomy is rather confused. Since natural sciences do not (in fact,
cannot) recognise the existence of rules and norms within their data,
it is to be expected that TG professes to be investigating regularities,
not rules. This is indeed what is asserted in most methodological sta-
tements made within TG. But it has also been claimed that, due to
the degenerate quality of actual speech and to the novelty of utter
ances, there are in speech in the sense of Saussure's 'parole' no
regularities to be described (Chomsky 1966:32, n.8), - a claim which
is neatly contradicted for instance by the observation that there are
already regularities in the speech of eighteen-month-old children
(Slobin 1971:53).
As far as I can see, Chomsky's position here could be reconstruct
ed as follows. On the one hand, as a positivist, he cannot accept
(known or knowable) rules as the subject matter of grammar. On the
other, he is well aware that he is dealing with intuitive knowledge
and that, from this point of view, eventual regularities exhibited by
actually occurring utterances are simply irrelevant; so much is indeed
evident from his repudiation of statistical considerations in grammar
(cf. 9.1. below). Thus, he comes to the conclusion that TG has to do
neither with rules nor with regularities, but only with particular
(correct) sample sentences. However, a natural science not investi
gating regularities in nature is clearly a contradiction. It is asto
nishing that Chomsky has managed to avoid facing this very simple truth.
To be sure, using a vocabulary so vague as to make useful discussion
impossible, Chomsky has always maintained that TG aims at discovering
some 'deep' or 'basic' regularities of language (e.g., Chomsky 1965:5).
What he has in mind is simply that TG attempts to make generalisations
concerning the correct sentences of particular languages. (Notice
that all sciences, empirical and nonempirical alike, make 'generalisa
tions'; cf. 10.0. and 11.0. below.) Within TG, generalisations are
expressed through rules of grammar. From this, it can be seen that
the only type of rules which TG allows for are grammatical rules, i.e.,
components of theoretical descriptions, about which native speakers
mostly have absolutely no antecedent, intuitive knowledge. This has
several unnatural consequences. For example, although rule and correct
ness are correlative concepts, TG is forced to maintain that native
speakers are aware of the correctness of sentences without being (able
to become) aware of the rules determining their correctness. Hence,
bound by its allegiance to positivism, TG ignores the existence of
linguistic rules as socially given, as normative phenomena, and inter
prets languages simply as infinite sets of sentences (i.e., correct
sentences). Sentences are straightforwardly taken to be 'objects' of
some kind, and they are supposedly investigated by means of methods
identical with those used by natural scientists in the investigation
of their data.
It is generally acknowledged by nonpositivists that one of the
most harmful consequences of the positivistic world-view is the objec
tification, or the 'reification', of entities belonging to the human
sphere. Equating men with things, positivism treats intentions, goals,
and values on a par with theoretical concepts of physics by simply em
bedding them in presumed cause-effect relationships. This attitude is
entirely inadequate, because it draws an absolute distinction between
people as research objects and people as researchers, and ignores the
role of the latter. As for TG, this reificatory tendency becomes appa
rent in the above-mentioned fact that the social and institutional
aspects of language are discarded and language is defined as a set of
objects (i.e., sentences). Within this static and reificatory frame
work there is no way to represent the obvious connection between sen
tences and the intentional, rule-governed actions of uttering them.
Moreover, within this same positivistic framework it remains incom
prehensible why this particular set of objects (i.e., sentences), as
distinguished from sets of objects investigated by standard natural
sciences, requires its own type of knowledge (i.e., linguistic intui
tion). But within a nonpositivisti c or hermeneutic framework lin-
guisticintuition is seen to be a special case of the 'agent's knowledge',
i.e., man's knowledge about his own (possible) actions and the (social)
rules governing them.
I suspect that TG's inability to perceive the existence and the
nature of rules of language is due, not only to its allegiance to po
sitivism, but also to its emphasis on artificial languages whose sen
tences are just strings of a's and b 's. In contradistinction to nat
ural languages (and to languages of logic and mathematics), such lang
uages cannot of course be used in any meaningful sense of the word.
Consequently there can be no rules (of use) connected w i t h these langu-
ages, and they can only be defined as ( i n f i n i t e ) sets of sentences, gen-
erated or not by grammars each of which i n turn c o n s i s t i n g of a ( f i n i t e )
set of grammatical rules. On the other hand, i t would be perverse to
define natural languages in the same way, given t h a t they are spoken
and w r i t t e n , i.e., used, according to s o c i a l l y v a l i d and c o n t r o l l e d
rules. I f we have to give a d e f i n i t i o n of natural language, then i t is
most n a t u r a l , and in keeping w i t h a long t r a d i t i o n in l i n g u i s t i c s , to
i d e n t i f y i t w i t h a set of social r u l e s . Nevertheless, under the i n f l u -
ence of the a r t i f i c i a l s i m p l i c i t y of the ab-languages, TG p e r s i s t s in
d e f i n i n g natural languages too as sets of sentences.
8.0. LANGUAGE AND GRAMMAR
Metascience deals with the dialectics between scientific descrip
tion and its object. Once we know the object, we know what type of de
scription it admits of. What still remains to be done in the matter is
to analyse those descriptions which are used in fact to describe the ob
ject, and, if possible, to suggest ways of improving upon them.
As I have pointed out earlier in this study (p.16), I do not think
that whatever is investigated by a theory, exists for us only qua the
subject matter of the same theory. In this sense, then, language is in
dependent of linguistics, as is confirmed by the self-evident fact that
there were languages long before there were grammars to describe them.
Yet even if we grant that language pre-exists all particular grammatic
al theories, it is clear that we cannot scientifically speak about it in
its pregrammatical state without using a more or less theoretical appa
ratus of some kind. It is certainly the case that we possess atheore-
tical knowledge of language, but the concept 'atheoretical' is not it
self atheoretical, given that it is interdefinable with the concept
'theoretical', which is itself theoretical. Even more obviously, con
cepts like 'rule-sentence' and 'unfalsifiabi1ity!, which are needed to
analyse language, are of a theoretical nature. Consequently we are
forced to make the following distinctions: First, there is the dis
tinction between language and its theoretical analysis (this latter
being, to repeat, different from grammar). Second, there is the theo
retical (i.e., metascientific) analysis of the relation between langu
age, as theoretically analysed, and grammar.
In 4.0.-5.0. I was almost exclusively concerned with the theory of
language, i.e., with the analysis of what will in the grammatical con
text turn out to be the object of grammatical descriptions. I reached
LANGUAGE AND GRAMMAR 193
the conclusion that language is not an entity definable in spatiotem

poral terms, which means that in whatever way grammatical descriptions
are constructed, they must be fundamentally different from empirical
theories. In 9.0.-11.0, 1 shall investigate in detail the relation
ship between grammatical descriptions (of the TG variety) and language,
as previously analysed. In the present chapter I shall outline the
basic characteristics of this relation.
8.1. The Basis of the Difference between Natural Science and

Human Science: Observer's Knowledge vs. Agent's Knowledge
To understand an action is to understand why and how it is done,

which implies that one might under suitable circumstances be able to
do it oneself. To understand a rule is to be able to follow it one
self, i.e., to do the same as those whose rule-following behaviour has
revealed the existence of the rule concerned in the first place. This
means that when we understand actions or rules, we must in a sense iden
tify ourselves with those who are doing the actions or following the
rules. This simple truth underlies the widely misunderstood 'method of
Verstehen', and is sufficient to expose the qualitative difference be
tween actions and rules, on the one hand, and events and regularities,
on the other. We can never come to understand events and regularities
investigated by the natural scientist, for we are obviously unable to
identify ourselves with planets or molecules for instance. This inabi
lity of ours is not simply due to the fact that planets and molecules
'differ too much' from us, but is, rather, a purely logical point. It
is a conceptual impossibility for us to learn to follow for example
regularities exemplified by the observable behaviour of people under
hypnosis. And the reason is, of course, that it is a conceptual im
possibility (to attempt) to learn consciously and intentionally to do
something in which consciousness and intentionality seem to be entire
ly lacking.
We could sum up the preceding passage by saying that our knowledge
of events and regularities is observer's (or 'Outsider's') knowledge,
whereas our knowledge of actions and rules is agent's (or 'insider's')
knowledge. This d i s t i n c t i o n was p r e l i m i n a r i l y introduced p.40. The
basis f o r the p e c u l i a r nature of agent's knowledge can be seen i n the
f a c t t h a t , as I have noted i n connection w i t h my W i t t g e n s t e i n - i n t e r -
p r e t a t i o n , man's r e l a t i o n to his actions i s not e m p i r i c a l , but con-
ceptual. A c t i o n s , qua a c t i o n s , must be conscious, i . e . , (potential-
l y ) understood and known by those who perform them; t h i s knowledge i s
u l t i m a t e l y based on the c o l l e c t i v e i n t e r p r e t a t i o n c a r r i e d out by the
whole s o c i e t y . Now, since man cannot be separated from his a c t i o n s ,
and actions must be known by the one who performs them, i t follows
t h a t knowledge about one's a c t i o n s , whether rule-governed or n o t , is
i n f a c t the p r i n c i p a l type of self-knowledge. (And, a f t e r what has
been said in 4 . 2 . , i t should be c l e a r t h a t self-knowledge can be se-
parated from n e i t h e r knowledge of others nor knowledge by o t h e r s . )
To those who are puzzled by the d i f f e r e n c e between observer's know-
ledge and agent's knowledge, i t seems perhaps less s u r p r i s i n g that
self-knowledge should be q u a l i t a t i v e l y d i f f e r e n t from the knowledge
of the e x t e r n a l , observable w o r l d .
In the h i s t o r y of philosophy there is a long t r a d i t i o n which
d i s t i n g u i s h e s , in one way or another, between observer's knowledge and
agent's knowledge, and m a i n t a i n s , i n p a r t i c u l a r , t h a t only the l a t t e r
represents genuine or certain knowledge. For P l a t o , genuine knowledge
is comparable to the s k i l l of an a r t i s a n : the man who knows something
is the one who can produce the object of his knowledge ( c f . , e.g.,
H i n t i k k a 1974b). On the strength of the same p r i n c i p l e , Hobbes t r i e s
to demonstrate a basic s i m i l a r i t y between geometry and p o l i t i c a l scien-
ce:
The science of every subject is derived from a precognition of

the causes, generation, and construction of the same; and con
sequently where the causes are known, there is place for demon
stration, but not where the causes are to seek for. Geometry
therefore is demonstrable, for the lines of figures from which
we reason are drawn and described by ourselves; and civil phi
losophy is demonstrable, because we make the commonwealth our
selves. But because of natural bodies we know not the construc
tion, but seek it from the effects, there lies no demonstration
of what the causes be we seek for, but only of what they may be
(Hobbes quoted from Neuendorff 1973:33).
It is clear that the word 'cause' is being used ambiguously here.

The 'causes', or generative principles, of geometrical figures are the
rules in conformity with which they are drawn, not the motives which
make someone draw such figures in such and such ways. By contrast,
the causes of political behaviour are precisely those motives or pas
sions which make people behave in the ways they do.
Vico also characterises in a similar vein the general difference
between natural science and human science:
But in the night of thick darkness enveloping the earliest anti

quity, so remote from ourselves, there shines the eternal and
never failing light of a truth beyond all question: that the world
of civil society has certainly been made by men, and that its prin
ciples are therefore to be found within the modifications of our
own human mind. Whoever reflects on this cannot but marvel that
the philosophers should have bent all their energies to the study
of the world of nature, which, since God made it, He alone knows;
and that they should have neglected the study of the world of na
tions, or civil world, which, since men had made it, men could
come to know. This aberration was a consequence of that infirmi
ty of the human mind by which, immersed and buried in the body,
it naturally inclines to take notice of bodily things, and finds
the effort to attend to itself too laborious ; just as the bodily
eye sees all objects outside itself but needs a mirror to see it
self (Vico 1968[1744], §331:96-97).
In the l a s t sentence of t h i s q u o t a t i o n , Vico draws a t t e n t i o n to

the general lack of ( s e l f - ) r e f l e c t i o n i n s c i e n t i f i c t h i n k i n g . This
c r i t i c i s m has not y e t l o s t i t s a c t u a l i t y today, given t h a t p o s i t i v i s m
is s t i l l the p r e v a i l i n g philosophy of science, and t h a t "dass w i r Re-
f l e x i o n verleugnen, ist der Positivismus" (Habermas 1968:9). - Vico's
p o s i t i o n recurs i n an almost i d e n t i c a l form w i t h i n the c l a s s i c a l Ger-
man hermeneutics ( c f . the Dil they-quotations i n 2 . 4 . ) .
The t r a d i t i o n a l notion of agent's knowledge (by whatever name i t
i s c a l l e d ) contains an element of t r u t h , but i t c l e a r l y cannot be ac-
cepted as i t stands. The d i f f e r e n c e between actions and events has
been demonstrated here in d e t a i l , e s p e c i a l l y i n 4 . 0 . Nevertheless, it
i s obvious t h a t f a c t u a l actions i n v e s t i g a t e d by empirical sociology are
located i n space and time j u s t l i k e f a c t u a l events i n v e s t i g a t e d by na-
t u r a l science. This means t h a t they share w i t h the l a t t e r the
properties stemming from man's i n e v i t a b l e epistemologi cal l i m i t a t i o n s
vis-ã-vis space and time. On a priori grounds, then, knowledge about
regularities of actions is just as hypothetical as knowledge about re
gularities of events. However, from the general creativity and histo
rical i ty of human behaviour it follows, in addition, that knowledge
about regularities of actions is more hypothetical or uncertain than
knowledge about regularities of events (cf. the Weber-quotation p.26).
Therefore, if genuine knowledge is equated with certain knowledge, na
tural science represents genuine knowledge to a higher degree than em
pirical sociology or political science.
In other words, even though I understand people's actions in a way
in which I cannot hope to understand physical events, I know what people
have done or will do with much less certainty than I know which events
have occurred or will occur. The same argument applies even to myself.
It may be true that in many instances I know with a very high degree of
certainty what I have done. But I am much less sure about my future
actions. This is a joint result of my free will and my being subject
to unforeseeable physical and emotional influences. However, although
the knowledge about actions which have been, are, or will be done is
necessarily uncertain, and more uncertain than the knowledge about past,
present, or future events, there is one type of knowledge of actions,
or one type of agent's knowledge, which is absolutely certain, i.e.,
more certain than knowledge of events. It is, of course, the knowledge
of what ought to be done, or the knowledge of rules. That is, normative
knowledge is a type of agent's knowledge, and in an extremely large num
ber of representative cases it is absolutely certain (cf. 5.3.-4. above).
The difference between invariably uncertain observer's knowledge and
agent's knowledge qua certain knowledge, i.e., normative knowledge, is
manifested as the difference between empirical hypotheses and (unfalsi-
fiable) rule-sentences (cf. 6.0. above).
All actions, rule-following or not, are subject to the disturbing
influence of unpredictable physical or emotional factors, but rules
themselves are not. Therefore such factors, while making knowledge of
the factual occurrence of actions quite uncertain, cannot have a simi
lar effect upon knowledge of rules. Similarly, since rules exist as
objects of common knowledge, and not in space and time, (subjective)

knowledge about them, once they have been learned, cannot be inhibited
by man's epistemological limitations vis-à-vis space and time. Further
more since rules do not exist in space and time, knowledge about them
cannot be in any sense observational; therefore it is appropriate to
call it 'intuitive'. This non-observational nature is precisely the
defining property of so-called 'practical knowledge' (cf. Anscombe
1958:49-57): I cannot observe the action which I intend to perform.
And when my intended action is a normative one, I can just as little
observe those rules in conformity with which I intend to perform it.
True, there is an element of observation involved when I under
stand an action which has been performed in my presence, given that
each factual action has its physical substratum. However, in each
particular case knowledge of actions, or agent's knowledge, is first
and foremost non-observational, because it is acquired through under
standing which pertains to such unobservable entities as intentions,
meanings, or rules (cf. 5.2.-3. above).
To sum up: Agent's knowledge is certain, and hence 'superior' to
observer's knowledge, only when it is about what ought to be done. Even
more narrowly, this 'ought' does not comprise the ethical aspect of
actions; rather, it is the 'ought' which applies to actions purport
ing to conform to such clear-cut rules as rules of language, logic, or
geometry (cf. the Hobbes-quotation above). On the other hand, even if
other types of agent's knowledge are much less certain, they remain
qualitatively different from observer's knowledge.
The distinction here at issue is often conceptualised as that
between 'practical' knowledge and 'theoretical' (or 'speculative' or
'contemplative') knowledge (cf. Hintikka 1974c). This dichotomy has
its own merits but it is clearly different from the one which I have
in mind, given that there are, on the one hand, grammatical or logical
theories based on agent's knowledge and, on the other, physical theories
based on observer's knowledge.
In Hintikka's opinion, it is pointless to try to base the 'quest
for certainty' on the notion of agent's knowledge (or 'maker's know
ledge 1 ). Such a view is comprehensible because Hintikka pays no atten-
19 8 GRAMMATICAL THEORY AND METASCIENCE
tion to rules. It is for the same reason, I think, that when dis
cussing Plato's concept of episteme, he fails to connect it with the
concept of anamnesis (cf. 8.3. below). And yet, for Plato, only 'forms'
or 'ideas' are objects of genuine knowledge, and this knowledge is
acquired, not by sense-impression, but by recollection (cf. Plato 1963b,
§73-76). Finally, Hintikka subsumes both natural science and human
science under the label of 'maker's knowledge':
... modern experience may be said to demonstrate how little truly

intentional action there is even among the phenomena studied in
Vico's "New Science" - that is, in language, culture, history,
literature, and politics. Man's mastery over his physical envi
ronment has opened a much larger scope for maker's knowledge than
the most rudimentary control he exercises over his society or his
culture (Hintikka 1974c:83).
If there is "little truly intentional action" among human phenomena,

there surely is even less among physical phenomena. More seriously, the
'behaviour' of physical things can be predicted and manipulated better
than the behaviour of human beings, but I am not concerned with this
self-evident truth. From my point of view, there is precisely the
same amount of intentionality in language for example as there is in
logic or geometry, conceived as the subject matters of the correspond
ing* disciplines: I can intentionally utter a correct sentence just as
I can intentionally make a valid inference or draw a (correct) geomet
rical figure. By contrast, there is, to repeat, not the smallest bit
of intentionality within the physical reality; what is intentional, is
the behaviour of the one who is investigating or manipulating the phy
sical reality. Moreover, in one decisive respect the control over the
social reality is, after all, better than that over the physical reali
ty. The laws of nature are immutable. But it is possible for man to
change the regularities governing social behaviour (cf. 6.3. above).
This is the idea behind the certainly unobjectionable exhortation that
man should strive to minimise all that makes him a merely natural being,
and to become a self-controlling, wholly human (or 'historical') being.
8.2. The Two-Level Nature of the Human Sciences :

Atheoretical vs. Theoretical
It has become amply clear that the difference between actions and
rules on the one hand, and events and regularities on the other, is
that between intentionality (or consciousness) and the lack of it.
The category of actions is considered here as representative of the
larger category of social behaviour, which means that social behavi
our qua social is at least potentially conscious of itself. Further
more, science is a social activity which follows its own procedural
rules, and is at the same time anxious to improve them. From this,
it follows that a science investigating any type of social behaviour
must, of necessity, involve tuo types of consciousness and of know
ledge, viz. that of the research objects on the one hand, and that
of the researchers on the other. For obvious reasons, I call these
two types of knowledge 'atheoretical' and 'theoretical', respective
ly. Together, they make up the 'two-level' character of any social
or human science.
Normative human sciences like grammar, logic, or philosophy con
centrate upon a given body of atheoretical knowledge, i.e., upon an
atheoretical conceptual system or institution. Empirical human scien
ces like psychology or sociology concentrate upon what is done as a
matter of fact, as atheoretically understood by means of concepts pro
vided by such systems or institutions. (It goes without saying that
behaviour so understood is explained by means of theoretical concepts
which are not available to the research objects, i.e., ordinary peop
le.) In conformity with my interest in the metascience of grammar,
I am concerned here primarily with the 'atheoretical - theoretical'
distinction insofar as it applies to normative human sciences.
The natural sciences are identifiable as one-level theories:
their research objects either have no consciousness or are treated
as if they had none. In other words, the natural sciences involve
only the consciousness of the researchers. This implies that the
difference between the layman's atheoretical knowledge about regula
rities in nature and the natural scientist's corresponding theoretic
al knowledge is one not in kind, but in degree. In fact, the know
ledge possessed by each of them is equally hypothetical, and hence
falsifiable by new events. A confirmation of this one-level charac
ter can be seen in the fact that, although there are well-known dif-
ferences between types of universal hypotheses, e.g., between simple or
inductive generalisations and theoretical hypotheses, these differences
are to a large extent treated as philosophically secondary. For example,
Hempel (1965) illustrates his theories of confirmation and explanation
with hypotheses as simple as "All swans are white" and "All pieces of
metal expand when heated".
In other words, the relation of theoretical thinking to atheoretic-
al thinking differs according as it holds within natural science or with
in human science; this difference could be expressed succintly as follows:
In natural science theoretical thinking replaces atheoretical thinking;
they both pertain to the same reality, but in different ways. In human
science, by contrast, theoretical thinking vervains to atheoretical
thinking, either exclusively, as in grammar for example, or partially,
as in empirical sociology. From the point of view of theoretical phy
sics, atheoretical thinking about physical events and regularities pos
sesses no scientific interest whatever; and the reason is of course that
such thinking is very often incorrect. By contrast, a man capable only
of atheoretical thinking is the ultimate authority on several questions
which are an inseparable part of what human sciences are investigating,
for instance: Does he think that the food prices are too high? or: How
did he understand such and such an action?
The distinction between atheoretical and theoretical is clearest
in a descriptive normative science like grammar, where atheoretical
knowledge is (to a large extent) certain and theoretical knowledge is
(as always) uncertain or, to put it in verbal terms, where atheoretic
al rule-sentences are unfalsifiable, and theoretical grammars are fal-
sifiable. However, in addition to descriptive human sciences, there
are also prescriptive and critical human sciences, whether normative or
not. Such sciences do not just accept atheoretical thinking as it is,
but try to improve it or in part even to reject it.
By way of a summary, we may quote Schutz' (1962:5-6) characterisa
tion of the essential difference between natural and human (or social)
sciences:
The facts, data, and events with which the natural scientist has
to deal are just facts, data, and events within his observational
field but this field does not "mean" anything to the molecules,
atoms, and electrons therein. But the facts, events, and data
before the social scientist are of an entirely different struc
ture. His observational field, the social world, is not essen
tially structureless. It has a particular meaning and relevance
structure for the human beings living, thinking, and acting the
rein. They have preselected and preinterpreted this world by a
series of common-sense constructs of the reality of daily life,
and it is these thought objects which determine their behaviour,
define the goal of their action, the means available for attain
ing them - in brief, which help them to find their bearings
within their natural and sociocultural environment and to come
to terms with it. The thought objects constructed by the social
scientists refer to and are founded upon the thought objects con
structed by the common-sense thought of man living his everyday
life among his fellow-men. Thus, the constructs used by the
social scientist are, so to speak, constructs of the second de
gree, namely constructs of the constructs made by the actors on
the social scene, whose behaviour the scientist observes and
tries to explain in accordance with the procedural rules of his
science.108
Atheoretical knowledge as here defined is roughly identical with

what hermeneutic philosophers cal1 Vorverständni s. Instead of using
the term 'atheoretical', representatives of the analytical philosophy
have variously spoken of 'presystematic' , 'preanalytic', or 'prescien-
tifie' knowledge. A large part of atheoretical knowledge is not
'knowledge that', but 'knowledge how', i.e., knowledge which cannot be
readily verbalised. A. favourite example is the knowledge about how
to ride a bicycle. It is commonly assumed that linguistic knowledge
too is of this type (e.g., Lewis 1969:63-64, and Robinson 1972:19).
The incorrectness of this view is evident already on quite formal
grounds: the knowledge about how to speak is normative in character,
whereas the knowledge about how to ride a bicycle is not. More con
cretely, I have already shown that this is indeed the case, viz. that
(unfalsifiable) rule-sentences about atheoretical linguistic knowledge
are quite easy to formulate.
As Schutz points out in the quoted passage, the theoretical con
cepts utilised by a social scientist must ultimately be based upon,
or tied to the atheoretical concepts that are utilised by the people
investigated by the scientist. This means, more precisely, that the
criteria used to determine the identity of the basic units of a social
science must be those provided by atheoretical thinking. Within lin
guistics. this requirement is self-evident because it is of course not
the linguist, but the native speaker, who decides which successions
of sounds are or are not utterances of his language, which utterances
are similar to or different from each other, etc.109 Within other
social or human sciences the requirement in question may seem more
controversial, since it could be taken as a threat to the theoreti
cian's autonomy vis-à-vis his subject matter. However, the following
illustration by Winch of the determining role of atheoretical concepts
seems compelling enough to me even if we have to admit that sometimes
the distance between theoretical and atheoretical concepts may be much
greater:
For example, liquidity preference is a technical concept of eco

nomics: it is not generally used by business men in the conduct
of their affairs but by the economist who wishes to explain the
nature and consequences of certain kinds of business behaviour.
But it is logically tied to concepts which do enter into busi
ness activity, for its use by the economist presupposes his
understanding of what it is to conduct a business, which in
turn involves an understanding of such business concepts as
money, profit, cost, risk, etc. It is only the relation be
tween his account and these concepts which makes it an account
of economic activity as opposed, say to a piece of theology
(Winch 1958:89) .
Mehtonen (1971) has correctly pointed out that Winch's position

in effect contains the far-reaching implication that practice deter
mines the nature of theory. The same view is espoused by the Erlangen
school which constructs sciences like physics and logic out of atheo
retical practice (cf. 2.5.-6. above). However, the relation of deter
mination can be here only a partial one - a certain type of practice
does not with causal necessity bring about a certain type of theory -
and so we are left with the problem of distinguishing between what is
determined and what is not, and of explaining why a theory qua undeter
mined is the way it is. As regards the social sciences, moreover, it
is clear that in societies with conflicting social classes and, hence,
with conflicting forms of practice, there obtains the necessity of a
choice between these different alternatives (cf. 6.3. above). Thus
there remains the question both about those reasons and about those
causes, i.e., unconscious and/or external factors, which lead to the
choice of one alternative over another.
In the light of the preceding discussion, it is clear that a pu
tative human science which would disregard people's own criteria of
conceptualising and classifying phenomena would eo ipso disqualify it
self as a human science and be, instead, a natural science only accident
ally investigating human beings, i.e., objects that would in some other
context be identifiable as human beings. Biology, for instance, meets
this characterisation. However, the interesting thing is that, con
trary to the explicit intentions of positivistically-minded psycholo
gists and sociologists, it is simply impossible for them to treat human
beings strictly on a par with inanimate objects. As Taylor (1964) in
particular has pointed out, even within behaviourism, which does every
thing in its power to imitate the methodology of natural science, and
with apparent success, the phenomena under study are in the last ressort
classified, or interpreted, according to criteria borrowed from the gen
eral atheoretical knowledge which the behaviourists share with their re
search objects (cf. 2.1. above). This is the only possible reason why
certain objectively measurable features of external behaviour are taken
as defining the notions 'motivation' and 'learning' for example, and
not just some arbitrary notions 'X' and 'Y'. More strikingly, even with
in animal psychology the psychologists cannot help projecting the gener
al atheoretical notions of 'deprivation', 'gratification', etc. into the
supposedly purely observable behaviour of animals (Taylor 1964:63-71 and
87-97).
There must be a constant mediation between atheoretical and theore
tical because, on the one hand, theory has grown out of atheoretical
thinking and, on the other, the scientist can describe new social phe
nomena only if he has acquired the atheoretical knowledge which, for
the people involved, constitutes these phenomena as what they are. This
mediating link between atheoretical and theoretical is provided by under
standing. Hence, understanding proceeds both horizontally and vertic
ally, so to say. In the vertical direction it connects two different
types of agent's knowledge. The emancipatory p o t e n t i a l of sciences
l i k e psychoanalysis or c r i t i c a l sociology resides p r e c i s e l y i n the
f a c t t h a t people are able to understand the r e s u l t s of a science
which has been i n v e s t i g a t i n g them, and to change t h e i r behaviour and
t h i n k i n g in the l i g h t of such r e s u l t s . The necessity of a mediation
between a t h e o r e t i c a l and t h e o r e t i c a l proves t h a t the d i s t i n c t i o n be-
tween the two must be a r e l a t i v e one ( c f . 4 . 2 . 4 . above). Considered
from this p o i n t of view, the human sciences are one-level sciences,
because researchers stand i n an i n t e r n a l r e l a t i o n , or are p a r t l y iden-
t i c a l , w i t h t h e i r research o b j e c t s . By contrasts the natural sciences
are in this respect two-level sciences because they involve an abso-
1 ute s u b j e c t - o b j e c t - d i v i s i o n .
One can acquire the a t h e o r e t i c a l knowledge of a community only
by, in a sense, i d e n t i f y i n g oneself w i t h the members of t h i s communi-
t y , t h a t i s , by understanding the ( a t h e o r e t i c a l ) rules which they
f o l l o w in acting and, more g e n e r a l l y , in i n t e r p r e t i n g t h e i r life;
t h i s kind of understanding is out of question w i t h i n the natural
sciences ( c f . 8 . 1 . above). Nothing more special i s meant by the me-
thod of Verstehen:
The fact that in common-sense thinking we take for granted our ac
tual and potential knowledge of the meaning of human actions and
their products, is, I suggest, precisely what social scientists
want to express if they speak of understanding or Verstehen as a
technique of dealing with human affairs (Schutz 1962:56).
It is a gross misunderstanding to interpret Verstehen as requir

ing some mysterious, essentially subjectivistic powers of imagination,
which of course would be dispensable from the point of view of scienti
fic research aiming at intersubjective validity. This misconception
was first presented to the positivistic audience in Abel (1948), and it
is widely shared still today. For instance, in Nagel (1961: 473-85) it
is maintained that understanding pertains to 'subjective' or 'private'
states. It is certainly easy to criticise such a (fictitious) position,
considering that private states have been proved logically impossible
(cf. 4.0. above).
The simplest case of Verstehen is understanding the meaning of

some ordinary word or action. It should be clear that whether the
meaning of a word like 'boy' has been understood or not, can be de
cided by means of objective criteria. It is not required that the
meaning of 'boy' be understood in some mysteriously deep sense, but
only that it be understood. It is just as easy to find out whether
someone understands the meanings of words like 'time' and 'knowledge'
as they occur in some tyoical contexts. It is only when philosophers
try to capture the total or real meaning of 'time' or 'knowledge' that
difficulties arise (cf. 5.3. above). On the other hand, when we set
out to describe societies that differ markedly from our own, the role
of so-called 'empathy', or Einfühlung, obviously increases. But
there are still intersubjective methods of checking the correctness
of our description, although these methods are not as simple as the
confirmation and the falsification used within natural science. It
is interesting to note that in the investigation of ancient and/or pri
mitive societies, gaining the requisite atheoretical knowledge, which
is then to be subjected to a theoretical treatment, may prove to be
more difficult and methodologically more demanding, than conducting
the theoretical analysis itself, It seems to me that the situation
is not too different in certain areas of the study of art or, for that
matter, in many cases of psychoanalytic treatment.
In natural science the atheoretical - theoretical distinction corre
sponds to some extent to that between observational concepts and theore
tical concepts. The strategy of theoretical-cum-observational concepts
does not work within the human sciences (cf. 2.2. and 7.2. above), and
therefore the atheoretical - theoretical distinction has here a differ
ent meaning. Furthermore, the rule-conception operating with the atheo
retical - theoretical distinction must not be confused with a conception
which distinguishes between an 'internal' and an 'external' attitude in
regard to rules, in such a way that the former is characteristic of
those who act according to rules and thus know them 'from the inside',
whereas the latter is characteristic of those who are describing rules.
This conception is espoused in Miller & Isard (1967), for example, and
it has been defended more recently in Gumb (1972, esp. pp.48-53).
I do not think that the 'internal/external' account of rules is
defensible, even if it is partly correct insofar as it at least recog-
nis.es (but then also distorts) the social and normative aspect of rules.
As I have pointed out already, there can be no external attitude in
regard to rules: one who has not('internally', or intuitively) under
stood rules, cannot (theoretically) describe them at all. The most
that he can do is to describe observable (statistical) regularities
of sounds and movements. Therefore, description of rules can only
build upon their atheoretical, 'internal' understanding; it cannot be
separated from it as a totally different, unrelated attitude. This is
also why Gumb and others go astray when they think that, from the exter
nal point of view, the difference between rules and regularities is not
methodological, but merely heuristic: If rule-governed behaviour is re
ally viewed externally, i.e., as a series of recurrent events consist
ing in sounds and movements which are not (and could not be) understood,
then there is absolutely no difference, not even a heuristic one, be
tween rules and regularities; on the other hand, if one sets about to
describe actions which one has understood by relating them to rules
which one knows, then any 'external' point of view is a fiction (due,
once again, to an uncritical imitation of the methodology of the natu
ral sciences), and the difference between rules and regularities cer
tainly has the greatest methodological importance.
It might also be pointed out that espousing the standpoint of 'in-
tensional positivism', which admits the meaningfulness of actions and
thus separates them from events, does not significantly improve the po
sition of those who deny the methodological importance of the rule -
regularity distinction: If they have not understood that actions which
are in fact rule-governed, are rule-governed, then they have misunder
stood them; and if they start to describe these actions on the basis
of the above-mentioned misunderstanding, they are indeed describing
regularities, i.e., regularities of actions, but then, again, their
description has nothing to do with a description of rules. One either
understands or does not understand the rules involved; there is no third
possibility. One cannot gradually pass from regularities (whether of
events or actions) to rules;110 nor can one arrive at a description of
rules by describing regularities. In actual practice, Gumb and others
always make sure that they have ('internally') understood the rules which
they are going ('externally') to describe; but once the description has
begun, they, in the name of 'scientific methodology', pretend not to
have understood the rules, thus deceptively placing them on a par with
regularities.
The same type of argument has also been forwarded by Hart (1961:87):
If, however, the observer really keeps austerely to this extreme
external point of view and does not give any account of the manner
in which members of the group who accept the rules view their own
regular behaviour, his description of their life cannot be in terms
of rules at all, and so not in the terms of the rule-dependent no
tions of obligation or duty. Instead, it will be in terms of ob
servable regularities of conduct, predictions, probabilities, and
signs. For such an observer, deviations by a member of the group
from normal conduct will be a sign that hostile reaction is likely
to follow, and nothing more.
Nevertheless, I cannot accept the 'internal/external' account of

rules in that form either in which it is presented by Hart. For him,
the 'external' point of view is characteristic also of the one who under
stands the rules, but does not accept them:
[The] attitude of shared acceptance of rules is to be contrasted

with that of an observer who records ab extra the fact that a so
cial group accepts such rules but does not himself accept them.
The natural expression of this external point of view is not "It
is the law that..." but "In England they recognize as law ...
whatever the Queen in Parliament enacts..." (Hart 1961:99).
Within my conception this distinction between acceptance and non-

acceptance of rules is accounted for by noting that rule-sentences are
necessary while the rules which they refer to exist contingently (cf.
6.1.; esp. n.88). However, this distinction is of course entirely dif
ferent from that between understanding rules and observing regularities.
Hart seems to have confused these two distinctions, with the consequence
that his position comes rather close to Gumb's, after all.
It may be added that Hart's account of the 'internal' aspect of
rules is not quite satisfactory, either. He argues convincingly that
the normative element of rules cannot be explained away, for instance
in terms of behavioural regularities and/or sanctions (Hart 1961:6-12,
54-56, 86-88). Yet he is less convincing when he tries to explain what,
precisely, the normativity of rules consists in. He emphasises, again
correctly (p.56), that it is not a matter of subjective feeling; but
his answer, which is to the effect that it is a matter of 'critical re
flective attitude', does not quite succeed in establishing the basis
for the intersubjective validity of rules. In my opinion, such a re
sult is unavoidable as long as one does not come upon the notion of
common 'knowledge,
8.3. The Two-Level Nature of Grammar
In conformity with the fact that a human science cannot start un
til the relevant atheoretical knowledge has been acquired, a linguist
ic description cannot be made until the linguist has learned the langu
age to be described. As I have mentioned before, the use of informants
is nothing but a way of speeding up this language-learning process.
For the sake of illustration, consider the case of a linguist facing
a language unknown to him. As transformationalists have emphasised in
particular, a large part of the linguistic data available to the langu
age learner is incorrect in one way or another. Therefore, to be able
to write a grammar of the language in question, the linguist must be
capable of discarding incorrect or accidental forms and uses of langu
age; in other words, he must come to understand the rules of this langu
age. If we assume that he would be collecting his data in the manner
of a natural scientist, the result of his description could not possi
bly represent the language concerned in any relevant sense. First of all,
it is totally unclear by which criteria he would be identifying the ba
sic units in the data to be described. And even if we by-pass this in
itself insurmountable difficulty, it is still the case that, since in
actual speech there occurs a certain number of incorrect forms, and
since within natural science all occurrences are equally relevant, a
linguist collecting his data in the manner of a natural scientist could
not distinguish between correct and incorrect, or between rules and vio-
lations of rules (cf. 7.1. above). Consequently, his description would

be based on the average of all so far observed occurrences, whether cor
rect or incorrect, and would necessarily be of a statistical nature.
But an adequate description of linguistic rules cannot be statistical
(for qualifications of this claim, see 5.4. above).
In the course of language acquisition children must develop an abi
lity to understand rules, i.e., to distinguish between correct and in
correct uses of language. TG assumes that they are aided in this by a
full-fledged innate linguistic theory. This is unnecessary, and even
misleading, because normativity is by no means restricted to linguistic
behaviour but is, rather, characteristic of human behaviour in general,
so that the ability to distinguish between correct and incorrect must
be equally general,
The paradigmatic case of language description is the one in which
the linguist has learned, and tnus knows, the language which he is go
ing to describe. As far as typical games are concerned, it is undispu-
table that once they, i.e., their rules, have been learned, they are
learned in their entirety, and no new facts are relevant to them. One
may still learn new strategies to apply the rules, that is, ways to im
prove one's playing, but this is irrelevant to the knowledge of rules as
such. Now I claim that the same is true of language as well: once the
atheoretical knowledge of a language has been acquired, it is acquired
in its entirety, and no new facts are relevant to it. There are a few
rather obvious objections against this claim, but none of them is of
serious consequence: i) It is impossible to tell when, precisely, a
language has been learned. - This is true, but irrelevant (cf. p.37
above), ii) There are instances where one starts learning a language,
but never fully learns it. - Again, this is true, but irrelevant to
the present argument, iii) The number of words contained in the le
xicon of any given language is indefinite; therefore it can never be
said with certainty that someone knows all the words of a given langu
age. - This is true, and requires that "learning a language L" is de
fined as equivalent, roughly, to "learning the phonological, morpholo
gical, syntactic, and pragmatic rules of L as well as the 'central' part
of the lexicon of L". It would be unacceptable to say that a normal
English-speaking adult has not learned, and does not know, English
just because he does not know and still must learn new words related
to the latest technological advances for instance, iv) No one can
fully learn a language because all languages keep changing. - It
must be possible to abstract synchronic states in one way or another
from the diachrony of languages. It would be unacceptable to argue
that the ancient Romans had not learned Latin just because they had
not learned French.
What I have been saying here is plain common sense.to anyone who
has ever been describing a language which he knows - primarily, but by
no means exclusively, his native language. In such a case, it never
happens that while making the description, one discovers for example
new prepositions, new morphological affixes, or new grammatical pat
terns, of which one had no previous knowledge. At any given moment
one may be unaware of such linguistic facts, simply because man's po
wers of attention are limited, but once any such facts are brought to
his attention, he must admit that he in effect knew them already. This
claim is merely a restatement of my view that rules of language, as
previously defined, are what grammars are describing. On the other
hand, the (abstract) relations between rules are not part of one's
knowledge of language. Such relationships are expressed through theo
retical generalisations like the subject-raising transformation or the
A-over-A principle, of which no one could have any previous knowledge
and which therefore remain falsifiable3. Yet all such theoretical gen
eralisations and claims can be shown to be about a range of linguistic
(normative) facts, i.e., rules, each of which was known separately be
fore the description got started. This is the distinction between a-
theoretical and theoretical as it obtains in grammar. - So I repeat
that once the (atheoretical) knowledge of the rules of a given langu
age has been acquired, it is acquired in its entirety, and no new facts
are relevant to it.
The next question to be answered is: How is knowledge of this kind
investigated and described? Although this is obviously one of the central
questions of theoretical linguistics, it has received very little atten

