Cogsci 89 60 PDF

DIACHRONIC REGULARITY AND IRREGULARITY:
STRUCTURAL PARALLELS BETWEEN

SEMANTIC AND PHONOLOGICAL CHANGE
By
Kiki Nikiforidou and Eve Sweetser

Department of Linguistics
University of California at Berkeley
Cognitive Science Program
Institute of Cognitive Studies
June 1989
The production of the Berkeley Cognitive Science Report Series is supported by a

grant in cognitive science to the University of California at Berkeley from the
Alfred P. Sloan Foundation.
-2-
Abstract
Recent work in historical semantics challenges traditional views of meaning-change as

random and irregular, as opposed to the regular, lawlike character of sound change. In this
paper, we argue that not only are there major parallels to be observed between regular
behavior in semantic and phonological change, but further, these parallels seem to be
motivated by a realistic model of psychological categorization processes which are common
to the two domains.
Among the kinds of regularity which have been observed in the history of phonological
systems and which have parallels in the history of semantic systems are:
(1) It is possible to identify directions of change which are recurrent and natural;
(2) units can be observed to change "in parallel," maintaining the contrast relationships
between them, and
(3) completed change affects a unit completely (all instances of the unit are affected).
We examine some of the evidence about the structure of psychological categories in

the semantic and phonological domains, and attempt to show how such evidence motivates
observed diachronic patterns. Further, looking at very general principles of categorization
and generalization, we identify similar metonymic and analogical patterns in phonology and
semantics which are motivated by these general principles.
We also see phonology and semantics to differ in important ways, and suggest ways in
which the differences between the two may be due to the inherent differences between
sound and sense. But in other ways, we show that general cognitive constraints appear to
structure the history of both domains.
DIACHRONIC REGULARITY AND IRREGULARITY:

STRUCTURAL PARALLELS BETWEEN
SEMANTIC AND PHONOLOGICAL CHANGE1
Kiki Nikiforidou and Eve Sweetser

Department of Linguistics
1 Numerous colleagues have furthered the progress of this paper by giving us the benefit of their reactions
to earlier versions. We are particularly grateful to Claudia Brugman, Suzanne Fleischman, John Kingston,
George Lakoff, and John Ohala. And we thank Elizabeth Closs Traugott for the example provided by her
work. None of these people should be imagined to agree with all the opinions here expressed; nor should they
be blamed for the places where we have gone astray by ignoring their advice.
Nikiforidou & Sweetser: 3
0. Introduction.
We began this paper when, with some surprise, we reached the conclusion that we had
no proof that meaning change is less regular than sound change. We still lack full data on
the question of regularity in semantic change, but the last couple of decades have brought a
wealth of new cognitive and linguistic knowledge to bear on both semantics and phonology.
This new data tends more and more to call into question classical views of historical change,
and to highlight parallels between synchronic and diachronic structure in these two disparate
areas.
It is traditional to regard phonological change and semantic change as two poles of a
spectrum: the former is the stronghold of regularity and systematicity, the latter is the home
of whimsical and unsystematic phenomena. There are at least three ways in which regularity
has been observed to manifest itself in the phonological domain:
(1) It is possible to identify recurring directions in sound change (particularly for condi-
tioned sound change) which are more natural than others (e.g., palatalization before
front vowels).
(2) There are frequent examples of sound change where natural classes of phonological
units change in parallel, maintaining the contrast relationships between them (e.g., in
word-final devoicing of stops to the corresponding voiceless stops, maintaining con-
trast in articulation-point).
(3) When a phonological unit changes its value (i.e., when a phoneme changes its phonetic
shape) all examples of this unit are affected by the change: at least in theory, barring
irregularity, all instances of that phoneme (or all which are in the relevant environ-
ment) throughout the lexicon should be subject to the same change.
Semantic change on the other hand has not traditionally been thought to manifest any of
these aspects of regular behavior and therefore has been taken to be largely random and
arbitrary.
Such a polarized picture is significantly altered by the increasingly supported
hypothesis that much lexical semantic change can be seen as following ‘regular,’ directional
tendencies which show up again and again in the histories of individual lexical items or
even in the histories of semantically defined classes of words. The seed of this hypothesis
can be found in the work of early semanticists like Bréal (1900), Stern (1931) and Ullmann
(1963) and more systematically in recent work by Anderson (1982, 1986), Bybee (1985),
Bybee and Pagliuca (1985), Fleischman (1982a, 1982b), Hopper (1982), Kemmer (1988),
Sweetser (in press), Traugott (1982, 1987, 1989), and Wilkins (1981). We shall not review
this work in detail, but will draw on it in discussing some well-known examples of phono-
logical and semantic change. We will argue that interpreting these examples in the right
way has some startling implications for the notion of diachronic regularity.
Work on recurring directions in semantic change shows that semantic change may
behave regularly in the sense defined in point (1) above. This paper will be concerned with
making explicit some of the mechanisms which motivate such regular historical directions
in both domains. We will further argue that the aspects of regular diachronic behavior
described in (2) and (3) above can also be found in semantics if we reconsider the basis on
which comparisons between the two domains are usually drawn.
It is our purpose to show that there are salient parallels to be drawn between phonolog-
ical and semantic change processes (although semantic change has only recently begun to
be investigated with an eye to possible regularities). We argue that these parallels between
divergent domains reflect the existence of psychologically real categories and category-
formation processes operating similarly in both. We see both phonology and semantics as
constrained by limitations on category formation (including prototype structure, analogy,
and non-objective ‘feature’ parameters of contrast) which motivate regular diachronic and
synchronic behavior and determine those regularities which show up in both kinds of
diachronic change.
We do not wish to claim that there are no differences between semantic and phonologi-
cal change; instead, we argue that to the extent that the differences between them can be
seen as directly derivable from natural properties of the two fields, and to the extent that we
could set aside such differences to expose any commonalities, the possibility of drawing
comparisons and parallelisms between the two domains would be increased. We further
argue that the same set of traditional criteria of regularity support (and call into question)
claims of lawlike behavior in both.
The organization of the paper will be as follows: In section 1, and as a background to
the arguments which follow, we present briefly the kind of semantic and phonological
analysis that we take to be crucial for our argument. Section 2 will present a pair of case stu-
dies in phonological and semantic change, and will argue that there are aspects of semantic
structure which can be argued to parallel closely the classic syntagmatic and paradigmatic
forces influencing phonology and sound change. Section 3 will discuss the general question
of lawlike behavior in both domains. Section 4 will deal with the ways in which basic cog-
nitive mechanisms produce unidirectionality in semantic and phonological change. Section
5 will conclude by discussing some consequences of our claims for the possibility of expla-
nation for linguistic universals.
1. Feature analysis and cognitive structure in semantics and phonology.2

It has been noted before that the type of synchronic analysis dominating the field at a
given time inevitably constrains the possible theories of diachronic change. Weinreich,
Labov, and Herzog (1968), for instance, offer an extensive discussion of the effects of struc-
turalist and generative linguistics on diachronic theory, focusing on the phonological
domain. In semantics, the analysis of lexical meaning into features appears to open at least
one fascinating vista to the historical linguist: the possibility of investigating changes in
feature-structure at large, including both phonological ‘feature-bundles’ and semantic ones.
2 This section reviews current arguments against some traditional views of semantic and phonological
analysis. Since the review is done only as a background to the historical goals of this paper, we cannot pretend
to do more than cite some salient points in this active ongoing debate. Readers are referred to Lakoff (1987)
for a far fuller review of some of the semantic issues.
This would have the major advantage of making the two domains parallel both in syn-
chronic and diachronic analysis.
Ullmann (1974) and others have indeed proposed analyses of semantic change in terms
of change (addition/loss) of features. More recently, Bybee and Pagliuca (1985) see the
semantic changes occurring in grammaticalization processes as invariably involving the loss
or deletion of specific semantic features. Consider for example the diachronic semantic
statement below (from Ullmann, 1974), describing the change of Latin ponere ‘to place’ to
French pondre ‘to lay eggs’:
+‘place’--> +‘place’
+‘eggs’
Similarly, Bybee and Pagliuca in talking about the shift from root to epistemic senses in the
semantics of the modal verbs (a shift we shall discuss in detail below), suggest that the
change involves among other things the loss of the meaning feature ‘animate and willful
subject.’
1.1. Problems with binarity.

Analyzing lexical meaning in terms of binary semantic features has been common
practice among semanticists. The work of the structuralist school in meaning analysis is
well-known (cf. work by Ullmann and others) and so are the later contributions of Katz and
Fodor (1963) and Wierzbicka (1980). Even more recent research, such as that of Jackendoff
(1983, esp. chapters 9 & 10), uses a notion much akin to the semantic feature, namely that
of the ‘semantic function,’ to analyze the meaning of spatial and other semantic fields. The
idea of using feature analysis in semantics is, at first glance, an enticing one, offering
improved rigor in formalizing the description of both synchronic and diachronic aspects of
meaning.
However, extensive semantic literature has recently been devoted to the major
difficulties which ensue from understanding lexical semantics in these terms. The standard
understanding of binary features (cf. Katz & Fodor, 1963) is that they define Boolean condi-
tions on category-membership; by checking down a list of features we can determine
whether some entity or event should be talked about with that word. If and only if all the
feature-values are correct, the entity or event falls in that category. However, such classical
‘if-and-only-if’ approaches to category-structure are inconsistent with the gradient character
of many semantic parameters (e.g., color contrasts; see Kay & MacDaniel, 1978) and the
evidence that category structure involves prototypes (Lakoff, 1987; Mervis and Rosch,
1981; Rosch, 1978). A color may be MORE or LESS red, meaning that the word red applies
to it more or less well—there is no simple cutoff with only indubitable members of the
category red on one side, and only indubitable nonmembers on the other. Further, other
categories (not necessarily defined by gradient parameters) nonetheless show gradient
membership: a robin is a ‘better’ bird than a turkey or an ostrich, and so forth. Speakers
appear to define categories relative to some prototype member or group of members, and
degree of resemblance to these prototypical members will determine degree of membership
in the set (Coleman & Kay, 1981).3
3 Recent work by Brugman (1988) and Lakoff (1987) further demonstrates that many categories exhibit
more structure than just the simple relationship between the prototypical and the other members; to account for
There is also evidence that even granted the possibility of feature analysis for a given
lexical category, different features or parameters are differentially weighted with respect to
category-membership decisions. For example, Coleman and Kay (1981) reported that of the
three parameters relevant to subjects’ decisions about whether a described speech act was or
was not a lie, factual falsity was the least influential. The speaker’s belief in the falsity of
his/her statement was the most important factor, and the speaker’s intent to deceive the
hearer had a weight intermediate between that of the other two factors. The classical feature
theory however does not provide the means to talk about the differential weighting of a
given feature or parameter.
Finally, work by Fillmore (1976, 1985), Sweetser (1987) and others has shown that
word-meaning often exists only embedded in a larger context or ‘frame’ (see the ‘breakfast’
example in section 2.3). This frame itself has structure, independent of the word’s meaning,
and cannot simply be included in a list of the word’s semantic features. But the frame is not
itself a binary feature, although a crucial element of lexical meaning; and it is also not an
objective aspect of meaning, which poses yet further problems for most feature analysts.
1.2. Problems for objectivist semantic analysis.

The bulk of the theoretical work on semantic features (Jackendoff, 1983 is an excep-
tion) has assumed that they are not only binary but also objective in nature. But, as just
mentioned, speakers’ subjective framing of lexical meaning is often an important com-
ponent of semantic structure. Fillmore’s (1976) discussion of the word bachelor shows, for
example, that culture-dependent understanding of marriageable age (not identical with phy-
sical puberty) is part of the frame with respect to which bachelor has meaning. Similarly,
the lexical choice of a preposition on rather than in in phrases like on the bus versus in the
bus reflects (among other factors) the speaker’s framing of the situation as involving either a
vehicle operating "in service" or a vehicle seen as a simple container (Fillmore, 1985): one
cannot say of children playing in an abandoned bus that they are "on the bus" without parti-
cipating mentally in their imaginary travels over an imaginary bus route. Since the spatial
relationship between the passenger and the bus is the same in both cases, such an example
poses problems for an understanding of in and on as encoding objectively defined spatial
configurations.
A further challenge for objectivist semantic analysis comes from recent work on meta-
phor (Brugman, 1983; Lakoff and Johnson, 1980; Lakoff and Turner, 1989; Sweetser, 1987,
in press). This work has seriously challenged the possibility of describing lexical meaning
and polysemy structures in terms of objective features, arguing that in many cases the mean-
ing of a given lexical item can be adequately understood only in terms of the metaphorical
structuring of one sense in terms of another, often more concrete, meaning. For example,
when we use rise, fall, go up or go down in contexts such as My income rose/fell last year,
we use words whose primary senses have to do with verticality and physical motion to talk
this kind of data, these researchers use the idea of radially structured categories, characterized by one or more
prototypical members and structured relationships between the rest.
about changes in abstract quantities (‘more’ and ‘less’ in a non-physical domain). Such uses
are pervasive, and structure polysemy relationships in our lexicon at large (cf. Lakoff &
Johnson, 1980). There is no definable objective criterion by which we could judge abstract
quantities such as income to be similar to physically higher or lower vertical locations:
analysis in terms of objective features will thus fail to explain natural and pervasive aspects
of lexical meaning.
The difficulties of describing meaning change in terms of objective features are evident
from standard examples. Cross-linguistically, it is common for words referring to physical
vision to come to mean ‘knowledge’ and ‘know’ (via a stage of meaning both ‘see’ and
‘know’).4 In describing a meaning change of this kind should we fabricate a semantic
feature [+psychological] or [ physical] to talk about knowledge being mental vision? If so,
what are the objectively definable features SHARED by knowledge and vision (to the exclu-
sion, for example, of other sensory modalities), which remain stable over the domain-shift
achieved by changing the value of the one feature [+psychological]?
Research has been therefore accumulating which shows that analyzing meaning in
terms of binary, objective semantic features cannot always account for the complexity of the
data. Together with other cognitive mechanisms that will be discussed in later sections,
metaphorical analysis gives a way of talking systematically about synchronic polysemy (a
pervasive phenomenon, though neglected in the literature until recently), as well as about
diachronic developments arising out of this polysemy. Historical semanticists agree that
polysemy is a necessary intermediate stage before a word changes (adds or loses) meaning.
A semantic theory which allows for the operation of such (apparently universal) cognitive
mechanisms as metaphor is inherently better equipped than standard feature analysis to deal
with these complex cases of relationships within lexical meaning. By virtue of this, such a
theory is also better equipped to deal with directional tendencies in semantic change, which
in turn greatly contribute to the notion of regularity in diachrony.
1.3. Cognitive category structure in phonology.