tion from the part of linguists, probably because of the misconception
that only the 'context of justification' is worthy of scientific treat
ment (cf. pp. 74-75). It is not suprising, then, that only philosophers,
more precisely those in the Wittgensteinian tradition, have devoted de
tailed attention to the question how knowledge of language is analysed
(cf., e.g., Hare 1971, Cavell 1971a and b, and Henson 1971). They ag
ree that, when we analyse our language for philosophical purposes (though
the same is also true of a purely linguistic investigation), we are not
looking for any new facts. Rather, all the facts are given once and
for all, and the question is what we are to make of them.
We know all these facts in an intuitive and unsystematic way, but
we wish to come to know them in some more disciplined way. We feel that
there are problems connected with these intuitively known facts, but we
cannot justify this feeling, let alone find solutions to the problems
felt, until we have rearranged and systematised the facts involved. We
are looking for a descriptive system which would 'explain' (i.e., ex-
plain3) the facts by giving a somehow coherent account of them, viz. by
bringing them into an illuminating, and hence 'true', relationship with
one another. Because any such system is theoretical, it is also falsi -
fiable 3 . If it can subsequently be demonstrated as unfalsifiable3, or
necessarily true, it ceases to be a theory and turns into an 'instrument
of language' (cf. p.l54above).
As we just noticed, any attempted systematisation is a creative act,
i.e., it brings into existence something new viz. something which is of
the theoretical order. It seems perfectly proper to say, then, that the
analysis brings about new 'facts', which may in turn become the object
of analysis. But these new facts are of a different kind than the ini
tial, intuitively known or atheoretical facts. In this sense it remains
true that the analysis does not require looking for new, as yet unknown
facts (that is, facts of the same kind as the known ones), but making
the available knowledge explicit. The actual description may be carried
out with varying degrees of formalisation of course (cf. 11.0. below).
It is clear that we have to do here with a process of coming to
know better what one already knows in an indubitable, but merely intui
tive way. To this end, as Wittgenstein himself notes, one has to re-
mind oneself of the way in which expressions are used and of the diffe
rent constructions in which they occur, or may occur. The same idea,
that one has not to search for new facts, but to remind oneself of
those which one knows already, is also emphasised for instance in Hare
(1971:237), Cavell (1971a:147) and (1971b:184), Henson (1971:214), Sear
le (1969:13-14), and Vendler (1967:18-19). Notice in particular that
here one is not asked to remember how he or someone else has as a mat
ter of fact used a certain word for instance, something which would
be a more or less empirical question, but, rather, how this word is to
be used, which is an entirely different, i.e., conceptual and normative,
question (cf.Henson 1971:325, n.5).
This process of sharpening one's intuitive knowledge has been per
tinently characterised by Specht (1969:132-33), who in this context coins
the term 'immanent reflexion':
So far as there are rules of usage for a word they must ... be de
rivable from the use of the word alone. Wittgenstein thereby makes
exclusive use of a method which we wish to call "immanent reflection
on linguistic use", the possibility of which rests on the fol
lowing fact: as children we are brought up to use the words of the
language exactly as they are normally used in the linguistic com
munity. The consequence is that all members of the community re
ally do follow the same rules. Without this homogeneity of lin
guistic use our language would lose its character as a general me
thod of communication. Now, if, as an adult, one is asked about
the rules of linguistic usage one only needs to reflect on how
words are used in everyday linguistic practice. One already uses
words, of course, in accordance with the rules that have been in
culcated into one, and can, therefore, recognise them by reflect
ing on one's own linguistic usage. This reflection is "immanent"
to the extent that one does not need to go beyond what one alrea
dy knows of linguistic use.
The method of immanent reflection is known under different names,

e.g., "rehearsal of usage" (Black 1962:90) and "exhibition analysis"
(Korner 1957:763). Because it necessarily pertains to what is first
learned and then remembered, it remains, as a method of gaining (theo
retical) knowledge, absolutely different from comparable methods with
in the empirical sciences. First of all, the memory of past experiments
and observations plays an indispensable, but nevertheless only partial,

role in the empirical data-collection: Empirical science is essential
ly science of space and time. Therefore it is necessarily open towards
the future; but it is logically impossible to know, and hence to remem
ber, future events. Moreover, it is vractieally impossible to know most
of the past events; therefore, they cannot be remembered, but must be
empirically discovered. (Notice that remembering an event is not a
form of empirical discovery; cf. Hare 1971:225 and 229.) Secondly, it
is not only the case that natural science does not collect its data on
the strength of memory about what has happened; it is, if possible, even
more obvious that natural science does not collect its data on the
strength of memory about what ought to happen, or about any other norma
tive matters. Yet it is only in this way that sciences like grammar,
philosophy, or logic, in brief: all normative sciences, collect their
data. We are once again confronted here with the nonempiricai nature
of normative sciences (cf. 6.0. above).
Hare (1971:239) has pointed out that the role of recollection in
linguistic analysis, e.g., in the analysis of the meaning of 'right',
offers a certain justification to Plato's concept of anamnesis:
Plato is right in implying that in recognising that such

a proposition [i.e., "It is always right to give a madman
back his weapons which he entrusted to us when sane"] is
not analytic we are relying on our memories. It is an exam
ple of the perceptive genius of that great logician, that
in spite of being altogether at sea concerning the source
of our philosophical knowledge; and in spite of the fact
that his use of the material mode of speech misled him as
to the status of the analyses he was looking for - that
in spite of all this he spotted the very close logical
analogies between philosophical discoveries and remember
ing. He was wrong in supposing that we are remembering
something that we learnt in a former life … What we are
actually remembering is what we learnt on our mother's
knees, and cannot remember learning.
Moreover, Plato's claim that genuine knowledge, or episteme, is

the result of recollection is directly supported by the fact that it
is only knowledge of rules which is absolutely certain (cf. 5.3. above),
and that such knowledge, i.e., knowledge about what one ought to do, is
indeed actualised through recollection (as we have just seen). It goes
without saying that this genuine knowledge (about what one ought to do)
is a case of agent's knowledge. For Plato, anamnesis pertains to ideal
norms for 'similarity', 'good', 'beautiful', 'right', etc., norms which
the actual reality can only approximate. As Husserl and Lorenzen, among
others, have pointed out, such norms are man-made constructions, whether
they pertain to ethics or to logic (cf. 2.5.-6. above).
From what has been said so far, important consequences can be drawn
concerning the scope and the limits of the notions of intuitive know
ledge and immanent reflection. It is clear that when a philosopher and a
linguist who are native speakers of English for example start to inves
tigate their native language, their respective subject matters are iden
tical. It is also clear that, whatever their more specific purposes,
there is absolutely no difference between the respective methods which
they apply in order to gain a better understanding of their linguistic
intuitions, because they cannot help relying on immanent reflection.
However, their attitudes differ as to the scope of this shared method.
Linguists study language for its own sake, and are accordingly equally
interested in forms and meanings of a language. By contrast, philoso
phers focus their attention on selected aspects of pragmatics and seman
tics, occasionally also of syntax. Philosophers neglect the study of
morphology and phonology entirely, because their interest in natural
languages derives from the fact that language is the inevitable medium
through which problems about man and the world must be approached. That
is to say, for those who are equipped with language, the world is inse
parable from language (but not from any particular language), as is evi
dent from the fact that sentences like "'X' means 'Y'" and "an X is an
Y" are freely interchangeable. This merging of language and the world
is illustrated by Specht's (1969:148) remark that an a priori sentence
like "All bachelors are unmarried" is about objects in the world, but
"depends exclusively for its truth value on the rules of usage of the
linguistic sign which signifies the object". Now, since knowledge of
language is of an intuitive nature, and language is inseparable from the

world, it follows, somewhat suprisingly, that knowledge of the world is
intuitive, too. However, the contradiction here is only apparent. Know
ledge of natural events cannot of course be intuitive. What is intuiti
ve about the knowledge of the world is, rather, its specifically lin
guistic component, that is, knowledge of the concepts expressed or creat
ed through language. If we, for the sake of clarity, consider concepts
used within the natural sciences, we perceive at once that, because con
cepts, in contrast to things and events subsumable under them, are hu
man constructions, it is only logical that they should be part of in
tuitive knowledge, i.e., agent's knowledge (cf. p.43 above).
The notions of intuitive knowledge and immanent reflection are va
lid not only within sciences like grammar and philosophy which deal,
though in different ways, with rules of language; rather, and more gen
erally, they are valid within all sciences dealing with any types of
rules of human behaviour. Some such normative sciences, e.g., logic,
have a greater prescriptive emphasis that others, e.g., grammar (cf.
6.3. above). Since knowledge of rules is analysed through immanent
reflection, and since there are rules determining the correct use of
any concepts, I conclude that immanent reflection is the general method
of conceptual analysis,
Knowledge of linguistic rules is a case of atheoretical, intuitive
knowledge. This knowledge is directly expressed in corresponding rule-
sentences. Theoretical generalisations about this knowledge, or about
the rules which are its object, are formulated by means of grammatical
hypotheses, e.g., (ordinary-language interpretations of) transformati
ons. I use the term 'grammatical hypothesis' instead of 'grammatical
rule' on purpose, because the latter term is apt to create confusion.
Grammatical hypotheses express results of immanent reflection upon athe
oretical linguistic knowledge. Consequently, rule-sentences are atheo
retical, whereas grammatical hypotheses are theoretical. This termino
logy is based on the stipulation that a grammar of a language L, con
stituted by a set of (nonempirical) grammatical hypotheses, is consi
dered here as a theoretical description of L. This stipulation is ar-
bitrary insofar as there is a perfectly good sense in which it can be
said that a description of L constituted by a set of (atheoretical)
rule-sentences describing the rules of L is also a grammatical descrip
tion of L; indeed, traditional school-grammars sometimes approximate
such an atheoretical notion of grammar. A grammar consisting merely
of rule-sentences would be unfalsifiable3, because the latter are un-
falsifiable3 . But, to repeat, I am considering grammars of L as theo
ries, and hence falsifiable3 descriptions, of L. Similarly an axioma
tic system of propositional deontic logic for instance is a (falsifi-
able 3 ) theory of the logic of obligations, permissions, and prohibi
tions .
If writing a grammar of L would mean just enumerating rule-senten
ces describing rules of L, linguistics would be a very trivial under
taking. However, knowing the rules of L is only a precondition for
writing a theoretically interesting grammar; a mere list of (unfalsi-
fiable3 or necessarily true) rule-sentences is certainly not such a
grammar. Rule-sentences simply describe atheoretical knowledge which
exists prior to any attempts at describing it. By contrast, grammat
ical hypotheses express theoretical knowledge which has been created
through immanent reflection upon atheoretical knowledge, and which did
not exist prior to attempts at creating and describing it.
This general distinction can be clearly seen in the standard case
of language description: A linguist sets out to describe a language
that he knows, i.e., of which he possesses atheoretical, intuitive
knowledge, but when his description proceeds, it produces new, theo
retical knowledge about which he has no previous intuition. From the
fact that I know1 something, it by no means follows that I also know2
how to describe this knowledge-, of mine in the best possible way. Ki-
parsky (1968:172) makes a similar point in the following way:
For example, most linguists would agree that two rules of the
form
X → Y
Ζ → Y
if not separated in the ordering by any other rules, should be
combined by factoring out their common right hand side as follows:
We would say that the braces represent a linguistically significant

generalization about these two rules. But how do we know that they
do? … There are no conscious a priori ideas of generality that
we can appeal to here in the way that we can appeal to intuitions
that reflect features of structural descriptions, such as ambigui
ty and synonymy. The processes of normal language learning being
unconscious, we have absolutely no ideas about the form of gram
mars, though we have clear ideas about the forms of sentences which
grammars account for.
In fact, Kiparsky's example illustrates the tentative character of

our theoretical or grammatical knowledge even more effectively than he
intended it to do. That is to say, today it is agreed that the use of
braces, or curly brackets, does in fact not represent a "linguistically
significant generalisation". On the contrary, "curly brackets are an
admission of defeat, since they say that no general rule exists and that
we are reduced to simply listing the cases where a rule applies" (La-
koff 1971a:291; cf. also McCawley 1972:508-13). Accordingly, both par
ticular theoretical descriptions and the general principles underlying
them are only tentative and subject to revision. This - it should be
noted - is a direct consequence from the fact that, at least initi
ally , our theoretical knowledge is not, and could not be, of the intui
tive nature.
Due to its general positivistic outlook, TG has never been able to
acknowledge the two-level nature of linguistics, which means that TG has
not been able to distinguish between atheoretical rules known by the
native speaker and theoretical-grammatical 'rules' constructed by the
linguist. This confusion has led to the generally accepted view that
native speakers (consciously) know the rules of their language only in
an unreliable and incomplete way. As a justification for this positi
on, transformationalists customarily refer to the incontestable fact
that speakers do not know the 'rules' contained in the (transformati
onal) grammar, i.e., the theoretical description, of their language.
But this 'justification' has no force whatsoever, because it is self-
evident that knowing atheoretical rules does not entail knowing their
theoretical descriptions, although the reverse must be true (cf. the
Schutz-citation p.201 above).
As an example of the above-mentioned fallacy, consider how Slobin
(1971:54) characterises what he takes to be the generally accepted no
tion of 'rule':
But there are even more stringent tests, or definitions of a rule.

Later on in his development, the child will demonstrate a normati-
ve sense of rules - that is, he will be able to judge if an ut
terance is correct with respect to some linguistic standard ...
When a child stops and corrects himself one can infer that he is
monitoring his speech against some notion of correctness ...
Perhaps a more stringent test of the sense of grammaticality is

met when the child detects deviations from the norm in the speech
of others. Three-year-olds are also heard to correct the speech
of other children (and even of their parents), though the chrono
logical relation between self-correction and correction of others
has not been established.
The most stringent criterion of grammatical judgement is response

to a direct question. One can ask the child, for example, if it
is 'better' or 'more correct' to say 'two foots' or 'two feet'.
Up to now, Slobin seems to have been discussing equivalents of my

(atheoretical) rules: His rules are normative phenomena exhibited or
instantiated by people's actions; moreover, from the fact that as sim
ple a rule as the one for correct plural endings in English is used to
illustrate these rules, it is clear that they are known or knowable by
the native speaker. But Slobin (p.55) then adds:
Note that I have left out the most stringent test for the existen
ce of rules, namely: Can the individual state the explicit rule?
As I pointed out before, using this as evidence, of course, we
would all fail the test. Since no complete and adequate grammar
of English (or any language) has yet been written, in fact none of
us knows the rules of English according to this criterion. We can
follow them and use them implicitly, but we can state them only
rarely, imperfectly, and with uncertainty.
From (atheoretical) rules of English, which even the average speak

er is able to formulate in corresponding rule-sentences, if given pro
per help, we have now suddenly moved to (theoretical) 'rules' of some
imaginary 'complete and adequate' grammar of English, which even the best
linguists are unable to formulate. And from the fact that the latter
are unknown, the conclusion is retrospectively drawn that the former
must be unknown, too., When Slobin admits that speakers are able to re
cognise deviations from the norm, but denies that the norm itself can
be known, he is asserting that contradiction which, as I have mentioned,
is generally characteristic of TG: although (in)correctness and rule
are correlative concepts, it is maintained that the one can be known
while the other cannot. To put it differently, the 'normative sense
of rules' does not really make sense, if rules are identified with
'rules' of grammar which are not known by anyone, not even by the lin
guists. Briefly, the practitioners of TG do not seem aware of the fact
that they are using the term 'rule' in widely different senses.
TG's one-level conception of linguistics has moved Postal (1968b:
274-75) to draw a comparison between 'implicit' knowledge of language
and the layman's knowledge about the regularities governing food diges
tion: in his opinion, writing a grammar is in every way similar to
discovering the biochemical processes involved in digestion. The ex
tent of the positivistic indoctrination within TG can be seen from the
fact that analogies so patently false have gone unchallenged. And yet,
a special effort is needed in order not to see that digestion is some
thing that happens to us, whereas speaking is something that we oursel
ves do. Similarly, it takes a lot of training until one is able to dis
miss the obvious fact that, since there are right and wrong ways of
speaking, speaking is a normative activity whereas digestion is an obser
vable, natural phenomenon; and it goes without saying that knowledge a-
bout norms cannot be compared to knowledge about observable events.
8.4. The Ontological Reality of Grammatical Descriptions

It is undeniable that rules exist, in some sense; and in 5.1. (above)
I have shown what I take their mode of existence to be. It is undeni
able, then, that true rule-sentences refer to something which exists, or
possesses 'ontological reality'. It is also clear that, assuming a gi
ven grammatical hypothesis to be true, it possesses ontological reality
at least in the sense that it makes a generalisation about, or systema
tises our knowledge of, a set of linguistic rules. The question is whe
ther there is a stronger or more direct sense in which grammatical de
scriptions can be said to be ontologi cally real.
In the American 'taxonomie' linguistics this question was concep
tualised as a controversy between 'God's truth' and 'hocus-pocus' posi
tions, as Householder put it in 1956. The rise of TG gave it new actu
ality, in view of TG's claim that the referents of grammatical descrip
tions are features of unconscious psychological mechanisms; in brief,
linguistic grammars and mental grammars were supposed to be either iden
tical or at least closely related. As formulated by TG, this position
contains two obvious fallacies (which were briefly indicated pp.82-84
and 116).
First: A synchronic grammarian analyses conscious knowledge, mostly
his own. This knowledge is about the concept "correct sentence in L",
which may be further subdivided into concepts "correct sound combina
tion in L", "correct case ending in L", "correct relative clause con
struction in L", etc. The theoretical-grammatical description at which
he arrives is the result of immanent reflection upon such (conceptual)
knowledge. This method of conceptual analysis, or explication, is ful
ly legitimate from the scientific point of view; logic and philosophy
(of science) for example rely exclusively on it. There is just as
little reason for trying to support descriptions of conscious linguis
tic knowledge by appealing to unconscious mechanisms as there is for
trying to support descriptions of conscious logical or philosophical
knowledge by similar appeals. From Saussure to Harris, in fact, lin
guists were content to describe language as an autonomous, non-psycho
logical entity. Today's Montague-grammarians consider language in the
same perspective.
Second: It is of course possible and indeed interesting to try to
enlarge one's point of view by trying to find out in what relation,
precisely, linguistic grammars stand to mental grammars, As we just
noticed, TG makes the hypothesis that the two types of grammars are
closely similar. It ought to be self-evident that this psycholinguis
tic hypothesis must be tested on the basis of new, independent eviden-
ce provided, above all. by psycholinguistic experimentation, and not

on the basis of those very same grammatical descriptions which, in the
first place, gave rise to the psycholinguistic hypothesis in question.
Curiously enough, TG has mostly proceeded precisely in this circular
fashion. For example, consider how Katz (1967:82-83) characterises
TG's notion of (causal) psycholinguistic explanation:
It is necessary to explain why the speaker says this [i.e. knives]

rather than knifes. The mentalist, I have argued, explains this
fact of English pluralization by crediting the speaker of English
with a linguistic description that contains both the kind of rules
and the kind of ordering restrictions that Bloomfield mentions
[i.e., "We can describe the peculiarity of these plurals knives,
mouths, and houses by saying that the final f, θ, s of the under
lying singular is replaced by v, d, z before the bound form is
added ... ; thus, the plural of knife adds not -s, but -z: 'first'
the -ƒ is replaced by -v, and 'then' the appropriate alternant -z
is added"]. The mentalist asserts that an English speaker says
knives rather than knifes because sentences whose underlying syn
tactic form is ... knife + pZ ... are produced by using such rules
and ordering restrictions to pass from this syntactic form to its
phonological realization knives.
In conformity w i t h Katz's reasoning, the f a c t t h a t an English speak

er says 'the man', and not 'man t h e ' , is causally explained by c r e d i t
ing him w i t h the corresponding r u l e which is p a r t of his psychological
competence and is u l t i m a t e l y rooted in his neuro-physiological mecha
nisms. Or when we notice t h a t in Swedish the i n d e f i n i t e a r t i c l e and
the d e f i n i t e one are d i s t r i b u t e d thus: 'en f l i c k a ' - ' f l i c k a n ' ('girl'),
we may again ' e x p l a i n ' t h i s by saying t h a t Swedish speakers have i n t e r
nalised a n e u r o - p h y s i o l o g i c a l l y based r u l e which causes them to say
'en f l i c k a ' and ' f l i c k a n ' . I t should be clear t h a t Katzian 'explanati
ons' are i n no way d i f f e r e n t from pseudo-explanations r i d i c u l e d a l r e a
dy by M o l i è r e : a drug causes people to f a l l asleep, 'because' i t pos-
sesses a ' v i r t u s d o r m i t i v a ' , or the power to cause people to f a l l a-
sleep.
More r e c e n t l y , constant c r i t i c i s m has forced TG to give up the most
obvious a b s u r d i t i e s of i t s p o s i t i o n . That i s , TG has recognised t h a t
psychological experimentation is relevant to the question of psycholo-
gical reality.' According to the current p o s i t i o n as expounded i n
Fodor & a l . ( 1 9 7 4 , chap.5), only one h a l f of l i n g u i s t i c grammars of the
TG variety is psychologically real; that is, surface structures and
deep structures are psychologically real whereas 'transformations con
necting the two are not. Moreover, in this context the term 'deep
structure' may be interpreted in different ways depending on the ex
perimental make-up. It seems that, when the attention of test persons
is directed to linguistic detail, evidence supports a notion of deep
structure determined by fixed lexical units and logically transparent
grammatical relationships. On the other hand, in experiments where
test persons are given less instructions and which, therefore, resem
ble spontaneous use of language more closely, evidence supports a notion
of deep structure where the meaning of the sentence, in whatever way
it is formally expressed, is the determining factor. In any case, even
if transformations are not psychologically real, the psychological real
ity of deep structures shows, according to Fodor and others, that, to
this extent, TG has been 'dramatically succesful', even from the psy
chological point of view. It seems clear to me, however, that this
assessment is based on a misinterpretation of the facts.
If we demythologise the notion of deep structure, we see that the
deep structure of a given sentence is essentially a paraphrase of this
same sentence, a paraphrase constructed according to more or less fixed
rules.112 For instance, the sentences "John hit his wife in a garden
with a hammer" may have as its deep structure a tree diagram containing
the simpler ('abstract') sentence "John has a wife", "He hit her", "He
used a hammer", "This happened in the garden". (Of course, this ana
lysis can be refined ad libitum.) Paraphrases, or deep structures, of
this kind are rather obvious; they amount to "saying the same thing dif
ferently", viz. more simply. Because they are so obvious, it would have
been extremely surprising, and scientifically interesting, if they had
not been psychologically real. On the other hand, transformations con
necting deep structures with surface structures, as well as their spe
cific order, are not obvious, and therefore it would have been scienti
fically interesting if they had turned out to be psychologically real.
But, as Fodor et al. admit, they did not. Because, with respect to the
scientific interest, the psychological reality of deep structures and
that of transformations are not at all on an equal footing, it follows

that, even if it could be said that one half of TG descriptions is psy
chologically real, it is in any case the less interesting half.
However, a further point is that TG descriptions cannot be divided
into two halves so to speak, Within TG the notions of deep structure
and transformation are defined in terms of each other, i.e., they are
conceptually interdependent. Now, since transformations are psycholo
gically unreal, it cannot be claimed that deep structures, qua entities
conceptually dependent on the former, are psychologically real. Or, if
such a claim is made - as it has been done -, this must mean that the
term 'deep structure' is being used in a new sense. In its new sense
'deep structure of a sentence S' is simply identical with 'meaning of S ',
the term 'meaning' being used in a general sense unrelated to any par
ticular theory of grammar. Consequently the claim of the psychological
reality of deep structures boils down to the claim that the meanings of
sentences, in whatever way they are formalised, are psychologically real.
But this is a trivial result indeed.
Even today, the following confusion, connected with the central
notion of 'knowledge', continues to debilitate TG's general methodolo
gical position. On the one hand, there is conscious knowledge as the
standard object of conceptual, including grammatical, analysis. On
the other hand, there is unconscious 'knowledge' as a dispositional
property, the existence of which can only be hypothetico-inductively
inferred from its behavioural manifestations. Because TG has chosen to
use one and the same term to refer to these fundamentally different en
tities, it continually confuses them and remains unable to make a viable
distinction between grammar and (experimental) psychology.
Even if the answers given by TG are not very encouraging so far,
it remains true that ontological support for grammatical descriptions
can be sought in psychological experimentation. However, such a sup
port is in no way necessary, and under no circumstances can it replace
the role of grammatical description (cf. 7.4. above).
So here we are left with the problem of the (non-psychological)
ontological reality of grammatical descriptions. For semantic de
scriptions, there seems to be an easy answer: The semantics of a sen
tence is the analysis of the state of affairs to which the sentence
(i.e., its syntax or 'form') purports to refer; the analysis is onto-
logically real to the extent that it captures this (possible) state of
affairs as it really is (cf. Itkonen 1969a and b, and 1970b). For in
stance, the semantic description of "John is taller than Bill" is on
tologically real insofar as it succeeds in analysing the state of af
fairs consisting in John's being taller than Bill. Now, there may be
some doubt as to whether this kind of analysis is really a proper task
of linguistic, or grammatical, description. More importantly for the
present discussion, different true descriptions of one and the same
state of affairs are always possible, and therefore there remains the
question whether such differences too are ontologically relevant. This
question, i.e., the question of the ontological reality of theoretical
(non-empirical) descriptions, is precisely the one which I raised in
the beginning of this section.
In my opinion it can be plausibly argued that the ontological real
ity of grammatical descriptions resides solely in the fact that they
systematise rules of language. However, I do not think that this view
exhausts the whole truth. I admit that the qualifications which I am
going to offer are of a highly speculative character. Nothing of what
I say elsewhere in this book depends on the truth or the falsity of
what follows below.
It should be borne in mind that the validation of proposed theo
retical descriptions is entirely different in the natural sciences and
in grammar. In the former case, the basis of validation is constituted
by events which, even when they are initiated by us, 'happen in the
world', i.e., outside us, and which we therefore cannot come to under
stand, in the precise sense of the word. Accordingly, when we are deal
ing with an empirical hypothesis formulated in a technical language that
bears no, or very little, resemblance to natural languages, we can only
be said to understand this hypothesis insofar as we are able, in prin
ciple, to find out. whether or not it is ultimately validated by obser-
vable events. In particular, to be able to devise an empirical theore

tical hypothesis that is confirmed by the truth of test predictions is
to understand this hypothesis.
In the case of grammar - and of normative (hermeneutic) sciences
in general - the subject matter of the theoretical description is
something that we have first understood and that we intuitively know
since then. It could be argued that the same principle of understand
ing which applies to the subject matter ought to apply, by way of exten
sion, to its theoretical description, too. That is to say, it should
not be sufficient that we, as it were, merely observed that our atheo-
retical knowledge can be systemati sed in such and such ways according
to criteria based on some preconceived notion of simplicity; rather,
the systematisation itself should be illuminating in such a way that
it produces understanding which is situated on the same continuum as
the initial, atheoretical understanding, only higher. It seems to me
that in this particular respect TG descriptions vacillate somewhere be
tween philosophical analyses and theoretical hypotheses of natural sci
ence: On the one hand, it is undeniable that both TG descriptions and
philosophical analyses are, broadly speaking, systematisations of atheo
retical, intuitive knowledge; on the other hand, while it is the expli
cit purpose of philosophical analyses to extend the domain of our under
standing, by eliminating conceptual confusions and by solving puzz
les for instance, TG quite generally uses 'theoretical concepts' which
are not (intuitively) understandable in themselves, but are justified
solely because they simplify the description in the sense of reducing
the number of grammatical rules needed for generating correct sentences.
(Here it would be awkward, even if factually correct, to speak of 'gram
matical hypotheses' instead of 'grammatical rules', or interpretations
of such rules.) I would say that only those grammatical rules which
either have or may receive an intuitive backing and which in this re
spect resemble the 'rules' of good philosophical analyses may claim to
have ontologically real counterparts in 'language itself'. Since only
'illuminating' rules of grammar are said to refer to something existing
we see that in this context questions of exristenoe ave inseparable from
226
questions of value. Grammatical rules which systematise and simplify
the data in more or less unintelligible or 'opaque' ways may be quite
useful for certain purposes, but they do not seem to possess ontolo
gical reality in the sense here specified. - The preceding discussion
is related to the requirement of 'naturalness' within the metatheo-
ry of TG. Initially it was assumed that simplicity and (intuitive)
naturalness simply coincide, but since then it has become more and
more evident that paying due attention to the naturalness of descrip
tions may necessitate important methodological changes.
I have outlined here a difference between two very general ap
proaches to grammatical rules. It could perhaps be said that according
to the one view, grammatical rules are only means to ends, while accord
ing to the other, they have an intuitive significance of their own. This
is of course an oversimplification, but I nevertheless maintain that
something like the difference here outlined is discernible in the cur
rent practice of language description. The two viewpoints might be
illustrated by the following quotations. On the one hand, grammatical
categories are simply theoretical concepts:
Motivated now by the goal of constructing a grammar, instead of a
rule of procedure for constructing an inventory of elements, we no
longer have any reason to consider the symbols ΝΡ, S e n t e n c e , VP
and so forth, that appear in these rules to be names of certain
classes, sequences, or sequences of classes, and so on, of concrete
elements. They are simply elements in a system of representa
tion which has been constructed so as to enable us to characterize
effectively the set of English sentences in a linguistically
meaningful way (Chomsky 1964a:216).
Instead of viewing the phonological section of the grammar as an

inventory of phonemes, constructed by procedures of analysis based
on such rather arbitrary principles as biuniqueness, we can con
struct the simplest possible grammar of the appropriate form and
consider the phonemes of this language to be the elements that
appear in the representation of utterances on the appropriate le
vel in this grammar (ibid. p.220).114
On the other hand, even the theoretical description itself must

somehow be intuitively acceptable:
That is, we know ahead of time in some sense what we want to come
out with as a result of our analysis. This is one sense in which'
it can be said that linguistic analysis tries to account for the

linguistic intuition of the native speaker (Bach 1964:151).
We have to do here with two different conceptions of the nature

of conceptual analysis, or explication. The former is expounded for
example in Quine (1960:257-62); it is closely connected with the use
of formal logic as a tool of analysis. The latter is represented by
Wittgenstein-inspired (linguistic) analysis which continues the clas
sical philosophical tradition (cf. 11.1. below).
The discussion of this topic is made very difficult by the inevi
table vagueness of the crucial notion of the intuitive acceptability
or naturalness of grammatical descriptions, as well as by the fact
that either the one or the other of the two notions of grammatical
rule here discussed is only seldom applied in a conscious and con
sistent way, so that there is an almost infinite number of unclear or
intermediate cases. In any event, I accept the possibility that (cer
tain types of) grammatical hypotheses could refer to, or directly re
present, something existent in language itself, i.e., in the (poten
tial) knowledge of language.
9.0. THE METHODOLOGY OF GRAMMAR
In this chapter I will concern myself chiefly with the methodolo
gy of TG. However, I shall carry out my analysis at a level of abstrac
tion where everything said about TG also applies to any other general
method of providing theoretical-grammatical descriptions for natural
languages. I shall prove here that TG does not satisfy the criteria
of empirical explanation or of empirical testability. Examples of this
will be given in 9.5. (below). In 10.0. and 11.0. (below) I shall prove
that TG satisfies those criteria of 'explanation' (i.e., explanation3 )
and 'testability' (i.e., testability3) which are characteristic of lo
gical and philosophical analysis. Therefore reading the two last-men
tioned chapters is essential for fully understanding what is going to
be said here.
9.1. General Remarks
Since TG aspires to be an empirical science, modelled upon the

example given by the paradigmatic natural sciences, TG descriptions
ought to make use of the methods of explanation, prediction, and test
ing, as they have been specified by the positivistic philosophy of
science (cf. 1.0. above). Now, if TG really were a psycholinguistic
theory of faculté de langage - as i t often claims to be -, then it
might qualify as an empirical science. The processes of language ac
quisition and use go on in space and time, and it is only natural to
try to explain them by postulating some general psychological regula
rities which involve non-observable (i.e., theoretical) and partly in
nate mechanisms, and by adducing actual language-learning situations
which function as antecedent conditions; together the regularities and
the antecedent conditions might be used to tentatively explain any
THE METHODOLOGY OF GRAMMAR 229
number of p r o t r a c t e d explanandum-events c o n s i s t i n g i n t h a t a given c h i l d

becomes a f l u e n t speaker of a given language. N o t i c e , however, t h a t
even i n t h i s instance TG could not possibly count as a natural (or em-
p i r i c a l · , ) science, because the notion of fluency is dependent on the
notion of ' c o r r e c t n e s s ' , and because correctness cannot be d e f i n e d , or
o p e r a t i o n a l i s e d , i n spatiotemporal terms ( c f . 7 . 1 . above).
On t h i s i n t e r p r e t a t i o n , t h e n , TG would be an empirical 2 theory
i n v e s t i g a t i n g a mixture of s p a t i o t e m p o r a l i t y and n o r m a t i v i t y , i . e . , a
set of f a c t u a l normative a c t i o n s . S o c i o l i n g u i s t i c s too is a science
of t h i s t y p e ; but of course, TG has never viewed i t s e l f as a s o c i o l i n -
guistic theory.
Whatever i t s methodological claims may be, however, TG i s first
and foremost a grammatical theory. And even i f i t should come to con-
t a i n a genuinely e x p e r i m e n t a l - p s y c h o l i n g u i s t i c component, as envisaged
in Fodor e t a l . (1974), i t is clear t h a t t h i s component cannot replace
the grammatical one. Now, as grammatical d e s c r i p t i o n s , TG d e s c r i p t i o n s
cannot make use of empirical methods of explanation and t e s t i n g , f o r
the simple reason t h a t t h e i r subject m a t t e r , c o n s t i t u t e d by the rules
of a given language, is not spatiotemporal. Notice also t h a t rules de-
termine p r i m a r i l y the correctness of sentences, or sentence-types, and
not of u t t e r a n c e s , or sentence-tokens, which means t h a t i n grammar
types are primary w i t h respect to tokens. In empirical science, by con-
t r a s t , tokens, as objects of e x p l a n a t i o n , are primary w i t h respect to
types.
Compare Hempel's (1965:423) following observation:
However, given this notion of explaining a particular occurrence
of a solar eclipse or of a rainbow, etc., one can speak deriva-
tively of a theoretical explanation of solar eclipses or rain
bows in general: such an explanation is then one that accounts
for any instance of an eclipse or a rainbow. Thus, the notion
of explaining particular instances of a given kind of occurrence
is the primary one.
As a supposedly empirical theory, TG claims to be investigating

space and time, but - as will be seen from what follows - it has
been forced to admit, implicitly, the falsity of this claim.
Within p o s i t i v i s m , explanations are divided i n t o two d i s t i n c t
types, namely deductive-nomological (= D-N) explanations and ( i n d u c t i -
ve-)statistical explanations. As the name i n d i c a t e s , i t i s the charac-
t e r i s t i c property of the l a t t e r t h a t (some of) the laws which occur in
them are of s t a t i s t i c - p r o b a b i l i s t i c form ( c f . Hempel 1965:376-93). D i f -
f e r e n t objects of research - such as planets or diseases - require,
or may r e q u i r e , d i f f e r e n t types of e x p l a n a t i o n . Consequently i t beco-
mes an empirical q u e s t i o n , which type of explanation is appropriate in
which ( e m p i r i c a l ) science. In other words, the adequacy of the one
or the other type of explanation must be experimentally established.
In t h i s c o n t e x t , i t is extremely revealing to note t h a t , assuming t h a t
TG is an empirical science, i t is the only empirical science which ex-
cludes s t a t i s t i c a l considerations on a pviovi grounds:
Evidently, one's ability to produce and recognize grammatical

utterances is not based on notions of statistical approximation
and the like. The custom of calling grammatical sentences those
that 'can occur', or those that are 'possible', has been respon
sible for some confusion here. It is natural to understand 'pos
sible' as meaning 'highly probable' and to assume that the lin
guist's sharp distinction between grammatical and ungrammatical
is motivated by a feeling that since the 'reality' of language
is too complex to be described completely, he must content him
self with a schematized version replacing 'zero probability,
and all extremely low probabilities, by impossible, and all high
er probabilities by possible'. We see, however, that this idea
is quite incorrect, and that a structural analysis cannot be un
derstood as a schematic summary developed by sharpening the blur
red edges in a full statistical picture (Chomsky 1957:16; emphasis
added) .
I t is r e a d i l y d i s c e r n i b l e t h a t the inadequacy of s t a t i s t i c a l ex-

planations is not established here upon the foundation of experimenta-
t i o n , but of insight or intuition. By way of comparison, consider the
absurdity of an astronomist who would 'see' t h a t s t a t i s t i c a l considera-
t i o n s are ' e v i d e n t l y ' i r r e l e v a n t in dealing w i t h c e l e s t i a l phenomena.
Since TG cannot adopt the ( i n d u c t ! v e - ) s t a t i s t i c a l type of expla-
n a t i o n , i t obviously has to adopt the D-N t y p e , if, in conformity w i t h
i t s self-imposed p o s i t i v i s m , i t wishes to be considered as an e m p i r i -
cal theory. This is why I have omitted the analysis of statistical
explanations (cf. 1.2. above). Curiously enough, representatives of TG

have not been able to identify TG 'explanations' with D-N explanations;
therefore it was left to linguists outside TG to do this (cf. 3.7. above).
However, the fact that the alleged similarity between TG 'explanations'
and D-N explanations is established on a priori, or intuitive, grounds,
shows, precisely, that we do not have to do here with genuine D-N ex
planations. That is, I entirely agree with TG that the statistical
point of view is irrelevant within grammar (cf., however 5.4. above).
But contrary to what TG assumes, this fact only proves the nonempirical
nature of grammatical descriptions: We see that because rules determine,
and are known to determine, the correctness of sentences qua conceptual
possibilities, any statistical concern with utterances either exemplify
ing or failing to exemplify correct sentences is evidently irrelevant.
More concretely, the D-N model cannot apply to TG descriptions,
because the 'individuals', i.e., space-time points or regions, referred
to by 'a', 'b', etc., which occur in D-N explanations, have no role to
play within TG. When we are analysing institutions by analysing the con
cept 'correct (result of) action', or in particular 'correct sentence',
we may validly deal with actions or sentences tokens of which have never
occurred in the intersubjective spatiotemporal world. We may 'explain'
an action (or a sentence) which has never been performed (or uttered)
by relating it to some general properties of correct actions (or sen
tences) of the same or of a different category. And we may 'confirm'
our analysis by means of nonexistent actions (or non-uttered sentences),
i.e., by showing that actions (or sentences) which according to our
analysis ought to be correct are indeed correct, irrespective of whether
or not they have ever been or will be exemplified in space and time. We
can do this, because we know the rules which determine the correctness
of actions (or sentences), and because knowing the rules means knowing
an infinite number of possible correct tokens of actions (or sentences).
Compare this situation to that prevailing in the empirical sciences: it
would be absurd to say that an empirical scientist confirms his theory
on the basis of nonexistent events: a thought experiment is not an ex
periment (Wittgenstein 1958: I, §265). It is clear that natural scien
tists make use of thought experiments too, but these are parasitic upon
genuine experiments, which aim at discovering what really goes on in
space and time. By contrast, grammarians rely solely on thought ex
periments; or, what comes to the same, any apparently genuine experi
ments conducted by grammarians are parasitic upon thought experiments.
That is, if I notice that my grammar generates a sentence 'X s , I know
whether 'X' is correct or incorrect and, accordingly, whether my gram
mar has been confirmed3 or disconfirrned3, irrespective of whether I have
ever observed the spatiotemporal occurrence of any utterance of 'X'.
Nothing is changed if I make such utterances occur by uttering 'X' my
self, or if I subsequently hear someone else utter 'X'. Such spatio-
temporal data are simply irrelevant for the construction and the test
ing of grammars. Grammars describe what we know, i.e., what we have
once learned, and what we now remind ourselves of (cf. 8.3. above).
In my opinion, the reason why the above-mentioned facts about
the nature of grammatical descriptions have always been overlooked by
TG (and by practically all other schools of linguistics as well) can
be seen inter olia in the peculiar character of acts of speaking. While
sitting in my arm-chair, I can think of any type of correct action, rang
ing from the correct way of playing football to that of getting married.
In practically all instances the distance from imagining a (correct)
action to performing it is considerable, It may even be that, in spite
of a lot of energy, time, and movement in space, I am prevented from
performing the (correct) action which I was thinking of. Yet there is
one type of rule-governed action with respect to which the difference
between thought and actual performance is almost nonexistent. Of course,
this is the (locutionary) act of speaking. Since we can straight away
utter all the correct sentences we can distinctly imagine, we are in
clined to think, erroneously, that when we are describing a language,
we are dealing with empirical, observable events only. When describing
the game of football for instance, we are not likely to commit the ana
logous fallacy, precisely because the distance between thought (of cor
rect performance) and actual performance is much greater. The same fal
lacy arises also in connection with writing generative grammars: when
a grammar generates a sentence 'X', it is possible to construct an ex-
piicit derivation which ends with a written utterance, or inscription,

of 'X'. So it may seem that grammars deal only with observable data,
i.e., black marks on a white paper. However, the notion of generating
(written) utterances is secondary with respect to the notion of genera
ting sentences (cf. the discussion of Friedman in 7.3. above).
What I have said in this section, and in previous chapters, may
suffice to demonstrate the inapplicability to TG of the empirical no
tions of explanation, prediction, and testing. For the sake of com
pleteness, however, I shall explore these matters in greater detail be
low.
9.2. Explanation and Prediction

So far I have been concerned only with the metascientific analy
sis of grammars, understood as descriptions of particular languages.
Their relation to the notion of universal linguistic theory will be
discussed in 9.4. (below). However, in approaching the problem of
explanation in TG we should mention already at this point the role of
universal linguistic theory, in as much as it is concerned with the
notion 'explanatory adequacy'. A grammar is said to be 'descriptive
ly adequate', if it correctly describes the native speaker's linguis
tic intuition or, in more controversial terms, if it correctly descri
bes the tacit competence of the idealised native speaker. A linguis
tic theory is descriptively adequate, if it provides a descriptively
adequate grammar for every particular language. Finally, a linguis
tic theory possesses explanatory adequacy, if it is capable of select
ing for each language the 'right' descriptively adequate grammar oyer
other possible grammars of the same type (cf. Chomsky 1965:24-25).
First of all, it should be noted that the distinction between de
scriptive and explanatory adequacy has very little theoretical or prac
tical value. It is certainly impossible to describe adequately one
particular language while using concepts which have no general appli
cability whatever. That is, describing one language adequately re
quires some knowledge about language in general, whereas any adequate
general theory of language must, of necessity, build upon descriptions
of particular languages. We have here to do with a clear case of the
hermeneutic 'circle of understanding'. Furthermore, it is a truism
about the nature of scientific activity that scientists are intent up
on making their theories as general or 'universal' as possible. Further
more, since the problem of constructing an 'evaluation measure' for se
lecting the right descriptively adequate grammar for each language is
practically identical with the problem of defining the notion of lin
guistically significant generalisation' (Chomsky 1965:42), it simply
follows that the best descriptions or theories, both at the level of
one language and at the level of all languages, are always the most
general ones. Hence the 'descriptive' level and the 'explanatory' le
vel are inseparable in practice. As one example among many, consider
Chomsky's (1964d:41) requirement of the recoverability of deleted
elements. This 'general condition on TG' is supposedly part of the
theoretical apparatus which for the first time makes explanatory ade
quacy possible; but since this condition is needed, inter alia, to pre
vent elliptical sentences of English from being infinitely ambiguous,
it must be part of every descriptively adequate grammar of English,
regardless of whether or not we are concerned with writing grammars
for any other language; but then again it might be argued that noticing
the relevance of this condition (and the irrelevance of some other theo
retically possible conditions) requires some knowledge about language
in general.
Secondly, the question of descriptive vs. explanatory adequacy is
irrelevant in the present context anyway, since TG has made it eminent
ly clear that it views grammars as theories of particular languages,
that is, theories strictly in the positivistic sense. It is of course
the defining property of such theories that they explain and predict
observable events and are tested, i.e., confirmed or disconfirmed, on
the basis of such events.
As a consequence, the topic here is the notion of explanation in
(synchronic) grammar. I shall concentrate on establishing the truth of
three specific claims. First, the D-N model does not apply to TG; in
particular, in addition to the fact that the data to be 'explained'
are not spatiotemporal, the requirement of the conceptual independence

between antecedent conditions and explanandum-'events' cannot be ful
filled either. Second, 'explanation' in TG is identical with (nonempi
rical) generalisation, Third, a TG description is a self-referential
description (equipped with a metalinguistic interpretation): it 'ex
plains' its data by showing them.
I shall discuss these three claims in relation to two phenomena
which, within the TG tradition, are commonly assumed to be 'explained'
or 'explainable' by grammar, viz. the correctness of sentences and the
distribution of sentential properties. - Because of the structural
identity between explanation and prediction, the results achieved in
analysing the former will hold true of the latter too.
Suppose, for the sake of argument, that a statement to the effect
that a sentence 'X' is a correct sentence of a language L, might quali
fy as a genuine explanandum in Hempel's sense. What would the corre
sponding explanans look like?TG has never given a clear answer to this
question. Wang (1972) proposes an explanans where the roles of univer
sal hypotheses and statements of antecedent conditions would be ful
filled, respectively, by universal sentences involving grammatical ca
tegories and sentences assigning words to lexical categories, This
proposal is unacceptable because, first of all, a sentence like "man
is a noun" is not about space and time, and hence does not qualify as
a statement of antecedent conditions. Nor can it be defined in spatio-
temporal terms, for example by means of Kasher-type instance functions
(cf. 7.2. above). The same is true, of course, of the putative expla
nandum "The sentence 'X' is a correct sentence of L". In effect, with
in grammar there is no room for statements about particulars; rather,
all grammatical hypotheses, in whatever way they may be formalised, are
equally universal. This is directly confirmed by the fact that in old
er variants of TG words were introduced by means of the same type of
rules as any other grammatical categories, i.e.,
man
woman
It was just a technical innovation that words were later introduced as
parts of a 'lexicon'. Changes in the ways of describing language do
not of course affect the nature of language itself. Moreover, the de
velopment of generative semantics has at least called into question the
use of a self-contained lexicon. - My position here is further support
ed by the close analogy between generative grammars and axiomatic sys
tems of logic (cf. 10.0 below), given that the latter contain no state
ments about particulars.
Secondly, if the correctness of sentences is what grammars are re
quired to explain, it is clear that, irrespective of their specific for
malisations, the substance of grammatical 'D-N explanations' must be
this:
Now, it is clear that a's having the property complex F. is not the
cause of, or more generally does not empirically determine the fact that
a is correct sentence of L; rather, the former fact entails the latter.
That is, to have the property complex F. is to be a correct sentence of
L. Consequently the second premise entails the conclusion all alone,
which means that the condition of conceptual independence is not met:
it is not possible that the 'statement of antecedent conditions' is true
and the 'explanandum' is false.
To give a simple example, to notice that in the form the man the ar
ticle precedes the noun is to notice that this form is correct. A form
in which the article follows the noun is incorrect by conceptual neces
sity (cf. pp.158-60). The same argument applies to all more complex
cases: We cannot notice that an utterance 'Y' has all the properties of
a correct utterance, without noticing that the utterance 'Y' is correct.
Similarly we can neither observe nor think of a closed figure with three
lines without observing, or thinking of, this figure as a (correct) tri-
angle. By contrast, we can think of a piece of metal being heated,

without being forced to think of it as expanding as well. This is
why the notion of antecedent conditions applies in physics, but not
in grammar or geometry for example.
Next, let us consider the case where a grammatical description
'explains', not just the correctness of sentences, but the distribu
tion, among correct sentences, of some syntactic phenomena. I shall
analyse an example from Lakoff (1971). Referring to R. Lakoff's stu
dies, Lakoff claims that the distribution of the subjunctive and of
the negative particles non and ne in Latin can be explained by pos
tulating abstract predicates not appearing at the surface. For in
stance, the similarity in form and meaning between the sentence "Im
pero ut venias" and "Venias", on the one hand, and "Impero ne venias"
and "Ne venias", on the other, is supposedly explained by showing
that they can be derived from a common deep structure (cf. the figu-
re).116
Lakoff (1971:289) sums up the rationale of postulating abstract

deep structures and, by the same token, of providing grammatical expla
nations, in the following way:
If the same syntactic phenomena that occur in sentences with cer

tain overt verbs occur in sentences without those verbs, and if
those sentences are understood as though those verbs were there,
then we conclude (1) a rule has to be stated in the case where
the real verbs occur; (2) since the same phenomenon occurs with
the corresponding understood verbs, then there should be a single
general rule to cover both cases; (3) since we know what the
rule looks like in the case of real verbs, and since the same rule
must apply, then the sentences with understood verbs must have a
structure sufficiently like that of those with the overt verbs
so that the same general rule can apply to both (emphasis added).
I shall return to this example below. Here I consider it only

from the perspective of the D-N model. If in connection with the cor
rectness of sentences the data to be explained are not spatiotemporal,
they can be so even less in the present, more complex case. What about
the requirement of conceptual independence between the data to be ex
plained and their antecedent conditions? The substance of the 'expla
nation' which Lakoff has in mind must be something like this:
i) "For all χ and y, if χ and y are sentences having the
same deep structure, then they are alike in meaning and
at least partly alike in form."
ii) "a and b are sentences having the same deep structure".
"Therefore a and b are alike in meaning and at least partly
alike in form."
Again, it is clear that the second premise entails the conclusion

all alone. In the grammatical tradition within which this 'D-N expla
nation' is formulated, for two sentences to have the same deep struc
ture means that they are alike in meaning. Lakoff expressly stipulates
that the sentence b, where the overt verb is lacking, must be understood
as containing it, i.e., as similar to the sentence a which contains it
(cf. above). Secondly, and analogously, it is necessarily the case that
the sentences a and b having the same deep structure are at least part
ly alike in form: Lakoff expressly stipulates that there must be the
same syntactic phenomena (viz. the subjunctive and the negative partic
les) occurring in both a and b. (This is not to deny that in other con
texts a common deep structure may be postulated for sentences with appa
rently no formal similarities.) Consequently, the requirement of con
ceptual independence between the second premise and the explanandum can
not be met here, either.
To sum up, it is not only the case that what TG descriptions purport
to explain is not of spatiotemporal, but of conceptual nature; in addi
tion, it stands in a conceptual, necessary relation to everything which
might be taken as its 'antecedent conditions'. Therefore any attempt
to force TG descriptions into the D-N model amounts to an (impossible)
attempt to split one and the same concept, or one and the same piece of
knowledge, into two.
Now, if TG descriptions are not (empirical) D-N explanations in the
sense of 1.2. (above), what are they then? This question brings us to
my second claim, viz, that in TG 'explanation' simply equals 'generalisa
tion' .
In TG it is customarily said that a grammar explains a sentence by
generating it. In this context the terms 'explain', 'predict', 'produce',
'deduce', 'generate', etc.. are used synonymously. Botha ' s (1968:62) follow
ing remark is characteristic:
It can be said that Lees' grammar of English nominalisations explains
nominal compound types in that it is possible to deduce them from
the rules, i.e., the explanatory principles contained in the gram
mar .
Kanngiesser has discussed this topic in some detail. He accepts

the thesis of the structural symmetry between explanation and predic
tion, and formulates the aims of TG in the following terms:
Da nun kein logischer Unterschied zwischen der Herleitung einer

Aussage und einer Prognose besteht, lässt sich zu Recht, wenngleich
etwas verkürzt, sagen, dass die Erklärung eines sprachlichen Sach
verhalts, eines Datums in seiner Voraussage besteht; ...
Nun sind es zweierlei Leistungen, die ein linguistisches System in

Hinblick auf seine prognostische Kapazität vollbringen muss: es
muss erstens die Klasse der in einer Sprachgemeinschaft möglichen
grammatischen korrekten [sic] Sätze voraussagen, d.i. e r z e u g e n , und
es muss zweitens die Klasse der Strukturbeschreibungen prognosti
zieren, die diesen Sätzen zuzuordnen sind, und die als Spezifika
tionen der Art zu interpretieren sind, in der die Sprecher/Hörer
der Sprachgemeinschaft diese Sätze verstehen (Kanngiesser 1972:
8).
He then sets up an artificial language L consisting of the senten

ces ab and aba and answers the question as to what it is to explain
these sentences:
Um die obigen Sätze … erklären zu können, muss offenbar eine
Theorie von L, in linguistischer Terminologie und mit einer ge
wissen Vereinfachung: eine Grammatik G von L existieren, die die
Sätze ... erzeugt, also eine Prognosededuktion ermöglicht, und
die es weiterhin gestattet, die für die Sätze ... einschlägigen
Strukturbeschreibungen zu prognostizieren (ibid.).
Now the following grammar G1 fulfils these requirements:
G1: S → SS (p 1 )
S → a (p 2 )
S → b (p 3 )
In other words, this grammar generates the sentences ab and aba

(as well as all other combinations of a's and b's) and automatically
assigns structural descriptions to them. The sentence aba may receive,
e.g., the following versions of derivation and structural description:
S (given)
S S (Ρ 1 )
a S (P 2 )
a S S (P1)
a b S (P 3 )
a b a
(p 2 )
I think that, as far as it goes, Kanngiesser's interpretation of
the notion 'explanation in TG' is quite correct. But notice that in
precisely the same way we could 'explain' for instance any string con
sisting of one or more zeros: for instance, the string 000 is 'explained'
as follows:
G2: X → XO (q 1 ) X (given)
X → 0 (q 2 ) X 0
X 0 0 (q 1 )
0 0 0 (q 2)
It should be clear that such 'explanations' have very little to do
with the way in which observable events are explained on the basis of
their antecedent conditions and some hypothetical regularities. There
fore it is quite clear, again, that TG does not use such methodological
ly central terms as 'explanation' and 'prediction' intheir customary em-
pirical sense, as defined in 1.2. (above).117 Yet it is not too dif
ficult to find a common denominator for D-N explanations1-2 and TG-
type explanations3. To put it somewhat metaphorically, in both in
stances we have to do with bringing order into (apparent) disorder.
The main difference lies in the fact that in the one case the disor
der obtains among space-time events and facts, while in the other case
it obtains among conceptual entities exemplified by strings of sym
bols. At a higher level of abstraction, again, it is easy to see that
in both cases the levels of disorder and order might be called, re
spectively, 'atheoretical' and 'theoretical'. That is, order is achiev
ed by postulating a set of theoretical-descriptive devices which are
able to show that things which previously seemed disconnected actual
ly belong together. As was pointed out earlier (p.211), theory is
creative insofar as it brings into existence something which did not
exist before, viz. something which reveals the coherence of, and thus
simplifies, the atheoretical disorder which did exist before. There
fore an empirical theory explains1-2 an event only when it also ex
piains1-2 an indefinite number of other events. Similarly, it makes
sense to say that the grammar G2 'explains3' the string of zeros 000
only because it also explains3 all similar strings and thus shows their
connection with one another.
A theory creates order precisely because it does not record each
event or each string of symbols separately, but rather says something
about all events or strings of symbols, or at least about as many events
or strings of symbols as possible. That is, a theory creates order by
being general. A theory is the better, the more general it is; its
greater generality is manifested as its capacity to make a greater num
ber of generalisations. In TG, in turn, "we have a generalisation when
a set of grammatical rules about distinct items can be replaced by a
single rule (or, more generally, partially identical rules) about the
whole set, ..." (Chomsky 1965:42). For instance, the grammar G1 is
more general than any grammar generating all combinations of a's and
b's by means of more than three rules.
It is the basic purpose of a TG description, then, to present in
a maximally simple and general way the relations of similarity and dif-
ference between the correct sentences of a given language. As we have
seen in 3.6. (above), it is the merit of TG, as against earlier Ameri
can linguistics, to accept more abstract sentential relations, e.g.,
the similarity between the active and the passive, and the difference
between seemingly identical 'easy'- and 'eager'-constructions, as well
as to provide more explicit methods of formalisation. Consequently a
TG description resembles, not only a logical derivation, but also a
proof of geometry: In an analogous manner, it is the basic purpose of
geometry to present the similarities and the differences between dif
ferent (correct) figures in a maximally simple and general way. In
empirical science, the nearest analogue to TG descriptions is not ex
planation of events, but classification of (empirical) types. Even
if the data to be classified has been obtained experimentally, the
classification itself is no longer an experimental undertaking; ra
ther, it consists in presenting a given body of (empirical) knowledge
in a maximally simple and general way. It follows that a (taxonomie)
classification 'explains' its data in precisely the same sense as a
TG description, e.g., the grammar G1 above, does. - From the above,
it also follows that the axiornatisation of an empirical science is ra
ther similar to grammar-writing.
In light of the preceding discussion, it is self-evident in which
sense grammars can be said to 'explain' the correctness of sentences.
It is the same sense in which systems of logic can be said to 'explain'
the validity of their theorems (cf. 10.0. below). Therefore it only
needs to be shown here that Lakoff's example of grammatical explanation
agrees with my interpretation of the TG-type explanation. Remember
that the distribution of the subjunctive and of the morphemes non and
me was meant to be explained by the postulation of an abstract predi
cate 'ego impero'. Now, the explanatory import of such predicates
resides precisely in the fact that they permit us to express generali
sations about sets of sentence-types which would otherwise remain un
noticed and unexpressed. (To be sure, it is debatable whether postu
lating additional, abstract structures is the only or even the most
adequate way of expressing generalisations in grammar; cf. n.125.) In
other words, a generalisation is made here by dispensing with those

phrase structure rules (or node admissibility conditions) which would
have generated a separate deep structure for "Venias" or for "Ne veni
as".
Consider the two sentences "Impero ut venias" and "Venias", and
let "(ego) impero", "tu ven-", and the subjunctive be referred to,
respectively, by A / a , B/b, and c/c, where the capital letters and the
small ones stand for deep forms and surface forms, respectively. At
the surface level, then, the two sentences are represented as abc and
bo. If we did not postulate a common deep structure for them, they
would be generated by the following two sets of rules:
1) S → ABC→abc
2) S → BC → bc
The postulation of a common deep structure ABC makes the rule S → BC

superfluous. The generation of bc now looks like this:
2') S→ABC → BC → bc
We see that the gain in simplicity, viz. the elimination of the rule
S → BC, is counterbalanced by a loss, viz. the addition of the rule
ABC → BC. However, it can be safely assumed that deletion transfor
mations are extremely common, or 'independently motivated', so that
the gain here outweighs the loss. It is clear, then, that Lakoff's
'explanation' is a 'generalisation' precisely in Chomsky's sense: two
different grammatical rules are replaced by a single one (plus an in
dependently motivated rule). The meaning of these formal exercises is
to show that apparently disparate things belong in fact together, or,
in a however modest way, to 'bring order into (apparent) disorder'.
Let us turn, finally, to my third claim, viz. that TG offers
s e l f - r e f e r e n t i a l 'explanations' of its data. Above, it became evi
dent that TG explains its data, i.e., the sentences of a given langu
age, by presenting them with all their various properties in the maxi
mally general way. In this respect TG-type descriptions were seen to
resemble taxonomie classifications of empirical science. However,
there i s also a clear d i f f e r e n c e here. For example a g e o l o g i s t who is
g i v i n g a c l a s s i f i c a t o r y d e s c r i p t i o n of his data i s not manipulating
d i f f e r e n t types of m i n e r a l s , but sentences r e f e r r i n g to the former.
By c o n t r a s t , when a grammarian ' e x p l a i n s ' the correctness of some
sentences, he i s l i t e r a l l y manipulating these same sentences ( o r , r a -
t h e r , i n s c r i p t i o n s representing them). For i n s t a n c e , we ' e x p l a i n ' a
given s t r i n g of a's and b ' s , by means of the grammar G1, by showing
how one can, applying the rules of G1 , gradually a r r i v e a t an i n s c r i p -
t i o n of p r e c i s e l y t h i s s t r i n g . In the same vein the s i m i l a r i t y be-
tween the sentences "Impero ut venias" and "Venias" i s ' e x p l a i n e d ' by
showing t h a t i t is from a common source t h a t one a r r i v e s at the cor-
responding i n s c r i p t i o n s .
I t i s c l e a r , t h e n , t h a t a TG explanation i s a picture, con-
s t r u c t e d according to maximally general p r i n c i p l e s , and t h a t , i n addi-
t i o n , the elements of the p i c t u r e , i n p a r t i c u l a r parts of natural
language sentences (or i n s c r i p t i o n s ) , represent nothing but themselves
A d e r i v a t i o n (or ' e x p l a n a t i o n ' ) of a theorem i n l o g i c i s a s i m i l a r
' p i c t u r e ' ; I s h a l l have more to say about t h i s t o p i c i n 10.0, (below).
A TG d e s c r i p t i o n i s s e l f - r e f e r e n t i a l i n the way of a p i c t u r e ,
not of a sentence; t h a t i s , i t does not admit of t r u t h - v a l u e . How-
ever, i t is always meant to be accompanied by a ( m e t a l i n g u i s t i c ) na-
t u r a l language i n t e r p r e t a t i o n , even i f the l a t t e r i s mostly not s p e l -
led out e x p l i c i t l y , l e t alone mechanically assigned. This i n t e r p r e t a -
tion admits, of course, of t r u t h - v a l u e . I t claims t h a t the sentences
described are c o r r e c t , t h a t the d e s c r i p t i o n expresses such and such
generalisations, etc. More g e n e r a l l y , the metagrammar is referential
- i t speaks about the ( s e l f - r e f e r e n t i a l ) grammar - and therefore
resembles ordinary (non-empirical) theories.
When discussing TG, I always assume that the distinction between
grammar and metagrammar is made as indicated here, and when I speak
simply of 'grammar', I mean 'grammar-cum-metagrammar'. It is rather
pedantic, and even a little dishonest, to argue that since TG fails
to make this distinction in a way which would meet the most rigorous
requirements of explicitness, it does not make the distinction at all
(cf. p.79above). This is tantamount to substituting, without any jus
tification, one's own aims and standards for those of someone else. It
is true, of course, that TG is largely guilty of the same arbitrary re-
interpretation of the earlier history of linguistics. For instance,
American 'taxonomie' linguistics is viewed as trying, and failing, to
do precisely that what TG does. Postal (1964) is only the most bla
tant example of this wide-spread attitude.
It has sometimes beensaid that available grammars are not expla
natory because the types of descriptive devices which they make use of
are in no specific way suited for the description of natural language;
it is, rather, only by adding otherwise unmotivated, and hence ad hoc,
constraints that such devices can be made to generate, even in a very
rough outline, the correct sentences of a given natural language. By
saying this, it is implied that inventing more adequate types of de
vices (or 'rules') would automatically render grammars truly explana
tory (cf. Peters 1972). This is an interesting line of thought, but
it is clear that the resulting (generative and generalising) explana
tion., must not be confused with empirical explanation1-2
9.3. Testing
A decisive criterion by which the empirical or nonempirical nature
of a description is to be judged is the presence or absence of concep
tual independence among the data which the description deals with. In
practice this question falls together with the question of the spatio-
temporality of the data. On closer inspection, however, four possibi
lities can be distinguished here. If the data are constituted by se
parate spatiotemporal events, like Fa and Ga or Ga and Gb in 1.2. (a-
bove), then their mutual conceptual independence, and the empirical
character of proposed explanations and confirmations (or falsifica
tions) is guaranteed. If it can be shown, however, that "Fa" and
"Ga" or "Ga" and "Gb" refer to one and the same event (or fact), then
there is no conceptual independence, and the proposed explanations are
not empirical in spite of the fact that they deal with space and time.
If the data are not spatiotemporal at all, and hence are conceptual,
then the nonempirical character of correlated explaining and testing
is guaranteed. But there are still two possibilities: Either there is,
or their is not, conceptual independence between the data. For instan
ce, 'father' is dependent upon 'male', and 'father' and 'male parent' are
dependent upon each other. By contrast, concepts like 'cat' and 'stone',
not to speak of concepts like 'two' and 'pleasant', are certainly inde
pendent of each other: there might conceivably be stones without there
being cats, and vice versa.
Since I have In previous chapters demonstrated the conceptual cha
racter of grammatical data and thus the nonempirical character of gram
matical descriptions, I am in this chapter mainly concerned with the
question of conceptual independence vs. dependence. In the previous
section I have shown that in grammatical descriptions the relations be
tween antecedent conditions and facts to be explained, or, in terms of
our D-N model,between Fa and Ga, is not conceptually independent. In
this section I shall analyse the relation between old evidence and new
one, or between Ga and Gb.
According to TG's standard position, a grammar is constructed on
the basis of a given corpus of a language L, and it is tested by find
ing out whether it generates all and only correct sentences of L, in
addition to those contained in the corpus, and assigns correct struc
tural description to them. The 'all' and 'only' aspects correspond,
respectively, to my 'explanatory' and 'predictive' types of testing
(cf. 1.2. above): On the one hand, we think of a correct sentence and
ask whether the grammar generates it; on the other, we make the grammar
generate a sentence and ask whether it is correct.
Before we go any further, TG's conception of testing must be qua
lified in two respects. First, the notion of correct structural de
scription is not as unproblematic as it might seem. We possess intui
tive (and atheoretical) knowledge about an enormous number of differ
ent structural features of correct sentences; but since structural de
scriptions of these same sentences are results of theoretical analysis,
and since we are not supposed to possess intuitive knowledge about
theoretical matters, it follows that we could not properly speak of
correct structural descriptions. In practice, a structural descrip

tion can be divided into roughly two parts, one about which we have
intuitive knowledge and the other about which we have not: if for
example a sentence has two different readings, a structural descrip
tion is (known to be) correct insofar as it represents this fact;
but it may represent this fact in many different ways, and normally
we are not able to decide which way is the intuitively 'correct' one.
If the requirement of the correctness of theoretical descriptions
is to be taken seriously, however, then it supports my claim that
systematisations of intuitive knowledge should, as extensions of
this knowledge, be themselves intuitively acceptable (cf. 8.4.
above).
Secondly, from TG's point of view the relation between the gram
mar of L and the (correct) sentences of L does not simply consist in
that the former is tested against its capacity to generate the latter.
In connection with very complex sentences the order of reasoning is
rather the inverse one: We have no intuition about their correctness,
but we can convince ourselves of it, if we can show that these senten
ces are generated by recursively applying grammatical rules which have
already been confirmed by the fact that they generate simpler, intui
tively correct sentences. This 'clear case principle' amounts to the
claim that grammars are decision procedures for the correctness of
complex (and otherwise intuitively unclear) sentences. In such instan
ces, then, it is not sentences which say something about the grammar,
but the other way around, the grammar which says something about sen
tences. - The clear case principle rests on an assumed analogy be
tween logical induction and what might be called 'grammatical induc
tion'. However, it is shown in Itkonen (1976a) that this analogy is
nonexistent. It follows that grammars cannot be used as decision pro
cedures for correctness and that recursivity loses almost all of its
justification in grammar.
Now, TG's standard conception of grammatical testing can be il
lustrated as shown on the next page.
In this presentation we have assumed that both test claims are

true. Are we now entitled to say that our grammar G3 has been empiri
cally confirmed? Obviously not. We must consider the full consequen
ces of the fact that the corpus does not simply consist of factually
occurring utterances, but of factually occurring correct utterances.
Utterances like ba, aab, and aba have been left out of consideration
because we, as fluent speakers of L3, intuitively know that they are
incorrect. Therefore the utterances in the corpus are utterances
exemplifying an intuitively known rule according to which correct sen
tences of L3 consist of any number of a's followed by an equal number
of b's. This means that G3 does not in reality describe (correct) ut
terances, but (the knowledge of) the above-mentioned rule which deter-
mines the correctness of utterances only via the correctness of sen
tences. Consequently, establishing a corpus of factually-made correct
utterances is just an idle ceremony. In actual practice no transfor
mationalist or Montague-grammarian ever bothers to base his descrip
tion on a real corpus. Their sample sentences are rather in the style
"Sincerity may frighten the boy" or "Every man such that he loves a
woman is a man".
The important question here is whether the new 'confirmatory' evi
dence constituted by the sentences (rather than utterances) aabb and
aaabbb is or is not conceptually independent of the old evidence ab,
aaaabbbb, aaaaabbbbb contained in the corpus and constituting the ba
sis of G 3 . It is not too difficult to see that the answer must be ne
gative. The correctness of ab is correlative with the rule which makes
it correct, and which we know. The same is true of the correctness of
aabb (and of any other sentences of L 3 ) . Hence, it is the rule in
question which provides the conceptual tie between ab and aabb , or be
tween the old evidence and the new one. In other words, if I try to
'test' G3 by finding out whether the 'predicted' sentence aabb is cor
rect or not, I am using as the criterion of the correctness of aabb,
and hence of the truth of G 3 that very same rule which I intuitively
know, and which I describe with my grammar G3.Since one and the same
entity is here both the subject matter of the description and the cri
terion of its truth or falsity, there can be no new, conceptually in
dependent evidence which could empirically either confirm or di scon-
firm the description. The truth of this claim is established in ab
straction from the fact that empirical confirmation or disconfirmation
is here out of question anyway, because the 'entity' here at issue is
a piece of (conceptual) knowledge,
To give another simple example, suppose that on the basis of the

'corpus' the man, the woman, I construct the following fragment of the
grammar of English :
G4: NP → the + Ν
Suppose further that I attempt to 'test' this grammar by picking out

the unit girl from the lexicon and making the 'prediction' that the
girl is a correct form. Of course, I know in advance that my 'pre
diction' is true (and hence is no genuine or empirical prediction),
because I know the rule in question. Similarly, I cannot meaning
fully ask whether G4 generates the correct form the boy, since I
know in advance that it must do so. It would be rather silly to
say that forms like the girl and the boy merely 'confirm' G 4 . -
What I am doing here is just to reaffirm the basic difference be
tween rules and regularities.
It would be a serious error to claim that the grammars G3 and G4
are 'too simple'. What we have here is a matter of principle. TG has
never claimed that there is a conceptual difference such that some
grammatical descriptions are empirically testabie, whereas others are
not. And if this is so, then each description has the same claim to
methodological relevance, It is only by starting from the simplest
possible cases that one can hope to make transformationalists see
their errors, which they 'nave not been able to see in connection with
more complex ones. Now, since rules of language constitute the sub
ject matter of any grammar (cf. 7.0. above), each grammatical genera
lisation is about a greater or smaller number of rules of language and
must be ultimately analysable as a set of descriptions of single ru-
les. 119 And since, as we just have seen once again, the description
of a rule of language is nonempirical (in the sense both of not being
about space and time, and of not satisfying the requirement of concep
tual independence), it follows that each grammatical generalisation
must be analysable as a set of nonempirical descriptions. This is
surely tantamount to saying that grammatical generalisations, and hence
grammatical descriptions, are nonempirical.
As regards the question of empirical testability, then, maximally
simple grammars like G3 and G4 are fully representative of grammars in
general: as descriptions of rules of language, all grammars from the
simplest to the most complex are nonempirical. On the other hand,
there is of course an important respect in which G3 and G4 are not re
presentative of grammars in general: As descriptions of a single rule
they are known to be true, which means that they are unfalsifiable (as
well as unverifiable). In just the same way an equally simple grammar
G5 would be KNOWN to be false, and would thus also be unfalsifiable
(cf. 6.1. above):
G5: NP → Ν + the
In view of their unfaLsifiability, it is not only the case that such

grammars are not empirically testable, but also that they are not non-
empirically testable, either. By contrast, it goes without saying
that ordinary scientific grammars are (nonempirically) testable 3 , for
the simple reason that we do not know whether they are true or false.
To avert any possibility of misunderstanding, I present the distinc
tions here at issue in the following diagram:
(empirically) testable1-2 (nonempirically)
testable3
simple grammar G1-5 NO. Reason: normative- NO. Reason: truth

conceptual nature of or falsity is
subject matter known
ordinary scientific NO. Reason: normative- YES. Reason: truth

grammar G. conceptual nature of or falsity is not
subject matter known 119a
A grammar like G4 describes one single rule (which could be de

composed into an indefinite number of more concrete rules; cf. p. 161),
whereas any complete grammar of English describes perhaps 100,000 rules
or more. Now we cannot say simply that the difference in testability3
between G1-5 and G1 is a direct result of the difference between the
degrees of complexity of their respective data. A grammar of English
which merely lists rules of language and/or correct forms exemplifying
them is not contrary to facts; it is even known to be true. Yet it is
uninteresting. Traditional school-grammars are generally not critici
sed for making claims which are literally false (although this occurs
too). From the scientific point of view, their defect is, rather, that
they make claims which are too obviously true, i.e., claims which re
main at a VERY concrete level, and thus are nor felt to 'explain' any
thing. To put it otherwise, school-grammars make only true claims,
but not all true claims which there are to be made; they stand up to
the 'predictive' testing (so to say), but not to the 'explanatory' one.
In contrast to these grammars, a grammar which makes strong theo
retical generalisations about a large number of rules is interesting
and informative, but likely to be false in one way or another. The
more systematic the grammars are, the more difficult it becomes to
modify some of their (nonempirical) hypotheses so as to make them
better fit the facts, without unwittingly modifying some other hypo
theses with the result that they lose their correspondence with the
facts. It is well known that "the amount of empirical information
conveyed by a theory, or its empirical content, increases with its
degree of falsifiabi1ity" (Popper 1965:113). But this principle is
true, not just of empirical theories, but of every kind of theory.
Both in physics and in logic, for instance, it is an equally valid
truth that the more information a description supplies, i.e., the
more phenomena or 'entities' it accounts for in a systematic and gen
eral way, the more possibilities there are that the description could
be (shown to be) false.
It follows, then, that for a grammar to be testable 3 , two things
are required. First, the grammar must purport to describe a relative
ly large number of rules of language. Second, it must be non-trivial,
i.e., it must attempt to make explicitly formalised generalisations
about its data, G3 is an example of a grammar which meets the second
condition, but not the first one, while an average school-grammarmeets
the first condition, but not the second one. As a result, neither of
them is testable3 .
In keeping with the gradual character of all distinctions having
to do with language (cf. 4.2.4. above), there extends a hard-to-define
middle ground between the two extremes, viz. grammars which are defi
nitely untes tables3 and those which are definitely testable 3 . For in
stance, if we clearly delimit a not-too-large area of a certain langu
age and propose a theoretical description of it, it is probable that
our description will be (shown to be) false in some respect, and also
that its revised or amended versions will be false as well, But since
our subject matter is clearly delimited, it is possible that after a
certain number of modifications we are able to convince ourselves and
others that our description is not false, i.e., that it generates all
120
and only those entities which it was meant to generate.
This does not yet mean that the description is the best possible
one, or that it will not be refuted, because we have not proved that
there cannot be a more general, i.e., scientifically more interesting,
description of the same subject matter. It is far from clear whether,
and when, this latter objective can be achieved. Yet there seem to be
some indisputable cases where this has been done. For instance, it took
a long time to invent a grammar like G3 for a language like L 3 , but once
it has been invented, it seems impossible to conceive even of the possi
bility of a more general description for the same subject matter. If
this result can be accepted, I do not see why it could not be generali
sed, at least in principle, to more complex cases. An analogous exam
ple is given, in a different domain, by the axiomatisation of propositi-
onal logic, which certainly was the result of a long succession of tri
als and errors, but has a long time already been accepted not only as
unfalsifiableo, or untestableo, but also as the most general (axioma
tic) description of its subject matter. It is however self-evident
that in view of the extreme complexity of its subject matter, no gram
mar which both describes a given natural language as a whole and is in
tent upon maximal generality can ever be (shown to be) unfalsifiable3.
The opposite situation would be a practical, even if not logical, im
possibility. Consequently scientific natural-language grammars will
remain testable3.
In the previous paragraphs it became evident that there are two
principal reasons why one grammatical description is inferior to an
other. First, it may be true, but less general. Second, it may be
more general, but (partly) false. These two types of defects are un
derlain by two distinct psychological factors, viz. the grammarian's
lack of insight or his lack of attention. Lack of insight is involved
when the grammarian presents his data truthfully, but simply does not
come up with any interesting generalisations. An instance of this
kind was discussed above in connection with school-grammars. On the
other hand, we have to do with lack of attention when the grammarian
overlooks or is mistaken about aspects of his own linguistic knowledge,
or when he fails to notice inconsistencies obtaining in his own descrip
tion. The defects due to lack of attention are rrristakes: one overlooks
something that one knows in fact, More particularly, it is practically
impossible that one should be mistaken about the content of single rules
of language, but it is quite usual that one overlooks such rules when
they conflict with a tentative generalisation that one is in the process
of formulating (for examples, cf. 9.5. below). Mistakes are corrected
by means of r e c o l l e c t i o n : one reminds oneself of something which one
knows in fact, but just was not aware of at the moment of making the
generalisation (cf. 8.3. above).
The role of recollection in grammatical analysis points to the li
mited nature of grammatical data. However many rules one language may
contain, their number must be finite; to know a language is to know all
of its rules; once a language has been learned, it is known in its enti
rety. This fact raises the question of conceptual independence in a new
form. As we have seen above, there can be no new, conceptually indepen
dent evidence for grammars like G3 and G4 which describe respectively
one single rule, because new sentences exemplifying the rule are con
ceptually interdependent with old ones; and insofar as ordinary gram
mars are broken down into their irreducible components, viz. claims
about single rules, the same is true of them too. However, since or
dinary grammars make generalising claims about classes of rules, we
may ask, in addition, whether these rules are conceptually independent
of one another. That is to say, a typical grammatical claim singles
out a class of rules and states that some property is true of it, viz.
of the entities exemplifying any of the rules belonging to the class.
Let 'A', ' A ' , and ' A 1 ' stand for, respectively, class of rules,
rule, and exemplification of rule, e.g., sentence. Then their inter
relations can be presented as follows:
If the grammatical claim is "A is B", it may be tested both at the le
vel of rules and at that of exemplifications of rules. In connection
with G3 and G4 it became evident that if is old evidence, can
not be conceptually independent new evidence, because they both exemp
lify the same rule A 1 . However, whether, for instance is or is not
conceptually independent new evidence with respect to , depends on
whether or not the rule A 2 is conceptually independent of A-,. Within

the theory of grammar, this is a relatively new type of question. In
stead of any further discussion of it here, I simply point out the fact
that analogous questions, as well as methods for providing answers to
them, are well-known in the domain of philosophical analysis.
Irrespective of whether rules A1 .. .An are or are not conceptually
independent of one another, there is an obvious sense in which their
totality, and, more generally, the totality constituted by all rules
of a given language, does not meet the condition of conceptual inde
pendence. In empirical science it is axiomatic that, when a proposed
description is being (empirically) tested, the test criterion, i.e.,
the criterion of the truth or falsity of the description, must be con
ceptually independent from the object of the description. Now, when
a grammarian has written a grammar of a language L, what is it that
he compares his grammar with, in order to find out its truth or fal
sity? It is the language L itself. Or when a semanticist has given
an analysis of the kinship terminology of L, what is it that he 'tests'
his grammar against, in order to find out its truth or falsity? Again,
it is the kinship terminology of L itself. It would surely be absurd
to say that the kinship terminology of L which the semanticist has
been analysing is conceptually independent from the kinship termino-
logy of L against which he is 'testing' his analysis. But if this is
the case, it follows that, at the level of totality, there can be no
conceptually independent evidence for or against descriptions of a
given language.
It may seem possible to deny this conclusion by claiming that the
same argument applies to natural science as well: physics for example
describes the physical reality, and is in turn tested against it. This
is, however, a rather superficial, and indeed misleading, analogy. As
objects of research, a language and the physical reality are given to
us in thoroughly different ways. Once a language has been learned, it
is (atheoretically) known in its entirety, even if, because of its com
plexity, no one can be simultaneously aware of it in its entirety; but
one can always remind oneself of what one has learned, and thus knows.
By contrast, the physical reality extends infinitely both in space and
in time; one can remind oneself neither of what happened in some other
galaxy one million years ago, nor of what will happen there in one mil
lion years. More concretely, physics does not describe the physical
reality and then test its description against this same reality. Ra
ther, physics bases its description on definite space-time events and
then tests its description against different space-time events. It is
an elementary conceptual truth that difference in (purely physical)
space and time entails conceptual independence. On the other hand, to
repeat, semantics bases its description on the kinship terminology of
L for instance and then tests its description against this same termi
nology. It is an elementary conceptual truth that an entity X is not
conceptually independent of itself. To be sure, the parts (here:
rules) of X may or may not be conceptually independent of one another.
In light of what precedes, we see that the inferiority of one de
scription vis-à-vis another has different roots in empirical science
and in grammar. To be sure, empirical scientists can display lack of
insight just as grammarians can; and the former can, just as well as
the latter, make mistakes about what they know or about the consis
tency of their descriptions. But there is one very important respect
in which empirical scientists cannot be mistaken whereas grammarians
(as well as philosophers and logicians) can. When the former make ge
nuine predictions which then turn out to be false, it would be inaccu
rate to say that they were mistaken. It is a matter of conceptual ne
cessity that the outcome of an experiment cannot be known in advance
(intuitively or otherwise), which is precisely why genuine predictions
as well as universal hypotheses tested on their basis are empirical.
But, of course, one cannot make a rrristake about something that one does
not know. On the other nand, it is the basic assumption of all (syn
chronic-grammatical work that the grammarian does know the language
which he is going to describe. (What he does not know are true theo
retical generalisations about the language.) At any given moment he
can make any number of mistakes about his data but, as opposed
to the empirical scientist, he always could have known his data. This
is precisely why he is making mistakes, instead of just being wrong.
And when he corrects a mistake, he reminds himself of some rule of
121
language which he has known all the time. (Here I have been mere
ly showing the relevance of the discussion in 8.3. to the question of
empirical testability.)
At this stage, there remains only one possible argument for the
claim that grammatical data are conceptually independent in the re
quired sense, after all. It is the argument which says that gramma
rians investigate, not language, but linguistic intuition. On this
view, more precisely, the object of grammatical description is not a
set of entities, be it sentences or rules of L, such that they exist
objectively in the sense of being independent of any particular speak
ers of L. Rather, this object is a set of entities (namely intui
tions) of such a kind that they exist objectively in the sense of being
mental responses of particular speakers of L. If this view is correct,
then it is quite clear that grammatical testing satisfies the condition
of conceptual independence, and even of empiricalness. Intuitions of
one person are temporally distinct happenings, and those of different
persons are spatially distinct as well. Therefore if the grammarian
describes one set of intuitions and tests his description against an
other set, the testing has at least the appearance of being empirical.
In 5.2. (above) I presented the general reasons why I reject the
view that what grammarians investigate is not a given language, but a
set of intuitions (cf., in particular, n.73). This view implies intev
alia that descriptions of rules are (or should be) of statistical cha
racter, and that grammar is in reality a branch of experimental psy
chology. The erroneousness of such implications was demonstrated in
7.3.-4. (above). Here I shall show that the implications of the view
at issue for the notion of testing are equally unacceptable.
If this view were correct, then a grammar like G4 would be an em
pirical theory. It is based on my intuitions according to which 'the
man' and 'the woman' are correct, and it predicts that according to a
future intuition of mine 'the girl' will be correct too. This is an
empirical prediction. It is thoroughly possible that for whatever,
perhaps pathological, reason I will find 'the girl' incorrect, and
'girl the' correct. But this means, then, that my grammar G4 has been
falsified, or has been shown to hold true only with a certain degree
of probability. This consequence is certainly unacceptable, and there
fore the view which it is a consequence of is unacceptable too. How
ever, its unacceptable i ty can be made still clearer.
On this view, even descriptions of maximally simple rules of
language would be empirical. Consider the description which says that
the expression 'the man' is correct. This can be taken as a claim
about either the expression-type 'the man' or about corresponding ex
pression-tokens. In both instances, I may have several temporally
distinct intuitions about 'the man', and it is an empirical possibi
lity that according to some of them 'the man' is incorrect. Similar
ly it is an empirical possibility that I shall sometimes find a sen
tence like "John is easy to please" incorrect. Here, at the latest,
it has become evident that the view which I am discussing fails to
make an absolutely crucial distinction. It is at least conceivable
that one and the same piece of metal sometimes expands and sometimes
does not, when heated; this is why the non-statistical character of
the relevant laws is an empirical matter. But it is simply inconceiv
able that expressions and sentences like 'the man' and "John is easy
to please" would be sometimes c o r r e c t and sometimes i n c o r r e c t (within