It would seem that the enrichment of our understanding of lexical categories entailed
by this kind of research makes semantics look very different from phonology, and removes
the possibility of investigating parallels between synchronic structures or diachronic
developments in the two domains. However, such ‘non-classical’ categorization
phenomena also occur in our phonological categories. Jaeger (1980a, 1980b) provides
experimental evidence to the effect that phonemes are psychologically real categories, as are
at least some traditional phonological features. In some instances, she demonstrates that
these phonological categories manifest prototype structure. For example, talking about the
possible extensions of the [kh] category, she reports that experimental subjects trained to
recognize the [kh] sound showed an unambiguous tendency to respond (in a subsequent
experiment) to the [k] in the [sk] cluster as an instance of the same category. On the other
4 Cf. Sweetser, in press.
hand, their response to the word final [k#] was less clear; subjects responded less strongly
and systematically to this as a member of the same category as [kh]. [kh] thus served as a
central, prototypical member of a category which speakers extended to a broader category
corresponding to something like /k/; not all members of this /k/ category are equal, in that
subjects clearly treated new examples of the original [kh] allophone as ‘best’ members of
the category, followed by [k] in [sk] and by [k#]. Phonological category structure in this
case looks a lot like Rosch’s analysis of psychological category structure.
Although many theoretical phonologists still claim that phonological categories can be
exclusively described in terms of binary features, data like Jaeger’s makes some of the dis-
tinctions look less binary. Jaeger reports that of subjects who successfully formed a category
[+anterior] or [ anterior] based on feedback to their classification of other sounds, some
included almost all [w]’s as [+anterior] and excluded them from [ anterior], but others
classified [w]’s as [ anterior]. Similarly, speakers who had formed a category [+sonorant]
and were successfully accepting undoubted sonorants and rejecting clear non-sonorants,
when asked to classify [h], showed significant variation and disagreement; [h] is presumably
a borderline case. Moreover, certain phonological parameters, such as vowel height, have
long been problems for binary analysis, and may well need at least a multi-valued phonolog-
ical treatment of an evidently gradient phonetic variable.
Yet another form in which category structure manifests itself is through some well-
known implicational universals. Languages elaborate certain areas of their phonology more
than (or temporally prior to) others. Stop systems, for example, are never less elaborated
than fricative systems (stops are possibly more central members than fricatives of the
category of consonants) while no language has /f/ without also having /s/ (here, /s/ is mani-
fested as a more central member than /f/ of the class of fricatives). Such implicational con-
straints are found in semantics as well. No language, for example, has a more elaborate sys-
tem of distinctions in its future tense system than in its past system. Berlin and Kay (1969)
and Kay (1975) have shown that languages do not acquire a basic color term centered on
pink before acquiring a term centered on red. Similarly, Kemmer (1988) has demonstrated
that middle voice systems begin with ‘central’ middle semantics and may gradually spread
to less central but related senses.5
To posit simple Boolean categories in phonology or in semantics would entail attribut-
ing to chance the striking similarities between the ways that change takes place in the two
domains. While feature theory in semantics and phonology may successfully delimit
categories, it does not provide any way for talking about intra-category structure such as the
prototype structures which we have said influence sound and meaning change. The motiva-
tions for /s/ being an earlier fricative than /f/, or for a red term being earlier than a pink
term, may be more directly accessible to physical and perceptual explanations than the rea-
sons for directions of change in middle voice senses. But all these cases demand an
5 Anderson’s (1982, 1986) discussions of ‘mental maps’ argue that semantic change tends to be between
cognitively ‘adjacent’ meanings; this presumably overlaps with prototype structure in semantics, inasmuch as
cognitively adjacent concepts will tend to be grouped together in psychological categories.
explanation in terms of non-classical psychological categories.6 Nor does classical feature

theory explain the fact that subjects in the experiments of Rosch (1978) and Jaeger (1980)
seem to be able to impose boundaries on fuzzy categories and domains—boundaries which
may themselves change depending on the context of the categorizing task.
This discussion of cognitive motivations for category (re)structuring is not intended to
say that change will never follow an unexpected path, but that there are paths of change
which are overwhelmingly predominant. Those paths are often explicable by reference to
the structure of cognitive categories in the relevant domains. Claiming that historical
change follows regular directions is not a claim that such changes will necessarily occur in
any given case, only that IF some change occurs, it will be in the established direction. The
notion of regularity we are arguing for is the one outlined in the introduction: regular (i.e.,
recurring) directions of change, maintenance of contrasts after the change, and ‘global’
change for all subinstances of a given unit.
The ‘non-classical’ aspects of category structure in semantics thus appear to match
similar aspects of phonological categories. Semantics is of course a richer domain than pho-
nology in that it is not limited by the rather narrow physical constraints on pronunciation
and perception imposed by our physical production and reception apparatus. But like pho-
nology, meaning is limited by human perceptual and cognitive capacity, and (re)structured
by whatever general principles underlie that capacity. Some of the differences between the
two domains will be taken up in a later section. Here, we have argued that the same kinds
of cognitive categorization processes are at work in both domains, and that these processes
affect broad directions of change inasmuch as those can be discerned from universals.
2. Structural parallels between sound change and meaning change.
2.1. Two case studies in parallel.

Having clarified the background understanding of semantics and phonology presumed
by our analysis, we would now like to present a couple of very basic data sets, one from
phonological change and one from semantic change, to exemplify some of the parallelisms
between the two domains.
A well-documented grouping of semantic shifts is the development of epistemic senses
from root senses of English modal verbs (cf. Shepherd, 1981, 1982; Traugott, 1982, 1989;
Sweetser, in press). During a relatively restricted period, may took on a meaning of possi-
bility, must took on a meaning of certainty, will added epistemic likelihood senses to its wil-
lingness and futurity senses, and can (later and less completely) added possibility to ability
6 Most of the recent historical work cited at the start of this paper—Traugott, Fleischman, Bybee, Ander-
son, Hopper and Kemmer, for example—assumes systematic influence of cognitive structure on semantic
change, and in turn constitutes further confirmation of such an assumption. What we are arguing here is not
simply that these linguists are correct, but that current cognitive theory provides convincing motivation for
their assumptions.
among its senses. The earlier (‘root’) meanings of these words, still maintained alongside
the newer ones, have to do with completely distinct semantic domains: personal obligation,
ability, willingness, and (in another development from the root sense of may) permission.
Sweetser (1982, in press) argues that the connection between the two sets of senses is sys-
tematically structured by a metaphor; our epistemic processes are understood and described
in terms of more social concepts such as obligation and permission. Thus, for example, John
must be home by ten o’clock would have a reading involving a root sense of must, which
would mean something like ‘John is obliged to be home by ten,’ and a reading involving an
epistemic sense of must, which would mean something like ‘I (the speaker) am OBLIGED TO
CONCLUDE that John is home by ten.’7 The particulars of this analysis are not here at stake.
What is at stake is the intuitive perception that the connection between root and epistemic
modal senses is not random: that must could NOT just as well happen to mean obligation and
possibility, rather than obligation and certainty.
What then are the regularities concerning the grouped shift of modal verbs to epistemic
senses? First and most importantly, there is a natural polysemy relationship between root
modal senses and epistemic senses: cross-linguistically, such polysemy is common, and in
English it exists in other lexical domains besides the modal verbs (Sweetser, 1982, in
press).8 It is as natural and as ordinary for a single word to have root and epistemic modal
senses as for [b] and [p] to be allophones of one phoneme. Second, there are regular
correspondences between senses in the two domains: ability and permission correspond to
possibility, while obligation corresponds to certainty. When a word adds an epistemic sense
to a root modal sense, it will be the epistemic sense CORRESPONDING to the relevant root
sense. And apparently, when one modal verb adds an epistemic sense, the others have an
added motivation (beyond the general tendency) to change in a parallel fashion.
Suppose we now examine the group of sound changes known as Grimm’s Law.
Diagrammatically, these changes can be represented as:
bh b p f
dh d t
gh g k
gwh gw kw w
7 A past tense example like John must have been home by ten is unambiguously epistemic (‘I am obliged to
conclude that he was home by ten’), since it is very hard to imagine a world in which one can be presently ob-
liged to do something in the past.
8 We are aware that there exist past treatments of English modals as essentially two sets of homonymous
lexical items, unrelated in meaning. In such an account, there would be only a historical rather than a syn-
chronic relationship between the two senses of the English modals. Sweetser (1982, 1986, in press) has argued
that a synchronic polysemy relationship now exists for the two senses of the English modals: but for the
present argument it is sufficient to note that the historical relationship makes a stage of synchronic polysemy
necessary AT SOME POINT. There cannot have been a sudden development of a set of epistemic modal verbs,
unrelated to but homonymous with the root modals.
The following are some obvious, but nonetheless important, generalizations about these
facts. One, the shifts are interconnected: we would have a far less likely set of shifts if we
instead grouped /b/ /p/ with /d/ /dh/ and /k/ /g/. The voiceless stops change in
parallel, as do the other series. Two, the voiceless stops change to the CORRESPONDING
(i.e., homorganic) voiceless fricatives, the voiced stops to the corresponding voiceless stops,
and so forth. We would indeed be surprised to see /b/ /t/ grouped with /g/ /p/ and /d/
/k/, rather than the groupings we do find.9
The usual explanation for the parallelisms observed in the groupings of changes such
as the Grimm’s Law shift runs as follows. First of all, speakers give preference to generali-
zation in their grammars (cf. Chomsky & Halle, 1968; Kiparsky, 1968b): a rule saying that
all voiceless stops become fricatives is more general than one saying that /t/ becomes / /.10
Second, such a rule is only general if the change or changes it describes are all shared by all
affected segments: other features of the segments, not changed in a shared fashion, are
assumed to remain unchanged. This means the stops change to corresponding fricatives,
rather than to some random ones.11
What emerges saliently from these two case histories is the structural parallelism
between the semantic change case and the phonological case. In each case, a group of units
with certain common parameters are seen to change in parallel with respect to one particular
shared parameter; the other parameters or features of each individual unit remain in some
way fixed. The result is that the contrast between the units in question remains the same
before and after the change: /b d g/ contrast with each other in the same way that /p t k/ do,
and root must contrasts with root may as epistemic must contrasts with epistemic may.
Semantic change, like phonological change, thus shows regularity of the type described
under (2) at the beginning of this paper. The fact that we understand the phonological
parameters involved in such parallelism much better than we understand the semantic
parameters (e.g., we understand voicing much better than epistemicity) does not mean that
we should ignore the semantic cases—rather the reverse.
9 Note that these observations are completely independent of one’s analysis of the Grimm’s Law data. One
could equally well accept Hopper’s (1973) or Gamkrelidze and Ivanov’s (1973) reanalysis of the Indo-
European stop system, in which case the first Germanic sound shift would be described quite differently than
Rask or Grimm ever described it. But the description would still have the kind of regularity we are talking
about.
10 As will become evident in section 4.3, we see category structure and analogy as motivating factors which
underlie historical trends towards generalization. That is to say, speakers do not abstract just any kind of regu-
larity from the data presented to them, so it would be insufficiently explanatory to speak simply of a trend to-
wards ‘generalization’ or ‘simplicity’ in the abstract.
11 In a more phonetically based terminology, this is natural because there are regularly shared acoustic and
articulatory parameters which give rise to groupings of segments, and which will cause related misperception
and mispronunciation phenomena.
2.2. Phonemes and morpheme-senses.

Grimm’s Law involves phonemes changing their phonetic forms ‘in tandem,’ while
our semantic example involved a group of semantically related but contrasting morphemes
changing their meanings ‘in tandem.’ Interestingly, for the parallelism to emerge, the units
we need to treat as equivalents in the two domains are the phoneme in phonology, and the
morpheme-meaning (what Langacker, 1987, would call the ‘semantic pole’ of a morpheme)
in semantics. It has not been traditional to treat these two units as equivalents; indeed, accu-
sations of irregularity in semantic change all tacitly assume that ‘regular’ semantic change
would involve all morphemes which share some particular semantic feature changing
together, as a parallel to all instances of a phoneme changing their phonetic forms. But this
would essentially posit a semantic parameter of contrast (a feature) as being the analogue of
a phoneme, rather than setting up an analogy between parameters in the two domains, and
between phonemes and morpheme-meanings. The suggestive parallelism of our two exam-
ples would thus be lost.
In this section, we will discuss the question of choosing equivalent levels of units in
phonology and semantics: to what extent are phonemes and morpheme-senses comparable
units?12 From a formal point of view, both the phoneme and the morpheme-sense can be
thought of as clusters of parameters. Values on these parameters distinguish the relevant
unit from surrounding units in the phonological or semantic system; both phonemes and
morpheme-meanings are embedded in systems of paradigmatic contrasts. In both cases, the
relevant parameters of contrast and the relevant categories are abstracted by speakers from a
set of actual instances of use—an observed range of actual pronunciations or actual uses.
Further, in both cases, it appears that speakers successfully group together distinct subc-
lasses of phenomena as members of these larger classes. Allophones are recognized as ‘the
same’ phoneme, subsenses of a polysemous morpheme as instances of ‘the same’ mor-
pheme.
Following our hypothesis about the phoneme-morpheme match in generality, all
instances of a phoneme changing to another phoneme are exactly parallel to all instances of
a given morpheme changing to or acquiring a new meaning (for example, there is no irregu-
lar treatment of morpheme meanings in the sense that we do not find random use of a mor-
pheme with new and old senses; the morpheme ‘globally’ acquires a new meaning, or shows
structured variation between subsenses).13 Thus the diachronic statements in (1), (2) and (3)
below all express the same level of generalization:
(1) Proto-Hellenic /a:/ Attic-Ionic /e:/
12 See Lakoff and Brugman (1986) for a comparison of arguments about lexical semantic and syntactic
structure.
13 We do sometimes find old senses of a morpheme retained only in specific fixed phrases, as in (bank) tell-
er, which retains an old sense of tell meaning ‘count.’ However, it is also the case that in phonology a fixed
phrase may undergo some phonological reduction not undergone by the relevant sounds in the system at large.
(2) Latin habeo, Eng. have ‘physically hold’ (I.-E.) ‘possess’ (Latin, English)
(3) Tzeltal ?isim ‘corn’ ‘grain’ (Berlin, 1972)
The change in (1) is an instance of unconditioned vowel fronting while the changes in (2)
and (3) are instances of abstraction and broadening/generalization respectively. The possi-
bility of embedding what appear to be isolated changes in more general and (at least for
semantics) possibly unidirectional trends will be taken up in sections 3 and 4. The point we
want to make here is that if indeed all instances of a phoneme ‘match’ all instances of a
word (as the correct comparison unit), then the regularity of (1) is matched exactly by that in
(2) and (3).
Naturally, there are evident differences between a morpheme changing meaning and a
phoneme changing shape, as well as parallels. A case in point is the difference between
polysemy and allophony, if we take both phenomena to involve subcategories beneath the
level of morpheme-meaning or phoneme. A morpheme apparently must go through a stage
of polysemy in order to change senses, and new and old senses can apparently coexist
indefinitely, not as rivals but as co-components of an integrated system. But although allo-
phonic variation may be a result of many sound changes, it is by no means an inevitable
result. Still, both cases can be seen as involving the restructuring of categories, and the
differences may well be due at least in part to the fact that a morpheme-meaning is
inherently part of a form-function mapping in a way that a phoneme is not. There is a stable
identity to the morpheme (as determined by the form) despite changing mappings between
that form and some grouping of senses, while there is no similar stable identity to a
phoneme despite changes in form—two unconditioned forms of a phoneme conflict directly
in a way that two senses of a morpheme do not, although conditioned allophones do not
conflict.
This section has thus addressed point (3) from the introduction, and argued that in the
same way that all instances of a phoneme in different lexical items are all regularly affected
by a sound change, similarly, meaning change has a global effect on a morpheme meaning.
2.3. Paradigmatic structuring in sound and meaning.

Both morpheme-meanings and phonemes exist as units partly by virtue of their rela-
tionships to other contrasting units. We have argued in section 1 that the parameters which
distinguish phonemic categories from each other, or morpheme-senses from each other, can-
not be objective, nor invariably binary. Here we will discuss the paradigmatic relations
which structure the two domains, with a view to further clarifying the claim that phonemes
and morpheme-meanings change in parallel ways.
Regular paradigmatic relations are much harder to establish in the domain of lexical
and morphemic semantics than in phonology, largely due to the immense quantitative
disparity between the two domains. The hugely greater number of semantic parameters and
semantic units, compared to the relatively tiny number of linguistically relevant parameters
and units in phonology, means that it is impossible for a researcher to deal with all the ways
in which morpheme-senses can potentially contrast.
However, paradigmatic relations between lexical meanings have been discovered. The
theory of semantic/lexical fields represented most prominently by Trier (1931) and Porzig
(1967) was devoted precisely to demonstrating the structural relationships between mean-
ings and their effect on semantic change. Trier’s (1931) example of the different German
words for ‘knowledge’ is to the point: in Middle High German, a shift took place between
two lexical systems for referring to knowledge. The earlier system contrasted kunst (all
high, courtly knowledge, including knowledge about social behavior) with list (lower and
technical knowledge, including immoral magical knowledge) and wisheit (a general term,
but also a synthetic one, emphasizing the integration of different aspects of the human intel-
lect). The later system contrasted kunst (no longer courtly, but still non-technical) with wiz-
zen (which includes more technical knowledge) and with wisheit (now no longer a synthetic
term, but referring only to spiritual wisdom, in contrast with more mundane crafts or
knowledge). Both the relevant parameters of contrast and the senses of the individual words
changed between the two stages of the language.
American structuralist work (e.g., Lounsbury, 1964) has also treated the vocabulary of
particular domains (kinship terms, pronouns, etc.) as analyzable in terms of a limited
number of relatively orthogonal meaning parameters, and in such cases the presence or
absence of a contrasting category is often crucial to the analysis of a morpheme’s meaning.
Studies of the synchronic variability and diachronic developments of color terms in different
languages also make clear the structural relationship that exists among the meaning scopes
of ‘adjoining’ color terms (Berlin & Kay, 1969; Kay & McDaniel, 1978). The color terms
are a particularly interesting case, because no analyst could now argue that their meanings
are structured SOLELY by contrast: on the contrary, the focal colors around which other hues
are classified are determined by the physiology of vision. Yet in any individual system of
color terms, the classification of a specific hue (for example ‘blue’) will depend not only on
the hue itself but also on the number of categories so far given names in the language. Even
a language with two color terms covers the whole color chart with those terms. In cases
such as these, viewing the relevant meanings as participating in paradigms is crucial to both
their synchronic description and to understanding the way they change.
Hoenigswald (1960) is an attempt to characterize morphological and lexical semantic
change in explicitly structural terms. Semantic change is described in terms of the emer-
gence or disappearance of linguistically relevant contrastive meaning-parameters. In his
discussion of the possible ‘replacement patterns,’ we find complete parallelism with the pat-
terns of phonological change discussed in the literature. Analogous to phonemic merger, for
example, we find morphemic merger resulting in semantic change: Latin had a contrast
between ‘maternal uncle’ (avunculus) and ‘paternal uncle’ (patruus) which was subse-
quently lost in French (oncle refers to both). Phonemic splits (phonologization) on the other
hand, are paralleled by morphemic splits; for example, older English calf was used for both
the domestic animal and its meat. The latter meaning was eventually taken over by the
French loanword veal, creating a contrast between the ‘animal’ and the ‘meat’ meanings.14
14 The répartition of calf and veal is part of a broad pattern involving French loan words in English:
pork/pig and beef/cattle are parallel cases.
Not all lexical items obviously participate in salient and regularly structured paradig-
matic sets. But phonemes also differ in their degree of integration into the phonological
contrast-system. There are many languages (Proto-Indo-European probably among them)
where for example the only fricative is s or h. In such languages, the /s/ or /h/ would be less
integrated into the system of contrasts than in a language where a larger fricative inventory
included contrasts such as voicing or place of articulation. Finally, just as phonemes can
participate in different paradigmatic classes, according to the relevant parameter(s) (for
instance, either place or manner of articulation or both can be the abstracted features charac-
terizing a paradigm), so can morphemes. Fillmore’s (1976) observation about the possible
meanings of the word ‘breakfast’ can serve to illustrate the point. ‘Breakfast’ can be used in
either of the sentences below but in each case it is understood as a member of a different
paradigmatic set or, in Fillmore’s terms, of a different frame:
(1) Breakfast served all day. (relevant parameter = food varieties)
(2) The Wongs have chicken soup for breakfast.
(relevant parameters = place in meal schedule and time of day)
(3) After sleeping from 3 to 10 PM, Anne ate breakfast at 10:30 PM and left for her night
shift at the hospital. (relevant parameters = place in schedule, and possibly foods)
The system of salient cognitive contrasts is crucial in structuring synchronic sound and
meaning systems, and hence change in those systems. And the individual category structures
appear to be influenced not only by particular contrast parameters, but also by the global
structure of the paradigm of contrasts.
2.4. Syntagmatic factors influencing change.

In attempting to understand and motivate phonological change, linguists have found it
useful to examine both the relationship of the phoneme to other units in the phonological
system and also the context of the phoneme in particular sound-sequences. We argue that
semantic shifts are similarly rooted in both semantic paradigmatic structure and the context
of use of a particular morpheme. The important point here is that the character of the expla-
nation has depended on the character of the change. Articulatory and perceptual explana-
tions have generally been accepted for conditioned sound change,15 while paradigmatic
explanations have often been proposed for unconditioned sound change.
15 For instance, ‘ease of articulation’ (cf. Paul 1880) has often been cited as a cause of assimilation in par-
ticular. But if ‘ease of articulation’ is the motivating cause of sound change, why do some dialects perversely
retain the ‘less easy’ pronunciation? A more precise way to talk about articulatory influences on assimilation
is to refer to the independence of different parts of the articulatory apparatus: phonetic nasalization is thus
often due to the imprecise timing of velar movement with respect to articulation of stops or vowels neighbor-
ing a nasal sound. The simple ‘ease of articulation’ factor becomes even harder to defend when processes like
dissimilation enter the picture, which in an obvious sense undo the predicted assimilation patterns. For a dis-
cussion of the perceptual factors underlying historical assimilation and dissimilation processes, see Ohala
(1981, 1983).
Our perception of phonological category-membership is strongly influenced by the

context of adjacent sounds. For example, a minimally nasalized vowel which is NOT in a
nasal environment may be clearly categorized as phonemically nasal, simply because there
is no contextual factor to which speakers can attribute nasality. Conditioned sound change
appears to result in large measure from the lexicalization of natural mispronunciations and
misperceptions which depend on the syntagmatic sequence of sounds being produced or per-
ceived. It is not clear that syntagmatic structure is relevant in precisely the same way to
meaning change: in particular, the definition of context in semantics does not require linear
adjacency, as we assume that phonological context does. However, there is a plausible
semantic analogue to conditioned sound change. From the point of view of cognitive
linguistics, we would argue that categorization is done in general with respect to a frame. In
semantics, categorizing tasks also show different results depending on the set of contrasts
which the speaker sees as relevant, and which are evoked by the presence of a semantic and
pragmatic context. For example, categorization of speech acts as lies (Sweetser, 1987)
depends crucially on the extent to which the user of the word lie sees the speech event
referred to as being a prototypical informative speech act.
We might thus take the syntagmatic context of a morpheme-meaning to be the seman-
tic and pragmatic information which is correlated with a use of the morpheme. And there
are cases where this context appears to shape meaning change, just as phonetic context
shapes sound change.
Nasalization of vowels adjacent to nasals is typically non-phonemic, since listeners
attribute nasal quality in such contexts to the inexact timing of the speaker’s velar lowering
and raising for the nasal consonant, rather than taking it as intentional nasalization of the
vowel. But speakers may reanalyze this situation, and attribute the nasality to the vowel
itself, rather than to the context. This typically happens in conjunction with loss of the
nasal, an important piece of evidence that speakers tend to treat predictable parts of the
sound sequence as non-contrastive. (In the earlier form of this putative language, the
vowel’s nasal quality was predictable; in the later form, the presence of a nasal stop
becomes predictable data, and hence ignorable.) Historical change can potentially occur in
this way, as a result of listeners’ perception and interpretation of sounds relative to the syn-
tagmatic context.
Similarly, Stern (1931, 185ff.) observes that words meaning ‘rapidly’ frequently come
to mean ‘immediately.’ The context here is the frequent correlation between rapid action
and immediate results. Stern hypothesizes that in cases such as (a) below, there was no
ambiguity, since there is nothing in the utterance explicitly mentioning results or telicity.
Rapidity is thus simply attributed to the writing process, and the possibility of immediate
results remains a possible pragmatic implicature, but not a cogent one which is likely to
become conventionalized. (This would presumably be parallel to a vowel with no nasal
quality and no nasal consonant in the immediate vicinity.)
(a) He wrote quickly.
(b) When the king saw him, he quickly rode up to him.
(c) Quickly afterwards he carried it off.
In (b), which involves an imperfective predicate which has been bounded and made telic by
the phrase ‘up to him,’ the adverb can still be understood to modify the speed of the riding.
But the telicity makes the pragmatic implicature of immediate results a very firm concomi-
tant of the semantics of rapidity. That is to say, immediacy of results is a contextually
predictable added feature of rapidity in the context of certain kinds of predicates (this is
parallel to a phonetically nasalized vowel in the context of a nasal stop). Speakers
apparently reanalyzed examples like (b), attributing the ‘immediate results’ reading to the
adverb’s lexical semantics, rather than to the pragmatic context. They were then free to pro-
duce examples like (c), where the adverb quickly modifies afterwards, and can therefore
only be construed as meaning something like ‘immediately’ or ‘soon.’ These would be
parallel to speakers reanalyzing nasality as a phonemic characteristic of a vowel.
We could therefore argue that lexical meaning is influenced by the way speakers inter-
pret it relative to its syntagmatic context of other neighboring lexical elements, and of con-
comitant pragmatic implicatures. In semantics, just as in phonology, historical change can
result from listeners’ reanalysis of contextual factors, which by this means become attri-
buted to the contextually affected unit rather than to the context.
2.5. Abstraction in sound and meaning.