a given synchronic s t a t e of language); t h i s is why the n o n - s t a t i s t i c a l
character of grammatical d e s c r i p t i o n s i s a nonempirical , a priori mat-
t e r ( c f . 9 . 1 . above). In other words, i t is not possible t h a t an ex-
122
pression or a sentence is sometimes correct and sometimes incorrect;
but it is possible that it seems sometimes correct and sometimes in
correct. This distinction is essential; it is the distinction between
(objective) common knowledge and (subjective) intuition. It is also
clear that the former, viz. what is correct, is the subject matter of
grammar, while the latter, viz. what seems correct, is (part of) the
subject matter of psycholinguisties.
When distinguishing between what is the case and what seems to be
the case, we have a genuine analogy between grammar and natural science.
The latter investigates physical reality, not our (subjective) observa
tions of it, as can be seen from the fact that each time when there is
reason to believe that our observations do not accurately reflect the
physical reality, we dismiss them. This is so even if it is through
our observations (and our logical reasoning) that we get hold of the
physical reality. (Of course, there is then another, psychological
science which investigates our observations, whether accurate or not.)
The same holds, mutadis mutandis, in grammar too. We investigate lan
guage, not linguistic intuition, although it is through the latter (as
well as through logical reasoning) that we get hold of the former.
The view of intuitions as grammatical data can be shown to fail
not only in connection with testing, but also, retrospectively, in
connection with(D-N) explanation. Even though the properties of a
correct sentence entail, and do not empirically determine, its correct
ness, it is conceivable that someone's intuitions about the former
differ from his intuitions about the latter. But this makes the rela
tion between those properties and the correctness empirical just as
little as the fact that someone may fail to notice the mutual entail
ment between 'father' and 'male parent' makes the relation between
the two empirical. And here as elsewhere it is with sentential pro
perties, not our intuitions thereof, that we are dealing.
I t is in connection w i t h t e s t i n g , I t h i n k , t h a t the d i f f e r e n c e be
tween empirical sciences and normative sciences l i k e grammar, philosophy,
and l o g i c becomes most e v i d e n t . Each science makes use of some kind of
t e s t i n g , but empirical science tests i t s theories by confronting them,
e x p e r i m e n t a l l y , w i t h space and t i m e , whereas normative science testsits
theories by confronting them w i t h concepts, or w i t h (common) knowledge.
Compare W i t t g e n s t e i n ' s (1967 I , §98) dictum: " I can calculate in the me
dium of i m a g i n a t i o n , but not experiment",
The de facto non-experimental nature of grammar has been sometimes
admitted i n a d v e r t e n t l y by t r a n s f o r m a t i o n a l i s t s :
That is, we know ahead of time in some sense what we want to come
out with as a result of our analysis. This is one sense in which
it can be said that linguistic analysis tries to account for the
linguistic intuition of the native speaker (Bach 1964:151; cf. also
the Chomsky-quotation p.81).
Here the analogy to logical and/or philosophical analysis hardly

needs to be pointed out. In discussing the analysis of rules of be
haviour (including rules of language) which one is able to follow one
self, Hare (1971:230) notes that "it is in this same way that a logi
cian knows, before he sets out to investigate the logical properties
of the concept of negation, what concept he is going to investigate".
By contrast, it would be absurd to say of empirical scientists
(among which Bach nevertheless counts himself) that, when they under
take an investigation, they already "know in some sense what they want
to come out with as a result of their analysis" or investigation. The
impossibility of such a view is shown (if it needs any showing) for in
stance by Kaplan's (1964:145) following remark which sums up the fun
damental characteristic of empirical science and, by implication, the
fundamental difference between empirical or experimental science and
normative science:
It is experimentation that expresses the basic empiricism of scien

ce. The scientist cannot lead us into nature's secrets unless he
will risk having her slam the door in his face; experiment knocks
on the door. The cardinal principle of experimentation is that we
must accept the outcome whether or not it is to our liking.123
Because experimentation expresses the true nature of empirical

science, particularly of natural science, a science which refuses to
make use of experiments is nonempirical and rather comparable to a
normative science like grammar. For us it is not quite easy to con
ceive of a nonempirical natural science. However, it seems that
Aristotelian physics was precisely this kind of science. It was a
(theoretical) explication of the everyday knowledge about the physi
cal reality and did not call this knowledge into question, as Mittel-
strass (1974b:64) explains:
Erfahrung wird von Aristoteles definiert als ein begriffliches

(allgemeines) Wissen, das aus "Erinnerungen" hervorgeht und da
bei als ein "Wissen des Besonderen" in exempl arischer (nicht et
wa induktiver) Weise gleichzeitig eine "Wahrnehmung des Alige
meinen", an dieser Stelle repräsentiert durch den Begriff (nicht
den generellen Satz), leistet. Insofern genügt dann auch "eine"
wiederum sogenannte "Erfahrung", d.h. hier ein Einzelfall, auf
den die Ausgangsunterscheidungen zutreffen, um ein solches be
griffliches Wissen zu vergegenwärtigen. Theoretische Sätze, also
z.B. physikalische Sätze, sind in diesem Zusammenhang trotz ihres
hochkomplexen Charakters Explikationen eines elementaren Erfahr
ungswissens, d.h. "in der Erfahrung" getroffener Unterscheidungen,
die stets als gemeinsam gelernte und beherrschte Unterscheidungen
gelten. Alle zentralen Sätze der Aristotelischen Physik … las
sen sich in dieser Weise verstehen.
I t is only l o g i c a l that the r o l e of recollection should be the

same i n A r i s t o t e l i a n (nonempirical) physics as in conceptual analysis
in general. 1 2 4 Mittelstrass' account is f u r t h e r in agreement w i t h
Feyerabend's (1975, chap.5) claim t h a t the p r i n c i p a l merit of Coperni-
can and Galilean physics as compared w i t h A r i s t o t e l i a n was to introduce
the idea that laws of nature are hidden and must be discovered by a
125
j o i n t e f f o r t of imagination and experimentation. The d i s t i n c t i o n
between ( e m p i r i c a l - ) e x p l a n a t o r y and ( n o r m a t i v e - ) e x p l i c a t o r y sciences
w i l l become c l e a r e r i n 11.0. (below).
Moreover, even though normative science is fundamentally different
from empirical science, i t is rather s i m i l a r to axiomatisation of the
latter. In the case of a x i o m a t i s a t i o n , t o o , we do not t r y to expand
our knowledge through the t e s t i n g of empirical hypotheses. Rather,
the universal (or s t a t i s t i c a l ) hypotheses contained i n the theory to
be axiornati sed are assvmed to be t r u e , and t h e i r mutual l o g i c a l rela-
tionships are e x p l i c i t l y stated by p o s t u l a t i n g axioms and p r o v i d i n g
d e f i n i t i o n s in such a way t h a t l o w e r - l e v e l hypotheses can be deduced
from h i g h e r - l e v e l ones, the axioms c o n s t i t u t i n g the top of t h i s h i e -
rarchy. Axiornatisation is thus systematisation of an e x i s t i n g body
of knowledge, whether empirical or conceptual; and i n each case sys-
t e m a t i s a t i o n produces new, t h e o r e t i c a l knowledge.
I have been emphasising the methodological differences between
empirical science and normative science, because i n the theory of
grammar a t l e a s t , those differences have been almost completely over-
looked; more c u s t o m a r i l y , i n f a c t , t h e i r existence has been e x p l i c i t -
l y denied. However, I am d e f i n i t e l y not claiming t h a t w i t h i n e m p i r i -
cal science new knowledge is gained only, or even p r i n c i p a l l y , through
observation of space and time or through experimentation. Especially
more advanced natural sciences develop, in the f i r s t p l a c e , through
the agency of c r e a t i v e t h e o r e t i c a l a n a l y s i s , which makes use of the
general method of conceptual a n a l y s i s ; compare Campbell's (1952:88)
statement:
At the present time, in the more highly developed sciences, ...

it is very unusual for a new law to be discovered or suggested
simply by making experiments and observations and examining the
results (although cases of this character occur from time to
time); almost all advances in the formulation of new laws fol
low on the invention of theories to explain the old laws.
In the same vein, it is important in my view to show that all

sciences make use of some variants of predictive and explanatory test
ing, and that the notion of 'generalisation' plays a similar role in
all sciences (cf. above, and 10.0.-11.0. below). Nevertheless, it
still remains true that empirical science is ultimately based upon
experimentation or, more generally, upon the fact that one has to
accept what happens in the natural course of events even though one
cannot know in advance what this will be like (cf. the Kaplan-quota
tion above). In this respect, which I consider fundamental, norma
tive science qua analysis of conceptual knowledge is different from
empirical science.
9.4. Universal Linguistic Theory
It might be argued that even if, in light of the three preceding

sections, the grammar of a natural language cannot be empirically test
ed by confronting it with the same language which it describes, the
related theory of grammar is nevertheless bound to contain empirical
hypotheses in the sense that it is an empirical question whether the
principles of analysis applying to the description of the language in
question will also apply to the description of other languages. Since
any general theory of grammar, or universal linguistic theory, would
thus be falsifiable by data coming from different languages, the con
struction of such a theory would apparently be an empirical task, even
if the construction of a grammar would not.
This seemingly conclusive argument does not show, however, that
universal linguistic theory and empirical science have the same logical
structure. It is a contingent fact that there have been, and are, so
many languages in the world, and that we have not been, and are not, in
a position to learn all of them. However, it is theoretically possible
that, if only there were sufficient documents concerning all these
(states of) languages, we would be able to learn and to know them. By
contrast, even if there existed one single natural regularity in the
world, we could never come to know it in the same way (cf. 6.1. and
8.1 . above).
Suppose that all languages of the world which have existed so far
are known, and that this knowledge has been written down in grammars.
In that case the task facing the grammarian would be, from the metho
dological point of view, exactly similar to that whicn is facing him
when he starts to describe his native language or any group of langu
ages known to him sufficiently well. The only difference with respect
to grammar-writing concerns the amount of data to be described and the
corresponding level of abstractness: it is question of making generali
sations not just about rules of one language, but also about hypotheses
in different grammars and thus, indirectly, about rules of different
languages. The following remark by Chomsky (1965:46) is fully in ag
reement with my position:
Thus the most crucial problem for linguistic theory seems to be to

abstract statements and generalisations from particular descrip
tively adequate grammars and, wherever possible, to attribute
them to the general theory of linguistic structure, thus enrich
ing this theory and imposing more structure on the schema for
grammatical description.
It is obviously impossible to construct a universal theory without

depending on some preliminary ideas about the form of such a theory,
which may then be revised or discarded in the course of actual theory-
construction. However, any eventual universal theory will be reached
as a result of the generalisations made from different grammars. Can
it be said that a tentative universal theory so constructed explains
grammars of new languages and thus these languages themselves, as TG's
notion of 'explanatory adequacy' suggests (cf. p. 233)? This can be said
only to the extent that a theory for instance which systematises card
games of our Western civilisation by generating descriptions of correct
game-performances can be said to explain hitherto unknown card games of
other civilisations insofar as it succeeds in generating the correspond
ing descriptions of game-performances. So much is clear, in any event,
that this is not a case of empirical explanation.
I repeat (cf. p. 228) that if TG is taken to be a general theory
of language acquisition and use, then it is an empirical theory, which,
to be sure, contains the theory of grammar, i.e., nonempirical theory,
as an autonomous component.
9.5. Appendix: Examples Taken from the Transformationalist

Literature
The claims which I have made in the previous sections will be

further substantiated here on the basis of certain representative TG
descriptions. Let us first consider Kiparsky & Kiparsky (1971). The
authors note that a distinction illustrated by the following two groups
of verbs can be made among the predicates which take sentences as their
objects:
A B
regret assert
ignore believe
resent conclude
These two types of predicates d i f f e r in the f o l l o w i n g respects:

i) Only A-predi cates can have as t h e i r object the noun fact with
e i t h e r a gerund or a t h a t - c l a u s e :
the fact that she is late.

He
the fact of her being late.
the fact that she is late»

*He
the fact of her being late.
ii) Gerunds can be objects of A-predi cates, but not freely of B-pre-
dicates:
Everyone Joan's being completely drunk.
Everyone Joan's being completely drunk.
iii) Only B-predicates allow the accusati v e - a n d - i n f i n i t i ve c o n s t r u c t i o n :
Mary to have been the one who d i t it.
Mary to have been the one who d i t it.
The foregoing differences are of s y n t a c t i c nature, Semantically

the d i f f e r e n c e between A- and B-predicates rests on the f a c t t h a t when
the speaker asserts a sentence containing an Α - p r e d i c a t e , he presupposes
t h a t the t h a t - c l ause or i t s e q u i v a l e n t , f u n c t i o n i n g as the o b j e c t , ex
presses a true propositions or refers to a f a c t , whereas no such pre
supposition is made when a sentence containing a B-predicate is assert
ed. For t h i s reason, A- and B-predicates are c a l l e d factïve and non-
factive, respectively.
The Kiparskys present the f o l l o w i n g ' t e n t a t i v e explanatory hypo-
t h e s i s ' , which presumably accounts f o r the above-mentioned s y n t a c t i c
and semantic data: The presupposition of a complement i s r e f l e c t e d i n
i t s ( s y n t a c t i c ) deep s t r u c t u r e , which means t h a t the underlying form
of f a c t i v e and n o n - f a c t i v e clauses w i l l be, respectively:
NP
F i r s t of a l l , i t is important to r e a l i s e t h a t the data which

the Kipars kys propose to explain are c o n s t i t u t e d by rules of language,
in the f i r s t place rules about the constructions in which particular
predicates can or cannot occur. The s t a r t i n g p o i n t i n the d e s c r i p t i o n
of the verbs assert, believe, and conclude f o r example is the i n t u i t i v e
knowledge about the rules which determine separately f o r each of these
verbs t h a t they cannot have as t h e i r object a that-clause preceded by
the noun fact nor a gerund e i t h e r w i t h or w i t h o u t the noun fact, where-
as they can have as t h e i r object the a c c u s a t i v e - a n d - i n f i n i t i v e construc-
tion. I t is obvious t h a t these are rather concrete r u l e s , j u s t like
a l l rules p e r t a i n i n g to d e f i n i t e s i n g l e words or expressions ( c f . 6.2.
above).
What does i t mean to say t h a t the Kiparskys' analysis 'explains'
anything? F i r s t , i t shows t h a t apparently d i f f e r e n t i)-A-sentences
and i i ) - A - s e n t e n c e s are in reality similar. This is done by postu-
l a t i n g a common deep s t r u c t u r e f o r them. As always, such a postulate
r e s u l t s i n making a g e n e r a l i s a t i o n . To i l l u s t r a t e : Factive sentences
of the type " I r e g r e t the f a c t t h a t John is i l l " are closest to the
postulated deep s t r u c t u r e Now, the other f a c t i v e sentence-types
can be derived from t h i s basic type w i t h the aid of the f o l l o w i n g op-

tional transformations:
THE METHODOLOGY OF GRAMMAR 26 7
Second, in a d d i t i o n to 'explaining' (or simply shoving) the i n t e r -
relatedness of the four sentence-types in q u e s t i o n , t h i s analysis also
' e x p l a i n s ' the incorrectness of i)-B-sentences and i i ) - B - s e n t e n c e s :
They lack the r e q u i s i t e deep s t r u c t u r e , which d i r e c t l y represents i)-
A-sentences and is the input f o r gerund-formation and fact-deletion
transformations leading ( i n t h i s order) to i i ) - A - s e n t e n c e s . By the
same token, the analysis 'explains' the surface s i m i l a r i t y and the
semantic d i f f e r e n c e of f a c t i ve and n o n - f a c t i v e that-clauses as in
" I regret t h a t John is i l l " and " I believe t h a t John is i l l " : The
former are derived from via f a c t - d e l e t i o n , whereas the latter-
are derived d i r e c t l y from
The preceding considerations ' e x p l a i n ' the data presented above

under the headings i ) and i i ) . The 'explanatory' import stems from
the g e n e r a l i s a t i o n which has been achieved when a set of s y n t a c t i c and
semantic f a c t s are described i n a uniform way, v i z . with the aid of
two p l a u s i b l e deep s t r u c t u r e s and two r e l a t i v e l y natural transformation
rules. However, the data presented under i i i ) still have to be account-
ed f o r : Why i s i t t h a t the a c c u s a t i v e - a n d - i n f i n i t i v e construction may-
occur only i n n o n - f a c t i v e complements? Or, to put i t more t e c h n i c a l l y ,
why does the ' s u b j e c t - r a i s i n g ' t r a n s f o r m a t i o n , i l l u s t r a t e d below, ap-
ply to n o n - f a c t i v e complements, but not to f a c t i ve complements?
"He believes Bacon to be the real author", but: *"He ignores Bacon to
be the real author".
The d i f f e r e n c e between these two sentences is 'explained' by r e l a -
ting i t to the s o - c a l l e d complex noun phrase c o n s t r a i n t , which pre-
vents transformations from taking c o n s t i t u e n t s out of a sentence S in
the c o n f i g u r a t i o n . This c o n s t r a i n t , which i s needed f o r
the d e s c r i p t i o n of many d i f f e r e n t constructions (see below), can be

used to ' e x p l a i n ' why f a c t i v e complements cannot have the accusative-
a n d - i n f i n i t i v e construction: Since t h e i r underlying s t r u c t u r e
is of the type and is thus subject to the complex noun
phrase c o n s t r a i n t , the subject of the embedded clause cannot be taken

out of S and be made i n t o the object of the upper sentence. This ana-
l y s i s r e q u i r e s , o b v i o u s l y , t h a t the s u b j e c t - r a i s i n g transformation
precedes / a c t - d e l e t i o n t r a n s f o r m a t i o n , - Here the impression t h a t
something has been 'explained' r e s u l t s , once again, from a g e n e r a l i -
sation: The data presented under i i i ) are shown to conform to a gen-
eral constraint which is required i n the grammar f o r the d e s c r i p t i o n
of many other constructions as w e l l ; and since the data conform to the
c o n s t r a i n t only on the assumption t h a t the proposed deep s t r u c t u r e s
f o r f a c t i v e and n o n - f a c t i v e complements are c o r r e c t , i t seems natural
to think t h a t they are i n f a c t c o r r e c t , or at l e a s t t h a t the 'hypo-
t h e s i s ' about t h e i r correctness has been ' c o n f i r m e d ' .
On account of what has been said above, the pronoun it which
may occur in connection w i t h f a c t i v e predicates can be i n t e r p r e t e d as
an optional reduction of the fact'.
I r e g r e t i t t h a t John is ill.
I believe i t t h a t John is ill.
The f a c t i v e it must not be confused with the e x p l e t i v e it which
i s introduced in. the place of an extraposed that-clause, regardless
of whether the predicate to which the that-clause i s assigned e i t h e r
as i t s subject or as i t s o b j e c t , i s f a c t i v e or n o n - f a c t i v e :
I t is r e g r e t t e d t h a t John is i l l .
It is believed that John is ill.
As a result, the Kiparskys are ready to refute Rosenbaum's (1967)
analysis according to which the factive it and the expletive it are
derived from the same source, viz. an i t which is assumed to be part
of the deep structure of all noun clauses. The Kiparskys' position
is further supported by certain facts concerning relativisation and
so-pronominalisation. It is to be noted that Rosenbaum's analysis
is not refuted, or 'disconfirmed' , in the sense that it would have
been shown to be unable to generate all and only correct sentences of
the types here discussed (if only because it could always be supple
mented with ad hoc rules). Rather, it is refuted because it misses
one or more generalisations, or fails to describe the data under study
in the most general and simplest way. For example, it fails to account
for the similarity of the sentences "I regret the fact that John is
ill" → "I regret it that John is ill" → "I regret that John is ill",
as contrasted with the incorrectness of "I believe it that John is
ill".
By now it has become succifiently clear that in Kiparsky &
Kiparsky (1971) the term 'explanation' is used in the sense indicated
in 9.2. (above). It signifies a generalising systematisation of in
tuitively known, conceptual data, i.e., rules of language. An ever
higher degree of generalisation is attained when more and more data
are described with the aid of fewer deep structures and grammatical
rules. It is important to realise that 'explanatory' systematisation
consists, not just in rearranging the intuitive data, but in postula
ting some additional, theoretical structure which puts the data into
a new, unifying and revealing perspective.
Next, we may ask how the Kiparskys' analysis is supposed to be
tested. Remember that a description may be tested in either of two
ways, namely, by asking whether it is false or whether it is less gen
eral, i.e., less 'explanatory', than some other description. It is
not clear to what extent the Kiparskys' analysis can be disconfirmed
by showing that it is (partly) false. This is due to the fact that
their subject matter has been clearly delimited so as to be easily
'surveyable'. Therefore they can convince themselves that the de
scription has no false consequences. In fact, its untestabi1ity is
hightened by the fact that the authors dismiss disconfirmatory evi
dence. That is, they admit that there are numerous exceptions to
their principal generalisation, i.e, predicates which are semanti-
cally factive but syntactically non-factive, or vice versa.125 The
Kiparskys apparently think that they can afford to ignore these ex
ceptions as long as their number is, intuitively speaking, not yery
high. This is a problematical assumption. Furthermore, the Kiparskys
admit that their data do not conform to the intuitions of a good num
ber of English speakers, but then they add: "We have chosen to de
scribe a rather restrictive type of speech (that of C K . ) because it
yields more insight into the syntactic-semantic problems with which
we are concerned" (1971:348, note).
This "rather restrictive type of speech" happens to be such that
it makes precisely those distinctions which the Kiparskys' 'explana
tory hypothesis' requires to be made: since it (in part) creates the
problems, it certainly "yields more insight" into them than does a
dialect which does not recognise their existence, Lakoff (1972:653,
n.l) uses a similar argument to convince his readers that his descrip
tions are not based on nothing: "The argument ... does not depend on
the particular examples given being correct for all dialects, but
only on the existence of examples of this sort for some dialects."
This formulation leaves open the possibility that examples taken
from different dialects are summarily lumped together and subjected to
a falsely unifying treatment. In addition, existence in itself is not
a yery secure criterion: when examples are claimed to exist only in
very restricted dialects, ultimately in idiolects, there is the dis
tinct possibility that they do not exist, as a matter of fact. As
the impossibility of private languages has shown, whatever seems cor
rect in such a case, is correct, with the consequence that the notion
of a language ceases to apply. - Insofar as intuitions differ here
in fact, we have to do with a decrease in the social control of lan
guage. In such a case (statistical) description of actual use of
language and/or psycholinguistic experimentation is needed in order
to ascertain what the facts are (cf. 6.4. above).
With regard to the second type of testing, the situation is re

latively simple. The Kiparskys' analysis can be 'disconfirmed' in
the same way that they have 'disconfirrned' Rosenbaum's analysis, name
ly, by showing that certain generalisations have been missed. We
might say that the inferiority of Rosenbaum's analysis (assuming
that it has been established) is due to lack of insight, rather than
to lack of attention (cf. p,253). His theoretical description does
not literally conflict with those facts which it purports to cover,
but the alternative account proposed by the Kiparskys covers a wider
range of facts.
To take another example, let us consider Ross's discussion of
Chomsky's 'A-over-A principle' (as Ross calls it), which is formula
ted as follows:
What this principle asserts is that if the phrase X of category

A is embedded within a larger phrase ZXW which is also of cate
gory A, then no rule applying to the category A applies to X
(but only to ZXW) (Chomsky 1964c:931).
The A-over-A principle is a (theoretical) generalisation out of,

and hence an 'explanation' of, a set of more or less obviously inter
related syntactic data. To illustrate: Elements of relative clau
ses may not be questioned or relativised; compare "I chased the
boy who threw a snowball at our teacher", but: "Here is the snowball
which I chased the boy who threw at out teacher". Secondly, an NP
which is exhaustively dominated by a determiner cannot be questioned
or r e l a t i v i s e d out of the NP which immediately dominates t h a t defer-
miner; compare "Whose book did you f i n d ? , but *"Whose did you f i n d
book?"
However, the A-over-A principle is 'disconfirmed' by showing
that it conflicts with, or excludes some perfectly correct construc
tions, like "Who would you approve of my seeing?" or "The book which
I lost the cover of is on the table". This deficiency has given Ross
the reason to construct his own 'complex noun phrase constraint' which
can handle also the latter cases and thus replaces, together with some
other independently needed constraints, the A-over-A principle.
Since Chomsky does not present the above-mentioned correct con
structions as exceptions to his A-over-A principle, we must ask why
he formulated it so as to exlude them. Do they constitute new dis
confirmatory evidence of which he could not possibly have been aware
at the time when he formulated his principle? Of course not. In rea
lity he knew these constructions all the time, and the fact that his
principle excludes them is simply due to his lack of attention (cf. p.
253),which is, by the way, quite understandable, given that his sub
ject matter is so complex as not to be surveyable any longer. When
Chomsky or any other linguist becomes aware of the (correct) construc
tion "Who would you approve of my seeing?", he reminds himself of its
existence (cf, p.212 and 254).
For comparison, suppose that I am struck by the fact that people
who make frequent insinuations tend to be malevolent and mendacious,
with the result that, when I am making the semantic analysis of to in
sinuate, I set up the following implication: "a insinuates that it
is not true that p". But then my analysis may be 'disconfirmed' simply
by pointing out semantically correct sentences where something that is
de facto true is said to be insinuated by someone. Consequently I
have overlooked quite an undeniable aspect of the meaning of to insi
nuate, which means that the defect in my analysis is due to my lack of
attention. I would maintain that even if my example is artificially
simple, it is not qualitively different from Chomsky's A-over-A prin
ciple. And since my example constitutes a case of conceptual analysis,
so does Chomsky's principle too.
To pursue the same example, it could be said that my imaginary
analysis of to insinuate makes the (false) 'prediction' that sentences
where something de facto true is said to be insinuated are semantically
incorrect. Analogously, Chomsky's A-over-A principle makes the false
'prediction' that sentences like "Who would you approve of my seeing?"
are incorrect. To give another example, Ross (1968:1) requires that
eyery adequate grammar of English should make for instance the follow
ing 'predictions' about the sentence "A gun which I had cleaned went
off":
The constituent a gun which I had cleaned is a constituent of the

same kind as the constituent I. Similarly, went off is the same
type of constituent as had cleaned, and neither is of the same type
as I, a, or off.
In other words, a grammar of English must be able to 'predict'

that (what we would call pronouns and verbs) belong to different gram
matical categories. But just as well we might require of a semantic
theory that it 'predict1 the difference between things and events, or
of a mathematical theory that it 'predict' the difference between odd
and even numbers.
The peculiarity of TG's use of the term 'prediction ' can be clear
ly seen also in connection with redundancy rules, which are usually
said to 'predict' redundant semantic, syntactic, and phonological fea
tures (cf. Chomsky 1965:164-70). Thus, on the basis of the feature
"(+human)" we may presumably 'predict' the features "(+animate)" and
"(+concrete)". But according to the general consensus among philoso
phers and methodologists, what we have here is not a relation of (em
pirical) prediction, but of (conceptual) entailment.
The confusion between the two last-mentioned concepts can be
rather well observed in a passage by Katz, where he states that, like
any scientific theory, the semantic component of TG makes empirical
predictions: for example, it predicts that sentences like "I saw an
honest stone" are semantically anomalous (Katz 1966:174). But what
this means is only that TG must be capable of giving a formal expres
sion to the analytic truth that a property like honesty which is
only predicable of humans is not predicable of non-humans, e.g.,
stones. If a description of this kind is 'empirical', then so is each
and every philosophical or conceptual analysis.
In TG's view, then, it is a matter of empirical description that
a property, e.g., "unmarried", which is a constitutive part of a con
cept, e.g., "bachelor", is predicable of all things exemplifying
this concept, yielding "All bachelors are unmarried"; for surely we
can 'predict' that if this bachelor is unmarried, so is also the next
one, just as we can 'predict' animacy on the basis of humanness (as
shown above). But, of course, there is nothing empirical about a
d e s c r i p t i o n of these matters. " A l l bachelors are unmarried", equiva-
l e n t to " A l l unmarried men are unmarried", is an instance of the lo
gical truth , and a d e s c r i p t i o n which says so i s , to
t h a t e x t e n t , necessarily t r u e , whereas a d e s c r i p t i o n which denies it
is necessarily false. TG's p o s i t i o n here amounts to the claim t h a t
a n a l y t i c a l philosophy is an empirical science, which only would
suggest t h a t TG simply does not understand the meaning of the term
'empirical' ( c f . also p.80 above).
Notice t h a t i t w i l l not do to claim t h a t i t is only the semantic
component of TG which is s i m i l a r to conceptual a n a l y s i s , as p r a c t i s e d
by a n a l y t i c philosophers, while the other components are s i m i l a r to
empirical t h e o r i e s . F i r s t , such an i n t e r p r e t a t i o n is definitely
ruled out by a l l representative statements expounding TG's methodo-
logy. Secondly, f o r a long time i t has been overwhelmingly clear
t h a t there i s no hard-and-fast d i s t i n c t i o n between semantics and
syntax. Therefore, while conceptual analysis and empirical science
must be kept a p a r t , the whole of TG must be i n t e r p r e t e d as conceptual
a n a l y s i s , or e x p l i c a t i o n .
I t is p r e c i s e l y i n connection w i t h Katz's semantics t h a t I f o r
the f i r s t time became aware of the i n c o m p a t i b i l i t y of the e m p i r i c a l -
science model and the d e s c r i p t i v e p r a c t i c e of TG. In a paper w r i t t e n
i n 1968 I showed t h a t , contrary to Katz's c l a i m s , semantic markers
(or meaning components) cannot be i n t e r p r e t e d as hypothetical con-
cepts, to be e m p i r i c a l l y confirmed or disconfirmed on the basis of
t h e i r a b i l i t y to p r e d i c t such semantic r e l a t i o n s as a n a l y t i c i t y , ano-
maly, and ambiguity. I n s t e a d , they are concepts which may be used
to explicate our i n t u i t i v e notions of a n a l y t i c i t y , e t c . I concluded:
Wenn man bewusst darauf verzichtet, die Komponenten zu definieren

(statt z.B. die Analytizität als den primitiven Begriff zu wählen),
und auf ihrer Grundlage die anderen Begriffe explizit definiert,
so ist dies natürlich eine übersichtliche und legitime Art und
Weise, die vorhandenen Kenntnisse zu systematisieren; sie kann
aber nicht empirisch verifiziert werden, solange sie implizit
wenn auch nicht explizit zirkulär bleibt und keine unabhängige
Evidenz vorhanden ist (Itkonen 1970a:9)
In other words, I anticipated here the basic tenet of 9.3. (above):
In semantic (or more generally, grammatical) analysis there is no new,
conceptually independent evidence. The semanticist is given a set of
intuitively known meaning-relations, and all he can do is to give a
systematic description theoreof, for instance using meaning components
as primitive descriptive concepts. From the above, it also follows
that not just empirical testability, but also empirical (D-N) expla
nation is impossible within grammatical analysis (cf. Itkonen 1972a
and b ) . 1 2 7
10.0. GRAMMAR AND LOGIC
Because I consider TG as representative of grammatical theory i n
g e n e r a l , I have to r e s t r i c t here my a t t e n t i o n to axiomatic l o g i c , when
comparing grammar and l o g i c , given t h a t generative grammars are d e f i n -
able as axiomatic systems ( c f . below). That i s to say, TG-type se-
mantics too i s described axi ornatically or s y n t a c t i c a l l y , i . e . , by gen-
erating sentences of 'Semantic Markerese' (Lewis 1972:169-70) which
are r e l a t e d to sentences of English f o r instance as the meanings of the
latter. This procedure i s problematical, but I can dispense w i t h cri-
t i c i s i n g i t here. The reason i s t h a t what I have to say about the no-
tions of l o g i c a l ' t e s t i n g ' and ' e x p l a n a t i o n ' , as well as t h e i r r e l a -
t i o n to t h e i r grammatical e q u i v a l e n t s , i s general enough to apply not
only to axiomatic l o g i c , but also to model-theoretic or semantic l o g i c
and to Lorenzen & Lorenz's game-theoretical logic.
There are also considerable differences between grammar and l o g i c ,
l a r g e l y due to t h e i r d i f f e r e n t research i n t e r e s t s . I can omit discus-
sing t h i s question here, because I have presented my views on i t i n
Itkonen (1976a) .
10.1. The Basts of the Similarity 'between Generative Grammars

and Systems of Logic
An axiomatic system consists of a set of axioms and of a set of

rules of inference by means of which theorems are derived from axioms
and/or from theorems previously derived. Generative grammars, from
type 0 to type 3, are subsumable under the general notion of axiomatic
system (Wall 1972:197-212). The basic similarity of axiomatic systems
can be brought out by comparing two systems, one in the rewriting no
tation and the other in the language of predicate calculus, both of
GRAMMAR AND LOGIC 277
which generate the same set of theorems, viz. all and only sentences
of the type a n b n , or the language L3 of 9.3. (cf. the figure below).
Both derivations, when interpreted, say the same thing: aabb 'is' an
S or belongs to the class s.
'Axiomatic system' is a concept which contains 'axiomatic theory'

as a special case. In connection with the former a rule of inferen
ce is merely a procedure for mechanically passing from strings of
symbols to other strings, whereas in connection with the latter the
strings must admit of truth-value and rules of inference are truth-
preserving procedures for passing from strings to other strings. The
fact that generative grammars are a subclass of axiomatic systems is
not, as such, any conclusive proof of the similarity between grammar
and logic, because the best-developed natural sciences have also been
axiomatised. What must be shown, in addition, is that there is a
significant difference between axiomatised natural science, on the one
hand, and grammar and logic, on the other.
The axiomatisation of a natural science generates a set of univer
sal and/or statistical hypotheses as its theorems. These sentences
purport to refer to something in the external world, namely higher-
level or lower-level regularities obtaining in that particular domain
of reality which is investigated by the science to be axiomatised. If
such regularities exist, the theorems, or hypotheses, are true; other
wise they are false. If they are false, then one or more of the high
er-level theorems from which that were derived are - by Modus Tollens-
also false, and must be modified accordingly. In other words, it is
the purpose of the axioms and of the rules of inference to generate
(empirically) true sentences as theorems, and the criterion for the
truth of a sentence lies outside it, i.e., in the external world.
Let us now compare this situation to that prevailing in gramma
tical theory. The sentences generated by the grammar of a natural
language do not, of course, refer to anything in the external world
which would be the subject matter of linguistics in the same way as,
for example, regularities of such and such a type are the subject
matter of physics. Rather, these sentences, with their characteris
tic meanings and presuppositions, are themselves the subject matter
of linguistics. That is to say, it is the purpose of a natural-
language grammar to generate, not tvue sentences, but. (syntactically
and semantically) covvect sentences, as its theorems; and the cri
terion for the correctness of a sentence lies in the sentence it
self, not outside it.
In the last-mentioned respect, logic seems at first glance to
occupy an intermediate position between grammar and natural science.
Just like the axiomatisation of a natural science, an axiomatic sys
tem of logic is not interested in generating just correct sentences.
Rather, it is in both cases a precondition for the sentences to be
generated as theorems that they be (formally) correct or 'well-form
ed'. On the other hand, a system of logic is not interested in gen
erating sentences which are simply true. Rather, it purports to
generate sentences which are valid, or logically true. Validity is
truth in 'all possible worlds', which means that reference to the
external world has no bearing on the validity of sentences. Conse
quently, and most importantly, the criterion for the validity of a
given sentence, just like that for the correctness, lies in the
sentence itself. (Horeover, since validity is given a formal defi
nition, logic may concentrate directly upon sentence-formulae, in
stead of dealing with sentences exemplifying formulae.) In this
decisive respect, logic and grammar are similar to each other and
different from natural science.
By noting that sentences and formulae themselves are the re-
spective subject matters of grammar and logic, I merely reaffirm

the self-referential character of grammatical and logical descrip
tions (cf. p. 243).
I have noted before that self-referential grammars are equipped
with referential metagrammars which provide natural-language inter
pretations for the former. A metagrammar is a theory which claims,
either truly or falsely, that such and such rewriting rules express
generalisations about a language L and generate, or approximate to
generating, all and only correct sentences of L. Now, since meta
grammars deal with truth and falsity, and not just with correctness
or incorrectness, it may seem that they are just empirical theories
among others. However, this is not so; or it is so only on the as
sumption that the same argument has turned logic as well into an em
pirical science. As we have just seen, a metagrammar claims, either
truly or falsely, that the correlated grammar generates all and
only correct sentences of L. But in precisely the same way the meta-
logic of a given logical system claims, either truly or falsely, that
the system generates all and only valid formulae of a given logical
language.
Instead of saying that grammars and systems of logic deal with
sentences and formulae, we might as well say that they deal with the
(normative) common knowledge constituting sentences and formulae, or
constituting those rules which make sentences and formulae correct
and valid respectively. This knowledge is a case of agent's know
ledge: it is acquired through common participation in the institu
tions of (correct) speaking and (correct) inferring. Logic tries to
formalise the rules of inferring as they are intuitively known by
everybody; this is the meaning of the somewhat shorthand expression
that what logic does is to formalise logical intuition. (The analo
gous claim that it is linguistic intuition what grammarians are de
scribing, must be understood in the same way; cf. 5.2. above). It
is in this precise sense, then, that logical intuition can be said
to precede, and to constitute the subject matter of, any type of
formal logic, whether syntactic as in axiomatic logic, semantic as
in model-theoretic logic, or pragmatic as in Lorenzen & Lorenz's game
theoretical logic. Pap has clearly expressed this primacy of logical
intuition (even though he ignores its pragmatic or institutional ori
gin):
The test of whether the range of a statement is universal [i.e.,

whether a statement is logically true] is constructed on the
assumption of the logical truth of the laws of contradiction and
excluded middle. This assumption is formally reflected by the
'convention' to assign either Τ or F, but not both, to each ele
mentary proposition, and the 'convention' to include in every
state description a given atomic statement or its negation but
not both. It follows that at least the logical validity of these
two simple and fundamental laws is not known as a result of such
a formal test. Incidentally, this simple consideration shows
that the claim that our knowledge of the laws of logic is purely
formal, not intuitive as claimed by many traditional logicians,
must be taken with a grain of salt (Pap 1958:157).
Moreover, i t i s c l e a r l y wrong to say, w i t h Rescher (1966:54),

t h a t the contrast between the axiomatic method and the model-theore
tic one is simply t h a t between an i n t u i t i v e approach to l o g i c and
a n o n - i n t u i t i v e one:
...little has been gained by Carnap's disguise in semantic termi

nology of the Leibnizian conception of 'truths of reason' as
truths holding in all possible worlds. The definition is still
as circular as it ever was: a possible world can be nothing
else that a world conforming to the laws of logic (Pap 1958:151).
Although both grammar and logic are based on intuition, their re
spective attitudes towards it are different. Formal logic tries to
overcome the inherent limitations of intuition, and to this end it
devises more and more effective methods of calculating. On the other
hand, it would not make sense to say that this is what grammar too is
aiming at (for details, cf. Itkonen 1976a). This is precisely the
difference between descriptive normative sciences like grammar and
prescriptive normative sciences like logic.
10.2. Testing
A grammar is 'tested' by finding out whether it generatesalland

only correct sentences of L (with their 'correct' structural descrip
tions). The same requirements are imposed on axiomatic systems of
logic, except that the crucial concept is not 'correctness' but 'vali-
dity'. Once a formal definition of validity has been given, it is

required that the system in question be both 'sound' and 'complete'.
It is sound, if it generates only valid formulae, and it is complete,
if it generates all valid formulae. The system is then 'tested' by
finding out whether it generates all and only valid formulae. More
over, even if the system is provably sound and complete, it may still
be tested with the view to finding out whether validity as defined
within the system corresponds in each case to the intuitive notion
of validity, that is, whether all theorems of the system (which have
been proved as formally valid) are intuitively valid, and whether all
intuitively valid truths formal i sable in the language in question are
theorems of the system. No general proofs concerning these last-men
tioned properties of axiomatic systems can be given. Such a proof
would require a formal definition of intuition, or of intuitive vali
dity, which either is a contradiction in terms or leads directly into
infinite regress (cf. p.24).
The axiomatic systems of the most familiar types of logic, i.e.,
propositional logic and predicate logic, have been proved both sound
and complete. Therefore it would be pointless to test them with a
view to investigating their soundness, i.e., the formal validity of
their theorems, or their completeness, i.e., the theoremhood of their
formally valid formulae. However, befove these proofs had been given
this kind of testing was perfectly reasonable. Moreover, it is gen
erally agreed that formal validity in propositional and predicate
logic corresponds to, and deviates from, intuitive validity in con
trollable and defensible ways. For instance, once the truth-func
tional definition of material implication has been accepted, the in
tuitive oddity of formally valid formulae like
does nothing to undermine the 'truth' or, rather, the adequacy of the
axiomatisation of propositional logic. Similarly, once the rules
governing the introduction and the elimination of the quantifiers
have been accepted, the formal validity of an intuitively non-obvious
formula like does not constitute counter-evidence
against the system of predicate logic but, on the contrary, enhances
its importance. Consequently, it hardly seems reasonable to test
systems of p r o p o s i t i o n a l or predicate l o g i c by comparing t h e i r formal
notion of v a l i d i t y w i t h the corresponding i n t u i t i v e n o t i o n . In b r i e f ,
as regards t e s t i n g , these systems (which have become 'instruments of
language' in the sense given on p.154) o f f e r only a l i m i t e d analogy
to natural-language grammars. Therefore I s h a l l turn to a d e f i n i t e l y
more i n t e r e s t i n g example, namely deontic l o g i c , as defined and deve-
loped by von Wright.
Deontic l o g i c i n v e s t i g a t e s l o g i c a l r e l a t i o n s between the basic
v a r i e t i e s of norm, namely o b l i g a t i o n s , permissions, and p r o h i b i t i o n s .
The domain of deontic l o g i c is not as w e l l - d e f i n e d as t h a t of propo-
s i t i o n a l or predicate l o g i c , because our i n t u i t i o n s about normative
r e l a t i o n s cannot be d e l i m i t e d in advance in any d e f i n i t i v e way. Con-
sequently, an analysis of the ordinary-language counterparts of deon-
tic formulae proves to be indispensable and t h e r e f o r e , contrary to
what is the case in p r o p o s i t i o n a l or predicate l o g i c , the f a c t t h a t
ordinary language r e f l e c t s (our knowledge of) l o g i c only i m p e r f e c t l y ,
c o n s t i t u t e s a genuine obstacle to l o g i c a l analysis.
Propositional deontic l o g i c is based on p r o p o s i t i o n a l l o g i c . The
essential a d d i t i o n is the deontic operator ' 0 ' , which stands f o r obli
gation. A formula l i k e '0p' means " I t i s o b l i g a t o r y t h a t p" or,
more p r e c i s e l y , "One ought to see to i t t h a t p". von Wright's so-
c a l l e d Old System of Deontic Logic contains the f o l l o w i n g two axioms:
A1 -(0p&0-p)
A2 0(p&q) = (0p&0q)
Al i s the deontic counterpart of the law of c o n t r a d i c t i o n , and