The final point of convergence between the phonemic and morphemic levels is that at
both levels we are dealing with abstractions, in one case of a set of phonetic realizations, in
the other over actual uses of a morpheme in context. Different relationships are possible
between a phoneme and its allophones: for example, one of the allophones may appear to be
the basic form of the phoneme, or the phoneme may have a more abstract form, not identical
to any particular allophone. Controversy has on occasion arisen (e.g., Kiparsky, 1968a;
Hyman, 1969) as to whether a phoneme MUST always have a basic form identical to some
particular allophone; or, indeed, whether a phoneme has any ‘basic’ form, or is simply a set
of surface alternate forms (Hooper, 1976). Of course, there is always the possibility that
allophonic variation is minimal or zero.
Regardless of one’s stance on the phonological issues, it is evident that closely parallel
questions preoccupy semanticists with respect to morpheme-meanings. Does some particu-
lar sense of a polysemous morpheme have to be basic, or can we have abstract senses which
are not identical to any of the specific subsenses? The latter case is generally known in the
literature as the abstractionist position. Jackendoff (1983, chapters 9 & 10) offers many
examples of this approach in his analysis of spatial and non-spatial (temporal, possessive,
existential, etc.) expressions. The former is argued for in detail by Brugman (1988), Lakoff
(1987), Traugott (1986a).16
Should the analyst look for a ‘basic’ unifying sense of a polysemous word, or is it
sufficient to talk about what groupings of senses co-occur in connection with a single form?
16 See also Sweetser (1986) for a discussion of the criteria for deciding between an abstractionist and a po-
lysemy analysis in a given case.
We would argue that this is an empirical issue, and that individual cases may vary with
respect to the presence of an abstract unifying sense, but that it is certainly not universally
possible to produce satisfying abstractionist analyses for all morpheme meanings. Horn
(1985) argues conclusively that morphemes such as and and not have unified (and
inherently quite abstract) meanings, and that pragmatic principles of interpretation account
for the differences in meaning observable in examples like (a) and (b).
(a) Chris isn’t happy, she’s sad. (entails ‘Chris is not happy’)
(b) Chris isn’t happy, she’s ecstatic. (entails ‘Chris is happy’)
On the other hand, we have previously mentioned that many examples of metaphorically
based polysemy present serious difficulties for abstractionist analysis, since it is extremely
difficult to find any ‘shared’ features of meaning in such cases.
But setting aside the question of when to treat a morpheme as polysemous (rather than
highly abstract in meaning), the important point here is that polysemy (which has not always
received the proper amount of attention) implies the recognition of a generalization in
semantics parallel to the phoneme-allophone distinction. And just as with allophonic varia-
tion of a phoneme, the range of polysemy of a morpheme may be extensive or minimal; but
in either case there is cognitive categorization of the allophones or the subsenses of the mor-
pheme into a related grouping, along parameters abstracted by the speaker from observed
use.
2.6. Syntagmatic and paradigmatic influences on regularity and

irregularity in change.
For phonology, it has been possible to set forth a relatively complete set of relevant
parameters (allowing, of course, for disagreement among researchers as to the exact set and
their precise characterization).17 It may be inherently impossible to delimit a full set of
semantic parameters relevant to semantic categorization and semantic change, both because
of the immense variety of entities and events to be categorized and because of the local
nature of many of the parameters (as male/female is only relevant within the class of
animals, so leafy/needle-bearing is only relevant within the domain of trees, and so on).18
But multiplicity and locality of relevant parameters do not prove that the parameters
are unrelated. As we have shown in section 2.3, there are in fact systematic paradigmatic
oppositions in semantics. Although the parameters involved in such oppositions may not be
classical objectivist binary features, progress has been made towards understanding and
characterizing them. And the ways in which they combine, or shift their combinations,
show parallels to phenomena long recognized in phonological change.
In phonology, the structure of the paradigm itself has often been claimed to influence
phonological change. Thus, for example, a pervasive phonological contrast parameter (such
as voicing in English) may influence the likelihood of a new phonemic contrast emerging
from allophonic status: the phonologization of English /v/ and / / may have been abetted by
the parallelism between the f/v or / contrast and others such as p/b, t/d. Martinet (1952,
1955) offered the notion of ‘paradigmatic ease’ as an explanation for unconditioned sound
changes which could not be explained by any notion of syntagmatic ease of articulation.
Without wishing to overemphasize the role of the phonological paradigm as some purely
formal structure, it seems clear that restructuring of a phonological system will depend in
important ways on which sound contrasts are already salient for speakers; i.e., on the
psychological categories already formed and the relevant contrastive parameters determin-
ing them.
In a parallel fashion, there may be different sorts of explanations to be given for dif-
ferent kinds of semantic change. Related groupings of semantic shifts seem (as in the case
of coordinated phonological shifts) to involve both basic structural facts about morpheme-
meanings (by which we simply mean what meanings are similar or related), and broader
syntagmatic relationships. Martinet’s ‘push chains’ and ‘drag chains’ in phonology may
well have analogues in semantics. For example, the idea of répartition in semantics (cf.
17 Ladefoged and Halle (1988) is a striking tribute to the significant amount of understanding of phonologi-
cal structure shared between researchers with widely differing approaches.
18 We have above indicated that we do not think semantics is reducible to a limited number of meaning
primes, although we have also indicated agreement with analysts who treat lexical meaning as universally con-
strained by cognitive factors. (Constrained is not ‘determined’; there is room for cultural variation.) We will
not here address some related issues of importance: for example, whether synonymy is possible cross-
linguistically, or the extent to which any language exhaustively covers human cognition. It is interesting to
point out that the former question is paralleled by a rather difficult question in phonology, namely to what ex-
tent we should identify phonemes with each other cross-linguistically despite minor phonetic differences.
Bréal, 1900) essentially involves push chains. The idea is that words which are identical in
sense are somehow naturally redivided and given distinct senses.19 By the same principle,
words which are too similar in sense are ‘pushed’ apart to avoid running into each other’s
space (a simpler way of putting this is that speakers are looking for meaning differences
when they see differences in form). A very clear example of such a synonymy-avoidance
chain can be seen in the Sanskrit and Armenian vocabularies for the equine domain (Wat-
kins, 1970): the Armenian word for ‘horse’ (ji) is cognate with the Sanskrit higher poetic
word for a heroic ‘steed’ or ‘courser’ (háya), while the Armenian word for ‘donkey’ (e:s) is
cognate with the Sanskrit word for ‘horse’ (áśva). Whichever of the semantic shifts occurred
first (‘steed’ to ‘horse’ or ‘horse’ to ‘donkey’), there is a clear relationship and coordination
between them. The systematic CONTRAST (here, a relationship on a scale of ‘nobility’)
between the two words has been maintained, even while the actual meanings have changed
over time, rather as the Great English Vowel Shift changed actual vowel heights but in
many cases left the height contrasts between phonemes unaffected.
We shall return in section 4.4 to the question of how systematic contrasts are mani-
fested in analogical developments. It does seem clear that more salient and entrenched
parameters of contrast are more likely to spread analogically, so that in this respect also we
can view paradigmatic structure as being deeply influential on directions of change.
Throughout the preceding discussion, we have tried to make it evident that there are
apparent differences as well as parallels between the domains of sound and meaning. It is
not yet clear, however, what are genuine non-parallelisms between the behavior of the two
kinds of unit, and what are simply apparent problems raised by the imprecise and rudimen-
tary nature of our understanding of semantics. Thus, for example, the English modals are
not QUITE regular in their shift to epistemic senses: can takes on an epistemic meaning only
in negative and interrogative contexts. One possible reason for irregularity of this kind is
the richness of the semantic domain: there are so many more factors which are potentially
relevant to semantic categorization than to phonological categorization, that multiple forces
may be at work to categorize morphemes together and to categorize them differently.
Another possibility is that a Verner has not yet come along to show us the tidy subregularity
exemplified by this (and presumably a wider class) of ‘exceptions’ to a broader generaliza-
tion.
Yet another point to be considered is that the number of phonological regularities that
have been discovered is much greater than the number of regular semantic shifts known to
linguists. This may be simply due to the fact that we have been looking for phonological
regularity for a much longer period of time. Again, however, we are leaving open the possi-
bility for a greater potential for irregularity in the semantic domain, given the complexity of
the parameters involved and what we take to be the non-feature-like character of meaning.
19 cf. Malkiel (1982) for broad treatments of ways in which two words’ proximity in sense may influence
their phonological development as well, and ways which phonological proximity may influence either seman-
tic or phonological development.
We should also recall the irregularities which do occur in sound-change, and which
usually go unmentioned in discussion of regular phonological change. Semantics itself is
sometimes an influential factor in such irregularities: for example, the /f/ in English four is
irregular and is apparently due to its semantic proximity to five (which has a regular /f/).
Learned forms from an older stage of the language may sometimes displace their own legiti-
mate heirs, dialectal borrowing may give the same speakers two (phonologically distinct)
forms of the same morpheme, and so forth. We are willing to admit these explanations for
irregular results of phonological change. It is true that we cannot count on such explana-
tions turning out to account for all irregular meaning shifts; but if Verner had not been look-
ing for regularity he would not have found it, so it behooves us not to discount the possibil-
ity that there are rules about apparently exceptional quirks in an otherwise regular grouping
of meaning-shifts.
In the rest of this paper we will address these aspects of regular behavior that have to
do with recurring and directional trends (point 1 in our introduction). In section 3 we will try
to show that recurring (and in this sense regular) trends occur in semantics as well as in pho-
nology, while section 4 will take up the issue of directionality and the cognitive mechanisms
that underlie unidirectional tendencies.
3. General trends and spatio-temporal unities: semantic and

phonological laws.
A recurrent point in the arguments against the existence of regularity in semantic
change is the observation that semantic change affects only individual words and thus can-
not be considered anything but sporadic and unsystematic. For example, Hock (1986)
writes that ‘by and large, semantic change operates in a rather random fashion, affecting one
word here (in one way), and another form there (in another way). Given the ‘fuzzy’ nature
of meaning, this is of course not surprising.’20
The sporadic nature of meaning change has not gone unnoticed. Characteristically,
Saussure’s prime examples of the accidental and particular nature of diachronic events come
from semantic change (although he claims that the same is true of morphological and
phonetic—i.e., phonological—change). That French poutre ‘mare,’ for instance, has come
to mean ‘piece of wood, rafter’ "is due to particular causes and does not depend on other
changes that might have occurred at the same time. It is only one accident among all those
registered in the history of the language." (Saussure, p. 93)21
Saussure’s position departs rather interestingly from the Neogrammarian position
which is dominant in modern phonological theory. While the Neogrammarian hypothesis
treated sound change as regular and exceptionless, Saussure sees the main dichotomy as
20 Hock may be among the few to believe still that meaning is completely unstructured and fuzzy, since
most modern semanticists agree that one can analyze meaning in a structured way (although the ways of
analyzing semantic structure can, as we have seen, differ among semanticists). However, he is certainly not
alone in believing that semantic change proceeds in a random and unsystematic way.
21 All Saussure references are to Saussure (1915/1959).
being between synchronic facts, which evidence a certain regularity, and diachronic facts,
which don’t; thus for Saussure, the correct way of viewing a diachronic ‘phonetic’ (i.e., pho-
nological) law is to see it as multiple manifestations of a single particular and thus isolated
fact, even if it affects all the words containing that given ‘phonic feature.’ In his words
(Saussure, p. 93), ‘we can speak of law only when a set of facts obeys the same rule, and in
spite of certain appearances to the contrary, diachronic events are always accidental and par-
ticular.’
There are (at least) two ways of interpreting Saussure’s characterization of diachronic
shifts as ‘isolated and accidental’: One interpretation may be taken to refer to his belief that
all instances of change are due to particular causes, not depending on other changes that
may be occurring at the same time; moreover, the causes may not be discoverable and
observable, since change does not reflect any intentional ‘move’ on the part of language
speakers.
Many linguists have remarked before on the apparent conflict between Saussure’s syn-
chronic view of the system as full of systematic paradigmatic pressure (e.g., if a word X did
not exist, its functional load would go to its competitors) and the diachronic claim that all
changes are isolated and unsystematic (see Ullmann, 1967, for an extensive review of the
relevant literature). Indeed, his celebrated example of the chess metaphor (p. 88), as Saus-
sure himself points out, illustrates perfectly the effect that a change can have on the system
as a whole; it is in this respect that it is hard to conceive of change as completely unsys-
tematic. Moreover, the influence of internal paradigmatic factors on change is evidence that
our ability to discover external causes is independent of whether there are observed regulari-
ties in diachronic shifts.
There is a second possible aspect, however, in Saussure’s characterization ‘isolated,’
namely that diachronic events, in his view, are each ‘one of a kind’; that is they are all of the
type exemplified by the change of poutre, of a particular character and restricted to a partic-
ular time and place. The rest of this section is devoted to addressing this last point.
In summarizing the literature on semantic change up to that time, Ullmann (1967, esp.
chapters 3 & 4) correctly points out that all the debate on the possibility of stating semantic
laws was done in the context of either the laws of natural sciences or that of sound-laws.
This created the following assumptions and expectations regarding phonetic and semantic
regularities:
(1) that all sound laws are of what Ullmann calls the ‘spatio-temporal’ variety, i.e., that
any phonological change took place in a limited period X and within a definite area Y.
But as Ullmann points out, such an assumption was already disputed (even in phonol-
ogy) by the work of Grammont and other phoneticians on assimilation and dissimila-
tion;
(2) that all linguistic laws should conform to this pattern, and
(3) that semantic developments in general do not offer themselves to this kind of formula.
The spatio-temporal requirement is of particular interest here. What Ullmann is refer-
ring to is of course the fact that, in general, diachronic phonological shifts are described as
taking place in a specific geographic/dialectic area and within a specifiable period of time.
Well known examples here include the Great English Vowel Shift and the numerous phono-
logical rules proposed to account for the differences between PIE and the daughter
languages.
This way of characterizing diachronic laws appears to be collapsing two kinds of gen-
eralities which should be kept distinct: first, general recurring trends (e.g., assimilation or
dissimilation) and secondly, specific instances of these trends at the level of the specific
phoneme or morpheme-meaning. One can think of this difference as a type-token kind of
distinction: on one hand, we have general processes characterizing a number of specific
shifts; on the other, we have the specific changes themselves instantiating the general princi-
ples within particular spatiotemporal boundaries (Grassmann’s law, for example, is both an
instance of dissimilation and a particular phonological change characterizing the passage
from IE to Greek). We believe that this is an extremely important distinction, pertaining to
the question of regularity in both phonological and semantic change. It addresses directly
Saussure’s objection to potential regularity in diachrony and the objection of other historical
linguists to the possibility of stating semantic laws.
Turning now to the semantic domain, it seems possible to view the documented seman-
tic shifts in exactly the same way, establishing a two level distinction; on the one hand,
semantic change appears to affect a specific lexical item or a specific class of lexical items
in a specific time and place. The shift of the English modals from content to epistemic
meanings, and Stern’s example of English adverbs changing from the meaning ‘rapidly’ to
the meaning ‘immediately’ (before 1400) can be thought of as instances of completed and
geographically definable changes. At the same time, however, the content-to-epistemic shift
is seen as exemplifying general, natural tendencies which recur in the history of words.
Thus, Sweetser (in press) talks about this particular change as an instance of the general and
pervasive metaphor ‘mind in terms of body’ (that is, our understanding of internal/more
abstract meanings in terms of the external/physical world) while Traugott (1987, 1989)
views it as an instance of the general tendency for meanings to become more situated in the
speaker’s attitude towards the situation (in other words of the tendency towards pragmatici-
zation).
The ‘mind-as-body’ metaphor, for example, also encompasses the semantic shift of
perception verbs in English and in other IE languages from concrete, physical senses to
internal/mental senses (cf. the polysemy of English see, or of Greek oida ‘know’ which
comes from the same root as the verb ‘to see,’ the shift of verbs meaning ‘hear’ coming to
mean ‘obey’ etc.). In Traugott’s terms, on the other hand, the shift of the English modals is
an instance of a general tendency towards greater ‘situatedness,’ a tendency which encom-
passes also the shift of while (and other such conjunctions) from temporal to concessive
meaning. The same tendency is exemplified by the semantic history of presuppositional
(scalar) terms (e.g., even, just, very, etc.) which in many languages start out with concrete
adjectival meanings (just < ‘fair,’ very < ‘true’ etc.). At this level of description, therefore,
and by virtue of being an instance of a regular (in the sense of recurring) tendency/trend, the
semantic shift exemplified by while ceases to be an isolated change in the Saussurian sense.
Although it is definitely restricted by spatio-temporal boundaries, it is also regular by virtue

of the fact that it instantiates a general shift, not constrained by such boundaries. We want to
suggest that this is true for both the semantic and the phonological domains; to the extent
that such recurring (in different languages, at different times) tendencies are discoverable,
they justify the characterization ‘regular.’22
For the semantic domain this is particularly important; as we noted above, viewing
semantic shifts in isolation may obscure the existence of regularity outside the specific
spatio-temporal boundaries. As Malkiel (1957-1958) observes, the split of Latin leo ‘lion,
lioness’ into French lion ‘lion’ and lionne ‘lioness’ will tend to be viewed as an insignificant
process unless one places it in the context of a long-range trend, in this case what Malkiel
calls hypercharacterization (the process by which vague, broader meanings split up again
and again into pairs of signs with different signifiers). In the same way, many apparently
isolated semantic shifts can be seen as falling under the umbrella of the general concrete-
to-abstract tendency in meaning change; Traugott (1986b) gives many examples of this
cross-linguistically pervasive tendency: spatial vocabulary acquiring temporal meanings,
developing into different kinds of connectives (e.g., but < butan ‘on the outside’) or giving
rise to speech act verbs or verbs of mental activity (e.g., assert < Lat. asserere ‘join oneself
to,’ deduce < Lat. ducere ‘lead’).
It is worth noting that such tendencies can be ‘captured’ at multiple levels of general-
ity. The shift of spatial motion verbs like go and come to tense or aspect markers, for
instance, is evidenced in many (genetically unrelated) languages at different periods of time
(cf. Fleischman, 1983; Bybee, 1985). This particular semantic and syntactic change is also
an instance of the more general trend of spatial terms acquiring temporal meanings. Finally,
it is also an instance of the ‘objective’ to more ‘subjective’ meaning shift, of the tendency
for meanings situated in the external/objective situation to change to meanings situated in
the internal/perceptual/cognitive world (Traugott, 1987, 1989). In yet another view, it is an
instance of our general metaphorical capacity to structure the more abstract in terms of the
more concrete (Lakoff, 1987; Sweetser, in press). In the latter view, the different levels of
generality noted above can be seen as metaphors stated at different levels of generality.
We have argued that if the phoneme and the morpheme-meaning are taken to be
matching units in their level of generality, salient parallels may be drawn between cases of
phonological and semantic change. We now suggest that by distinguishing between general
trends of change and specific instances of these trends, the possibility of discovering regu-
larities in both semantics and phonology is greatly increased.
22 The claim is not that semantic change will inevitably take place or that it will inevitably affect all poten-
tial candidates, but rather that if change occurs at all, it is much more likely to follow the direction predicted by
the metaphor or (in Traugott’s terms) by the unidirectional trend towards pragmaticization. The same is of
course true of phonological change as well.
4. Directions of metaphor, metonymy and analogy.

Having discussed some of the ways in which the domains of sound and meaning are
similar and different in category-structure, we will now treat some of the major cognitive
structuring mechanisms which shape the two domains in parallel ways. Recent work on
directions in semantic change has differed from earlier work in its focus on possible expla-
nations for directional trends. Traugott (1988) postulates metonymic structuring as an
important factor in certain types of unidirectional meaning change. Similarly, metaphorical
accounts of meaning shifts seem to inherently involve an explanation (as well as a descrip-
tion) of the semantic change in question. That is, such accounts rely on a general under-
standing of human cognition as inherently and pervasively involving metaphor and meto-
nymy. Analogy has long been treated as basic to linguistic change, and can presumably be
added to metaphor and metonymy as a cognitively basic aspect of linguistic structure.
4.1. Metonymic relations in phonology and semantics.