A2 regulates the d i s t r i b u t i o n of the o- o p e r a t o r . The rules of i n -
ference are the rules of s u b s t i t u t i o n and of detachment (= Modus
Ponens), as in p r o p o s i t i o n a l l o g i c , plus the ' r u l e of e x t e n s i o n a l i t y '
which says t h a t i f ' p ' and ' q ' are l o g i c a l l y e q u i v a l e n t , then '0p' and
' 0 q ' are l o g i c a l l y e q u i v a l e n t . F i n a l l y , the concept of permission is
defined as f o l l o w s : ' P ( - ) ' = '-0 - ( - ) ' ( c f . von Wright 1951 and 1971a)
The v a l i d i t y of a deontic formula may be decided by transforming
the formulae of p r o p o s i t i o n a l l o g i c w i t h i n the scope of the o-operators
i n t o t h e i r p e r f e c t conjunctive normal forms and by d i s t r i b u t i n g the
operators in accordance with A2. The r e s u l t i n g expression i s a con

j u n c t i o n of o-expressions each of which consists of one o-operator
having a d i s j u n c t i o n of p r o p o s i t i o n a l variables in i t s scope. Such
o-expressions are c a l l e d o - c o n s t i t u e n t s of the o r i g i n a l formula. The
l a t t e r expresses a t r u t h - f u n c t i o n of i t s o - c o n s t i t u e n t s , and i t s va
l i d i t y i s decided by t r u t h - t a b l e s , as in p r o p o s i t i o n a l l o g i c , w i t h
the r e s t r i c t i o n t h a t a l l possible o- c o n s t i t u e n t s cannot be true a t
the same time. Hence, the system i s deci dable. I t is also both
sound and complete.
How i s such a system of deontic l o g i c to be tested? I t is im
p o r t a n t to r e a l i s e t h a t t h i s question can be sensibly asked in s p i t e
of the f a c t t h a t the system has been proved both sound and complete.
In e f f e c t , the subsequent development of von Wright's Old System is
a good i l l u s t r a t i o n of how an axiomatic system of l o g i c may be modi
f i e d and r e f i n e d in response to ' d i s c o n f i r m a t o r y ' claims to the ef
f e c t t h a t i t does not generate something which i t ought to generate,
or generates something which i t ought not to generate. At a s u f f i
c i e n t l y high level of a b s t r a c t i o n , the analogy to the t e s t i n g of
natural-language grammars ( o f the TG v a r i e t y ) should be obvious.
As f a r as the l o g i c of absolute or unconditional norms is con
cerned, i t i s possible to prove in the Old System several intuiti
vely obvious formulae as well as formulae which are non-obvious but
acceptable on closer i n s p e c t i o n , and therefore extend l o g i c a l know
ledge beyond mere l o g i c a l i n t u i t i o n . However, as has been pointed
out by P r i o r (1954) and Chisholm (1963), i t is impossible to give an
adequate f o r m a l i s a t i o n of the notion of commitment o r , more general
l y , of r e l a t i v e or c o n d i t i o n a l norms, in the framework of the Old
System, von Wright o r i g i n a l l y used the formula to express
the c o n d i t i o n a l norm t h a t doingp commits one to doing q. Now, be
cause the equivalent formulae and are theo
rems o f , or generated by, the system, i t follows t h a t , w i t h i n it,
doing something f o r b i d d e n , i . e . , -p in the f i r s t and ρ in the second
formula, commits one to doing anything whatever, i . e . , q; but t h i s
is i n t u i t i v e l y unacceptable. In other words, although
is formally v a l i d , w i t h i n the system, i t is nevertheless intuitively
i n v a l i d , and therefore the system must be m o d i f i e d . Here as e l s e
where, i n t u i t i o n is the primary t h i n g , and a f o r m a l i s a t i o n i s deemed
a success or a f a i l u r e depending on whether we ( i n t u i t i v e l y ) know
t h a t i t represents adequately the i n t u i t i o n to be f o r m a l i s e d . (On the
other hand, as I j u s t noted, i t is true t h a t an adequate f o r m a l i s a
t i o n also reveals hidden l o g i c a l i n t u i t i o n , or produces new l o g i c a l
knowledge«)
In order two obviate P r i o r ' s and Chisholm's p a r t i a l 'disconfir
mation' of his systematisation of deontic l o g i c , von Wright i n t r o
duced the a d d i t i o n a l symbol ' / ' and replaced the monadic n o t a t i o n
of the type '0p' by a dyadic n o t a t i o n of the type ' 0 ( p / q ) ' . This
l a s t formula can be read: "One ought to see to i t t h a t p when q".
Accordingly, he replaced the two axioms of the Old System by the
f o l l o w i n g three axioms:
B1 -[0(p/q) & 0(-p/q)]

B2 0(p&q/r) = 0 ( p / r ) & 0 ( q / r )
B3 0(p/qvr) ≡ 0 ( p / q ) & 0 ( p / r )
Bl formulates Al i n r e l a t i o n to the condition q, while B2 and B3

determine the d i s t r i b u t i o n of the o-operator in accordance w i t h the
requirements of the new, dyadic n o t a t i o n . The r e s u l t i n g system i s
c a l l e d the New System of Deontic Logic. The decision procedure f o r
v a l i d i t y is modified in the obvious way: Tne formulae to the l e f t and
to the r i g h t of ' / ' are transformed r e s p e c t i v e l y i n t o t h e i r perfect
conjunctive and d i s j u n c t i v e normal forms, a f t e r which the o-operators
are d i s t r i b u t e d in accordance with B2 and B3. The r e s u l t i n g formula
i s a conjunction of 0-constituents of the type ' 0 ( . . . v . . . / ... & . . . ) '
and, subject to c e r t a i n r e s t r i c t i o n s , i t s v a l i d i t y is decided by t r u t h -
t a b l e s , by considering o - c o n s t i t u e n t s as atomic components. 128
I t can be shown t h a t the dyadic equivalents of the objectionable
formulae and are not theorems of the New
System, and, to t h i s e x t e n t , von Wright has succeeded in answering
P r i o r ' s and Chisholm's ' d i s c o n f i r m a t o r y ' c r i t i c i s m . However, some of
the theorems of the New System, t o o , are i n t u i t i v e l y i n v a l i d , i n s p i t e
of t h e i r formal v a l i d i t y . As Geach has p r i v a t e l y pointed out to von
Wright ( c f . von Wright 1971a:115), the system generates as i t s theo

rem the formula , which says, i m p l a u s i b l y , t h a t if
there is an o b l i g a t i o n to see to i t t h a t ρ under circumstances q,
then there is no o b l i g a t i o n to see to i t t h a t not-p under some other
circumstances v. That is to say, the New System too generates some
thing which i t ought not to generate and needs to be revised accord
ingly.
von Wright's answer to t h i s challenge is to modify Bl , which was
meant to be the dyadic reformulation of A l , i . e . , ' - ( 0 p & 0 - p ) ' . As he
sees i t i n von Wright (1971a), i t is not the purpose of Al to deny
t h a t there are the o b l i g a t i o n to see to i t t h a t ρ and the o b l i g a t i o n
to see to i t t h a t n o t - p . Rather, he now i n t e r p r e t s Al as making the
weaker claim t h a t , loosely speaking, not a l l deontic possibilities
(of the monadic type) can be true at the same time. As applied to
the simplest dyadic case, t h i s p r i n c i p l e produces the f o l l o w i n g r e
formulation of Bl :
Bl' -[0(p/p)&0(p/-p)&0(-p/p)&0(-p/-p)]
The c o u n t e r - i n t u i t i v e formula is not a theorem

of the amended New System containing B l ' in l i e u of B l . In t h i s sense
the disconfirmatory claim has been, again, disposed o f . However, t h i s
r e s u l t is obtained at the cost of allowing f o r the existence of genui
nely (and not j u s t apparently) c o n f l i c t i n g c o n d i t i o n a l norms. For i n
stance, both ' 0 ( ρ / ρ ) ' and ' 0 ( - p / p ) ' may be t r u e , which means t h a t , i n
such a case, one ought to see to i t t h a t the s i t u a t i o n ρ remains as i t
is, and one ought to see to i t t h a t the s i t u a t i o n ρ ceases to e x i s t .
This p r i n c i p l e is problematic, to say the l e a s t . In f a c t , i t has been
suggested t h a t the New System ought to be amended by r e v i s i n g B3, not
B l , which would both prevent the c o u n t e r - i n t u i t i v e formula from being
a theorem and exclude the existence of c o n f l i c t i n g norms (Føllesdal &
Hilpinen 1971:28-31).
The examples I have adduced here suffice to show that there are
several possible systematisations of deontic logic. They also make
it clear enough that, just like every transformational grammar, every
particular system of deontic logic is, and will remain, subject to
286 GRAMMATICALTHEORYAND METASCIENCE
(nonempirical) testing and, eventually, to di sconfirmation. Both in
linguistics and in deontic logic there is, then, no possibility to
give a general proof to the effect that a given description is the
best one or the definitive one. In deontic logic, this property
of axiomatic systems in rendered even more comprehensible by the
fact that there obtains considerable disagreement as to the correct
(intuitive) interpretation of certain formulae. One might mention the
much-discussed formulae and' , of
which the former (known as 'Ross's paradox') is, and the latter is
not, generated by the Old System. On intuitive grounds, this treat
ment of the two formulae seems unsatisfactory. That is, the former
of them seems to contradict the principle that a system ought to gen
erate only (intuitively) valid formulae, while the latter seems to
contradict the complementary principle that a system ought to generate
all (intuitively) valid formulae. It is therefore only logical that
von Wright (1968) should have constructed a system which does not
generate the first formula and generates the second. But again, his
position is not universally accepted.
Some linguists seem to be startled by the fact that there are
conflicting linguistic intuitions. They apparently think that this
fact seriously undermines the status of grammar as a respectable
science. Therefore I wish to emphasise, in addition to what I said
in 5.4., that this situation is by no means unique to grammatical
analysis. We meet the phenomenon of conflicting intuitions in logical
and philosophical analysis as well. The interesting question is, in
the first place, why this fact alarms grammarians so much and logicians
and philosophers so little. As exceptions to this rule, one might men
tion at least Naess (1952) and Mates (1971), who have recommended some
thing like an empirical approach to philosophical analysis, with, signi
ficantly, very little sympathetic response.
In conclusion, I repeat that the requirement that a system generate
all and only valid formulae may be taken in two senses: Either it con
cerns formally valid formulae, and in that case it may be proved that
the system fulfils or fails to fulfil the requirement, Or it concerns
intuitively valid formulae, and in that case no general proofs about

the generabil ity can be given., In the ideal case formal validity and
intuitive validity coincide as a matter of fact. But it cannot be
proved that they do so, because intuitive validity lies beyond formal
proof.
The testing of a logical (or grammatical) system as to its sound
ness and completeness is structurally similar to predictive and expla
natory testing, respectively (cf. 1.2.). At a sufficiently high level
of abstraction, then, all sciences are seen to make use of the same
notion of testing. At this level, 'testing' simply means finding out
whether the description is as it should be. Experimentation is of
course only one way of doing this.
10.3. Exp lana tion
Within TG, explanation i s equivalent to g e n e r a l i s a t i o n . A genera-

l i s a t i o n consists i n reducing apparent or ' s u r f a c e ' v a r i a b i l i t y to
'deeper' u n i f o r m i t y . Because a generative grammar has only one axiom,
i.e., the symbol S, i t i s s e l f - e v i d e n t t h a t i n TG making generalisa-
t i o n s manifests i t s e l f as decreasing the number, not of axioms, but
of rules of inference and, eo ipso, of deep s t r u c t u r e s as w e l l .
One of the main p r i n c i p l e s guiding the c o n s t r u c t i o n of systems
of l o g i c is to t r y to s i m p l i f y the system w i t h o u t unduly diminishing
i t s generative c a p a c i t y . I t is easy to see t h a t t h i s p r i n c i p l e is
i d e n t i c a l w i t h the attempt to replace a set of grammatical rules by a
s i n g l e r u l e , or the attempt at g e n e r a l i s a t i o n , as p r a c t i s e d i n l i n g u i s -
tics. Hence, although the terms ' e x p l a n a t i o n ' and ' g e n e r a l i s a t i o n '
are seldom used i n l o g i c , there obtains i n l o g i c too a p r a c t i c e cor-
responding to them. For i n s t a n c e , a f t e r i t was proved t h a t the f i f t h
axiom of Russell and Whitehead's o r i g i n a l system of p r o p o s i t i o n a l
logic, i.e., , need not be s t a t e d separately but can
be derived from the four o t h e r s , the r e s u l t i n g system w i t h only four
axioms was c l e a r l y more general than the previous one, since i t both
was simpler and retained the same generative capacity.
In l i n g u i s t i c s i t is a well-known f a c t t h a t making a generalisa-
t i o n that coverss e.g., a c e r t a i n number of sentence-types r e s u l t s in
a d e s c r i p t i o n which i s more t h e o r e t i c a l or more a b s t r a c t than the de-
s c r i p t i o n o f any of these sentence-types taken separately. In l o g i c ,
t h i s property of grammatical d e s c r i p t i o n i s to some extent r e f l e c t e d
i n the f a c t t h a t the most obvious or 'concrete' l o g i c a l t r u t h s do not
s u f f i c e to generate a l l v a l i d formulae. In Russell and Whitehead's
system of p r o p o s i t i o n a l l o g i c , f o r i n s t a n c e , such i n t u i t i v e l y obviou-
t a u t o l o g i e s as or ' p v - p ' do not f i g u r e among axioms but have to
be proved i n q u i t e roundabout ways, whereas the axioms contain such
a r e l a t i v e l y complex formula as . I t could be
said t h a t , as a r u l e , g e n e r a l i s a t i o n leads both to greater simplicity
and to greater abstractness of the e n t i r e system.
The notion of simplicity is notoriously a difficult one, and in
logic too it can be understood in different ways. If we take decreas
ing the number of primitive terms and axioms as the decisive criterion
of simplicity, then we ought to consider Nicod's axiomatisation of pro-
positional logic as the simplest and, hence, the most general one,
because he achieves it by means of one single connective, i.e., the
so-called 'stroke function' equivalent to '-(p&q)', one single axiom,
and a modification of Modus Ponens as the rule of inference (Copi 1967:
267-76). On the other hand, Nicod's single axiom is more complex than
any of the axioms of standard systems, and it contains five propositio-
nal variables instead of the three needed in standard systems. More
over, proving theorems in Nicod's system is an extremely complicated
undertaking. Therefore, at least from the practical point of view,
this system can hardly be deemed more general than Russell and White
head's. Analogous difficulties in interpreting the notion of simpli
city are familiar in grammar: should we for instance prefer one complex
transformation to two (or three or ...) simple ones?
Up to now I have been discussing the similarity of the notions of
explanation in grammar and in logic at a rather abstract level. How
ever, this similarity can be shown to hold in much greater detail. In
what follows, I shall compare the notion of logical explanation as de
fined in Hintikka (1969), with the notion of grammatical explanation
as defined in 9.2. (above).
The descriptive framework employed by Hintikka in his investiga

tions on epistemic logic is not axiomatic but model-theoretic or se
mantic. This difference is not directly relevant to the issue at
hand, however. Hintikka is concerned with the question how methods
of formal logic can be made use of in philosophical analysis, more
precisely, how epistemic logic may contribute to the clarification
of philosophical problems connected with the concepts 'knowledge'
and 'belief'. So we might be inclined to say that the explanandum
is philosophical while the method of explanation is logical. Such
a strict separation of philosophy and logic is, however, an illusion.
For instance, deontic logicians freely adduce philosophical arguments,
when debating about the correct interpretation of the formula
(cf. 10.2.).
By way of an example, Hintikka examines the analysis of the mean
ing of the expression 'knowing that one knows', as presented in Hin
tikka (1962). He notes that this expression has several actual or
'residual' meanings which are felt to be more or less closely related.
Providing a mere list of these meanings is scientifically not very
revealing. Rather, their interrelatedness may be explained by pos
tulating a 'basic' meaning from which the residual ones may be de
rived or 'predicted' by means of certain general considerations per
taining to contextual or pragmatic features connected with (the utter
ing of) the sentence 'a knows that he knows that p'. In accordance
with a view held by several philosophers, the basic meaning is taken
to be "a knows that p" simpliciter, and the residual meanings are
seen to arise out of the fact that using the roundabout expression
'a knows that he knows that p' is generally understood as aiming at
specific psychological effects, the exact nature of which may be de
termined only by considering the actual speech situation. These
facts are summed up in the figure on the following page.
This figure is closely similar to that on p,257: the basic mean

ing and the residual meanings correspond to the deep structure and the
surface structures, and the lines which might be thought to connect
the two types of meaning correspond to transformations connecting deep
and surface structures. The principal difference between the two de
scriptions consists in the fact that in the present case we have to do
with meaning whereas in the previous case we had to do with form. That
is to say, in Lakoff's example the (affirmative) meaning "I order you
to come" remains constant and the abstract syntax or form 'Ego imper-
tu tu ven-' is transformed, for instance, into the concrete syntax or
form 'Venias'. By contrast, in Hintikka's example the form 'a knows
that he knows that p' remains constant, and the basic meaning "a knows
that p" is transformed, for instance, into the residual meaning "a is
aware that he knows that p". Consequently the two descriptions, gram
matical and logical, are structurally identical, even if they differ
as to their content.
Hintikka (1969:5) characterises his method of analysis as follows:
A branch of logic, say epistemic logic, is best viewed as an ex
planatory model in terms of which certain aspects of the workings
of our ordinary language can be understood.
If our explanatory model is an appropriate one, and if we have

correctly diagnosed the pragmatic and other extra factors involved
in the different cases, we shall in this way be able to explain
what actually happens on the different ocasions of ordinary use
(op. cit., p.7).129
Thus, in logical analysis of ordinary discourse two or more en

tities are explained when, instead of being described separately,
they are derived from a common source, which means that a generali
sation has been made. By now, the analogy to grammatical explanation
ought to be overwhelmingly clear.
In carrying out his analysis, Hintikka is forced to rely exten
sively on less-than-secure logi co-linguistic intuitions. His discus
sion is interspersed with locutions like "x is not an unnatural thing
to say about y", "whoever utters x is not likely to deny y", "x comes
very close to one of the ways in which y could naturally be taken",
'it is much more unobjectionable to say χ than to say y" (Hintikka
1962:116-17). Given the general unreliability of such intuitions,
Hintikka reaches the conclusion that whether there is a genuine dif
ference in meaning or just a difference in use, can be decided only
by reference to the explanatory model: If the meanings in question
have different roles in the model, i.e., if they are different basic
meanings, then they are genuinely different meanings; if they can be
accounted for or explained in terms of basic meanings, i.e., if they
are different residual meanings, then they merely reflect different
uses of the words or expressions in question. There is an air of
circularity about this strategy, which is identical with the clear
case principle used in deciding the correctness or incorrectness of
intuitively unclear natural-language sentences, and same kinds of
objection apply to both: An explanatory model or a grammar decides
absolutely nothing, unless it is an adequate one; and this question,
i.e., the question of adequacy, cannot be decided formally by refe
rence to the model or to the grammar, but only by finding out whether,
in a great number of cases, the model or the grammar is intuitively
understood as being an adequate representation of unequivocally dif
ferent meanings or unequivocally correct sentences, that is, meanings
and sentences which are known to be what they are intuitively , and
not on the basis of a (circular) reference to the model or to the
grammar. Secondly, it still has to be shown that such absolute di
chotomies as 'correct - incorrect' or 'difference in meaning - dif-
ference in use' reflect, also in non-obvious cases, the nature of the

object of description and are not arbitrary categorisations imposed
by the method of description (cf. 4.2.4., 5.4., and Itkonen 1976a).
Just as 'testability', if understood in a wide enough sense, was
seen in 9.2. to be a necessary characteristic of every type of descrip
tion, so 'explanation', if used synonymously with 'generalisation', is
necessarily a desideratum for every type of description. Consider the
paradigmatic case of sociological explanations viz. Durkheim's (1960)
explanation of suicide. Durkheim noticed that suicide correlated,
inter alia, with being Protestant, unmarried, and city-dwell er. In
stead of simply listing these factors as causes of suicide, however,
he sought for a generalisation, i.e., a common or 'deep' cause from
which these three 'surface' causes along with the effects, i.e., actual
cases of suicide, could be derived. He was able to identify the common
cause as the lack of social cohesion.
Moreover, equating explanation with generalisation applies even
to the description of a particular historical action, for instance the
crossing of the Rubicon by Caesar, which is not meant to be explained
by appealing to general 'laws of history'. This one of Caesar's actions
can be described in a revealing or explanatory way only by subsuming it,
along with several others of his actions, under some unifying purpose
which, coupled with a set of beliefs, can be plausibly attributed to
him and from which the actions in question can be 'derived'. We might
even say that all these particular actions, with their 'residual' or
'surface' intentions, are derived from the one 'basic' or 'deep' inten
tion by means of certain general considerations concerning human ratio
nality and its liability to extraneous influences.
Generalisations are not restricted to phenomena within a single
discipline but may cut across several disciplines. It is plain that
in the reduction of chemistry to physics we have to do with a genera
lisation: two sets of theoretical laws are replaced by a single set,
plus a set of definitions correlating physical and chemical terms.
To give another example, the attempt has been made to reduce deontic
logic to alethic modal logic, or modal logic proper, by adding the
definition and the axiom '-NS' to the two axioms of

the modal system M, viz. and , where 'N'
stands for necessity (Hi 1 pi nen & Føllesdal 1971:19). The definition
says, roughly, that doing p is obligatory if and only if it is neces
sarily the case that not doing ρ implies liability to sanction, or
'S'. '-NS' says in turn that it is not necessary that one is liable
to sanction no matter what. Together with the axioms of the system
M, these two formulae generate the axioms of a standard system of
deontic logic. Once again, we have to do with a generalisation:
two sets of axioms are being replaced by a single set, plus two
formulae, one of which, i.e., the definition, functions as an ana
logue of the definitions encountered in the reduction of chemistry
to physics.
It is perhaps the best proof of the pervasiveness of the striv
ing after generality in science that when I am right now pointing
out similarities between the notions of 'generalisation' as exem
plified in grammar, logic, sociology, history, and natural science,
I am, again, making a (metascientific) generalisation.
11.0. GRAMMAR AND PHILOSOPHY
My thesis is that grammar, logic and (formal) philosophy make
use of one and the same method of conceptual analysis. It is the
purpose of this method to transform atheoretical, intuitive neces
sity into theoretical, formal necessity. Conceptual analysis as
here defined is identifiable as explication in the sense of Pap (1958),
albeit with the qualification that, as I wish to argue, it is not re
stricted to philosophy only. In addition, it is my claim that clas
sical philosophy, as represented by Plato, does not differ methodo
logically from modern formal philosophy.
11.1. The Methodology of Classical Philosophy
We have defined ' e x p l a n a t i o n ' i n grammar and l o g i c as i d e n t i c a l

w i t h ' g e n e r a l i s a t i o n ' , and we have noted t h a t the g e n e r a l i t y of a de-
s c r i p t i o n increases p r o p o r t i o n a l l y to i t s f a l s i t i a b i l i t y as well as
to i t s s c i e n t i f i c i n t e r e s t . These conceptions play a central role
also i n P l a t o ' s philosophy. I t goes w i t h o u t saying t h a t i t cannot
be question here of ernpirical generalisations or of empirical falsi-
fiability.
Let us consider the dialogue Meno. In the beginning, Meno asks
whether v i r t u e can be taught, and Socrates r e p l i e s t h a t , before an-
swering t h i s q u e s t i o n , one must f i r s t know what v i r t u e i s , or give an
analysis of ' v i r t u e ' . Meno's f i r s t t e n t a t i v e d e f i n i t i o n i s as f o l l o w s :
First of all, if it is manly virtue you are afer, it is easy to

see that the virtue of a man consists in managing the city's
affairs capably, and so that he will help his friends and injure
his foes while taking care to come to no harm himself. Or if
you want a woman's virtue, that is easily described. She must
be a good housewife, careful with her stores and obedient to her
husband. Then there is another virtue for a child, male or fe-
GRAMMAR AND PHILOSOPHY 295
male, and another for an old man, free or slave as you like, and
a great many more kinds of virtue, so that no one need be at a
loss to say what it is. For every act and every time of life,
with reference to each separate junction, there is a virtue for
each one of us, and similarly, I should say, a vice (Plato 1963c:
71e-72a).
Socrates does not object to this analysis on account of its being

directly false, but rather on account of its lack of generality. To
use our terminology, we might say that Meno's analysis is falsified not
on predictive, but on explanatory grounds: what it says is true from
the viewpoint of the contemporary Greek society (or at least not clear
ly false), but there are truths which it leaves unsaid, and these are
precisely the most important ones. They concern the 'deep' or 'basic'
virtue which, being common to the 'surface' or 'residual' virtues enu
merated by Meno, would show that these apparently disparate virtues
actually belong together. This is the same concept of 'explanation' as
in grammar (cf. 9.2.) or in logic (cf. 10.3.). In other words, Meno's
analysis is comparable to a school grammar, while Socrates seeks the
equivalent of a generalising, theoretical grammar. His position here
might be summed up by saying that it is false not to generalise.
Socrates demonstrates the prima facie need for generalisation in
the analysis of 'virtue' with the aid of a previous generalisation,
i.e., by making Meno admit that qualities like health, size, and strength
are the same for a man and a woman and the rest, and by generalising
this result to the quality 'virtue' as well. When Meno questions the
justifiability of this generalisation, Socrates demonstrates it more
concretely as follows:
SOCRATES: Well then, didn't you say that a man's virtue lay in
directing the city well, and a woman's in directing her house
hold well?
MENO: Yes.
SOCRATES: And is it possible to direct anything well - city or
household or anything else - if not temperately and justly?
MENO: Certainly not.
SOCRATES: And that means with temperance and justice?
MENO: Of course.
SOCRATES: Then both man and woman need the same qualities, justice
and temperance, if they are going to be good.
MENO: It looks like it.
SOCRATES: And what about your child and old man? Could they be
good if they were incontinent and unjust?
MENO: Of course not.
SOCRATES: They must be temperate and just?
MENO: Yes.
SOCRATES: So everyone is good in the same way, since they become
good by possessing the same qualities.
MENO: So it seems.
SOCRATES: And if they did not share the same virtue, they would
not be good in the same way.
MENO: No (Plato 1963c :73a-c).
At this point Meno offers his second definition, which is, this
time, meant to be a geneval definition of virtue: according to it,
virtue is "simply the capacity to govern men". This analysis is,
again, falsified, namely by pointing out, first, that it does not
apply to slaves and children and, second, that 'capacity to govern'
cannot suffice in itself, but must be reformulated as 'capacity to
govern justly but not otherwise'.
This latter point reveals a further difficulty. Justice is not
the only virtue, but there are also virtues like courage and tempe
rance. Therefore 'justice' cannot be used to define 'virtue' as a
whole. At the same time, a definition based on 'justice' alone is
not general enough since it leaves many types of virtue out of account.
The question remains the same as before: what is the 'deep' or 'basic'
virtue which underlies all these different types of virtue?
At this point Socrates offers an analogy from geometry: there
are many different shapes, but they all undeniably exemplify one and
the same concept 'shape'. Therefore there must be one general defini
tion of 'shape', i.e., a definition which captures the veal meaning,
or the essence , of 'shape'. Socrates proposes, first, to define 'shape'
as "the only thing which always accompanies colour". After Meno has
objected that 'colour' is no more primitive that 'shape' and hence may
not be used to define it, Socrates gives another definition: "Shape is
that in which a solid terminates, or more briefly, it is the limit of
a solid" (op. cit. 76a). - This analogy shows clearly that for Plato
analysis of concepts of whatever type is methodologically one and the
same undertaking. It may be added that Aristotle remained equally con-
vinced that, far from being just 'conventions' of some type, definitions,
disregarding all that is variable or accidental, express the general and
130
unchanging essence of things.
Next, Meno tries to analyse the concept of virtue along the lines
of the geometrical example. His third definition is as follows: virtue
means "desiring good things and being able to acquire them".
To begin with, Socrates falsifies this analysis by showing that its
first part, i.e., 'desiring good things', is empty, because no one de-
sires things which he thinks are evil.131 So Meno modifies his defini
tion: now virtue means simply "the power of acquiring good things".
This definition is again refuted by pointing out that the manner of ac
quiring good things like health, wealth, and high offices in the state
is all-important here. If good things are not acquired justly and
righteously, then the power of acquiring them is not virtue, but vice.
The definition is modified accordingly, but this modification is reject
ed, again, by pointing out that 'justice' is here used to define 'vir
tue', although it was agreed previously that the former is only one type
of the latter and therefore cannot be meaningfully used to define it.
At this point Meno gives up. His attempt to give a maximally gen
eral definition has each time led to falsification. It may be added
that the whole dialogue ends inconclusively. Instead of offering a
full-fledged definition of virtue, Socrates contents himself with
stating some of its general characteristics. On the one hand, it is
a form of knowledge: "good men cannot be good by nature". On the other
hand, and in at least apparent contradiction to the foregoing result,
the initial question whether virtue can be taught must be answered in
the negative, for instance on the common-sense evidence that the most
virtuous men of Greece often had rather unvirtuoussons.
As I noted on p. 144, Wittgenstein is opposed to the search of
basic or real meanings; he wishes to replace the latter by 'family
resemblances' between meanings. Wittgenstein's position constitutes a
meaning theory of its own, and its applicability must be determined
in different contexts separately. For instance, it may well be that
there is no single definition which could capture 'virtue' in its en-
tirety. By contrast, it would be simply false to say that there can be
no such definitions of geometrical concepts either.
I already have discussed Plato's conception of knowledge as recol
lection and have connected it with his conception of genuine knowledge
as agent's knowledge (cf. 8.1.). Here I wish to examine Plato's former
conception in relation to the task of conceptual analysis and, in par
ticular, to what is nowadays known as the 'paradox of analysis'. I also
briefly suggest my own solution to this 'paradox'.
Meno discovers an apparent paradox in that Socrates professes not
to know what virtue is, but is nevertheless willing to search for an
analysis of it. Socrates identifies Meno's argument as the one which
claims that "a man cannot try to discover either what he knows or what
he does not know. He would not seek what he knows, for since he knows
it there is no need of the inquiry, nor what he does not know, for in
that case he does not even know what he is to look for" (Plato 1963c:
80e).
The paradox of analysis, in turn, is as follows: On the one hand,
if the concept to be analysed, i.e., analysandum, and its proposed ana
lysis, i.e., analysans, are (semantically) different, the analysis has
failed. On the other hand, if the two are identical, the analysis is
trivial. Let A and Β stand for analysandum and analysans, respective
ly. If Α ‡ Β, the analysis has failed. If A=B, then (A=B)≡(A=A),
which means that the analysis is trivial.
The two paradoxes are clearly akin to each other in that they
both deal with the relation between analysandum and analysans, even if
the former paradox emphasises the question of the knowledge about the
two whereas the latter emphasises the question of the identity between
the two. A coherent account of the nature of conceptual analysis must
provide an answer to both of these questions simultaneously.
Plato illustrates his conception of knowledge as recollection by
letting Socrates elicit from one of Meno's slaves the true claim that
the side of an eight-foot square is the diagonal of a four-foot square.
Since Socrates does not teach him this truth, at least not by directly
telling it to him, it must be the case that in some sense he knew it
already before, even if he did not know that he knew it, and that he
has now merely become aware of his previous knowledge:
SOCRATES: What do you think, Meno? Has he answered with any opi
nions that were not his own?
MENO: No, they were all his.
SOCRATES: Yet he did not know, as we agreed a few minutes ago.
MENO: True.
SOCRATES But these opinions were somewhere in him, were they not?
MENO: Yes.
SOCRATES: So a man who does not know has in himself true opinions
on a subject without having knowledge.
MENO: It would appear so.
SOCRATES: At present these opinions, being newly aroused, have a
dreamlike quality. But if the same questions are put to him on
many occasions and in different ways, you can see that in the
end he will have knowledge on the subject as accurate as anybody's.
MENO: Probably.
SOCRATES: This knowledge will not come from teaching but from
questioning. He will recover it for himself.
MENO: Yes.
SOCRATES: And the spontaneous recovery of knowledge that is in him
is recollection isn't it?
MENO: Yes (Plato 1963c :86b-d).
So far, Plato's account can be accepted, but in the sequel he goes

definitely astray. Plato namely assumes that there are only two pos
sible ways in which someone's knowledge (of geometry or of virtue) can
be explained: either he has acquired it, and for Plato knowledge can
be acquired only through teaching; or he has always possessed it. Now,
since the slave in question knows geometry although no one has ever
taught it to him, it follows that he must have known it already before
he came to be what he is now, i.e., that "his soul has been forever in
the state of knowledge".
I t may be added t h a t the above-mentioned argument has no cogency
f o r A r i s t o t l e , because he does not accept P l a t o ' s r e s t r i c t i v e view of
the nature of the a c q u i s i t i o n of knowledge:
All instruction given or received by argument proceeds from pre

existent knowledge. ... The pre-existent knowledge required is
of two kinds. In some cases admission of a fact must be assumed,
in others comprehension of the meaning of the term used, and some
times both assumptions are essential. ... I imagine there is no
thing to prevent a man in one sense knowing what he is learning,
in another not knowing it. The strange thing would be, not if in
some sense he knew what he was learning, but if he were to know
it in that precise sense and manner in which he was learning it
(Aristotle 194lb:71a-l, 71a-i0, 71b-5).
In other words, there is no mystery in trying to make implicit
knowledge explicit. More precisely, one must, first, become aware of
that intuitive, atheoretical knowledge which one already has. This
is, for instance, knowledge about different ways of being virtuous or
about rules of a language L. Secondly, one must try to give a general,
theoretical analysis of the atheoretical knowledge once it has been
brought onto the level of consciousness, i.e., one has to provide a
definition of 'virtue' or write a grammar of L. This theoretical ana
lysis produces, in turn, knowledge which one never had before.
This last point shows the connection with the paradox of analysis:
If the theoretical, analysis, i.e., analysans, is something which we did
not know before, how can it be identical with the analysandum? And if,
on the other hand, the former is not identical with the latter, what
justification is there for calling it the analysans of precisely this
analysandum, and not of something else?
In practice the solution consists in relaxing or 'liberalising'
the condition of strict identity between analysandum and analysans.
All that is required is a sufficient degree of similarity between the
two (cf. Pap 1958, chap. 10). This solution is adopted also in what
follows (cf. 11.2.-3.). However, it seems to me that the philosophical
problem, i.e., the one concerning the simultaneous identity and dif
ference between analysandum and analysans, remains untouched by our
decision to regulate the selection of analysantia in such and such
ways. Moreover, I do not think that this problem can be solved at
all in that formal and static frame of reference which is generally
characteristic of analytic philosophy. What needs to be seen, is that
analysandum and analysans are not just concepts or expressions which
are being compared with each other; rather, they represent different
stages of a process. Analysandum represents a body of knowledge in
its prescientific, atheoretical state. Analysans represents a dif
ferent state of the same body of knowledge, viz. its scientific or
theoretical "tate. This explains why the two are simultaneously
identical and different; the relation between them is a conceptual
or necessary one. but it is not logical equivalence in the sense of
formal l o g i c . Rather, t h i s r e l a t i o n can be adequately characterised
only i n terms of Hegelian or dialectical logic. Hegel sums up his
t h i n k i n g by saying t h a t i t deals w i t h "das. werdende Wissen", know-
ledge i n the process of becoming (Hegel 1975:593). This aspect is
badly neglected i n a n a l y t i c philosophy, which is s t i l l today the pre-
v a i l i n g trend i n Anglo-Saxon philosophy.
I t may seem paradoxical t h a t the a n a l y s i s , while leading from
imperfect knowledge to more p e r f e c t knowledge, also leads from cer-
t a i n t y to u n c e r t a i n t y ( c f . p. 1 4 4 ) . However, the a t h e o r e t i c a l cer-
t a i n t y is about disparate and very concrete (normative) f a c t s , where-
as the t h e o r e t i c a l u n c e r t a i n t y is about t h e i r systematic totality.
As I noted b e f o r e , analysans, or theory, produces genuinely new know-
ledge. But since i t i s , at the same t i m e , only a new stage of the
old knowledge, i t is perhaps understandable t h a t Plato was led to
claim t h a t analysis t o o , and not j u s t i t s o b j e c t , is remembered.
Taken l i t e r a l l y , t h i s is obviously i n c o r r e c t , but i t i s s t i l l the
case t h a t atheory, or the set of r u l e s , shades off i n t o t h e o r y , or
the system of rules ( c f . p. 126).
In the next section the terms 'analysandum' and 'analysans' will
be replaced, r e s p e c t i v e l y , by the e s s e n t i a l l y synonymous terms 'ex-
plicandum' and ' e x p l i c a t u m ' .
11.2. The Concept of 'Explication'