In both phonological and semantic change, metonymic relationships are evident. As
we shall see, there is a major discrepancy between the two domains, in that metonymy is
bi-directional in at least some semantic domains, but apparently unidirectional is other
semantic domains and in phonology. Without claiming to explain these facts, we would like
to set forth some of the generalizations which seem to have emerged from recent research.
Classic discussions of semantic change (Bréal, 1900; Stern, 1931; Ullmann, 1967).
have given the names of broadening and narrowing to cases where a word’s meaning takes
on a more inclusive class of referents (subsuming the older class) or comes to refer to a sub-
class of its older class of referents. Examples include girl, which narrowed from meaning
‘young person’ to meaning ‘female young person,’ and stuff, which broadened from ‘cloth’
to ‘material, substance.’ Broadening and narrowing phenomena can be viewed as meto-
nymic in nature (cf. Lakoff, 1987; Traugott, 1988); the category-name of a subclass is used
to refer to the whole larger class of which the subclass was a part, or vice versa. It is often
said that broadening is more common than narrowing, although we know of no controlled
studies of developments throughout a language’s lexicon which have actually supported this
hypothesis.
The most confusing aspect of metonymic developments in lexical semantic change is
that two distinct varieties of semantic relationship have been referred to as metonymic. One
is the kind of broadening/narrowing mentioned above, where the part-whole relationship is
between the category and a subcategory; the other is cases like words for ‘hand’ coming to
mean ‘arm,’ where the individual referents stand in a part-whole relationship. That is, meto-
nymy occurs both in partonomies and taxonomies. The classic studies of semantic change
in taxonomic systems (Berlin, 1972; Berlin, Breedlove, & Raven, 1973) have shown that
folk taxonomies of plants and animals begin by naming the generic level (which seems to
correspond to a basic level category in Rosch’s sense (Mervis & Rosch, 1981; Rosch, 1978).
Generic level names often historically develop into names for EITHER the larger class of
which the genus in question is taken to be a salient representative (e.g., ‘oak’ ‘tree’) OR
the species taken to be a salient representative of the genus (note the English ‘oak’ for both
genus Quercus and species Quercus quercus). In fact, such developments are one of the
most common sources of higher and lower-level terms in folk taxonomies. The story is
rather different for partonomies, however. Examining developments of body-part terms,
Wilkins (1981) argues that while part for whole is frequently instantiated in semantic
change (‘foot’ ‘leg,’ or ‘finger’ ‘hand’), the reverse process is not observed. Although
other principles may be at work here (for example, ‘finger’ also does not come to mean
‘body’), the asymmetry observed by Wilkins suggests that development towards a more
inclusive sense is more natural than the opposite direction in this domain.
Narrowness and specificity are often seen as linked to concreteness, generality to
abstraction. However, it is important to keep the two claims distinct, and the two parame-
ters independent. There seems genuine support for the claim that words do not tend to
become more concrete in meaning: the reason for this appears to be (at least in part) that we
tend to metaphorically understand abstract domains in terms of concrete ones, and not vice
versa. Many cases of abstraction seem to be metaphorical, rather than metonymic in nature
(see 5.2).
But there are amply documented cases in taxonomic structures of words either gaining
or losing in specificity; this is presumably because the basic level of cognitive categoriza-
tion is an intermediate generic level. Further, such cases (e.g., ‘oak’ ‘tree’ or ‘oak’
‘Quercus quercus’) do not bring about a change in concreteness: a tree, an oak, and a
member of the species Quercus quercus could in fact all be the SAME concrete physical
object, referred to in more or less restrictive terminology. A more general term may simply
refer to a larger class of concrete objects. It has often been suggested that a more general
category has a simpler semantic representation (e.g., fewer features) than a more specific
one. But if Rosch is correct, we in fact find basic-level categories cognitively simpler than
either more specific or more general categories, because they involve experientially
motivated correlations in the relevant parameters. And they are in close metonymic rela-
tionship with their adjacent superordinate and subordinate categories, sometimes leading to
semantic change in either direction.
Wilkins suggests that part-for-whole is more common than whole-for-part because a
part implies the existence of some whole of which it is a part, while a whole need not have
parts. Of course, it is not some real, extant whole which is implied by a part, but some CON-
CEPT of a whole which frames our understanding of the part. Thus, an object can be ‘part’
of some particular variety of car, even if it was never used in the assembly of any whole car,
or even if no such cars ever made it to the assembly line. Given this idea of the whole as a
FRAME for the part (an idea strongly supported for body part terms by the work of Petruck,
1986), we might extend metonymy to include other kinds of framing. Thus Traugott (1988)
discusses examples of pragmatic strengthening which involve speakers attributing aspects of
the pragmatic frame to the lexical semantics of a morpheme used in that frame: for example,
a conventional inference becoming lexicalized. She treats Stern’s ‘rapidly’ > ‘immediately’
cases (see 2.4 above) as examples of metonymic reanalysis of the adverb. The adverb’s ori-
ginal meaning of ‘rapidly’ produced, in conjunction with telic predicates, a pragmatic impli-
cature that the intended goal of the quick action or event would be achieved in a short time.
The adverb was interpreted relative to a frame involving this implicature. From being a
contributing part of this pragmatic frame, the adverb was reanalyzed as contributing the
whole, including the implicature.
Metonymy involving frames is inherently partonomic rather than taxonomic, and so
joins our list of ways in which metonymy is likely to produce unidirectional change.23
Metonymy in general appears to be a powerful force for restructuring both sound and mean-
ing systems.
4.2. Metaphor.
Metaphor is, like metonymy, an apparently universal human cognitive mechanism. It
involves mapping one domain onto another, and its linguistic manifestation is frequently the
use of the vocabulary of the ‘source’ domain to talk about the ‘target’ domain.24 For exam-
ple, in our earlier example of verbs meaning ‘see’ coming to mean ‘know,’ knowledge is the
target domain, and is viewed as vision and hence talked about with vocabulary from the
domain of vision; this makes the relevant vocabulary polysemous, and potentially subject to
reanalysis as non-metaphorical elements of the target-domain vocabulary.
Metaphor regularly shapes semantic change in this way, and the routes of change set
up by metaphorical mappings are unidirectional because the metaphorical mappings are
inherently unidirectional.25 It has been argued that more cognitively basic and/or structured
domains tend to serve as source domains for mapping onto less basic or less structured areas
of cognition (cf. Lakoff and Johnson, 1980; Johnson, 1987). In particular, our understand-
ing of our everyday physical experience of space and motion in the concrete world provides
a basis for structuring more abstract and less clearly intersubjectively shared domains.
Given such an account of metaphor, regular trends in semantic change, such as the develop-
ment of temporal senses by spatial vocabulary (Fleischman, 1982a, 1982b; Traugott, 1982),
or the development of psychological and epistemic senses by physical vocabulary (Traugott,
1982, 1987, 1989; Sweetser, in press) are unidirectional because human experience of the
world naturally prompts us to metaphorically view these particular target domains in terms
of these particular source domains, and not vice versa.26
23 One might see as metonymic cases where nasalization or creaky voice start out as properties of a particu-
lar phoneme and spread to become properties of the syllable or the word.
24 Our own views on the metaphorical structuring of linguistic usages have been particularly shaped by
Lakoff and Johnson (1980), Johnson (1987), Turner (1987); some further description of one of the present
writers’ views can be found in Sweetser (1987, 1988, in press).
25 It has frequently been noticed that the same pair of domains may have reversed source-target relations in
two distinct metaphorical mappings. For example, ‘My memory banks are shorted out today’ is evidence of a
metaphor with computers as source domain and human beings as target domain, which apparently maps (at
least) aspects of the two domains such as memory and reasoning powers. But ‘My MacIntosh has been tem-
peramental lately’ shows human as source and computer as target domain, and appears to map human emo-
tional structure onto the causes of computer behavior. It is important to realize that these are not instances of
the SAME metaphor, and that each of the two metaphors is in itself unidirectional.
26 Such a claim says nothing about other possible, but less ‘built in’ metaphorical cognitive mappings, or
about other possible mechanisms which might motivate reverse directions of change. Metonymy is one such
mechanism, and metonymic time-for-space usages can be seen in examples like ‘She lives five minutes from
Because of the unidirectional tendency to view abstract domains in terms of concrete

ones, metaphor is often a motivation for morphemes’ acquiring more abstract meanings.
Why should metaphor, if it tends to use concrete vocabulary, tend historically to create
abstract meaning? We may compare this with the parallel paradoxical effect of euphemism.
In speaking euphemistically, we try to find a nice word (a word whose basic sense is ‘nice’)
to express a ‘not-nice’ meaning, and the historical result is often that the word is reanalyzed
as lexically meaning the ‘not-nice’ thing. In speaking metaphorically, we often try to speak
concretely, looking for a way to express abstract concepts in terms which will be close to
our shared experiential structure of the concrete world. The historical result is that the con-
crete vocabulary can lexically and conventionally take on the abstract senses, and become
abstract vocabulary.27
Metaphor is often based on experiential correlation (cf. Lakoff and Johnson, 1980),
but must be distinguished from it. Thus, for example, we experience a correlation between
quantity and verticality, in situations like pouring milk into a glass and watching the level
rise. But this is only a partial correlation: if we instead pour the milk onto the kitchen floor
or onto the grass on the lawn, it might take a great deal more milk than we are likely to have
to allow us to witness a rise in level rather than lateral spreading or absorption of the liquid.
However, although not an absolute correlation, the link between verticality and quantity is
apparently a salient one, and (perhaps via a metonymy of these cases for a broader set of
less correlated cases), a metaphorical mapping between the two domains is established at
large. The presence of such a systematic metaphorical mapping motivates our ability to say
‘my income rose/fell’ or ‘my supply of patience is low today.’ using the vocabulary of ver-
ticality to talk about abstract quantities which have no actual correlation with physical posi-
tions on a vertical dimension.
A primary level of structure, and one which may well universally remain constant
across metaphorical mappings (Brugman & Lakoff, 1988; Sweetser, 1988) is image-
schematic structure. This is an abstract and topological representation of aspects of mean-
ing, largely rooted in the understanding of physical experience mentioned above (cf.
Talmy, 1987). Thus, for example, Sweetser (1988) suggests that an image schema involv-
ing a proximal source, a distal goal, and an ego moving from source to goal is preserved in
the mapping of the deictic physical motion verb go onto the abstract temporal domain of
futurity.
Unlike metonymy, metaphor appears to be relevant to semantics exclusively, not to
phonology. However, as we shall see below, analogy is relevant to both phonology and
semantics, and analogy is often influential in the spread of metaphor through the lexicon (for
example, once one root modal verb has taken on an epistemic meaning, that fact adds to the
the BART station,’ where ‘five minutes’ is metonymic for ‘the distance travellable in five minutes.’
27 Meillet (1965) discusses the recurring phenomenon of ‘vivid’ vocabulary being used to reinforce and
strengthen speakers’ communicative intent, and the resulting recurring tendency for vocabulary so used to be-
come less vivid. See Sweetser (1988) for a rediscussion of some of the same issues from a different perspec-
tive, involving metaphor.
motivation for the other root modals taking on epistemic senses). Metaphor is thus an
important input to cognitive processes which are involved in both sound and meaning
change.
Further, it is not always easy to distinguish metaphor from analogy. Gentner and
Gentner (1982) and Gentner (1983) have said that analogy maps relations rather than pro-
perties; but, as we shall see below, some phonological analogies seem to involve properties
(or features).28 So it may be useful to treat metaphor and analogy in some ways as a unified
phenomenon; there may be generalizations about the way one domain is mapped onto
another which apply to mappings which we would call both metaphor and analogy.
4.3. Analogy in sound, meaning, and grammatical paradigms.

It is intuitively clear that analogy has (like metaphor) some natural directions: it is not,
in general, equally comfortable to talk about X in terms of Y or Y in terms of X, nor is it
equally likely that domain A or domain B could serve as a source for analogical reorganiza-
tion of the other domain. One area where we might find common limitations on historical
change, therefore, is in the area of structural limitations on plausible sources and targets for
analogical restructuring.
Analogy is the setting up of some equation (X:Y::X’:Y’) which identifies the contrast
between X and Y as ‘the same’ contrast which separates X’ and Y’. In linguistic change,
such incomplete equations (X:Y::X’:?) often result in a new form Y,’ which is said to be
analogically created. The classic example of analogy in linguistic change has often been
taken to be the sort of grammatical analogy discussed in Kurylowicz (1945-1949/1966). For
example, paradigmatic leveling is the result of grammatical analogy, as when a noun like
honos (gen. honoris) changes its nominative form to honor, by analogy with the invariant
stem-final r of nouns like color (gen. coloris). The analogy is color:coloris:: X :honoris.
Crucial to the structure of this analogy is the perception of a contrast (color differs from
coloris only in the presence or absence of an ending -is), and imposition of this contrast in
the structure of another domain (the form which would differ from honoris precisely by the
absence of -is would be honor).
Analogy is said to be (unlike phonological change) irregular in its operation: not every
Latin nominative ending in -os was affected by the process just mentioned. But such a per-
ception is dependent on the assumption that all instances of a phoneme changing form
would be parallel to all morphemes of some morphological class (or all lexical items of
some semantic class) undergoing some change. If the correct parallel is rather between a
single phoneme and a single morpheme-meaning, then analogy is not so unlike processes
which are regularly observed in phonological systems.
28 It also seems that some of Gentner and Gentner’s analogy examples map properties as well as relations:
for example, in her ‘atom-as-solar-system’ example, physical bodies (sun, planets) are mapped onto physical
bodies (nucleus, electrons). Unless properties are sometimes to be treated as one-place relations, which would
make the whole discussion vacuous, it seems that we cannot rigidly distinguish metaphor from analogy in this
way.
For example, in the second half of the 4th century C.E., Daco-Romance (the ancestor
of Romanian) underwent a sound change /i/ /e/ (thus nigr- ‘black’ negr-). Other
Romance dialects underwent the same change, including Italian. Italian, however, (together
with most of Romance), shows evidence of a parallel process /u/ /o/, while Romanian
does not (Latin ursus ‘bear’ > Italian orso, Romanian urs). We know the social and histori-
cal reasons for this: the Italian dialect area underwent the /u/ /o/ shift some fifty years
later than /i/ /e/, and during that fifty years the political connection between Rome and
Romania was broken. The theoretical points to be made are (1) that there is a coherence
between these two changes (Italian did something natural in adding the second one to the
first), and (2) that the coherence was nonetheless not of such a compelling nature as to force
Romanian into the second shift. Let us first examine the nature of the coherence.
The contrast between /i/ and /e/ is one of height (plus the elimination of contrastive
length in the later forms of Romance), and in particular between a lax high vowel and a
tense mid-vowel—or, if one sees vowel height as a non-binary phenomenon, between two
adjacent positions on the continuum. The contrast between /u/ and /o/ is the SAME differ-
ence, although between different vowels. We could set up a proportion i:e::u:o. Given the
small number of units involved, we can give a classical broadening analysis to the second
shift: we can say that rather than adding a second parallel rule, Italian broadened /i/ /e/ to
‘all high lax vowels go to the corresponding tense mid-vowels.’ But that word ‘correspond-
ing’ is a crucial one, involving precisely the perception that ‘the same difference’ is
involved in each case.
Now, regular though the combination of /i/ /e/ and /u/ /o/ may be, pure structural
regularity was not enough to force such a change on Romanian. /u/ /o/ is no more an
inevitable result of /i/ /e/ than the replacement of honos by honor is an inevitable result
of the analogy with paradigms like that of color (as we have said, not every Latin noun with
stem-final s/r variation was restructured analogically).
Regular sound change itself is apparently spread from one lexical item to another by
analogy. Speakers must be aware of two possible phonetic realizations of the ‘same’
phonemic unit, for this lexical diffusion to be possible: a speaker who simply learned some
of the conservative forms and some of the new ones, never knowing that two forms of the
same word existed, would have no particular motivation to lexicalize the new phonetic reali-
zation in new areas of the vocabulary.29 Lexical diffusion of phonological changes eventu-
ally results in regular substitution of one phone for another, IF competing changes do not
29 John Ohala (personal communication, 1988) has pointed out to us a way in which sound change could be
initiated, for at least one particular form, without any awareness on the part of the initiator. He recounts the
story of a linguist who told a tableful of other linguists the name of her undergraduate alma mater; the listeners
all heard ‘Chulane,’ although the speaker insisted that the word pronounced was ‘Tulane.’ The allowable
range of affrication and palatalization of /th/ for this speaker was evidently different from those of her hearers;
her /th/ probably overlapped both with their /th/ and with their /c/. If the speakers in question had copied her
v
pronunciation, and if they had (recognizing /c/ as a new, replacement realization for /th/) spread this new
v
pronunciation through the lexicon, sound change might have taken place in a regular fashion eventually. The
propagators of such a change, if not the initiator, would have to be aware of two systems.
bleed possible input from each other (cf. Wang, 1969), or competing social groups do not
succeed in imposing inconsistent dialectal forms on speakers.
It is quite revealing to compare the morphological and phonological examples above
with classic examples of meaning change. A case in point is the semantic development of
the English modals, already discussed above in another context. Epistemic senses have
spread through the English modal verbs until it is close to true to say that the modals
underwent parallel semantic changes (cf. Shepherd, 1981, 1982; Traugott, 1982, 1988,
1989; Sweetser, 1982). One can very well imagine setting up proportions like ‘root may :
root must :: epistemic may : X.’ As has been argued in more detail elsewhere (Sweetser,
1982, in press), these proportions are inherently metaphorical, in that the mapping of root
onto epistemic senses involves metaphorically talking about our mental processes in terms
of external physical and social processes or modalities.
The analogical development in the modal example is the EXTENSION of the metaphori-
cal polysemy from one lexical item to another which is perceived as having a similar basic
sense. Metaphor thus structures the possible directions of analogical change in meaning.
Metonymy could also be viewed as spreading analogically: it is reasonable to think that
speakers who are used to a particularly salient species sharing the name of its genus might
tend to extend this pattern to other genera, which had not previously shared names with their
most salient subcategories.
Such analogical proportions have not produced complete regularity in the semantic
paradigm of English modals, however: can has at best a restricted epistemic use in polar
environments (negatives and questions), as in ‘You can’t be from New York with that per-
fect Boston accent.’ The extension of the root/epistemic contrast to many different words
within the modal class has not forced a complete generalization of this metaphorical map-
ping; and although the mapping is motivated by the same metaphorical structuring in every
case, actual semantic change is dependent on morpheme-specific analogical extensions.
Independent of analogical factors, there are natural metaphorical reasons for modals to
take on epistemic meanings, just as there are natural metonymic motivations for the
‘rapidly’ ‘immediately’ shifts. Even when a new epistemic sense arises with the analogi-
cal help of other extant root/epistemic polysemy patterns, these non-analogical motivations
are present as well. Our claim, then, is not that either the ‘model’ polysemy structure or the
newly created one is motivated solely by semantic analogy. Rather, the parallelism in
behavior between semantically similar words may analogically reinforce natural tendencies
in the development of lexical polysemy—without imposing completely obligatory restruc-
turing on every possible target domain. Neither phonology nor semantics, then, REQUIRES
paradigmatic parallelism in historical shifting of units, but to some extent such parallelism
does occur in both domains, and may be viewed as structured by analogy.
It should be noted, finally, that the classic examples of grammatical analogy do involve
semantic as well as phonological ‘proportions.’ The honor/honoris example is crucially
dependent on the analogy-maker’s understanding that the formal contrast (presence or
absence of an ending -is) correlates with a semantic contrast (genitive vs. nominative), and
that that MEANING-contrast is ‘the same difference’ in the two different nominal paradigms.
To take another example, English has phonological identity between its agentive suffix -er
and its comparative suffix. But if we hear a child produce badder as a comparative of bad,
we assume this is on the analogical model of adjective pairs like big/bigger, short/shorter.
We cannot imagine that the same formal contrast in pairs of verbs and nouns (run/runner) is
relevant to badder, because the semantic contrast involved in such pairs is not the same as
the one involved in comparatives. The unique facet of grammatical analogy (and in particu-
lar, levelling of morphological paradigms) is then that they involve PAIRED analogical struc-
turing at two levels; the unique result is regularization of form-meaning mappings, of a kind
which is sometimes called levelling.
Returning to the examples of semantic analogical structuring, we may say that they are
a degenerate case of the sort of structure involved in paradigmatic levelling: in the case of
the modals, there is no regular contrast in form paralleling the contrast in meaning (the
whole point is that a single form has multiple related senses, and that the relationships
between the senses are parallel). With no form contrast present, we can reduce our structure
to a purely semantic analogy between parallel meaning-contrasts. Similarly, in phonologi-
cal analogical structures, there is not a meaning-parameter.
4.4. Directions of analogical spread.

In general, most linguists would agree that a pattern is more likely to spread analogi-
cally (all else being equal) if it is more pervasively present in the language; the more regular
it already is, the more likely speakers are to regularize other areas of the linguistic system
by analogy with it. For example, with the spread of epistemic senses among the modal
verbs in English, the initial changes were presumably motivated by broad tendencies to use
concrete vocabulary in extended epistemic senses; but the later shifts were also specifically
motivated by the growing root-epistemic polysemy pattern of the other modals. Connec-
tionist theories of cognitive structure would say that the stronger a pattern of connections
has been set up, the more likely it is to extend to new areas (cf. McClelland & Rumelhart,
1986).30
Kurylowicz (1945-1949/1966) lays out some proposed universals about the direction of
analogy. Although most of these have been challenged in various ways, and some of them
have been claimed to be mutually contradictory, Kurylowicz correctly observed that there
are more and less likely choices of source and target domains for an analogy. In language, a
root with less allomorphy is more likely to be the source for analogical restructuring of
another root whose paradigm is full of allomorphy, rather than the other way around. (For
example, a possible analogical restructuring of color/coloris to colos/coloris on the model of
honos or flos was not at all likely, while the restructuring of honos which did occur was
‘natural.’) Certain members of morphological paradigms are frequently observed to be
‘pivotal’ in analogical restructuring: for example, the third person singular and first person
singular forms in verbal paradigms are more likely to be the basis for analogically created
30 Connectionist work offers interesting prospects for both the understanding of metaphorical mappings
and, as here, the reasons why humans tend to generalize in analogical ways.
forms than the second person singular or the third person plural. The tendency to restructure
away from allomorphy presumably has to do with maximizing the one-to-one fit of form-
meaning mappings. The pivotal character of certain forms seems connected with their
unmarked status: for example, as well as being pivotal in morphological change, third per-
son singular verbal forms frequently have zero morphology.
The paradigmatic factors which appear to influence the direction of phonological
change could also be seen as analogical in nature. For example, when we say that the pres-
ence of a broad systemic voicing contrast in the English stop system may have abetted the
phonologization of the voicing contrast in the English fricatives (initially motivated by
French and dialectal English loan words), we are essentially talking about an analogical
spread of a distinctive feature from one domain to another. And it would probably be more
reasonable that the stop system be an analogical model for the fricative system than vice
versa, since stops are apparently more central to the consonant system than fricatives, and
languages usually have better developed stop systems than fricative systems.31 The stop
system is thus, for various reasons, more ‘pivotal’ to the system than the fricative system.
In phonological change, broadening (‘generalization’) of a rule’s environment or
domain of application is historically common. A classic example (Kiparsky, 1968b;
Niepokuj, 1985) is the change in the structure of the vowel-lowering rule in certain dialects
of Swiss German: originally these dialects lowered /o/ to [ ] before /r/, but the environment
was later generalized to include all coronals. Icelandic (cf. King, 1973) broadened its final
devoicing rule from oral stops to all stops including nasal ones. On the other hand, we do
not in general observe phonological rules ‘narrowing,’ or coming to apply to some subclass
of the entities to which they once applied. Analogical mechanisms spreading patterns to
more general application are part of the reason why broadening is so common. For example,
we could view the Swiss German case as involving an analogy between the /or/:/ r/ contrast
and a more general contrast of /o followed by coronal/ versus / followed by coronal/. In
order for this to be an analogical structuring, there must be metonymic structuring first: only
given the metonymic relationship between /r/ and the class of coronals is there any motiva-
tion for treating the two classes of sequence in the same way with respect to some sound
change. It is not clear why narrowing is so atypical in phonology: in particular, one might
expect the relationship between /r/ and the class of coronals to be a taxonomic one, not a
partonomic one, so that in principle either broadening or narrowing might be possible.
However, analogy itself tends to broaden rather than to narrow the domain of a generaliza-
tion (it extends the generalization to new cases). Perhaps the pervasiveness of analogy
accounts in large measure for our large numbers of broadening examples in phonology and
semantics.32
31 Connectionist theories of cognitive structure would add that it seems likely that the more cases of
phonemic voicing contrast were already in the system, the more likely such a contrast would be to spread to
some other part of the system.
32 Broadening of a morpheme’s meaning also inevitably involves using the morpheme in some contexts
where more conservative speakers would have found it incorrect. Narrowing, on the other hand, involves us-
ing the morpheme in a subset of the previously accepted contexts, so that none of the uses will actually be
Overall, then, we claim not only that similar cognitive factors of category structure and
paradigmatic/syntagmatic environment are relevant to structure and change in the two
domains of sound and meaning, but also that because of these parallels in cognitive struc-
ture, metonymic and analogical restructuring will occur in similar directions in the two
domains. The differences in patterns of restructuring are not fully explained, but may be
motivated at least in part by differences in the cognitive structure of the two domains.
5. Category structure, universals and directions of change.