I s h a l l make use of t h a t concept of ' e x p l i c a t i o n ' which is deve-
loped i n Pap (1958). In the course of e x p l i c a t i o n an i n t u i t i v e l y
known concept or conceptual system, i . e . , 'explicandum', which is r e -
f e r r e d to by the corresponding explicandum-expression(s), is replaced
by i t s redefined or reconstructed form, i . e . , ' e x p l i c a t u m ' , which again
is r e f e r r e d to by the corresponding e x p l i c a t u m - e x p r e s s i o n ( s ) . Expli -
candum-expressions belong mostly to ordinary language, whereas e x p l i -
catum- expressions are mostly p a r t of some t h e o r e t i c a l , formal language.
Often i t is needless to d i s t i n g u i s h between explicanda or e x p l i c a t a and
the corresponding expressions. An explicandum i s i d e n t i f i e d w i t h the
aid of s o - c a l l e d c r i t e r i a of adequacy, i . e . , sentences which are intui-
tively known to be necessarily true and in which the explicandum(-ex-
pression) occurs essentially. To give an example, if we explicate
the concept of knowledge, we may start identifying our explicandum
with the aid of the following sentence which is intuitively known
to be necessarily true: "If a knows that p , then ρ is true." (This
is precisely why to know is a facti ve verb; cf. 9.5.). The actual
process of explication consists in transforming these necessary
truths of the merely intuitive kind into necessary truths of the
formal or analytic kind. In the sentences functioning as criteria
of adequacy, this is achieved by replacing the explicandum, which
mostly has no, or very little, inner structure, by an appropriate
explicatum which has an explicit, articulated inner structure. Fre
quently it is also necessary to change the whole sentence-structure
of the initial criterion of adequacy. This means de facto that the
initial, non-explicated criterion of adequacy will be translated
into its explicated counterpart. If this new sentence which results
from replacing the explicandum by the explicatum (and from changing,
in the process, the structure of the original sentence) is formally
true, the explicatum is said to satisfy the criterion of adequacy
in question. Since the criterion of adequacy was originally used
to identify the explicandum, and since it is now satisfied by the
explicatum, it follows that the explicandum and the explicatum are
similar. Their similarity is guaranteed, above all, by the requi
rement that the explicatum must satisfy several criteria of adequacy
identifying the explicandum. Explication would of course lose its
point, if explicata could uncontrollably differ from explicanda.
Explication is an attempt to formalise intuition, more precisely,
that which is intuitively known (cf. 5.2.). Not only in logic (cf. p.
280 ) , but more generally in each type of conceptual analysis intui
tion is primary with respect to formalisation:
... it must be conceded that it is intuitive perception of ne

cessity of propositions which guides the selection of the mate
rial criteria of adequacy for a given explication, and that if
this is denied, explication appears either as circular or as
philosophically irrelevant (Pap 1958:416).
Yet, since it is only by means of modal judgements that we can

clarify concepts - e.g. 'is it logically possible that a man

should be wholly devoid of reasoning ability' we ask, in order
to make clear to ourselves the meaning of 'man' - faith in mu
tual understanding of basic modal terms is indeed an indis
pensable presupposition of all analytic philosophy. To him
who does not grasp the sense of 'possible' in which the exis
tence of immortal men is possible yet the existence of round
squares not possible, no analytic philosophy can be taught. ...
The distrust of the 'intuitional' basis of analytic philosophy
is rooted in nothing less than an imperfect understanding of
scientific method - in the broad sense of 'scientific' in which
analytic philosophy can be scientific (op. cit., p. 419).
For illustration. I present here one possible explication of the

meaning of "a knows that p" (cf. Pap 1958:295):
Explicandum: a knows that p (= 'q')

Criterion of adequacy: if a knows that ρ, then ρ is true (=
)
Explicatum: a has good reasons to believe that p, and a believes
that p, and ρ is true (= 'r&s&t')
Explication: a knows that p = def. a has good reasons to believe
that p, and a believes that p, and ρ is true (= 'q = r&s&t')
Criterion of adequacy after explication: if a has good reasons to
believe that p, and a belives that p, and ρ is true, then ρ is
true (= )
I f we take the c r i t e r i o n of adequacy as our s t a r t i n g p o i n t , the

same e x p l i c a t i o n can be presented as f o l l o w s :
Explication:
We thus e s t a b l i s h a d e f i n i t i o n a l equivalence between the c r i t e r i o n

of adequacy and i t s new, analysed or e x p l i c a t e d form. stands
here f o r a sentence which is merely i n t u i t i v e l y known to be necessarily
t r u e , whereas stands f o r (and i s ) a f o r m a l l y necessary t r u t h .
N o t i c e , on the other hand, t h a t the d e f i n i t i o n a l equivalence, whether
i t obtains between the two versions of the c r i t e r i o n of adequacy or be
tween the explicandum ' q ' and the explicatum 'r&s&t' is of course not
a f o r m a l l y necessary t r u t h , i . e . , a logical equivalence. This is so
already f o r the obvious reason t h a t the explicandum is normally an un-
analysed expression(-complex) of ordinary language. Moreover, remember
t h a t explicandum and explicatum represent d i f f e r e n t stages of the pro-
cess of a n a l y s i s : the l a t t e r i s a developed form o f , or an improve
ment upon, the former. I t would be c l e a r l y c o u n t e r - i n t u i t i v e to
claim t h a t the s t a r t i n g p o i n t and the end p o i n t of a process of
development ought to be l o g i c a l l y e q u i v a l e n t . On the other hand,
there is c l e a r l y a conceptual or necessary connection here between
explicandum and explicatum, p r e c i s e l y because they p e r t a i n , though
i n d i f f e r e n t ways, to one and the same concept. And, as in a l l good
d e f i n i t i o n s , t h i s connection is i n t u i t i v e l y acceptable. Yet, to r e
peat, i t cannot be i n t e r p r e t e d as a l o g i c a l equivalence.
Our example is q u i t e simple because i n i t the explicatum i s j u s t
the conjunction of those sentences which the explicandum e n t a i l s as
i t occurs in d i f f e r e n t c r i t e r i a of adequacy. (For i n s t a n c e , there
must be a c r i t e r i o n of adequacy l i k e the f o l l o w i n g : " I f a knows t h a t
p , then a (also) believes t h a t p " . ) Therefore i t is only t r i v i a l that
t h i s type of explicatum should turn c r i t e r i a of adequacy i n t o f o r m a l l y
necessary t r u t h s . The same method i s applied in a l l standard examples
of conceptual a n a l y s i s . For i n s t a n c e , 'male parent' as the explicatum
of the explicandum ' f a t h e r ' i s obtained by c o n j o i n i n g the consequents
of the two i n t u i t i v e l y necessary i m p l i c a t i o n s " I f α is a f a t h e r , a is
male" and " I f a i s a f a t h e r , a is a p a r e n t " . Formal l o g i c o f f e r s an
example of a more complex e x p l i c a t i o n : in i t a whole idiom which ad
mits o n l y , or mainly, of i n t u i t i v e necessity is t r a n s l a t e d i n t o a d i f
f e r e n t idiom which admits only of formal necessity. For i n s t a n c e , or
dinary language sentences about norms, which are only i n t u i t i v e l y ne
cessary, are turned i n t o formal t r u t h s by t r a n s l a t i n g them i n t o the
language of deontic l o g i c . Thus when, f o r i n s t a n c e , the sentence "If
one ought to do p and is not forbidden to do q , then one is permitted
to do both ρ and q" i s t r a n s l a t e d i n t o the language of von W r i g h t ' s
'Old System', i t becomes , which is q u i t e obviously
v a l i d w i t h i n t h i s system. Each i n t u i t i v e l y necessary deontic sentence
c o n s t i t u t e s a c r i t e r i o n of adequacy f o r corresponding l o g i c a l explica-
tion.
E x p l i c a t i o n r e s u l t s i n proposing an explicatum, which is tested
in the usual explanatory-cum-predictive way, namely by f i n d i n g out
whether it satisfies all, and only, those criteria of adequacy which
it is meant to satisfy. To the extent that it fails to do so, it is
(nonempirically) falsified. Of two equally complex explicata the one
which satisfies a greater number of criteria of adequacy is more gen
eral, and hence more explanatory.
Philosophy is a prescriptive science: it tries to make us think
better. Therefore it is possible that the explicatum will have re
percussions on the way in which the explicandum(-expression) is used
(cf. Hanna 1968:42-43). However, it is not the purpose of explication
to change our basic intuitions, but rather to develop them (cf. p. 44
above). It might well be said that explications 'educate' our in
tuitions in the sense that several aspects of the knowledge which we
have possessed all along become transparent to us only in an expli
citly formulated framework. Similarly, it is only in such a frame
work that several questions about our own intuitions can be intelli
gibly asked.
In general, it is not possible to state in any straightforward
way all those necessary sentences in which a given concept occurs es
sentially, and which could be used to identify this concept. Conse
quently, every explicandum offers in principle the basis for an in
definite number of possible criteria of adequacy. The selection be
tween these is to a certain degree a matter of decision, and depends
largely on the specific purpose of the explication in question. The
material implication as an explicatum of the conditionals of natural
language is a good example of an explicatum which can be accepted for
the purpose at hand, although it neglects most of the possible crite
ria of adequacy. The one criterion of adequacy for the explication
of conditionals which is crucial to scientific thinking, and which
is also satisfied by the material implication, is that, if the ante
cedent is true, the implication as a whole is true only in case the
consequent is true. To give another example, the different explica
tions of the concept of truth, as represented by 'correspondence
theory', 'coherence theory', and 'consensus theory', reflect differ
ences of opinion and of taste concerning the relevance vs. irrele-
vance of d i f f e r e n t possible c r i t e r i a of adequacy, because i t i s q u i t e
clear t h a t i ) a true sentence must in some sense correspond to r e a l i t y ,
ii) a true sentence must be compatible w i t h other true sentences (for-
mulated w i t h i n the same frame of r e f e r e n c e ) , and i i i ) the notion of
t r u t h persupposes t h a t people are able to agree upon what is or i s not
true.
On the other hand, because an explicandum is i d e n t i f i e d on the
basis of c r i t e r i a of adequacy, s e l e c t i n g a d i f f e r e n t set of criteria
f a c t u a l l y amounts to s e l e c t i n g a d i f f e r e n t explicandum. I t may also
happen t h a t analysts do not agree as to whether a given, c r i t e r i o n of
adequacy r e a l l y d e f i n e s , or i d e n t i f i e s , the explicandum under study.
Most often disagreements of t h i s kind are merely verbal i n s o f a r as
the disputants turn out to have d i f f e r e n t explicanda i n mind ( c f . Pap
1958:384-86). However, i n more subtle cases there is a p o s s i b i l i t y
of genuine disagreement ( c f . p. 286).
I have been claiming here t h a t conceptual analysis as p r a c t i s e d
by c l a s s i c a l philosophy and e x p l i c a t i o n are r e a l l y one and the same
thing. (For i n s t a n c e , the analysis " f a t h e r = male parent" is an exam-
ple of P l a t o ' s and A r i s t o t l e ' s 'real d e f i n i t i o n s ' . ) Quine (1960:257-
62) is s t r o n g l y opposed to t h i s view. In his opinion i t is not the
purpose of e x p l i c a t i o n to discover hidden meaning or to make e x p l i c i t
what we have known a l l along. His paradigmatic example of philosophica
analysis is Wiener's e x p l i c a t i o n of the ordered p a i r ( x , y) as {{x},
{y, ø}}, i.e., the class of those two classes whose sole members a r e ,
on the one hand, χ and, on the o t h e r , y and the empty c l a s s . It is
the m e r i t of t h i s e x p l i c a t i o n to reduce ordered pairs to o b j e c t s . As
another example, Quine mentions the e x p l i c a t i o n of natural-language
c o n d i t i o n a l s as material i m p l i c a t i o n s ( c f . above). What makes these
examples paradigmatic, i s t h a t they concentrate upon one s i n g l e aspect
of a d i f f i c u l t concept and show t h a t , f o r the purpose at hand, t h i s is
sufficient.
Quine's p o s i t i o n is open to two main o b j e c t i o n s . F i r s t , the con
t r a s t between the t r a d i t i o n a l analysis and the Wiener-type analysis is
f i c t i t i o u s ; i t depends merely on the number of those c r i t e r i a of ade-
quacy which one chooses as r e l e v a n t to the problem at hand. Wiener's

problem was such t h a t s a t i s t y i n g one ( c r u c i a l ) c r i t e r i o n was s u f f i c i e n t .
I t i s s e l f - e v i d e n t t h a t a d i f f e r e n t problem would have required a d i f -
ferent solution. Second, philosophical problems are characteristically
those whose s o l u t i o n requires the s a t i s f y i n g of several c r i t e r i a of
adequacy. In t h i s respect they d i f f e r from mathematical problems, i n -
cluding Wiener's problem: Quine e x p l i c i t l y admits t h a t there i s only
one s i n g l e c r i t e r i o n of adequacy here, and t h a t i t is imposed by "uses
in mathematics". Consequently, the ordered p a i r seems to be a r a t h e r
bad example of philosophical a n a l y s i s . Quine's other example is j u s t
as i l l - c h o s e n . I t goes w i t h o u t saying t h a t material i m p l i c a t i o n is
t o t a l l y unable to express the notions of natural necessity and l o g i c a l
e n t a i l m e n t , which are expressed by natural-language c o n d i t i o n a l s , and
of which the former, in p a r t i c u l a r , c o n s t i t u t e s a genuinely p h i l o s o -
phical problem. I t is only understandable, t h e n , that attempts at
more adequate e x p l i c a t i o n of c o n d i t i o n a l s are continuously made ( c f . ,
e.g., Sosa 1975) .
E x p l i c a t i o n ranges from simple analyses of s i n g l e concepts to en-
t i r e systems of l o g i c . In the former case formal necessity i s achieved
by r e p l a c i n g explicanda by e x p l i c a t a w i t h i n f i x e d s e n t e n c e - s t r u c t u r e s ,
whereas i n the l a t t e r case i t is achieved by t r a n s l a t i n g sentences of
an explicandum-language i n t o those of an explicatum-language. By way
of a summary, I quote Pap's (1958:388) s u c c i n t c h a r a c t e r i s a t i o n of the
method of e x p l i c a t i o n :
The conception of analysis outlined so far involves the steps (1)

intuitive perception of the necessity of propositions by means
of which the explicandum is, though not uniquely, identified (of
course, such 'perception' is inseparable from an understanding
o f words, especially in view of the inseparability of abstract
thinking from verbal images); (2) construction of an explicatum;
(3) test of adequacy of the explicatum, consisting in formal de
monstration of the pre-analytic entailments (criteria of adequacy)
by substituting explicatum for explicandum.
11.3. Grammars as Instances of Explication

Since grammatical descriptions are conceptual analyses, they must
be identifiable, more narrowly, as explications. This is plausible
already in view of the fact that grammatical analysis and philosophical
analysis practically coincide in one area, viz. the analysis of mean
ing. The foregoing analysis of "a knows that ρ", which was taken from
Pap (1958), could just as well be presented in a grammatical context.
Other examples can be freely added. It is a well-known fact that kin
ship terminology was one of the first areas where linguistic semantics
was systematically developed (for an overview, see d'Andrade 1970).
Compare this fact with the following, definitely philosophical state
ment:
The proposition, e.g., that all concepts of kinship relations that

happen to be meanings of predicates of the English language are
definable in terms of just the concepts male, female, and pavent
is knowable a priori , by reflecting on concepts. So is the pro
position that the primitive concepts of arithmetic are definable
in terms of logical constants alone; ... (Pap 1958:274).
It still remains to be shown, point by point, how the different

components of grammatical analysis correspond to the components of ex
plication set forth in 11.2. In grammar the explicandum is ultimately
the language to be described. More precisely, we describe a language
L by analysing the concept 'correct sentence in L'. Therefore 'langu
age L' is really a short-hand for 'all the rules which are relevant
to the correctness of sentences in L', and for instance the sentence
"In English the definite article precedes the noun" should read "So
far as the correctness of sentences in English is concerned, the de
finite article precedes the noun". Notice that in the expression
'correct sentence in English' the constituents 'correct sentence' and
'English' are so-called incomplete symbols: neither of them can be ana
lysed in itself, but their conjunction must be analysed as a whole.
The explicandum of a linguistic explication is identified with the
aid of an indefinite set of rule-sentences, which could be rendered
more formally as universal implications. As was shown in 6.1. (above),
rule-sentences are necessarily true (or false); their necessity is of
the non-formal, merely intuitive kind. Consequently, they function as
criteria of adequacy. Now, if the explicandum is the language L, then
it is obvious that the explicatum must be the grammar of L, i.e., G. .
In other words, GL is a theoretical analysis of L. It is clear, then,

that grammatical explication consists in replacing the atheoretical
reference to L by a theoretical reference to GL .
Because GL is an extended axiomatic system, rule-sentences, which
are either true or false, cannot be directly reformulated in, or trans
lated into, the language of GL . That is, GL shows what rule-sentences
say in an atheoretical way. It is the language of the metagrammar, or
of the theory of GL , which reformulates the rule-sentences: sentences
in this language are the explicated counterparts of rule-sentences,
i.e., the former say in a theoretical way the same thing which the
latter say in an atheoretical way.
If we concentrate here only on the 'weak generative capacity' of
GL, or the requirement that GL generate all and only correct sentences
of L, then, according to what has just been said, grammatical explica
tion consists in a systematic translation of sentences about 'correct
in L' into sentences about 'generable by G, '. The former sentences,
i.e., rule-sentences, are necessarily true (or false) in a merely in
tuitive way. If GL , or more precisely its metagrammatical , natural-
language interpretation, is to be an explicatum in Pap's sense, then
the latter sentences must be true (or false) in a formally necessary
way. This is indeed the case. It is a formally necessary truth that
an axiomatic system, whether extended or not, generates what it gen
erates. That is, a sentence like "An application of the rule 'S→aSB'
followed by an application of the rule 'S→ab' generates the string
aabb" is formally true, or analytical, in precisely the same way as
the sentence "'q' is derived from 'p=q' and 'p' by Modus Ponens".
More generally, the same is true of all sentences about the genera
tive capacity, both weak and strong, of natural language grammars. -
I repeat that there is no formally necessary truth within GL , but only
within the metagrammar which speaks about what GL generates or does
not generate.
If all true rule-sentences, which are intuitively necessary sen
tences about a language L, can be transformed in the way here discus
sed, and salva veritate , into formally necessary sentences about the
generative capacity of the grammar of L, and vice versa, then the
grammar satisfies its criteria of adequacy and is, consequently,
adequate. On the other hand, if there are true (atheoretical) sen
tences about L which cannot be transformed salva veritate into
corresponding metatheoretical, or metagrammatical, sentences, this
means that the grammar fails to generate something which it ought
to generate. Conversely, if there are true metatheoretical sen
tences about the forms generated, and the structures assigned, by
the grammar which have no counterparts among true atheoretical sen
tences, this means that the grammar generates something which it
ought not to generate. These are the two principal ways, explana
tory and predictive, of falsifying any proposed grammar, reformu
lated in terms of the present 'explicatory' framework.
Above, I have demonstrated the formal or structural identity
between explication and grammatical analysis. The terminology of
'explication' provides merely a reformulation of the general ratio
nale of writing generative grammars: Systematic translation of sen
tences about 'correct in L' (and other sentences about L) into sen
tences about 'generable by GL ' means the same thing as writing a
generative grammar of L. The only difference is that the former
formulation reveals, via the concept of explication, the connection
with logic and philosophy. The notion of explication provides, then,
the common denominator of all normative sciences, viz. grammar, logic,
and philosophy.
CONCLUSION
In this book I have analysed the nature of the science of gram
mar. After presenting some methodological (1.0.-2.0.) and historical
(3.0.) background, I set forth a theory of language (4.0.-7.0.) and a
theory of grammar (8.0.-9.0). The end result of my analysis is that,
as a science investigating (the knowledge of) rules, the science of
grammar is not an empirical, but a normative science comparable to
logic and philosophy (10.0.-11.0 . ). The normative sciences are cha
racterised by their use of the method of explication (11.0.).
By contrast, psycholinguistics and sociolinguisties are empirical
sciences. As sciences investigating human, and hence meaningful, be
haviour, however, they are qualitatively different from the natural
sciences. In linguistics there can be no question of directly adopt
ing the methods of natural science. Psycholinguistics and sociolin-
guisties constitute what might be called 'space-time linguistics'.
They are akin, respectively, to (experimental) psychology and (empi
rical) sociology, just like the science of grammar, which might be
called 'norm linguistics', is akin to logic and philosophy. It is in
my view the central task of theoretical linguistics, or general l i n
guistics , to work out the integration of norm linguistics and space-
time linguistics.
The difference between empirical science and normative science is
relative to the level of abstraction. At a sufficient high level of
abstraction, all sciences, including metascience, are seen to make use
of the same methods of 'explanation' and 'testing'.
The philosophy of science which my analysis requires is not posi
tivism, but hermeneutics (interpreted perhaps in a somewhat unorthodox
way). I t has proved to be necessary to r e j e c t the p o s i t i v i s t i c space-
time ontology and to adopt instead a continuum-like ontology extending
from physical space-time to consciousness and n o r m a t i v i t y . Within t h i s
framework a l l sciences, i n c l u d i n g l o g i c and philosophy, can be brought
in a rather natural way i n t o r e l a t i o n w i t h one another. The t r a n s i t i o n
from empirical social science to normative science i s provided by the
science which i n v e s t i g a t e s (spontaneous) conceptual change, or change
of r u l e s . There must be a s i m i l a r t r a n s i t i o n between physical science
and empirical social science, provided by an as y e t nonexistent science
i n v e s t i g a t i n g the emergence of meaning.
Chap. 1. THE IDEA OF 'POSITIVSM'
Notes 1-9
1
The classic presentation of this topic is Husserl (1954) . Von Wright
(1971b) and Mittel strass (1974b) characterise, from somewhat different
angles, the difference between the Aristotelian and the Galilean con
ceptions of science; cf. also here 9.3.
2
There are well-known difficulties with the notion of DN-model (cf.,
for example, Stegmüller 1974, chaps. 1-2 and 10). Yet no clearly
superior models of deterministic explanation seem to be forthcoming.
3
This claim is not uncontested, however. For instance, Radmtzky
(1970:i, p.169) objects to the use of 'explanation' according to which
"whenever a person has rejoined the question why something behaved in
a certain way with the comment that all entities of that kind behave
in this way, he had been producing an explanation sketch"; similarly
Harre (1970:16 and 20) .
4
'p' and 'q' stand for any sentences. 'F' and 'G' stand for any pro
perties or relations. 'x' and 'y' stand for any individuals. The
junctors '&', 'v', , '≡' , and '-' stand, respectively, for con
junction (= 'and'), inclusive disjunction (= 'and/or'), implication
(= 'if - then'), equivalence (= 'if, and only if, - then'), and ne
gation (= 'not'). The quantifiers '(x)' and '(Ex)' stand, respecti
vely, for universality (= 'all') and existence (= 'some'). For in
stance, the formula '(x) reads: "For all individuals x, if χ
has the property F, then it has the property G." The rule of univer
sal instantiation permits to infer any (for instance, 'Fa') from all
(i.e., '(x)Fx'). The rule of Modus Ponens permits to infer 'q' from
and 'p'. The rule of Modus Tollens permits to infer '-p' from
and '-q'. Cf. 2.6. below. - I adopt the convention according to
which ρ and Fa stand for the referents of the sentences 'p' and 'Fa'.
5
Cf., e.g., Hempel (1965:39-40), and Popper (1965:41). It is well
known that Popper tries to dispense with the notion of confirmation
by evaluating theories on the basis of their liability to falsifi
cation. However, it is difficult not to see the similarity between
confirmation and Popper's notion of 'corroboration' which is defined
as the ability of a theory to stand up to tests (cf. Popper 1965:265-
69).
. . ....
6
Peirce and Popper, for instance, have emphasised the falsifIability
as the defining property of scientific theories. However, this 'fal-
libilistic' principle holds true not just of empirical science, but
of any type of science, including 'pure' philosophy, cf. Moore (1912:
8-9): "The probability is, that hardly any positive proposition, which
can as yet be offered in answer to [fundamental ethical questions]
will be strictly and absolutely true. With regard to negative propo
sitions, indeed - propositions to the effect that certain positive
answers which have been offered, are false - the case seems to be
different. We are, I think, justified in being much more certain
that some of the positive suggestions which have been made are not
true, than that any particular among them is true,.."; cf. also
here 9.3. - If, in our example, the prediction would be of the form
'(Ey)Gay' (i.e., there is some individual y to which a has the rela
tion G), straightforward falsification of the corresponding theory
would not be possible. Therefore, to achieve greater generality,
'falsiflability' must be replaced by 'testability' (cf. 1.1.). I
wish to emphasise that, for instance, Lakatos' (1970) criticism of
'dogmatic falsificationism' does nothing to undermine the fact that
(intersubjective) testability is the hallmark of scientific rationa
lity, or the only thing that distinguishes science (whether empiri
cal or not) from poetry, or from magic.
7
Just think of a theory of mechanics pertaining to the behaviour of
round falling objects weighing less than 1 kg. Such a theory would
be uninteresting. That is, even if all sentences generated by it are
true, which means that it generates only true sentences, it still does
not generate all sentences (about mechanics) that are true.
8
Perhaps the most serious shortcoming of the 'standard' conception of
scientific theories has been its unwillingness to consider the con
struction of causal models, i.e., models for those actual mechanisms
that make things happen in the way they do happen (cf. Harre 1970).
9
The quotes around 'simply' are meant to indicate that this is in f act
no simple question at all. The problem is, to put it bluntly, that
we must have a way of discarding spurious or inconsistent thoughts.
Chap. 2. THE IDEA OF 'HERMENEUTICS'

Notes 10-19
10
The so-called 'psychology of personal constructs' has been purpose
fully developed in opposition to positivistic psychology. It takes
account of the reflexive structure of consciousness, both in psycho
logists and in their test persons, as well as of the de facto social
character of all psychological experiments (see Bannister & Fransella
1971) . This last point is also emphasised in the critique of positi
vistic psychology offered in Harre & Secord (1972, chaps. 2-3).
NOTES 315
When, due to the lack of time and resources, no proper psychoana

lytic or psychotherapeutic treatment can be offered, the same result
is striven for through the administration of suitable drugs. This
amounts to the attempt to obstruct one form of causality through an
other. Characteristically, the treatment of the latter type produces
poorer results.
12
Notice that this claim is fully compatible with the obvious fact that
different groups of people follow different sets of rules at differ
ent times and different places; cf. n.88.
13
Douglas (1973) offers an original selection of readings from the
field of the sociology of knowledge. Because of its generally non-
statistical character, 'ethnomethodology' is closer to sociology of
knowledge than to empirical sociology (cf. Turner 1974), The same
seems to be true of the Hymes-type 'ethnography of speaking'.
14
Cf. Lessnoff (1974:44-45): "It scarcely needs to be said that, for
an institution to function, those involved must know what the rules
are, and must expect one another to keep more or less to the rules.
Language is not only involved in all social institutions, it is it
self a social institution..."
Cf. Ryan (1970:145): "The concepts which individuals possess are
simply the mirror-image of the rules which shape their lives in their
society. ... Thus when we elucidate concepts we are elucidating the
possibilities of social life, and conversely when we explain social
life we elucidate the concepts available to members of that society.
When we began our account of philosophical argument we described it
as essentially a conceptual study. Now we see that the social sci
ences are permeated by conceptual considerations; ... it is the task
of social science to reflect on the concepts with which we make so
cial life intelligible... But this activity is much like that of phi
losophy through the' centuries..." - This important passage, which
refers to Winch (1958), elucidates the sense in which sociology of
knowledge, philosophy, logic, and grammar are, in my opinion, all
cases of conceptual analysis.
16
What I am saying here is intimately connected with the distinction
between rule and regularity (cf. 6.1.). Moreover, experimentation
and observation are in order when the existence of rules is less than
certain (cf. 5.3-4.).
17
This striving after the consistency of one's psyche is the basic ex
planatory principle employed by the so-called theory of cognitive
dissonance (or consistency); cf. Abelson et. al. (1968)
18
This does not mean, however, that formal logic in its present form
would be psychologically real. On the contrary, there exists a large
body of evidence showing that several principles of formal logic, be
ginning with the equivalence '(p∩q)≡ (-pvq)', are not psychologically
real; cf. Wason & Johnson-Laird 1972.
19
This means that Hintikka's (1974:157) following claim is mistaken:
"... the games I have defined are the only explicitly defined games
to be found in the literature so-called both in the sense of game
theory and in the sense of Wittgenstein's language games." In the
same context Hintikka equates 'language games' with "rule-governed
activities serving to connect our language to the reality it speaks
of". This is an accurate characterisation of Lorenzen's and Lorenz's
dialogical logic. - Wessel (1972) offers an excellent overview of
Lorenzen's and Lorenz's approach, fully recognising its philosoph
ical importance.
Chap. 3. 20-TH CENTURY LINGUISTIC THEORIES

Notes 20-37
20
I am using the 1962 edition.
21
'PTL' stands for Hjelmslev's Prolegomena, the 1963 edition.
22
The analogy to the general goals of TG, and especially to the notion
of 'evaluation procedure', should be obvious. - I have omitted here
certain inconsistencies and oddities in Hjelmslev's account; for de
tails, see Itkonen (1968).
23
Cf. Chomsky & Halle (1968:/6, note): "... in many significant re
spects we are following in the general line of Sapir's approach to
linguistic structure." See also Halle (1959:13-14); Chomsky (1964d:
69-70); Postal (1968:305-06).
24 , . . .
An historically accurate picture of Sapir which takes his contribu
tion to larger methodological issues into account is given in Hymes
(1975).
25
It is questionable whether Sapir had in fact read Saussure's Cours
at the time.
26
Cf. Mead (1934:6-7): "We want to approach language not from the stand
point of inner meanings to be expressed, but in its larger context of
co-operation in the group taking place by means of signals and ges
tures, Meaning appears within that process. Our behaviorism is a
social behaviorismo Social psychology studies the activity or be
havior of the individual as it lies within the social process; the
behavior of an individual can be understood only in terms of the be
havior of the whole social group of which he is a member, since his
individual acts are involved in larger, social acts which go beyond
himself and which implicate the other members of that group."
27
A few excerpts will be given here: "The formula that I would venture
to suggest is simply this: Religion is man's never-ceasing attempt
to discover a road to spiritual serenity across the perplexities and
dangers of daily life" (Sapir 1949h:347). "What constitutes spirit
ual serenity must be answered afresh for every culture and for every
NOTES 317
community - in the last analysis, for every individual. Culture

defines for every society the world in which it lives, hence we can
expect no more of any religion than that it awaken and overcome the
feeling of danger, of individual helplessness, that is proper to that
particular world. The ultimate problems of an Ojibwa Indian are dif
ferent as to content from those of the educated devotee of modern sci
ence, but with each of them religion means the haunting realization
of ultimate powerlessness in an inscrutable world, and the unques
tioning and thoroughly irrational conviction of the possibility of
gaining mystic security by somehow identifying oneself with what can
never be known" ( i b i d . ) .
PR
Today, it is the express purpose of the 'psychology of personal con
structs' to describe individual persons as self-conscious idiosyncra
tic or largely unique totalities; cf. Bannister & Fransella (1971).
For an attempted synthesis of anthropology and psychiatry, cf. Becker
(1971).
29
It is against this background that the following statement must be
understood: "The over-all purpose of work in descriptive linguistics
is to obtain a compact one-one representation of the stock of utter
ances in the corpus" (Harris 1961:366). TG has been particularly fond
of quoting this statement because, taken out of context, it wholly
distorts and trivialises Harris's conception of grammar. Bach (1965)
even goes so far as to quote this passage twice within the limits of
one rather short article.
30
Notice that in addition to the general principle of disconfirmability,
Harris is formulating here a precise equivalent to Chomsky's 'clear
case principle'.
31
Cf. Chomsky (1957:56): "In short, we shall never consider the ques
tion of how one might have arrived at the grammar whose simplicity
is being determined... Questions of this sort are not relevant to the
program of research that we have outlined above. One may arrive at a
grammar by intuition, guesswork, all sorts of partial methodological
hints, reliance on past experience, etc. ... But this problem is not
within the scope of our investigations here."
32
Lees goes on to claim (p.407) that the data of grammatical descrip
tions are of a physical nature.
33
Bach's two types of prediction, i.e., the derivability of all and only
sentences of L, correspond respectively to my 'explanatory' and 'pre
dictive' types of testing (cf. 1.2.).
34
In 2.0. I have tried to show that in the human sciences any absolute
distinction between scientists and their objects of investigation
must be an illusion; cf. further 8.1-2.
34a
The preceding discussion can be summed up by the following question:
Is there, in the TG opinion, any methodological reason why theoretical
physics cannot be called 'mentalistic physics'?
35
In a public discussion at Salzburg, August 1977, Gaberell Drachman
claimed that no transformationalist has ever let experimental data
falsify an otherwise plausible grammatical description. Motsch (1974:
118) makes use of the same argument in constructing his well-balanced
critique of TG.
Chomsky (1976:230, n.ll) has this to say about the nature of linguis
tic creativity: "Appropriateness is not to be confused with control,
nor can the properties of language use noted here (what I have else
where called 'the creative aspect of language use') be identified with
the recursive property of grammars. Failure to keep these very dif
ferent concepts separate has led to much confusion." Characteristi
cally, however, he fails to mention that he is the one who has created
all the confusion here; cf. Chomsky (1967:7): "The available evidence
shows that the output of this device is a system of recursive rules
that provide the basis for the creative aspect of language use and
that manipulate highly abstract structures."
37
The following statement expresses the 'standard' TG position: "A human
speaker has the ability to create and understand an unlimited number
of completely novel sentences." Consider how Householder (1969:888-
89) comments upon it: "So far the Bloomfieldian would generally have
agreed, but this was never one of his vital doctrines. He might ques
tion the word 'completely', which suggests that no sentence bears any
partial resemblance to any other sentence - a claim so obviously
false that Langacker must mean something else, though I cannot for the
life of me figure out what."
Chap. 4. PHILOSOPHICAL FOUNDATIONS OF THE CONCEPT OF LANGUAGE

Notes 38-64
For relevant passages, see Marx & Engels (1973:30-31,37); Dilthey

(1927:141,147); Mead (1934:223, and 223, η.25); Lauer (1965:155);
Rossi (1972:158).
39
This presumed rock-bottom of knowledge is described by Descartes as
follows: "... it is I who have sensations, or who perceive corporeal
objects as it were by the sense. Thus, I am now seeing light, hearing
a noice, feeling heat. These objects are unreal, for I am asleep; but
at least I seem to see, to hear, to be warmed. This cannot be unreal;
and this is what is properly called my sensation" (Meditations II, p.
71).
40
Saunders & Henze (1967:136-41) attribute this insight to Strawson
(1967:94-107)= However, the same idea is already expressed in Witt
genstein (1958:I,§398): "And this too is clear; if as a matter of
logic you exclude other people's having something, it loses its sense
to say that you have it." Cf. also Wittgenstein (1965:63-64,55). As
we have seen, the essentially same conclusion was reached already by
NOTES 319
Marx and Engels, Dilthey, Mead, Husserl, and Heidegger, among others.
It is conceivable that in other cultures the Cartesian fallacy
would have been easier to overcome or would not have come up at all.
In the Tanzanian countryside there is a proverbial Swahili saying
"Niko kwa kuwa tuko" (= "I am because we are"), which reflects the
traditionally strong solidarity between the members of a community
and constitutes, incidentally, the perfect formulation of Strawson's
semi-grammatical insight. (I am indebted to Dr. Raimo Harjula for
bringing this proverb to my attention.)
41
This crucial notion will be defined in 5.1.(below).
42
Cf. also Marx & Engels (1973:21): "Wie die Individuen ihr Leben äu
ssern, so sind sie."
43
The following discussion is strongly influenced by Winch (1958) .
44 . .
I take this to be an indisputable truth, born out by what we know
about the history of the earth. I know fully well that any attempt
to specify the relation between our concepts and the physical reality
'as it really is' involves formidable philosophical difficulties. I
decline to make any such attempt here, since I can draw my general
distinction between event and action, without being committed to ex
plain what 'events an sich1 are. - Ringen (1977:415, also n. 6)
disputes the dissimilarity between events and actions as here de
scribed, referring to the fact that descriptions of physical regular
ities, e.g. dispositional law statements, may contain conceptual rela
tions. My reply is that Ringen fails to distinguish between events
and descriptions of events. As human constructions, the latter may
quite obviously contain conceptual relations. However, this is not
true of the former. Cf. Mackie (1974:16-17): "Of course, we can find
descriptions of causes and effects such that the description of a
cause is analytically connected with the description of its effect,
but this is a trivial point on which no time need be wasted." Cқ. also
n.9.
45
The symbol 'Λ' stands for exclusive disjunction (= 'either - or').
'D' does not, strictly speaking, meet the definition of exclusive
disjunction, because it is conceptually impossible for the disjuncts
' 'B' and 'C' to be true at the same time. Consequently, 'D' is false
only when both 'B' and 'C' are false.
46 .
The point that he is trying to make here is not easy to grasp; it runs
counter to all ingrained atomistic habits of thinking. This is why I
have chosen to quote him at some length.
47 . . . . . .
The inseparability of mental phenomena and their environment is
graphically expressed by Marx & Engels (1973:30, n=2): "Mein Verhält
nis zu meiner Umgebung ist mein Bewusstsein."
48
Wittgenstein (1958:I,§285 and 261) vividly describes the tradition-
ist's predicament: "I want co keep a diary about the recurrence of a
certain sensation, To this end I associate it with the sign 'S' and
write this sign in a calendar for every day on which I have the sensa
tion. ... What reason have we for calling 'S' the sign for a sensa
tion? For 'sensation' is a word of our common language, not of one
intelligible to me alone. So the use of this word stands in need of
a justification which everybody understands. - And it would not help
either to say that it need not be a sensation; that when he writes 'S',
he has something - and that is all that can be said. 'Has' and
'something' also belong to our common language. - So in the end when
one is doing philosophy one gets to the point where one would like
just to emit an inarticulate sound."
49
Wittgenstein is discussing here the distinction between seeing and
seeing as, which is clearly analogous to, though not identical with,
that between observing and understanding.
Somewhat similar ideas are expressed in Donnelian (1968), where it is
shown that the meaning of an utterance is conceptually dependent upon
its context, the term 'context' being of course understood widely
enough to contain the previous usage of the words making up the utter
ance .
51
The same standpoint, namely that qualitative differences result from
the increasing complexity of the data, is advocated by modern systems
theory and marxist philosophy, for example.
52
From the irreducibility of intentional to physical, Quine (1960:216-
21) draws the surprising conclusion that the former must be ignored.
53
As always, I am not denying the existence of intermediate cases, as
in the language of schizophrenics, for example.
54
The 'logical syntax' is only an apparent exception. It is charac
terised by the fact that its marks can be manipulated as if they had
no meaning; but the whole point of manipulating them is that they do
have meaning, namely, a certain analogy to certain thought-operations.
55 .
Linguists have for the most part ignored the private-language argu
ment; and when they do consider it, they seem to have difficulties in
grasping its subtlety. For instance, Schnelle (1973:39-40) purports
to 'refute' Wittgenstein's position by constructing for himself the
following rule which is both private and applicable at best once:
"If I meet a lion in the wilds, I shall go directly to it and caress
it." This may or may not be a witty example; in any case it has
nothing to do with the epistemological questions raised by Wittgen
stein. The same judgment applies to Fodor (1975 : chap.2). Fodor's
putative defense of private languages rests on his identification of
'private language' with 'unconscious, largely innate cognitive struc
ture'. Yet there is nothing to justify such an identification. Just
as well one might call the internal structures of individual atoms
their 'private languages', at least if one accepts TG's general meth
odological line (cf. n.34a). Fodor's (1975:63) knock-down argument
NOTES 321
speaks for itself: "Hence, in general, any behavior whatever is com

patible with understanding, or failing to understand, any predicate
whatever. Pay me enough and I will stand on my head if you say
'chair'. But I know what 'is a chair' means all the same. - So
much for caveats."
The topic of this section has several far-reaching implications, only

a few of which I can mention here. Poetic language is often a result
of the tension between the triviality of the public language and the
impossibility of a private language. It is a natural urge for many
a poet to try to create an entirely new, and hence private language,
which would nevertheless be understandable to his audience. Now, the
logical proof that even a private language understandable only to the
poet himself is impossible, serves to illustrate the dilemma of those
who, like Rimbaud for instance, try to innovate language in a radical
way, and, sensing the impossibility of their attempt, may ultimately
be reduced to silence. In this respect their attitude resembles that
of mystics; cf. Wittgenstein (1969a:6.521 and 6.522): "Die Lösung des
Problems des Lebens merkt man am Verschwinden dieses Problems. (Ist
dies nicht der Grund, warum Menschen, denen der Sinn des Lebens nach
langen Zweifeln klar wurde, warum diese dann nicht sagen konnten,
worin dieser Sinn bestand.) Es gibt allerdings Unaussprechliches.
Dies zeigt sich, es ist das Mystische." Here one is reminded of
Buddha's 'answer' to the question concerning the meaning of life.
Furthermore, the impossibility of a private language that could at
the same time be used for communication explains, quite simply, why-
it is that when people who possess some truly extraordinary, or liter
ally private, gift try to explain this gift of theirs, for instance
in their memoirs, their accounts invariably sound disappointingly
trivial.
'Explanations of the latter type are widely discussed in the current

philosophy of action; cf., e.g., Harré & Secord (1972).
Compare this result with Chomsky's unsuccessful attempt to account

for creativity within his self-imposed methodological restrictions.
This distinction will be discussed in 5.2.(below).
Compare the following passages from Humboldt (1968[1836]:208, 212, and

206, in that order): Die Sprache wird durch Sprechen gebildet, und das
Sprechen ist Ausdruck des Gedankens und der Empfindung." - "Jeder
nun braucht die Sprache zum Ausdruck seiner besondersten Eigenthüm-
lichkeit: denn sie geht immer von dem Einzelnen aus, und jeder bedient
sich ihrer zunächst nur für sich selbst." - "Der wahre Vorzug einer
Sprache ist nur der, sich aus einem Princip und in einer Freiheit zu
entwickeln, die es ihr möglich machen, alle intellectuelle Vermögen
des Menschen in reger Thätigkeit zu erhalten, ihnen zum genügenden
Organ zu dienen, und durch die sinnliche Fülle und geistige Gesetz
mässigkeit, welche sie bewahrt, ewig anregend auf sie einzuwirken. In
dieser formalen Beschaffenheit liegt alles, was sich wohlthätig für
den Geist aus der Sprache entwickeln lässt."
60
For comments, see Itkonen (1968). However, the emphasis on the indi
vidualistic aspect of language is lacking in Hjelmslev's theory.
.
61
The same judgment applies to the theory of language presented m Lieb
(1970) for example, which takes idiolects of individual speakers as
basic units and constructs other linguistic entities out of them by
means of repeated application of the set theory. Here we have a case
where a method of description projects its own properties into the
data, instead of presenting the latter as they really are. Another
example of the inadequacy of formal logic as a method of conceptual
analysis will be presented in the end of 11.1. (below).
62
Curiously enough, they do not even mention Wittgenstein's name.
This Wittgenstein-inspired criticism of TG's position is very similar
to Marx's criticism of Feuerbach's conception of religion; cf. Hymes
(1974:121-22).
64
Here 'grammar' stands for any analysis of rules (of speaking). It
follows that part of what today is generally labelled as 'sociolin-
guistics' should, in my opinion, be subsumed under the notion of gram
mar. By contrast, 'sociolinguistics', as I understand this term,
stands for an empirical science necessarily characterised by statis
tical methods of description.
Chap. 5. THE CONCEPT OF LANGUAGE

Notes 65-84
Attempts to base the existence of rules solely on the existence of

rule-sentences are discussed in Itkonen (1974, chap.3). To give one
more example, Harre & Secord (1972:12 and 181) claim, too, that "rules
are propositions". - It might be advisable to read 6.2.(below),
before entering upon the present chapter. Otherwise much of what I
say about 'rules' and 'rule-sentences' may be difficult to understand.
For the notion of 'doing something under some description', cf.

Anscombe (1957:37-47).
67
This is the central concept of social philosophy. It is discussed
extensively in Mead (1934) and Schutz (1962) for example, and, more
recently, in Lewis (1969) and Schiffer (1972). In my terminology con
cepts and rules are 'objects' both of common (or social) and of sub
jective knowledge. Common knowledge literally constitutes concepts
and rules as what they are, whereas their (social) existence is inde
pendent of the subjective knowledge of any individual person; cf.
immediately below.
. . .
68
To qualify as incorrects an action must have a recognisable relation
to a possible correct action from which in doviates for whatever reas
on. If no such relation can be made out, an action cannot even be
called incorrect, Therefore it is impossible that a speech act for
NOTES 323
instance could be incorrect in each conceivable respect, i.e., that it

could violate all, or even most, of those rules which would determine
the correctness of its correct counterpart. This issue is familiar
to those who have tried to determine different degrees of (un)gram-
maticalness.
69 .
This is, I think, the precise meaning of the basic tenet on Winch
(1958), which has puzzled some commentators (e.g., Maclntyre 1967a),
that all actions are rule-governed. On the other hand, I must admit
uncertainty about the precise status of those 'rules' whose ubiquity
is taken for granted in Harre & Secord (1972), for instance, and more
generally in ethnomethodology.
70
This brief remark is only meant to show that I am aware of the fact
that the notion of common knowledge, as here defined, does not pay
sufficiently attention to the notion of conceptual innovation.
71
Such a philosophy of science was briefly outlined in 2.0.(above). Re
member Habermas's dictum: "Lack of reflection is positivism."
72
The inference says that from 'p or q' and 'not-p', we may validly in
fer 'q'. Within ordinary thinking, the necessary truth of 'p or
not-p' is self-evident; and yet, from a definite point of view it can
sensibly be called into question, as in so-called intuitionist or con
structivist logic.
73
The above point can also be made in the following way. If a concept
like 'father' is inherent to, or contained in, particular subjective
intuitions, then we have simply no way of knowing what 'father1 means
outside of those intuitions, i.e., what father' means tout court. An
intuitional account of the existence of concepts cannot help relying
on objective, not-just-intuitive meanings of concepts, and therefore
defeats itself. The same argument can be used to show, e.g., that
'good' cannot be defined as 'what X thinks is good' (cf. Maclntyre
1967b:29). In other words, 'transcendental idealism', which makes no
difference between subjective knowledge and its object, is certainly
false.
74
The meaning of this formula should be clear on the basis of n.4 and
45.
75
It is clear that the equivalence is not meant to hold between formula-
tokens. If we consider only the latter, i.e., physical marks on
paper, we have no right to think of them as logical entities.
75a
What we have here is Dilthey's and Weber's well-known distinction
between Zw eckverstehen and Aus drucksverstehen. - There are rules of
social behaviour which specify what one ought to do next. In such a
case, explaining why an action was done does not produce psychological
knowledge, or does so only if the explanation pertains to why the per
son in question followed those rules, instead of breaking them. - Jo-
hach (1974) presents a modern interpretation of Dilthey's work.
Collingwood ' s concept of 'thought' is here âe facto identical with
'social context' in the sense used in 4.2.(above), or with the process
which gives social contexts the meanings they have.
Intuition as used in linguistics is to be distinguished from intro-

spection as used in psychophysics. Ringen (1977:415-16) justifiably
criticises Itkonen (1974) for not explaining how the objects of intui
tion and introspection differ from each other. The foregoing discus
sion should make it clear that uses of intuition and introspection
result in quite different kinds of claims. Introspection is about
temporally definite happenings in one person's nervous system whereas
intuition is about entities that obtain, within rather broad stretches
of time, equally for each member of a community. It can be further
asked, of course, why there should be such a distinction, but this is
like asking Heidegger's (and Leibniz's) question, viz. why there is
something rather than nothing.
78
Of course, it is quite reasonable to argue that a rule should not ex
ist, for instance if it is morally repugnant.
79
When I speak here of 'certainty', I of course mean objective cer
tainty, not any kind of subjective or 'incorrigible' certainty. All
subjective knowledge is fallible and hence liable to correction
through social control. On the other hand, objective certainty, or
certainty at the level of common knowledge, does not mean immutable
or superhistorical certainty.
80
Cf. the Hare-quotation p.213.
81
However, this is not true of all rules of language; cf. 5.4.(below).
One has to distinguish between what might be called 'invariant' and
'variable' rules (of language). The former type of rules are known
with certainty; I also call them 'genuine rules'. When in this book
I speak simply of 'rules', I always mean invariant or genuine rules.
Elsewhere, e.g., in Itkonen (forthcoming b), I have analysed in detail
the relation between invariant and variable rules.
82
It should be evident that I am not using the term 'social control' in
the sense of an explicit system of sanctions. - In my opinion, lin
guistic intuition is a type of (subjective) pre-experimental know
ledge. Therefore it could not be claimed without contradiction that
intuition could be exhaustively defined as an object of experimenta
tion (cf. 2.1.). This does not mean that there are no checks on in
tuition. Such checks are provided by intuition's amenability to so
cial control. Hence, at the level of intuition, social control is my
counterpart to testability. The difference between the two may seem
trifling, but it has its philosophical importance.
83 . . .
Instead of linguistic change, we could speak, more generally, of
diachronic/geographic/social variation in language.
The topic under discussion is intimately related to the rise of 'vari-
325
NOTES
ationist' schools of linguistics, represented by Labov, Bailey,
Bickerton, and others. The argument of the present section 5.4. can
be summed up by quoting Labov's (1971:496) rationale for variationism:
"When you find that intuitions have dissolved into fragmentary idio-
synchratic views then it is quite possible that help from observation
may be useful." This formulation is, if anything, much too cautious;
for discussion, cf. Itkonen (forthcoming a and b ) .
Chap. 6. THE BASIS OF THE NONEMPIRICAL NATURE OF GRAMMAR

Notes 85-97
85
I use the terms rule and regularity' as they are generally used in
the theory of social science, for instance in Ryan (1970) and Lessnoff
(1974). This is not how linguists are accustomed to use these terms.
86
Ofcourse, my standard definition of 'empirical' is 'spatiotemporally
testable'. But the rule-sentences (which I am discussing right now)
are of the universal from, and universal sentences are tested by try
ing to falsify them.
87
For instance, the standard example of a necessarily true sentence,
viz. "All bachelors are unmarried" is o.bviously both universal and un-
falsifiable.
88
A similar type of argument is used by Hare (1971:232-33) when he de
fends his view that a statement about how the dance 'eightsome reel'
is correctly danced is a priori, i.e., universal, nonempirical, and
necessarily true (or false): "There is first the danger of thinking
that it could not have been the case that the eightsome reel was
danced in some quite different way. It is, of course, a contingent
fact, arising out of historical causes with which I at any rate am
unacquainted, that the dance called 'the eightsome reel' has the form
it has and not some other form. If it had some different form, what
my dancers have learnt in their childhood would have been different,
and what they would have learnt to call 'the eightsome reel' would
have been different too; yet the statement 'the eightsome reel is
danced in the following manner, etc.' would have had just the same
characteristics as I have mentioned (though the 'etc.' would stand for
some different description of steps and movements)."
89
Dahl (1975:3-4) makes precisely the same point in what he, oddly,
takes to be a criticism of my position.
90
Of course, terms like 'definite article' may be unknown to ordinary
people, but the referents of such terms are definitely known to them.
Besides, terms like 'definite article' are in any case dispensable
here. The rule in question could be formulated as "In English the
expression the man is correct, and man the incorrect", which contains
no technical terms. On the other hand, it is quite clear that rules,
which - as here defined - are fully accessible to conscious lin
guistic intuition, must be tied together by some more abstract or more
general psychological mechanisms which are not accessible to intui
tion, but can only be discovered by psycholinguistic experimentation.
91 .
The distinction between atheoretical certainty and theoretical uncer
tainty should not be too difficult to grasp (cf. also 5.3.above).
Surprisingly enough, there are linguists who find even this little bit
of conceptual differentiation 'confusing' (e.g., Dahl 1975:2).
92
To be sure, what we have here is not a genuine existential statement.
Since it is about the existence of a rule, it is implicitly of univer
sal character. In other words, grammar and empirical science are so
different that they offer no basis for genuinely analogous examples.
The statement that there is at least one utterance of such and such a
sentence in some unidentified region of space and time, would be genu
inely existential, but grammar is not concerned with such statements
(cf. 7.0.below).
93
In their debate with Cavell, Fodor & Katz (1971, esp. p.203) defend an
exceptionally rigid positivistic line. They go so far as to count
philosophy of language among the empirical sciences.
94 . . . .
This is not to deny that, even m linguistics, synchrony can be prop
erly understood only with reference to its counterpart, i.e., dia-
chrony. Moreover, even if the relations between different states of
language are not conscious, they are not purely atomistic either,
since linguistic changes normally display a functional or teleolo-
gical character.
95
It is clear that the 'truth' of such a rule-sentence cannot be known
with the same immediacy and certainty as the truth of rule-sentences
discussed in 6.1.-2.
96
The growth of scientific knowledge involves both a relative and an
absolute concept of truth: A sentence proposing a new methodological
rule may be more true than a sentence referring to an existing rule
in relation to an absolute truth which science approximates, but which
it never reaches. However, the view of science as developing in such
a rectilinear way has been called into question by Kuhn and Feyer
abend, among others.
97
Notice that the splitting of atoms and the creating of living cells,
for instance, take place within unchanging laws of nature.
Chap. 7. THE INELIMINABILITY OF LINGUISTIC NORMATIVITY

Notes 98-107
98
The foregoing also answers Sampson's (1976) criticism of Itkonen
(1975a). In fact, it was anticipated and answered, more fully than
here, already in Itkonen (1974:169-70); cf. also Itkonen (1976d: 56-60).
99
The 'index' of an utterance contains the 'rules of operation on the
NOTES 327
inscription' in question. The nature of these rules is not specified

more narrowly by Kasher. In any case, it hardly seems correct to
treat rules (of whatever type) and speakers equally as 'indexes' of
utterances. Moreover, this treatment loses sight of the fact that
rules determine primarily the correctness of sentences (in my sense),
not of utterances.
Kasher's account is espoused in Kanngiesser (1976:121), for instance.

A somewhat similar attempt to combine grammatical concepts with space-
time is made in Schnelle (1973:120-25).
101
To be sure, Labov's work has shown that, by introducing a set of in
dependent linguistic and social variables, we are able to fix quite
reliable conditional probabilities for selected subsentential phe
nomena, taken as dependent variables. But, to repeat, Kasher is not
at all interested in what really goes on in space and time. This
surely is a curious form of 'empirism'.
102
Cf. Schnelle (1976), who bases this claim on his interpretation of
Lorenz's (1970) concept of Lern- und Lehrsituationen.
103
There are further difficulties with Lewis's (1969) position. If he
really wishes to make linguistics empirical by eliminating the ref
erence to correctness in favour of the reference to mutual expecta
tions, then the actual existence of such expectations must be estab
lished. If this cannot be done in purely empirical terms, then Lewis
has failed. Now, suppose that a person A has some actual expecta
tions about what another person B will say. Shall we discover that A
has the expectations that B will say "It's raining", and not "It's
rainil" (or some such form), "I'm tired" and not "I'k tired" (or some
such form), and similarly for the whole incalculable number of cases
where one can say something either correctly or incorrectly? Or shall
we discover that A has only the general expectation that whatever B
is going to say he will speak correct English? I am certain that the
latter alternative is the true one. But this means, once again, that
we just cannot get rid of normativity.
104
The same is true also of Dretske (1974), Ringen (1975) and Lass
(1976), who - like myself - explicity argue for the nonempirical
character of grammar.
105
An analysis of Labov's descriptive practice shows that the relation
between grammar and sociolinguistics is analogous to that between
protophysics and physics, the former being the 'transcendental'
counterpart of the latter; cf. Itkonen (forthcoming a and b ) .
106
The emptiness of TG's concept of 'novelty' was exposed on p.86.
107
This is denied by Dougherty (1974:133), for instance, who claims that
"intuitions are observations". I submit that this formulation is
devoid of sense.
Chap. 8. LANGUAGE AND GRAMMAR
Notes 108-114
7 OR
Juntunen (1977) offers an analysis of the notions of atheoretical
knowledge utilised in the human and in the natural sciences.
109
Cf. Hymes (1974:11): "Ethnographic objectivity is intersubjective ob
jectivity, but in the first instance, the intersubjective objectivity
in question is that of the participants in the culture".
110
Of course, the learning of rules is gradual. But this does not mean
that the transition from regularities to rules is gradual. One
learns a rule by making successive hypotheses about it and by mod
ifying them in the light of new evidence. It is this transition from
wrong guesses to the right guess, and then to genuine knowledge,
which is gradual.
111
Chomsky is here still the exception; cf. 3.6.
112
Immler (1974:139) uses the term 'normierte Paraphrase1.
113
The question is, of course, what is the 'linguistically' meaningful
way.
114
Notice that the phonemes are not supposed to have any intuitive (or
experimental-psychological) reality. Rather, we, as it were, observe
their sudden appearance at a certain level of our theoretical de
scription.
Chap. 9. THE METHODOLOGY OF GRAMMAR

Notes 115-127
Notice that it is the number of utterances, not of sentences, which

is claimed to be potentially infinite. I expressly reject the notion
that the number of the sentences of a given language could be infin
ite (cf. Itkonen 1976a).
Today Lakoff has abandoned the view that grammatical explanations

must have the structure of (vertical) derivations; cf. n.125.
117
Kanngiesser (1976:148, n.25) tries to avert this conclusion by appeal
ing to Kasher-type functions; but this appeal is without force, as
we saw in 7.2.(above).
118
I use here my own terms 'predictive test claim' and 'explanatory test
claim' to explicate TG's notion of testing a grammar like G3 . It
will turn out immediately below that maximally simple grammars like
G3 are in reality untestable. (In this respect they are similar to
rule-sentences.) Therefore we do not have genuine test claims here.
NOTES 329
119
Notice that I am not saying here that theoretical-grammatical de
scriptions are reducible to rule-sentences (which would be something
like a nonempirical counterpart of 'strong reductionism'). What I am
saying is that a theoretical statement about (a fragment of) a lan
guage must be analysable into a set of (theoretical) statements about
rules.
119 a
This means, incidentally, that 'testability ' does not necessarily
entail conceptual independence: there may be conceptual connections
too complex for us to grasp.
120
As an example, one could mention the componential analysis of the
kinship terms of English by means of the three components 'male',
'female', and 'parent'; cf. p.308.
121
In Itkonen (1970a) I devised a quasi-experimental method for elic
iting new components of word meanings in new verbal contexts, How
ever, this method is experimental only in appearance. It merely
serves to direct our attention to semantic facts which we know in
principle, but which we might otherwise have overlooked.
122
Cf. Wittgenstein (1967 : III, §52) : "We do not acoept e.g. a multi
plication's not yielding the same result every time." - The relation
of normative science to experimentation will further occupy us in
10.0.(below).
Cf. Aristotle (1941c :980b-981a) : "Now from memory experience is prod
uced in men; for the several memories of the same thing produce fi
nally the capacity for a single experience. And experience seems
pretty much like science and art, but really science and art come to
men through experience".
124
It is also interesting to note that, in Feyerabend's (1975:81-82)
opinion, Galileo retained the notion of recollection, i.e., anamnesis,
for propagandistic purposes.
125
Such cases are obviously due to the operation of analogy. T. Itkonen
(1976) adduces several examples from Finnish morphology and syntax
that cannot be handled by TG-type discrete and vertical derivations,
but only by nondiscrete and multidirectional 'derivations', identi
fiable as sets of mutual analogical influences of varying force.
Taken as a whole, such an 'influence unity' constitutes a Gestalt
roughly in Köhler's sense. Strikingly similar views have been for
warded in Lakoff (1977), albeit without experimental evidence. For
criticism of Bever's (1974) notion of analogy, cf. Itkonen (1976a:
sect.13); for a general discussion of analogy, cf. Anttila (1974).
The following formulations and examples are taken from Ross (1968:
10-13).
127
Besides criticising the methodological self-understanding of Katzian
semantics, Itkonen (1970a) contains an attempt to make semantics an
empirical science. Yet such an attempt could not succeed; cf. n.121.
Chap. 10. GRAMMAR AND LOGIC
Notes 128-129
128
To illustrate, let us apply the above-mentioned transformations to the
the formula '0(p&q/pvq)'. We obtain: i) 0[(pvq)&(-pvq)&(pv-q/(p&q)
v(-p&q)v(p&q)] ii) 0[(pvq)/(p&q)v(-p&q)v(p&-q)]&0[(-pvq)/(p&q)ν(-p&q)
v(p&-q)]&0[(pv-q)/(p&q)v(-p&q)v(p&-q)] iii)[0 (pvq)/(p&q)]&0[(pvq)/
(-p&q)]&0[(pvq)/(p&-q)]&0[(-pvq)/(p&q)]&0[(-pvq)/(-p&q)]&0[(-pvq)/
(p&-q) ]&0[ (pv-q) / (p&q) ]&0[ (pvq) / (-p&q) ]&0[ (pv-q) / (p&-q) ]
7 9Q . . .
Hintikka's formulation is apt to suggest the misleading impression
that, when doing logic, he claims to be doing empirical psychology.
In reality, he is making conceptual distinctions, i.e., distinguish
ing between different types of (possible) use. Which types are actu-
alised on which occasions, is an entirely different question. In
other words, "what actually happens on the different occasions of
ordinary use" is not a concern of the logician, but of the psycho
logist .
Chap. 11. GRAMMAR AND PHILOSOPHY

Notes 130-131
130
Cf. Aristotle (1941c : 1031a,10-15): "Clearly, then, definition is the
formula of the essence, and essence belongs to substances either
alone or chiefly and primarily and in the unqualified sense." And:
"Each thing itself, then, and its essence are one and the same in no
merely accidental way ... because to know each thing, at least, is to
know its essence, so that even by the exhibition of instances it be
comes clear that both must be one" (op. c i t . , 1031b,15-25).
131
Of course, Socrates overlooks here the possibility that 'desiring
good things' might mean things which are good in spite of the fact
that one may not consider them as good.
REFERENCES
Abel, Theodore. 1948. "The Operation Called 'Verstehen'". American
Journal of Sociology 54.211-18.
Abelson, R. P. et al., eds. 1968. Theories of Cognitive Consistency.

Chicago: Rand McNally.
Andersen, Henning. 1973. "Abductive and Deductive Change". Language

49:4. 765-93.
d'Ändrade, Roy G. 1970. "Structure and Syntax in the Semantic Analysis

of Kinship Terminologies". C o g n i t i o n ; a Multiple View ed. by Paul L.
Garvin, 87-143. New York & Washington: Spartan Books.
Andresen, Helga. 1974. Erklärungsgehalt linguistischer Theorien.

München: Hueber.
Anscombe, G. Elisabeth M. 1957. Intention. Oxford: B. Blackwell.
Anttila, Raimo. 1974. Analogy. Helsinki: Publications of the General

Linguistics Department of the University of Helsinki, 1.
Apel, Karl-Otto. 1973a. Transformation der Philosophie, vol. 2:

Das Apriori der Kommunikationsgemeinschaft. Frankfurt/Main: Suhrkamp.
. 1973b. "Die Kommunikationsgemeinschaft als transzendentale

Voraussetzung der Sozialwissenschaften". In Apel 1973a:220-63.
. 1973c. "Noam Chomskys Sprachteorie und die Philosophie der

Gegenwart". In Apel 1973a:264-310.
Aristotle. 1941a. The Basic Works. Edited and with an Introduction by

Richard McKean. New York: Random House.
. 1941b. Posterior Analytics. In Aristotle 1941a:110-86.
. 1941c. Metaphysics. In Aristotle 1941a:689-926.
Bach, Emmon. 1964. An Introduction to Transformational Grammars. New

York: Holt, Rinehart & Winston.
. 1965. "Structural Linguistics and the Philosophy of Science".

Diogenes 51.111-28.
Bach, Emmon. 1974. Syntactic Theory. New York: Holt, Rinehart & Winston.
Bannister D., and Fay Fransella. 1971. Inquiring Man. The Theory of Per-
sonal Constructs. Harmondsworth, Middlesex: Penguin Books.
Barker, S. F. 1957. Induction and Hypothesis. Ithaca N.Y.: Cornell Univ.

Press.
Becker, Ernest. 1971. The Birth and Death of Meaning. New York: The Free
Press.
Bennett, Jonathan. 1973. "The Meaning-Nominalist Strategy". Foundations

of Language 10:1.141-168.
Berger, Peter, and Thomas Luckmann. 1966. The Social Construction of

Reality. New York: Doubleday.
Bever, Thomas G. 1970. "The Cognitive Basis for Linguistic Structures".

In Hayes 1970:279-362.
. 1974. "The Ascent of the Specious or There's a Lot We Don't

Know about Mirrors". In Cohen 1974:173-200.
Black, Max. 1962. Models and Metaphors. Ithaca N.Y.: Cornell Univ. Press.
Blalock, Hubert M. , Jr. 1964. Causal Inferences in Honex'perimental Re-

search. Chapel Hill: Univ. of North Carolina Press.
Bloomfield, Leonard. 1926. "A Set of Postulates for the Science of

Language". Language 2:2.153-64.
. 1935. Language. London: Allen & Unwin. (1st ed., 1933.)
. 1936. "Language or Ideas?". Language 12:1.89-95.
Bocheñski, I. M. 1971. Lie zeitgenössischea Denkmethoden. 5th ed.

München: Francke.
Böhme, Gernot, ed. 1976. Protophysik. Frankfurt/Main: Suhrkamp.
Botha, Rudolf P. 1968. The Function of the Lexicon in Transformational

Generative Grammar. The Hague: Mouton.
. 1971. hlethodological Aspects of Transformational Generative

Phonology. Ibid.
. 1973. The Justification of Linguistic Hypotheses. Ibid.
Boudon, Raymond. 1974. The Logic of Sociological Explanation. Transi.

by Tom Burns. Harmondsworth, Middlesex: Penguin Books.
REFERENCES 333
Bubner, Rüdiger. 1976. "Is Transcendental Herraeneutics Possible?" In

Manninen & Tuomela 1976:59-77.
Campbell, Lyle. 1976. Generative Phonology vs. Finnish Phonology. Albany:

State University of New York, mimeo.
Campbell, Norman. 1952. What Is S c i e n c e ? 2nd ed. New York: Dover.
Carnap, Rudolf. 1937. The Logical Syntax of Language. Transl, by Amethe

Smeaton. London: Routledge & Kegan Paul.
Cavell, Stanley. 1971a. "Must We Mean what We Say?" In Lyas 1971:131-65.

(Originally published in Inquiry 1, 1958. References are to the Lyas
edition.)
. 1971b. "The Availability of Wittgenstein's Later Philosophy".

In Lyas 1971:166-89. (Originally published in Philosophical Review
71, 1962. References are to the Lyas edition.)
Chihara, Charles S., and Jerry A. Fodor. 1966. "Operationalism and Ordi
nary Language: A Critique of Wittgenstein". In Pitcher 1966:384-419.
(Originally published in American Philosophical Quarterly 2, 1965.
References are to the Pitcher edition.)
Chisholm, Roderick M. 1963. "Contrary-to-Duty Imperatives and Deontic

Logic". Analysis 24.33-36.
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.
. 1961. "Some Methodological Remarks on Generative Grammars".

Word 17:219-239.
. 1964a. "A Transformational Approach to Syntax". In Fodor &

Katz 1964:211-45. (Originally published in Proceedings of the Third
Texas Conference on Problems of Linguistic Analysis ed. by Archibald
A. Hill. Austin: Univ. of Texas Press, 1962. References are to the
Fodor & Katz edition.)
. 1964b. Review of B. F. Skinner's Verbal Behavior (New York:

Appleton-Century-Crofts, 1957). In Fodor & Katz 1964:547-78. (Orig
inally published in Language 35, 1959. References are to the Fodor &
Katz edition.)
. 1964c. "The Logical Basis of Linguistic Theory". Proceedings

of the Ninth International Congress of Linguists ed. by Horace G.
Lunt: 914-78. The Hague: Mouton.
. 1964d. Current Issues in Linguistic Theory. Ibid.
. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.:

MIT Press.
Chomsky, Noam. 1966a. Cartesian Linguistics. New York: Harper & Row.
. 1966b. Topics in the Theory of Generative Grammar. The Hague:

Mouton.
. 1967. "Recent Contributions to the Theory of Innate Ideas".

Synthese 17:1.2-11.
. 1968. Language and Mind. New York: Harcourt, Brace & World.
. 1969. "Quine's Empirical Assumptions". Words and Objections.

Essays on the Work of W. V. Quine ed. by Donald Davidson and Jaakko
Hintikka, 53-68. Dordrecht-Holland: Reidel.
. 1972. Problems of Knowledge and Freedom. London: Fontana/
Collins.
. 1973. For Reasons of State. Ibid.
. 1975[1955]. The Logical Structure of Linguistic Theory. New

York: The Plenum Press.
. 1976. Reflections on Language. London: Fontana/Collins.
Chomsky, Noam, and Morris Halle. 1968. The Sound Vattevn of English. New
York: Harper & Row.
Chomsky, Noam, and Jerrold J. Katz. 1974. "What the Linguist Is Talking
about". Journal of Philosophy 71.347-67.
Cicourel, Aaron V. 1964. Method and Measurement in Sociology. New York:

The Free Press.
Cohen, David, ed. 1974. Explaining Linguistic Phenomena. New York: John
Wiley & Sons.
Cohen, David, and Jessica Wirth, eds. 1975. Testing Linguistic Hypoth-
eses. Ibid.
Collingwood, R. G. 1946. The Idea of History. Oxford: Clarendon Press.
Copi, Irving M. 1967. Symbolic Logic. 3rd ed. New York: Macmillan.
Coseriu, Eugenio. 1974[1958]. Synchronie, Diachronie und Geschichte.

Transl, by Helga Sohre. München: Fink.
Dahl, Östen. 1975. "Is Linguistics Empirical? A Critique of Esa Itko-

nen's Linguistics and Metascience". Gothenburg Papers in Theoretical
Linguistics 29.
Davidson, Donald, and Gilbert Harman, eds. 1972. Semantics of Natural

Language. Dordrecht-Holland: Reidel.
REFERENCES 335
Derwing, Bruce L. 1973. Transformational Grammar as a Theory of Lan-

guage Acquisition. New York & London: Cambridge Univ. Press.
Descartes, René. 1954. Philosophical Writings. Transl, by G. Elisabeth

M. Anscombe and Peter Geach. Edinburgh: Nelson.
Dilthey, Wilhelm. 1914. Gesammelte Schriften. Ed. by G. Misch et al.

vol. 1. Leipzig: Β. G. Teubner.
. 1924. Gesammelte Schriften vol. 5. Ibid.
. 1927. Gesammelte Schriften vol. 7. Ibid.
Donnellan, Keith. 1968. "Putting Humpty Dumpty together again". Philos

ophical Review 77.203-15.
Dougherty, Ray C. 1974. "What Explanation Is and Isn't". In Cohen

1974:125-51.
Douglas, Mary, ed. 1973. Rules and Meanings. The Anthropology of Every
day Knowledge. Harmondsworth, Middlesex: Penguin Books.
Dretske, Fred I. 1974. "Explanation in Linguistics". In Cohen 1974:

21-41.
Durkheim, Emile. 1960[1897]. Suicide. New Edition. Paris: Presses uni

versitaires de France.
. 1938[1895]. Les règles de la méthode sociologique. 9th. ed.

Paris: Alean.
Feigl, Herbert. 1970. "The 'Orthodox' View of Theories: Remarks in

Defense as well as Critique". In Radner & Winokur 1970:3-16.
Feyerabend, Paul. 1968. "How to Be a Good Empiricist". The Philosophy

of Science ed. by P. H. Nidditch, 12-39. London: Oxford Univ. Press.
. 1975. Against Method. London: NLB.
Fodor, Jerry A. 1975. The Language of Thought. New York: Crowell.
Fodor, Jerry Α., Thomas G. Bever, and M. F. Garret. 1974. The Psycholo
gy of Language. New York: McGraw-Hill.
Fodor, Jerry Α., and Jerrold J. Katz, eds. 1964. The Structure of Lan
guage. Englewood Cliffs, N. J.: Prentice-Hall.
. 1971. "The Availability of What We Say".'In Lyas 1971:190-

203. (Originally published in Philosophical Review 72, 1963. Refer
ences are to the Lyas edition.)
Friedman, H. R. 19/5. "The Ontic Status of Linguistic Entities". Foun-
dations of Language 13:1.73-94.
Grice, H. P. 1957. "Meaning". Philosophical Review 66.377-88.
. 1968. "Utterer's Meaning, Sentence-Meaning, and Word-

Meaning". Foundations of Language 4:3.225-242.
Gumb, Raymond D. 1972. Rule-Governed Linguistic Behcvior. The Hague:

Mouton.
Habermas, Jürgen. 1968. Erkenntnis und Interesse. Frankfurt/Main: Suhr-

kamp.
. 1970. Zur Logik der Sozialwissenschaften. Ibid.
. 1971a. "Vorbereitende Bemerkungen zu einer Theorie der

kommunikativen Kompetenz". In Habermas & Luhmann 1971:101-141.
. 1971b. "Theorie der Gesellschaft oder Sozialtechnologie".

In Habermas & Luhmann 1971:142-290.
Habermas, Jürgen, and Niklas Luhmann. 1971. Theorie der Gesellschaft

oder S o z i a l t e c h n o l o g i e ? Was leistet die Systemforschung. Ibid.
Halle, Morris. 1959. The Sound lattern of Russian. The Hague: Mouton.
Hanna, Joseph F. 1968. "An Explication of 'Explication'". Philosovhy of

Science 35.28-44.
Hare, R. M. 1971. "Philosophical Discoveries". In Lyas 1971:223-40. (A

shorter version was published under the title "Are Discoveries about
the Uses of Words Empirical?" in Journal of Philosophy 54, 1957.
References are to the Lyas edition.)
Harre, Rom, 1970. The Principles of Scientific Thinking. London: Mac

millan.
Harré, Rom, and P. F. Secord. 1972. The Explanation of Social Behavior.

Oxford: B. Blackwell.
Harris, Zellig S. 1957. "From Morpheme to Utterance". In Joos 1957:

142-53. (Originally published in Language 22, 1946. References are
to the Joos edition.)
. 1961. Structural Linguistics. 5th. ed. Chicago: Univ. of

Chicago Press.
Hart, H. L. A. 1961. The Concept of Law. Oxford: Clarendon Press.
Hayes, John R., ed. 1970. Cognition and the Development of Language.
New York: John Wiley & Sons.
337
REFERENCES
Hegel, G. F. W. 1975[1807]. Phänomenologie des Geistes. Frankfurt/Main:
Suhrkamp.
Hempel, Carl G. 1952. Fundamentals of Concept Formation in Empirical

Science. Chicago: Univ. of Chicago Press.
. 1965. Aspects of Scientific Explanation and. Other Essays in

the Philosophy of Science. New York: The Free Press.
. 1970. "On the 'Standard Conception' of Scientific Theories".

In Radner & Winokur 1970:142-63.
Henle, M. 1962. "On the Relation between Logic and Thinking". Psycho-
logical Review 69.366-78.
Henson, Richard. 1971. "What We Say". In Lyas 1971:204-22. (Originally-

published in American Philosophical Quarterly 2, 1965. References
are to the Lyas edition.)
Hilpinen, Risto, ed. 1971. Deontic Logic: Introductory and Systematic

Readings. Dordrecht-Holland: Reidel.
Hilpinen, Risto, and Dagfinn Fllesdal. 1971. "Deontic Logic: An Intro

duction". In Hilpinen 1971:1-35.
Hintikka, Jaakko. 1962. Knowledge and Belief. Ithaca, N. Y.: Cornell

Univ. Press.
. 1969. Models for Modalities. Dordrecht-Holland: Reidel.
. 1973. Logic, Language-Games, and Information. London: Oxford
Univ. Press.
. 1974a. Knowledge and the Known. Dordrect-Holland: Reidel.
. 1974b. "Plato on Knowing How, Knowing That, and Knowing

What". In Hintikka 1974a:31-49.
. 1974c. "Practical vs. Theoretical Reason - An Ambiguous
Legacy". In Hintikka 1974a:80-97.
. 1974d. "Quantifiers vs. Quantification Theory". Linguistic

Inquiry 5:2.153-17 7.
Hjelmslev, Louis. 1963[1943]. Prolegomena to a Theory o f Language.

Transl. by Francis J. Whitfield. 2nd ed. Madison: Univ. of Wisconsin
Press.
Householder, Fred. 1969. Review of Ronald W. Langacker's Language and

Its Structure, Language 45:4.886-97.
Hsieh, Hsin-I. 1975. "How Generative Is Phonology? (On Listing Phonolo-

gical Surface Forms in the Lexicon)". In Koerner 1975:109-44.
Humboldt, Wilhelm von. 1968 [1836]. Veber die Verschiedenheit des

menschlichen Sprachbaues und ihren Einfluss auf die geistige Ent-
wicklung des Menschengeschlechts. Bonn: Dümmler.
Husserl, Edmund. 1954. Die Krisis der europäischen Wissenschaften und

die transzendentale Phänomenologie. The Hague: Nijhoff.
. 1962[1913] . Ideas. General Introduction to Iure Phenome-

nology. Transl, by W. R. Boyce Gibson. New York: Collier Books.
Hutchinson, Larry G. 1974. "Grammar as Theory". In Cohen 1974:43-74.
Hymes, Dell. 1974. Foundations in Sociolinguistics. Philadelphia: Univ.

of Pennsylvania Press.
. 1975. "Pre-War Prague School and Post-War American Anthro

pological Linguistics". In Koerner 1975:359-80.
Immler, Manfred. 1974. Generative Syntax - Generative Semantik.

München: Fink.
Itkonen, Esa. 1968. "Zur Charakterisierung der Glossematik". Neuphilo-

logische Mitteilungen 69.452-72.
. 1969a. Semantics as an Analysis of language-Independent

Reality. Cambridge, Mass.: MIT, mimeo.
. 1969b. "Semantiikan ja syntaksin suhteesta" (= "On the Re

lation between Semantics and Syntax"). V i r i t t ä j ä 73:3.217-37.
. 1970a. "Zwei verschiedene Versionen der Bedeutungskompo

nente". Linguistics 59.5-13.
. 1970b. "An Epistemological Approach to Linguistic Seman

tics". Ajatus: Yearbook of the Philosophical Society of Finland
32.96-142.
. 1972a. "Onko synkroninen kielitiede luonteeltaan empii-

ristä?" (= "Is Synchronic Linguistics Empirical?"). Virittäjä 76:
1.1-14.
. 1972b. "Concerning the Methodological Status of Linguistic

Descriptions". Derivational Processes (= KVAL PM, Ref. No. 729), ed.
by Ferenc Kiefer, 31-41. Stockholm.
. 1972c. "On Grice's, Searle's, and Strawson's Concept of

Meaning". Ajatus 34.149-54.
. 1974. Linguistics and Metas cienee. (= Studia Philosophica

Turkuensia 2.) Kokemäki: Risteen Kirjapaino.
REFERENCES 339
Itkonen, Esa. 1975a. "Transformational Grammar and the Philosophy of

Science". In Koerner 1975:381-445.
. 1975b. Concerning the Relationship between Linguistics and

Logic. Bloomington: Indiana Univ. Linguistics Club, mimeo.
. 1976a. "The Use and Misuse of the Principle of Axiomatics

in Linguistics". Lingua 38:2.185-220.·
. 1976b. "Die Beziehung des Sprachwissens zum Sprachver

halten". Sprachtheorie und Pragmatik. Akten des 10. Linguistischen
Kolloquiums, Tübingen 1975, Band I ed. by Heinrich Weber and Harald
Weydt, 39-48. Tübingen: Niemeyer.
. 1976c. "Was für eine Wissenschaft ist die Linguistik

eigentlich?" In Wunderlich 1976:56-76.
. 1976d. Linguistics and Empiricalness : Answers to Criticisms.

Publications of the General Linguistics Department of the Univer
sity of Helsinki, 4.
. forthcoming a. "The Relation between Grammar and Socio-

linguistics". Forum Linguisticum.
. forthcoming b. "Qualitative vs. Quantitative Analysis in

Linguistics". Evidence and Argumentation in Linguistics ed. by
Thomas A. Perry. Berlin: de Gruyter.
Itkonen, Terho. 1976. "Syntaktisten vaikutusyhteyksien luonteesta"

(with English summary: "Non-Generative Phenomena in Finnish Morpho
logy and Syntax"). V i r i t t ä j ä 80:1.52-81.
Jacobs, Roderick Α., and Peter S. Rosenbaum. 1968. English Transfor

mational Grammar. Waltham, Mass.: Blaisdell.
Jakobovits, Leon Α., and Murray S. Miron, eds. 1967. Readings in the
Psychology of Language. Englewood Cliffs: Prentice-Hall.
Johach, Helmut. 1974. Handelnder Mensch und objektiver Geist. Zur

Theorie der Geistes- und Sozialwissenschaften bei Wilhelm Dilthey.
Meisenheim/Glan: A. Hain.
Joos, Martin, ed. 1957. Readings in Linguistics I. Chicago: Univ. of

Chicago Press.
Juntunen, Matti. 1977. "Den humanistiska forskningen som en vetenskaps

filosofisk kategori och begreppet samhällsvetenskap". Nordisk Forum
12:1.48-56.
Kac, Michal B. forthcoming. "Linguistics as a Science of Relevant Para

meters". Paper read at the XIIth International Congress of Lin
guists in Vienna, August-September 1977.
Kambartel, Friedrich, and Jürgen Mittelstrass, eds, 1973. Das normative
Fundament der Wissenschaft. Frankfurt/Main: Athenäum.
Kamlah, Wilhelm, and Paul Lorenzen. 1967. Logisehe Propädeutik. Mann

heim: Bibliographisches Institut.
Kanngiesser, Siegfried. 1972. Aspekte der synchronen und diachronen

Linguistik. Tübingen: Niemeyer.
. 197 6. "Spracherklärungen und Sprachbeschreibungen". In

Wunderlich 1976:106-60.
Kaplan, Abraham. 1964. The Conduct of Inquiry . San Francisco: Chandler.
Kasher, Asa. 1972. "Sentences and Utterances Reconsidered", Foundations

of Language 8:3,313-45.
Katz, Jerrold J. 1966. The Philosophy of Language. New York: Harper &
Row.
. 1967. "Mentalism in Linguistics". In Jakobovits & Miron

1967:73-84. (Originally published in Language 40, 1964. References
are to the Jakobovits & Miron edition.)
Katz, Jerrold J., and Bever, Thomas G. 1974. The Fall and Rise of
Empiricism. Bloomington: Indiana Univ. Linguistics Club, mimeo.
Kenny, Anthony. 1975. Wittgenstein. Harmondsworth, Middlesex: Penguin

Books.
Kiparsky, Paul. 1968. "Linguistic Universals and Linguistic Change".

Universals in Linguistic Theory ed. by Emmon Bach and Robert T.
Harms, 171-202. New York: Holt, Rinehart & Winston.
Kiparsky, Paul, and Carol Kiparsky. 1971. "Fact". In Steinberg &

Jakobovits: 1971:345-69.
Kockelmans, Joseph J. 1969. The World in Science and Philosophy. Mil

waukee: Bruce.
Koerner, E. F. K. 1973. Ferdinand de Saussure . Origin and Development

of his Linguistic Thought in Western Studies of Language . Braun
schweig: Vieweg.
Koerner, E. F. K., ed. 1975. The Transformational-Generative Para-

digm and Modern Linguistic Theory. Amsterdam: John Benjamins.
Körner, Stephen. 1957. "Some Remarks on Philosophical Analysis". Jour-

nal of Philosophy 54:758-66.
Labov, William. 1971. "Methodology". A Survey of Linguistic Science ed.

by William Dingwall, 412-97. College Park, Md.: Linguistics Program,
Univ. of Maryland.
REFERENCES 341
Labov, William. 1972. Socioliaguistic Patterns. Philadelphia: Univ. of

Pennsylvania Press.
Laing, Ronald D., and A. Esterson. 1964. Sanity, Madness, and Family.
London: Tavistock.
Lakatos, Imre. 1970. "Falsification and the Methodology of Scientific

Research Programmes". Criticism and the Growth of Knowledge ed. by
Imre Lakatos and Alan Musgrave, 91-196. London: Cambridge Univ.
Press.
Lakoff, George. 1971. "On Generative Semantics". In Steinberg &

Jakobovits 1971:232-96.
. 1972. "Linguistics and Natural Logic". In Davidson & Harman

1972:545-665.
. 197 7. "Linguistic Gestalts". Papers from the 13th Regional

Meeting ed. by Woodford A. Beach et al., 236-87. Chicago: Chicago
Linguistic Society.
Lass, Roger. 1976. English Phonology and. Phonological Theory. London:

Cambridge Univ. Press.
Lauer, Quentin. 1965. Phenomenology: Its Genesis and Prospect. 2nd ed.
New York: Harper & Row.
Leech, Geoffrey. 1974. Semantics. Harmondsworth, Middlesex: Penguin

Books.
Lees, Robert B. 1957. Review of Noam Chomsky's Syntactic Structures.

Language 33:2.375-407.
Lenneberg, Eric H. 1964. "The Capacity for Language Acquisition". In

Fodor & Katz 1964:579-603.
Leontiev, A. A. 1971. "Social and Natural in Semiotics". Biological and

Social Factors in Psycholinguistics ed. by J. Morton, 122-30.
London: Logos Press.
Lessnoff, Michael. 1974. The Structure of Social Science. London:

Allen & Unwin.
Lewis, David. 1969. Convention. Cambridge: Harvard Univ. Press.
. 1972. "General Semantics". In Davidson & Harman 1972:

169-218.
Lieb, Hans-Heinrich. 1970. Sprachstadium und Sprachsystem. Stuttgart:

Kohlhammer.
. 1974. "Grammars as Theories: The Case for Axiomatic Gram

mar". Theoretical Linguistics 1:1/2.39-115.
Linell, Per. 1974. Problems of Psychological Reality in Generative
Phonology. Reports from Uppsala University Department of Lin
guistics, 4.
. 1976. "Is Linguistics an Empirical Science? Some Notes on Esa

Itkonen's Linguistics and Metascience" . Studia Linguistica 30:1.77-
94.
Lorenz, Kuno. 1970. Elemente der Sprachkritik. Frankfurt/Main: Suhr-

kamp.
. 1973. "Die dialogische Rechtfertigung der effektiven Logik".

In Kambartel & Mittelstrass 1973:250-80.
Lorenzen, Paul. 1969a. Normative Logic and Ethics. Mannheim: Biblio

graphisches Institut.
. 1969b. Systematisches Denken. Frankfurt/Main: Suhrkamp.
Lorenzer, Alfred. 1973. Heber den Gegenstand der Psychoanalyse. Frank

furt/Main: Suhrkamp.
Lukes, Steven. 1968. "Methodological Individualism Reconsidered".

British Journal of Sociology 19.119-29.
Maclntyre, Alasdair. 1967a. "The Idea of a Social Science". Proceedings

of the Aristotelian Society, Suppl. 41.95-114.
. 1967b. A Short History of Ethics. London: Routledge & Kegan

Paul.
Mackie, J. L. 1974. The Cement of the Universe. A Study of Causation.

London: Oxford Univ. Press.
Manninen, Juha, and Raimo Tuomela, eds. 1976. Essays on Explanation and
Understanding. Dordrecht-Holland: Reidel.
Marcuse, Herbert. 1964. One-Dimensional Man. Boston: Beacon Press.
Martindale, Don. 1960. The Nature and Types of Sociological Theory.

London: Routledge & Kegan Paul.
Marx, Karl, and Friedrich Engels. 1973[1932]. Die Deutsche Ideologic

Werke, vol. 3. Berlin: Dietz.
Mates, Benson. 1958. "On the Verification of Statements about Ordinary

Language". Inquiry 1:161-71.
Mauthner, Fritz. 1923. Beiträge zu einer Kritik der Sprache, vol. 1.

3rd ed. Leipzig: F. Meiner.
McCawley, James D. 1972. "A Program for Logic". In Davidson & Harman
1972:498-544.
REFERENCES 343
Mead, Georg H. 1934. Mind, Self, and Society. Chicago: Univ. of Chicago
Press.
Mehtonen, Lauri. 1971. "Some Reflections on Theory and Practice in

Peter Winch's Idea of Science". Âjatus 33.274-79.
Miller, George Α., and Stephen Isard. 1967. "Some Perceptual Con-
cequences of Linguistic Rules". In Jakobovits & Miron 1967:219-31.
(Originally published in Journal of Verbal Learning and Verbal Be
havior 2, 1963. References are to the Jakobovits & Miron edition).
Mittelstrass, Jürgen. 1974a. Die Möglichkeit von Wissenschaft. Frank

furt/Main: Suhrkamp.
. 1974b. "Erfahrung und Begründung". In Mittelstrass 1974a:

56-83.
. 1974c. "Wider den Ding1er-Komp1ex". In Mittelstrass 1974a:

84-105.
Moore, G. E. 1912. Ethics. London: Oxford Univ. Press.
Moravcsik, J. M. E. 1967. "Linguistic Theory and the Philosophy of

Language". Foundations of Language 3:2.209-33.
Motsch, Wolfgang. 1974. Zur Kritik des sprachwissenschaftlichen

Strukturalismus. Berlin: Akademie-Verlag.
Naess, Arne. 1952. "Toward a Theory of Interpretation and Preciseness".

Semantics and the Philosophy of Language ed. by Leonard Linsky,
249-69. Urbana: The Univ. of Illinois Press.
Nagel, Ernest. 1961. The Structure of Science. New York: Harcourt,

Brace & World.
Neuendorff, Hartmut. 1973. Der Begriff des Interesses . Frankfurt/Main:

Suhrkamp.
Palmer, Richard E. 1969. Bermeneutics. Evanston: Northwestern Univ.

Press.
Pap, Arthur. 1958. Semantics and Necessary Truth. New Haven, Conn.:
Yale Univ. Press.
Peirce, Charles S. 1958. Collected Papers, vol. 7, ed. by A. W. Burks.

Cambridge, Mass.: Harvard Univ. Press.
Peters, Stanley. 1972. "The Projection Problem: How Is a Grammar to Be

Selected". Goals of Linguistic Theory, ed. by S. Peters, 171-88.
Englewood Cliffs, N. J.: Prentice-Hall.
Pitcher, George, ed. 1966. Wittgenstein. The Philosophical Investiga-

tions .· Garden City: Doubleday,
Rosenbaum, Peter S. 1967. The Grammar of English Predicate Complement
Constructions. Cambridge, Mass.: MIT Press.
Ross, John R. 1968. Constraints on Variables in Syntax. Bloomington:

Indiana Univ. Linguistics Club, mimeo.
Rossi, Sirpa. 1972. "Ueber das Problem der Verdinglichung bei Lukacs
und Heidegger". Ajatus 34.155-61.
Runciman, W. G. 1969. Social Science and Political Theory. London:

Cambridge Univ. Press.
Ryan, Alan. 1970. The Philosophy of the Social Sciences. London: Mac
millan.
Sampson, Geoffrey. 1975. The Form of Language. London: Weidenfeld &

Nicolson.
. 1976. Review of Koerner (1975). Language 52:4.961-65.
Sapir, Edward, no date [l92l]. Language. New York: Harcourt, Brace &
World, Harvest Book.
. 1949a. Selected Writings in Language, Culture, and Persona-

lity ed. by D. G. Mandelbaum. Berkeley: Univ. of California Press.
. 1949b. "Sound Patterns in Language". In Sapir 1949a:33-45.
. 1949c. "The Psychological Reality of Phonemes". In Sapir
1949a:46-60.
. 1949d. "Communication". In Sapir 1949a:104-09.
. 1949e. "The Grammarian and His Language". In Sapir 1949a:

150-59.
. 1949f. "The Status of Linguistics as a Science". In Sapir
1949a:160-66.
. 1949g. "Anthropology and Sociology". In Sapir 1949a:332-45.
. 1949h. "The Meaning of Religion". In Sapir 1949a:346-56.
. 1949i. "The Unconscious Patterning of Behavior in Society".

In Sapir 1949a : 544-59.
. 1949j. "Psychiatric and Cultural Pitfalls in the Business
of Getting a Living". In Sapir 1949a:578-89.
Saumjan, Sebastian K. 1971 [1965 ]. Principles of Structural Linguistics.

Transl, by J. Miller. The Hague: Mouton.
REFERENCES 345
Plato. 1963a. The Collected Dialogues. Edited by Edith Hamilton and

Huntington Cairns. Princeton, N. J.: Princeton Univ. Press,
. 1963b. Phaedo. In Plato 1963a:40-98.
. 1963c. Meno. In Plato 1963a : 353-84.
Popper, Karl R. 1963. Conjectures and Refutations. London: Routledge &

Kegan Paul.
. 1965 [1934 ]. The Logic of Scientific Discovery. New York:

Harper & Row.
. 1972. Objective Knowledge. London: Oxford Univ. Press.
Postal, Paul M. 1964. Constituent Structure, International Journal of

American Linguistics 30:1, part II.
. 1968a. Aspects of Phonological Theory. New York: Harper &

Row.
. 1968b. Epilogue to Jakobs & Rosenbaum 1968, pp. 267-89.
Prior, A. N. 1954. "The Paradoxes of Derived Obligation". Mind 63.64-

65.
Quine, Willard Van Orman. 1960. Word and Object. Cambridge: MIT Press.
. 1969. Ontological Relativity and Other Essays. New York:

Columbia Univ. Press.
Radner, Michael, and Stephen Winokur, eds. 1970. Minnesota Studies in

the Philosophy of Science, vol. 4. Minneapolis: Univ. of Minnesota
Press.
Radnitzky, Gerard. 1970. Contemporary Schools of Metascience. 2nd ed.

Göteborg: Akademiförlaget.
. 1976. "Popperian Philosophy of Science as an Antidote

against Relativism". Essays in Memory of Imre Lakatos ed. by R. S.
Cohen et al., 505-46. Dordrecht-Holland : Reidel.
Rescher, Nicolas, ed. 1967. The Logic of Decision and Action. Pitts
burgh: Univ. of Pittsburgh Press.
Ringen, Jon D. 1975. "Linguistic Facts". In Cohen & Wirth 1975:1-41.
. 1977. Review of Esa Itkonen's Linguistics and Metascience.

Language 53:2.411-17.
Robinson, W. P. 1972. Language and Social Behavior. Harmondsworth,

Middlesex: Penguin Books.
Saunders, John T., and Donald F. Henze. 1967. The Private-Language
Problem. New York: Random House.
Saussure, Ferdinand de. 1957. "Cours de linguistique generale (1908-

1909): Introduction". Ed. by R. Godei. Cahiers Ferdinand de
Saussure 15:6-103.
. 1962. Cours de linguistique générale. 5th ed. Paris: Payot.
Scheff1er, Isaac. 1963. The Anatomy of Inquiry. New York: Alfred Knopf.
Schiffer, Stephen. 1972. Meaning. London: Oxford Univ. Press.
Schlieben-Lange, Brigitte, ed. 1975a. Sprachtheorie. Hamburg: Hoffmann

& Campe.
. 1975b. "Metasprache und Metakommunikation". In Schlieben-

Lange 1975a:189-205.
Schneider, Hans J. 1975. Pragmatik als Basis von Semantik und Syntax.
Frankfurt/Main: Suhrkamp.
Schnelle, Helmut. 1973. Sprachphilosophie und Linguistik. Hamburg:

Rowohlt.
. 1976. "Kommentar zu Itkonen 1976c". In Wunderlich 1976:80-

81.
Schutz, Alfred. 1962. Collected Papers, vol. 1: The Problem of Social

Reality. The Hague: Nijhoff.
Searle, John R. 1969. Speech Acts. London: Cambridge Univ. Press.
Skousen, Royal. 1974. Substantive Evidence in Phonology. The Hague:

Mouton.
Slobin, Dan I. 1971. Psycholinguistics. Glenview, 111.: Scott & Fores-

man.
Sosa, Ernest, ed. 1975. Causation and Conditionals. London: Oxford

Univ. Press.
Specht, Ernst K. 1969. The Foundations of Wittgenstein's Late Philo-

sophy. Transl, by D. E. Waiford. Manchester: Manchester Univ. Press.
Stegmüller, Wolfgang. 1970. Probleme und Resultate der Wissenschafts-

theorie und Analytischen Philosophie, vol. 2: Theorie und Erfahr-
ung. Berlin: Springer.
. 1974. Probleme und Resultate der Wissenschaftstheorie und

Analytischen Philosophie, vol. 1: Wissenschaftliche Erklärung und
Begründung. 2nd ed. Berlin: Springer.
REFERENCES 347
Steinberg, Danny D., and Leon A. Jakobovits, eds. 1971. Semantics.

London: Cambridge Univ. Press.
Steinberg, Danny D., and Robert K. Krohn. 1975. "The Psychological Va
lidity of Chomsky and Halle's Vowel Shift Rule". In Koerner 1975:
233-59.
Stoutland, Frederick. 1976. "The Causal Theory of Action". In Manninen

& Tuomela 1976:271-304.
Strawson, P. F. 1963. Individuals. 2nd ed. Garden City: Doubleday.
. 1971. Logico-Linguistic Papers. London: Methuen.
Stroud, Barry. 1965. "Wittgenstein and Logical Necessity". Philosoph-

ical Review 74.504-18.
Taylor, Charles. 1964. The Explanation of Behavior. London: Routledge &

Kegan Paul.
Turner, Roy, ed. 1974. Ethnomethodology. Harmondsworth, Middlesex:

Penguin Books.
Uldall, H. J. 1957. Outline of Glossematics. Copenhagen: Universitets

førlag.
Wall, Robert. 1972. Introduction to Mathematical Linguistics. Englewood

Cliffs: Prentice-Hall.
Wang, J. T. 1972. "Wissenschaftliche Erklärung und generative Gramma

tik". Linguistik 1971. Referate des 6. linguistischen Kolloquiums ,
Kopenhagen 1971 ed. by K. Hylgaard-Jensen, 50-66. Frankfurt/Main:
Athenäum.
Wason, P. C. 1964. "The Effect of Self-Contradiction on Fallacious

reasoning". Quarterly Journal of Experimental Psychology 16.30-34.
Wason, P. C , and P. N. Johnson-Laird. 1972. Psychology of Reasoning.

Cambridge: Harvard Univ. Press.
Weber, Max. 1968. Gesammelte Aufsätze zur Wissenscha.ftslehre. 3rd ed.,

by J. Winckelmann. Tübingen: Mohr.
Weydt, Harald. 1975. "Das Problem der Sprachbeschreibung durch Simu

lation". In Schlieben-Lange 1975a:53-80.
Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca: Cornell Univ.

Press.
Wessel, Horst. 1972. "Eine dialogische Begründung logischer Gesetze".

Quantoren, Modalitäten, Paradoxien ed. by H. Wessel, 256-78.
Berlin: VEB.
Vico, Giambattista. 1968[1744]. The New Science. Transi, by Thomas G.
Bergin and Max H. Fisch. Ithaca, N. Y.: Cornell Univ. Press.
Winch, Peter. 1958. The Idea of a Social Science. London: Routledge &
Kegan Paul.
Wittgenstein, Ludwig. 1958. Philosophical Investigations. 2nd ed.

Transl, by G. Elisabeth M. Anscombe. Oxford: B. Blackwell.
. 1965. The Blue and Brown Books. 2nd ed. New York: Harper &
Row.
. 1967. Remarks on the Foundations of Mathematics. 2nd ed.

Transl, by G. Elisabeth M. Anscombe. Oxford: B. Blackwell.
. 1969a. Tractatus logico-philosophicus. 7th ed. Frankfurt/

Main: Suhrkamp.
. 1969b. On Certainty. Transl, by Denis Paul and G. Elisabeth
M. Anscombe. Oxford: B. Blackwell.
Wright, Georg Henrik von. 1951. "Deontic Logic". Mind 60.1-15.
. 1968. An Essay in Deontic Logic and the General Theory of

Action. Amsterdam: North-Holland Publ. Co.
. 1971a. "A New System of Deontic Logic". In Hilpinen 1971:
105-20.
. 1971b. Explanation and Understanding. London: Routledge &

Kegan Paul.
Wunderlich, Dieter, ed. 1976. Wissenschaftstheorie der Linguistik.

Frankfurt/Main: Athenäum.
INDEX OF AUTHORS
A. C.
Abel, T.: 204 Campbell, L.: 89
Abelson, R. P.: 315n.l7 Campbell, N.: 262
Andersen, H.: 89 Carnap, R.: 1, 68, 70, 83
d'Andrade, R. G.: 308 Cavell, S.: 166, 211-12, 326n.93
Andresen, H.: 89 Chihara, Ch.: 114
Anscombe, G. Ε. Μ.: 197, 322η.66 Chisholm, R. M.: 283-84
Anttila, R.: 89, 329n.l25 Chomsky, N.: 71-73, 75-87, 114, 117,
Apel, K.-0.: 21, 47, 64, 88-89, 144-45, 176-78, 181, 187-89, 226,
93, 123 230, 233-34, 241, 243, 260, 263,
Aristotle: 261, 296, 299, 306, 271-73, 316n.23, 317n.30, 317n.31,
313n.l, 329n.l23, 330n.l30 318n.36, 321n.57, 328n.lll
Ayer, A.: 91 Cicourel, A.: 27
Collingwood, R. G.: 140, 324n.76
B. Copi, I. M.: 288
Coseriu, E.: 150-51
Bach, E.: 77-78, 167, 227, 260,
317n.29, 317n.33
D.
Bailey, Ch.-J.: 325n.84
Bannister, D.: 314n.l0, 317n.28 Dahl, Ö.: 325n.89, 326n.91
Barker, S. F.: 8 Davidson, D.: 136
Becker, E.: 317η.28 Derwing, B. L.: 88, 187
Bennett, J.: 183, 185 Descartes, R.: 62, 91, 144, 318n.39
Berger, P.: 37 Dilthey, W.: 40, 43, 46-47, 67, 94,
Berkeley, G.: 91 129-30, 139, 195, 318n.38, 319n.40,
Bever, Th. G.: 79, 84-85, 109, 323n.75a
115, 117, 149, 153, 321η.125 Dingler, H.: 46
Bickerton, D.: 325η.84 Donnelian, K.: 320η.50
Black, M.: 212 Dougherty R. C : 84-85, 327η.107
Blalock, H. M.: 25 Douglas, M.: 315η.13
Bloomfield, L.: 68-71, 75, 84, 87, Drachman, G.: 318η.35
89-90, 176 Dretske F. I.: 89, 327η.104
Bocheñski, I. M.: 12 Durkheim, E.: 41, 129, 139, 292
Böhme, G.: 45
Botha, R. P.: 87, 89, 152, 187, E.
239
Engels, F.: 94, 318n.38, 319n.40,
Boudon, R.: 25
319n.42, 319n.47
Brentano, F. C.: 107
Esterson, A.: 31
Bubner, R.: 48
Buddha: 321n.56
F.
Feigl, H.: 14 Hsieh, Η-I.: 89
Feuerbach, L.: 322n.63 Humboldt, W. von: 62, 70, 118,
Feyerabend, P.: 16, 261, 326n.96, 321η.59
329n.l24 Hume, D.: 91
Fodor, J. Α.: 85, 114-18, 221-22, Husserl, E.: 37, 46-47, 53-54, 95,
229, 320n.55, 326n.93 108, 133, 214, 313η.1, 319η.40
Føllesdal, D.: 285, 293 Hutchinson, L. G.: 89, 187
Fransella, F.: 314η.10, 317n.28 Hymes, D.: 120, 174, 315n.l3,
Freud, S.: 30 316n.24, 322n.63, 328n.l09
Friedman, H. R.: 132, 182, 233
I.
G.
Immler, M.: 328n.ll2
Galileo: 3, 261, 313n.l, 329n.l24 Isard, S.: 205
Garrett, M. F.: 85, 115 Itkonen, T.: 329n.l25
Geach, P.: 284
Grice, H. P.: 119 J.
Gumb, R. D.: 205-07
Johach, H.: 323n.75a
Johnson-Laird, P. N.: 315n.l8
H.
Juntunen, M.: VI, 328n.l08
Habermas, J.: 20-21, 24, 30, 33,
47, 88, 120, 195, 323n.71 K.
Halle, M.: 316n.23 Kac, M. B.: 89
Hanna, J. F.: 305 Kambartel, F.: 16
Hare, R. M.: 166, 211-13, 260, Kamlah, W.: 45, 48-49, 52
324n.80, 325n.88 Kanngiesser, S.: 239-40, 327n.l00,
Harjula, R.: 319n.40 328n.ll7
Harman, G.: 136 Kaplan, A.: 260
Harré, R.: 8, 313n.3, 314n.8, Kasher, Α.: 178-81, 187, 235,
31411.10, 321n.56a, 322n.65, 327η.99, 327η.100, 327η.101,
323n.69 328η.117
Harris, Ζ. S.: 71-77, 89-90, Katz, J. J.: 79-80, 83-84, 109, 117,
165-66, 176, 220, 317n.29, 221, 273-74, 326η.93, 329η.127
317n.30 Kenny, A.: 111, 144
Hart, H. L. A.: 207-08 Kiparsky, C : 264-71
Hegel, G. F. W.: 47, 129-30, 301 Kiparsky, P.: 216-17, 264-71
Heidegger, M.: 95, 319n.40, Kockelmans, J. J.: 44
324n.77 Koerner, E. F. K.: VI, 58
Hempel, C. G.: 1, 4-5, 7, 12, Kohler, W.: 329n.l25
14, 46, 77, 79, 83, 156, Korner, S.: 212
181, 200, 229-30, 313n.5 Krohn, R. K.: 89
Henle, M.: 49 Kuhn, T. S.: 326n.96
Henson, R.: 211-12
Henze, D. F.: 91, 107, 318n.40 L.
Hilpinen, R.: 285, 293 Labov, W.: 138, 153, 325n.84,
Hintikka, J.: 51-52, 194, 197-98, 327η.101, 327n.105
288-91, 316n.19, 330n.l29 Laing, R. D.: 31
H j e l m s l e v , L . : 5 9 - 6 2 , 87, 8 9 - 9 0 , Lakatos, I.: 314n.6
118, 316n.21, 316n.22, 322n.60 Lakoff, G.: 217, 237-38, 242-43, 270,
Hobbes, T.: 194, 197 290, 328η.116, 329n.125
Householder, F.: 220, 318η.37 Lakoff, R.: 237
351
INDEX OF AUTHORS
Lass, R.: 89, 327n.l04 P.
Lauer, Q.: 318n.38
Leech, G.: 167 Palmer, R. E.: 31
Lees, R. Β.: 76, 317n.32 Pap, Α.: 17, 280, 294, 300-03, 306-08
Leibniz, G. W.: 324n.77 Peirce, Ch. S.: 9-10, 47, 314n.6
Lenneberg, Ε. Η.: 83 Peters, S.: 245
Leontiev, A. A.: 119 Plato: 194, 198, 213-14, 294-99, 301,
Lessnoff, M.: 315n.l5, 325n.85 306
Lewis, D.: 183-87, 201, 276, Popper, K. R.: 1-2, 8, 16, 36, 47,
322n.67, 327n.l03 77, 79, 89, 131, 146, 156, 252,
Lieb, H.-H.: 79, 322n.61 313n.5, 314n.6
Linell, P.: 89 Postal, P. M.: 77, 79, 83, 86, 219,
Locke, J.: 91 245, 316n.23
Lorenz, K.: 44, 46, 49-52, 276, Prior, A. N.: 283-84
316η.19, 327η.102
Lorenzen, P.: 42, 45, 48-52, Q.
214, 276, 316n.19 Quine, W. V. 0.: 128, 227, 306-07,
Lorenzer, Α.: 30 320n.52
Luckmann, Th.: 37
Lukes, S.: 127 R.
Radnitzky, G.: 21, 47, 313n.3
M.
Reichenbach, Η.: Ι
Maclntyre, Α.: 323n.69, 323n.73 Rescher, N.: 280
Mackie, J. L.: 319n.44 Rimbaud, Α.: 321η.56
Marcuse, H.: 172-73 Ringen, J. D.: 89, 319n.44, 324η.77,
Martindale, D.: 2 327n.104
Marx, K.: 94, 318n.38, 319n.40, Robinson, W. P.: 201
319n.42, 319n.47, 322n.63 Rosenbaum, P. S.: 269, 271
Mates, B.: 286 Ross, A.: 286
Mauthner, F.: 113, 129 Ross, J. R.: 271-72
McCawley, J. D.: 217 Rossi S.: 318n.38
Mead, G. H.: 37-38, 62, 64, 94, Royce, J.: 47
106-07, 316n.26, 318n.38, Runciman, W. G.: 27
319n.40, 322n.67 Russell, Β.: 91, 133, 154, 287-88
Mehtonen, L.: VI, 202 Ryan, Α.: 155, 157, 315n.15, 325n.85
Miller, G. A.: 205
Mittelstrass, J.: 4, 16, 44, 46, S.
261, 313n.1
Montague, R.: 220 Sampson, G.: 87-88, 176-77, 179,
Moore, G. E.: 314n.6 187-88, 326n.98
Moravcsik, J. M. E.: 117-18 Sapir, E.: 61-67, 69, 84, 89-90, 101,
Motsch, W.: 318n.35 173, 316n.24, 316n.25, 316n.27
Saumjan, S. K.: 87, 89-90
N. Saunders, J.T.: 91, 107, 318n.40
Saussure, F. de: 55-59, 62-63, 68, 81,
Naess, A.: 286 84, 89-90, 138-39, 189, 220,
Nagel, E.: 1-2, 5, 15, 79, 127, 316n.25
204 Scheffler, I.: 83
Neurath, 0.: 1, 68 Schiffer, S.: 322.67
Neuendorff, H.: 194 Schlieben-Lange, Β.: 126-27
Nicod, J.: 288 Schneider, H. J.: 120
Schnelle, H.: 21, 320n.55, W.
327n.100, 327n.102
Schutz, Α.: 37-38, 62, 117, Wall, R.: 79, 276
200-01, 204, 218, 322n.67 Wang, J. T.: 235
Searle, J. R. : 119, 212 Wason, P. C : 49, 315n.l8
Secord, P. F.: 314n.l0, 321n.56a, Weber, M.: 26, 35, 155, 323n.75a
322n.65, 323n.69 Weiss, A. P.: 68
Skousen, R.: 89 Weydt, H.: 89
Slobin, D. I.: 189, 218-19 Wessel, H.: 316n.l9
Socrates: 294-98, 330n.l31 Whitehead, Α.: 154, 287-88
Sosa, E.: 307 Wiener, Ν.: 306-07
Specht, E. Κ.: 212, 214 Winch, P.: 53, 62, 64-66, 202,
Stegmülier, W.: 13-14, 313n.2 315η.15, 319η.43, 323η.69
Steinberg, D. D.: 89 Wittgenstein, L.: 16, 23, 46-47, 64,
Stoutland, F.: 60 66, 93-97, 100-01, 104-08, 111-14,
Strawson, P. F.: 119, 318n.40 137-38, 141-49, 153-54, 227, 231,
Stroud, B.: 147 260, 297, 318η.40, 319n.48,
320n.49, 320n.55, 321n.56,
Τ. 322n.62, 322n.63, 329n.122
Wright, G. H. von: 4-5, 16, 60,
Taylor, Gh.: 106-07 20383,101-02,107,173,282-86,304,
Trubetzkoy, N. S.: 62
Turner, R.: 315n.l3 313n.l
U.
Uldall, H. J.: 61
V.
Vendier, Z.: 212
Vico, G.: 195
INDEX OF TERMS
a c t i o n 17, 4 1 , 60, 98-102, empirical, definition of 2-4, 16-18,
105-08, 115-20, 1 2 2 - 2 5 , 1 8 2 - 8 3 , 29, 36, 80, 146, 155-57
193-94, 319n.44
event 2-4, 25, 99, 115-16, 122-24,
anamnesis 198, 213-14, 298-99, 182-83, 193-94, 319n.44
301, 3 2 9 n . l 2 4
explanation
atomism, thesis of 16-18, 97-102, empirical 4-11, 24-26, 81, 84,
106-07, 160, 319n.46 89, 228-30, 292, 313n.2, 313n.3
nonempirical 65, 71-72, 78,
basic statement 233-45, 264, 266-69, 287-93, 295
in grammar 167-68, 250, 254
in natural science 2, 16, 167 explication 17-18, 35-36, 39, 42-48,
81, 188, 215, 220, 261, 272-75,
behaviorism 68-70, 105-07, 294-310, 315n.l5
114-15, 117, 203
falsification
causality 5, 15, 17, 21, 24-25, see testing
31-36, 83, 115-17, 195, 202-03,
221, 292, 314n.8, 315n.11, generalisation 73, 79, 189, 217,
319n.44 220, 239-43, 264-65, 268-69,
287-88, 291-93, 295-96
certainty
vs. uncertainty 40, 141-54, grammatical rule/hypothesis 72, 89,
194-97, 301, 324n.79, 326n.91 162, 181, 215-18, 224-28, 251-56,
vs. feelings of certainty 96, 329n.ll9
148
introspection 104, 324n.77
common knowledge 41-42, 96, 111,
122-31, 139-41, 182-86, 208, intuition
259, 322n.67, 323n.70, 324n.79 as recollection 104, 212-14, 222,
254, 257
conceptual analysis see also anamnesis
see explication linguistic 56, 69, 77, 131-41,
257-59, 280, 291-92, 324n.77,
confirmation 324n.82
see testing logical 133-36, 279-87, 291-92
philosophical 133-36, 286, 289,
disconfirmation 301-07
see testing
knowledge intuitive vs. formal 17, 280-87,
acquisition vs. mastery of 37, 302-10
40, 43, 64, 182, 186, 209-10,
299-300, 328n.ll0 norm
agent's vs. observer's 40, 43, see rule
190, 193-98, 214-15, 279, 298
atheoretical vs. theoretical normativity
103-04, 126, 144, 148, vs. factuality 89, 157, 185
151-54, 161-62, 166-68, irreducibility of 35, 125, 128,
179-80, 198-219, 241, 246-47, 149, 175-90, 207-08
269, 300-01, 326n.91,
328n.l08 observation
conscious vs. unconscious 36, vs. intuition 2-4, 57, 69, 77,
82, 96, 116-17, 217-27, 84-85, 197, 206
298-300, 325n.90 vs. theory 1-3, 12-16, 178-79
conventional vs. psychological vs. understanding 20, 22-23, 26-27,
134, 139-40, 323n.75a 63-64, 69, 92-93, 96-97, 105-08,
critique of 33, 47-48 114, 132, 193, 320n.49
experimental vs. pre-experimental
22-24, 42-43, 145-46, 188, operationalisation 2-3, 23-26, 29,
324n.82, 329n.l21 43, 64-65, 176-81
incorrigible, nonexistence of
137, 147, 324n.79 prediction
objective v s . subjective 57, see explanation, testing
91-96, 113-14, 127-28,
137-41, 259, 322n.67, 323n.73 psychological reality
of grammar 88-89, 116-17, 219-23,
language 328n.ll4
artificial vs. natural 24, 126, of formal logic 315n.l8
153, 169-70, 190-91
private vs. public 94, 109-13, recursivity 70-71, 86, 136, 183-84,
118-19, 126-27, 270, 320n.55, 247, 318n.36
321n.56, 322n.61
see also rule of language reflexion 20, 23-24, 33, 44, 47-48,
126-28, 132, 195, 212-16, 220,
logic 314n.l0
axiomatic 53, 154, 236, 253,
276-88 regularity
dialogical/game-theoretic/ see rule
pragmatic 48-54, 276, 316n.l9
model-theoretic/semantic 128, rule
276, 280 as known or knowable 37, 150, 161,
social origin of 48-49, 53-54, 186, 218, 263, 315n.l4
147, 159 in general 34-54, 111-12, 322n.65,
322n,67, 323n.69, 328n.ll0
mentalism 82-84, 114-15, 187, of language 57-58, 89, 112,
221-23, 317n.34a 122-68, 182-86, 189-91, 266,
324n.81
necessity vs. regularity 58, 156-66, 173-74,
conceptual vs. natural 17-18, 182-86, 189, 206-07, 263
97-104, 119, 124, 134, 140,
160, 183-84, 236-39, 245-50, rule-sentence 122, 148, 156-74,215-16,
254-57, 259, 273, 319n.44 249-52, 308-10, 322n.65, 325n.86
355
INDEX OF TERMS
science statistics, use of 134, 151, 161,
axiomatic 14, 79, 242, 261-62, 185, 209, 230-31, 270, 322n.64
277-78
critical/emancipatory 32-33, testing
38, 67, 172-74, 204 empirical 4-12, 29, 36, 69,
descriptive vs. prescriptive 84-85, 156-57, 313n.5
45-46, 53-54, 174, 200, 280, nonempirical 36, 39-40, 65,
305 71-72, 78-79, 156-66,
empirical vs. nonempirical 1-4, 245-262, 269-75, 280-87,
19, 44, 80, 153-74, 231, 294-297, 304-05, 310,
260-64, 273-75, 326n.92 314n.6, 328n.ll8
growth of 10-12, 74-75, 171-74,
326n.96 theoretical concepts
human vs. natural 21, 28-29, in natural science 13-15, 64-65,
32-33, 43, 55, 60-61, 64-65, 103, 177-78
85-86, 92-93, 187-88, 195, in grammar 73, 76-78, 178-81,
198-203 225-27, 328n.ll4
preconditions of 20, 45-48
value-free 67, 173-74, 225-26 understanding
see observation
sentence vs. utterance 82, 120,
132, 175-81, 190, 229, 233, variability
244, 248-49, 323n.75, 328n.ll5 of knowledge in general 16, 28-30,
147, 286, 306, 315n.l2
social control of language 151-54, 158-59, 270,
as precondition of knowledge 286, 324n.81, 324n.83, 324n.84
95, 111-13, 125-27, 151-54
vs. testability 324n.82

Grammatical Theory and Metascience PDF

Uploaded by

Copyright:

Available Formats

Grammatical Theory and Metascience PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Grammatical Theory and Metascience PDF

Uploaded by

Copyright:

Available Formats

GRAMMATICAL THEORY AND METASCIENCE

AMSTERDAM STUDIES IN THE THEORY AND

E.F.K. KOERNER, General Editor

Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board

Henning Andersen (Copenhagen); Raimo Anttila (Los Angeles)

Grammatical Theory and Metascience

INTO THE METHODOLOGICAL AND PHILOSOPHICAL

FOUNDATIONS OF 'AUTONOMOUS' LINGUISTICS

AMSTERDAM / JOHN BENJAMINS B.V.

Helsinki, December 1977 Esa Itkonen

TABLE OF CONTENTS VII

1.0. THE IDEA OF 'POSITIVISM' 1

1.3. Comparison with Peirce's Logic of Science 9

1.4. Theory and Observation 12

2.0. THE IDEA OF 'HERMENEUTICS' 20

2.4. Sociology of Knowledge 33

4.0. PHILOSOPHICAL FOUNDATIONS OF THE CONCEPT OF LANGUAGE 91

4.2. Refutation of the Traditionist Epistemology 94

4.2.1. The Axiomaticity of the Concepts of Person and Thing . . . . 94

4.2.2. Mind, Behavior, and Environment 96

4.2.3. Characterisation of Mental Phenomena; the Notion of

Employed in the Present Study . 108

4.2.5. The Impossibility of Private Languages 109

4.3. Implications for Linguistic Theory 113

4.3.1. Psycho Unguis tics 113

5.4. Rules of Language and Social Control 151

6.0. THE BASIS OF THE NONEMPIRICAL NATURE OF GRAMMAR 155

6.3. Two Different Types of Rule-Sentence 168

7.0. THE INELIMINABILITY OF LINGUISTIC NORMATIVITY 175

7.2. Grammatical Concepts Are not Comparable to Theoretical

7.3. Rules Are not Regularities of Non-Normative Actions 182

7.4. Grammatical Descriptions Cannot Be Replaced by Psycho-

7.5. The Position of Transformational Grammar vis-à-vis

8.0. LANGUAGE AND GRAMMAR 192

8.2. The Two-Level Nature of the Human Sciences: Atheoretical

vs. Theoretical 198

8.4. The Ontological Reality of Grammatical Descriptions 219

9.0. THE METHODOLOGY OF GRAMMAR 228

9.5. Appendix: Examples Taken from the Transformationalist

10.0. GRAMMAR AND LOGIC 276

and Systems of Logic 276

11.2. The Concept of Explication 301

INDEX OF AUTHORS 349

INDEX OF TERMS 352

with pure observation tout c o u r t ) . In most 'observational' sciences,

1.2. Explanation, Prediction, and Testing

According to standard positivism, "science is interested in estab­

The standard type of a D-N explanation is a causal explanation,

However, this principle cannot be taken in an absolute sense, as

tinction I want to make here is not primarily that between prediction

1.3. Comparison with Peirce's Logic of Science

We can sum up the methodological notions introduced so far by

In this example the prediction 'Ga' turns out to be false, i.e.,

ing that deduction determines what must be as a matter of logical ne­

However, all the three above-mentioned operations are clearly distinct

In predictive falsification (cf. A ) , we start from 'p' and derive

The preceding sketch of the basic methodological notions under­

what more r e a l i s t i c p i c t u r e can be gathered from analysing the i n t e r -

According to standard positivism, "science is interested in estab

ing that deduction determines what must be as a matter of logical ne

The preceding sketch of the basic methodological notions under

they contain, a prediction is always deduced jointly from several hypo

mas (1970) offers a good exposition of the hermeneutic modes of think

In the most general terms, hermeneutics says that it makes a dif

Just like experimental psychology, main-stream or Durkheimian so

possibility of psychoanalysis rests on the fact that the patient's langu

So macht denn jede gelegentliche (oder auch "philosophische") Rück

Die fundamentalen Voraussetzungen der Erkenntnis sind im Leben ge

La synchronie ne connaît qu'une perspective, celle des sujets par

In any event, Hjelmslev, with a view to avoiding the alleged weak

Observation shows us movements and sounds connected with human bo

Sapir concludes by making the proposal that psychiatry should re

It is ... often convenient to make the division into morphemic seg