Next, we would like to briefly address some of the inevitable consequences of our
claim for the understanding of universals and diachrony. The specific issue we wish to dis-
cuss has to do with a distinction often found in the literature, concerning possible motiva-
tions for formal and substantive universals.
Hyman (1984), discussing the issue of the nature of explanations given for formal and
substantive universals, suggests that most commonly a distinction is made between ‘inter-
nal’ and ‘external’ explanations: ‘If the problem to be accounted for is a syntactic one, an
internal explanation will propose an account in terms of the nature of syntax itself, while an
external explanation will attempt to relate the syntactic problem to phenomena outside the
realm of syntax (e.g., semantics or pragmatics). Similarly, if the problem is a phonological
one, an internal explanation will construct a theory of phonology to account for it, while an
external explanation will seek a relation with, say, articulatory or perceptual phonetics.’
(Hyman, p. 67).
At first glance, it might seem that such distinct types of explanations can be found in
semantics as well. For example, it has been claimed that binary feature analyses for seman-
tics and phonology are made possible by a formal universal motivated by the respective
natures of semantic and phonological theory. Similarly, a substantive semantic universal is
the development of root/content meanings before epistemic ones in the modal operators (cf.
Sweetser, in press; Shepherd, 1981, 1982 for the cross-linguistic evidence for this direction
of development), while a substantive phonological universal is the assimilation of nasals to
the place of articulation of stops or fricatives rather than the other way round. Motivating
both of these presumably could be some sort of external explanation (cf. Traugott’s prag-
matic explanation of the nature of the modal shift, discussed in section 3; the explanation for
the assimilation universal lies presumably in articulatory constraints).
However, our discussion of category-structuring principles in section 1 should have
made it clear that category structuring considerations motivate both formal and substantive
universals, which are thus unified in having a common type of explanation. This motivation
has to do with the way we seem to structure categories in terms of their prototypical
members and in terms of specific types of relationships (e.g., metaphorical or metonymic)
between members. Features (semantic or phonological), as the experimental evidence
unacceptable to conservative speakers: the lack of any violation may make the change unnoticeable, and there-
fore unlikely to be adopted by more speakers.
shows, may well reflect prototypically structured categories as well. It is in this respect that
the dichotomy between internal and external explanations for language universals breaks
down. We are not arguing that there are really no universals of a formal nature (such as
phonetic features) but rather that category structure can motivate both substantive and for-
mal universals. Perhaps many so-called formal universals should really be seen as higher-
order organizational properties of the cognitive underpinnings of language (features reflect
the human need to abstract relevant parameters of contrast), while so-called substantive
universals may often reflect more particular interactions between general cognitive princi-
ples. The universal fact that languages have /s/ if they have /f/ may well have solid acoustic
and perceptual motivations, but not at the level of generality at which feature categories are
motivated.
Such an understanding of universals would by no means imply that there is no contrast
between ‘formal’ aspects of the linguistic system (taking ‘formal’ in the sense of properties
of some particular language system which cannot be predicted from universal cognitive
principles) and human cognition in general. Hyman’s contrast between the tendency to
phonetically nasalize vowels next to nasals and the issue of phonemicizing this nasalization
in a particular language is a contrast well-taken. But phonetic nasalization is a universal,
while phonemicization of this phonetic nasalization is not a universal. Any true universal of
language must have its explanation in the structure of cognition. Language-specific
phenomena, on the other hand, are constrained and motivated by general cognitive structure,
but clearly not predictively determined by it.
We agree with Hyman in finding the internal-external distinction artificial and in seek-
ing a unified approach to formal and substantive universals. But we hesitate to make a divi-
sion between external and internal explanations, in the usual sense of these terms. If we
understand formal universals such as binary features to be the result of human perceptual
and cognitive structure, rather than the result simply of the external nature of the
phenomena so categorized, then it is not clear that such universals are inherently different
from universals which are based on our cognitive categorization of other phenomena such as
vowels adjacent to nasals, or topicality, or the root-epistemic shift. To label some of these
phenomena as ‘internally’ explicable, and others as ‘externally’ motivated only makes sense
if we see some categories (the ‘internally’ motivated ones) as having a genuinely different
sort of motivation from the others.
It might seem from this last point that we are at odds with those who seek to give a
functional/external explanation to language universals. However, we do not believe that
there is actually a contradiction. In fact, viewing these as contradictory motivations would
seem to entail a rigid separation of semantics and pragmatics, a view that we do not wish to
align ourselves with.33 Rather, the approach we are taking is that a given word or morpheme
serves a particular pragmatic function by virtue of having a particular semantics associated
with it. As a result, we do not, for instance, see Traugott’s ‘external’/pragmatic motivation
33 In this respect we share the views of a broad group of linguists, including, for example, and Fillmore,
Kay and O’Connor (1988), Jackendoff (1983), and Langacker (1987).
for the semantic shift exemplified by the modals (and other semantic classes of words—cf.
section 3) as being in opposition to Sweetser’s metaphorical, and therefore more
‘internal’/semantic account of the same change. It is by virtue of their acquiring epistemic
meanings (of the sort predicted by their root meaning and the structure of the relevant meta-
phor) that modal operators can serve pragmatic functions closer to the speech situation.34
As with the phonological universals discussed above, some metaphorical and meto-
nymic structures seem far more universal than others. For example, mapping space onto
time appears to be a true universal, although—as with phonemicization of phonetic nasal
spreading—it is a fact about a particular language that it has chosen to lexicalize a temporal
sense of some particular spatial term. Other metaphorical structures which may be per-
vasive in the lexicons of particular languages may nonetheless be anything but universal: the
Indo-European ‘text as woven fabric’ (etymologically, the English word text meant
‘woven’) metaphor, for example, may be quite culture-specific. The universal metaphorical
structures are presumably determined by human experience and cognition, while the
culture-specific ones are underdetermined by the human mind and body.
Both in Hyman’s ‘internal’ areas such as phonology and syntax35 and in ‘external’
areas like phonetics and pragmatics, the same universal cognitive factors seem relevant.
The crucial distinction, as we see it, is between fully determined aspects of linguistic struc-
ture, and aspects of language which show cross-linguistic variation because they are not
fully predictable from general cognitive capacities.
6. Conclusions.
In summary, we have argued that semantic and phonological change are as different as
sound and meaning—and as similar as human category-structures are to each other. Given a
level of analysis which takes the morpheme as parallel to the phoneme, the structural
characteristics of the two kinds of change are strikingly similar. Both can be seen as involv-
ing reformation of category structure, often by parallel processes of analogy (which may
partially overlap with metaphor) or metonymy.
To return to the central question of irregularity, the level of comparison suggested
above radically changes our understanding of regularity in semantic change. Phonological
change is regular in all of the senses mentioned in our introductory section: (1) there are
regular directions of change (particularly for conditioned change) which are more natural
than others; (2) there are instances where natural classes of units change in parallel, main-
taining the contrast relationships between them, and (3) when a unit changes its value (e.g.,
when a phoneme changes its phonetic shape), all sub-manifestations of the unit are affected.
The first two of these kinds of regularity are readily observable in semantic change, and we
have cited examples above. The third is the most difficult one traditionally: semanticists are
34 As we read Traugott (1989), she sees her analysis as opposed to Sweetser’s.
35 cf. Van Oosten (1984) for a discussion of the prototype characteristics of English subjecthood as a
category, and of the English passive construction.
asked to explain why, although all instances of a phoneme change shape together, all
instances of a certain semantic class of morphemes do not necessarily change meaning
together. Setting aside the fact that this statement of the problem represents an
oversimplified understanding of phonological change (ignoring, for example, lexical diffu-
sion), our reformulation would suggest that the comparison should be between change of all
instances of a phoneme (or all instances in the relevant context), and meaning change affect-
ing the uses of a morpheme-meaning globally (as stated above, it normally affects all uses in
the relevant semantic and pragmatic context, rather than affecting only particular phrasal
uses).
As with phonological change, we cannot predict meaning change: even the most
natural changes are not obligated to happen in particular languages at particular times,
although they will be statistically much commoner than the less natural ones. What we can
do is observe regular directions and regular groupings of changes, and use them as further
evidence in our attempts to understand the nature of the units and parameters involved in
semantic structure. Since semantic structure is less readily accessible and far less well
understood than phonological structure, investigation of historical semantic regularities is a
data-source which semanticists cannot afford to neglect. Our deepest hope is that this paper
will stimulate further additions to the corpus of thorough, theoretically situated studies of
diachronic semantic phenomena.
References
Anderson, B. L. (1982). The ‘Perfect’ as a universal and as a language-specific category. In Paul J.
Hopper (Ed.), Typological studies in language: Vol. 1. Tense-aspect: Between semantics and
pragmatics (pp. 227-264). John Benjamins: Amsterdam.
_______. (1986). Evidentials, paths of change and mental maps: Typologically regular asymmetries.
In Wallace Chafe & Johanna Nichols (Eds.), Advances in discourse processes: Vol. 20. Eviden-
tiality: The linguistic coding of epistemology (pp. 273-312). Norwood, NJ: Ablex.
Berlin, B. (1972). Speculations on the growth of ethnobotanical nomenclature. Language in Society,
1, 51-86.
______ & Kay, P. (1969). Basic color terms: Their universality and evolution. Berkeley: University
of California Press.
______, Breedlove, D. E. & Raven, P. A. (1973). General principles of classification and nomencla-
ture in folk taxonomy. American Anthropologist, 75, 214-242.
Bréal, M. (1900). Semantics. (N. Cust., Trans.). London: Heinemann.
Brugman, C. (1983). The use of body-part terms as locatives in Chalcotongo Mixtec. Survey of Cali-
fornia and other Indian languages (Report No. 4). Berkeley: University of California.
_______. (1988). The story of over: Polysemy, semantics, and the structure of the lexicon. New
York: Garland Publishing.
_______ & Lakoff, G. (1988). Cognitive typology and lexical networks. In G. Cottrell and S. Small
(Eds.), Lexical Ambiguity Resolution. Palo Alto: Morgan Kaufman.
Bybee, J. L. (1985). Morphology: a study of the relation between meaning and form. Amsterdam:
John Benjamins.
Bybee, J. L. & Pagliuca, W. (1985). Cross-linguistic comparison and the development of grammati-
cal meaning. In Jacek Fisiak (Ed.), Trends in Linguistics, Studies and Monographs: Vol. 29.
Historical Semantics/Historical Word-Formation (pp. 59-83). Berlin: Mouton Publishers.
Chomsky, N. & Halle, M. (1968). The Sound Pattern of English. Cambridge, MA: MIT Press.
Coleman, L. & Kay, P. (1981). Prototype semantics: the English word lie. Language, 57, 26-44.
Fillmore, C. J. (1976). Frame semantics and the nature of language. In S. R. Harnad, H. D. Steklis,
& J. Lancaster (Eds.), Origins and evolution of language and speech (pp. 20-32). New York:
New York Academy of Sciences.
_______. (1985). Frames and the semantics of understanding. Quaderni di Semantica, 4(2), 222-
254.
Fillmore, C., Kay, P. & O’Connor, M. C. (1988). Regularity and idiomaticity in grammatical con-
structions: The case of let alone. Language, 64(3), 501-538.
Fleischman, S. (1982a). The future in thought and language. Cambridge: Cambridge University
Press.
_______. (1982b). The past and the future: Are they coming or going? In M. Macaulay & O.
Gensler (Eds.), Proceedings of the Eighth Annual Meeting of the Berkeley Linguistics Society.
Berkeley, CA: Berkeley Linguistics Society.
_______. (1983). From pragmatics to grammar: Diachronic reflections on complex pasts and futures
in Romance. Lingua, 60, 183-214.
Gamkrelidze, T. V. & Ivanov, V. V. (1973). Sprachtypologie und die Rekonstruktion der gemeinin-
..
dogermanischen Verschlusse. Phonetica, 27, 150-156.
Gentner, D., & Gentner, D. R. (1982). Flowing waters or teeming crowds: Mental models of electri-
city (Report # 4981). Cambridge, MA: Bolt Beranek and Newman.
Gentner, D. (1983). Structure-mapping: a theoretical framework for analogy. Cognitive Science, 7,
155-170.
Givón, T. (1973). The time-axis phenomenon. Language, 49, 890-925.
Hock, H. H. (1986). Principles of historical linguistics. Berlin: Mouton de Gruyter.
Hoenigswald, H. (1960). Language change and linguistic reconstruction. Chicago: The University
of Chicago Press.
Hooper, J. B. (1976). An introduction to natural generative phonology. New York: Academic Press.
Hopper, P. (1973). Glottalized and murmured occlusives in Indo-European. Glossa, 7(2), 141-66.
________. 1982. Aspect between discourse and grammar: An introductory essay for the volume. In
P. J. Hopper (Ed.), Tense-Aspect: Between Semantics and Pragmatics (pp. 3-18). John Benja-
mins.
Hopper, P. & Martin, J. (1985). Structuralism and diachrony: The development of the indefinite arti-
cle in English. In A. G. Ramat et al. (Eds.), Proceedings of the VII International Conference
on Historical Linguistics. Amsterdam: John Benjamins.
Horn, L. (1985). Metalinguistic negation and pragmatic ambiguity. Language, 61(1), 121-174.
Hyman, L. (1969). How concrete is phonology? Language, 46, 58-76.
_______. (1984). Form and substance in language universals. In B. Butterworth, B. Comrie, & O.
Dahl, (Eds.), Explanations for Language Universals (pp. 67-86). Berlin: Mouton.
Jackendoff, R. (1983). Semantics and cognition. Cambridge, MA: MIT Press.
Jaeger, J. J. (1980a). Categorization in phonology: An experimental approach. Unpublished doctoral
dissertation, University of California, Berkeley.
________. (1980b). Testing the psychological reality of phonemes. Language and Speech, 23, 233-
53.
Jeffers, R. J. (1977). Morphological reanalysis and analogy: Two case histories from Latin and
Greek. Lingua, 41, 13-24.
Johnson, M. (1987). The body in the mind: The bodily basis of reason and imagination. Chicago:
University of Chicago Press.
Katz, J. J. & Fodor, J. A. (1963). The structure of a semantic theory. Language, 39, 170-210.
Kay, P. (1975). Synchronic variation and diachronic change in basic color terms. Language in
Society, 4, 257-270.
______ and McDaniel, C. K. (1978). The linguistic significance of the meanings of basic level color
terms. Language, 54, 610-646.
Kemmer, S. (1988). The middle voice: A typological and diachronic Study. Unpublished doctoral
dissertation, Stanford University, Stanford, CA.
King, R. (1973). Rule insertion. Language, 49(3), 551-578.
Kiparsky, P. (1968a). How abstract is phonology? In O. Fujimura (Ed.). Three Dimensions of
Linguistic Theory (pp. 5-56), Tokyo: TEC.
______. (1968b). Linguistic universals and linguistic change. In E. Bach & R. Harms (Eds.),
Universals in linguistic theory, New York: Holt Rinehart.
Kurylowicz, J. (1966). La nature des procès dits ‘analogiques.’ In E. Hamp, F. Householder, & R.
Austerlitz (Eds.), Readings in Linguistics II. Chicago: University of Chicago Press. (Reprinted
from Acta Linguistica, 1945-1949, 5, 122-138)
Ladefoged, P. & Halle, M. (1988). Some major features of the International Phonetic Alphabet.
Language, 64(3), 577-582.
Lakoff, G. (1987). Women, fire and dangerous things: What categories reveal about the mind. Chi-
cago: University of Chicago Press.
_______ & Brugman, C. (1986). Argument forms in lexical semantics. In V. Nikiforidou & M. Van
Clay (Eds.), Proceedings of the Twelfth Annual Meeting of the Berkeley Linguistics Society.
_______ & Johnson, M. (1980). Metaphors we live by. Chicago: University of Chicago Press.
_______ & Turner, M. (1989). More than cool reason. Chicago: University of Chicago Press.
Langacker, R. W. (1987). Foundations of cognitive grammar: Vol. 1. Theoretical Prerequisites.
Stanford, CA: Stanford University Press.
Lounsbury, F. G. (1964). The structural analysis of kinship terms. In H. G. Lunt (Ed.), Proceedings
of the Ninth International Congress of Linguists (pp. 1073-1093). The Hague: Mouton.
McClelland, J., & Rumelhart, D. E. (1986). Parallel distributed processing: Explorations in the
microstructure of cognition. Cambridge, MA: MIT Press.
Malkiel, Yakov. (1957-1958). Diachronic hypercharacterization in romance. Archivum Linguisticum,
9, 79-113; 10, 1-36.
______. (1982). Croisement, empiètement, bousculade de verbes latins en hispano-roman (SPLEN-
DERE, EXPANDERE, EXPENDERE). Revue de linguistique romane, 47, 271-97.
Martinet, André. (1952). Function, structure, and sound change. Word, 8(1), 1-32.
_____. (1955). Economie des changements phonetiques: traité de phonologie diachronique. Berne:
Editions A. Francke.
Meillet, A. (1965). L’évolution des formes grammaticales. In Librairie Honoré Champion (Ed.),
Linguistique Historique et Linguistique Générale (Collection Linguistique publiée par la
Société de Linguistique de Paris, No. 7, pp. 130-148). Paris.
Mervis, C. B. & Rosch, E. H. (1981). Categorization of natural objects. Annual Review of Psychol-
ogy, 32, 89-115.
Niepokuj, M. (1985). A rule’s progress: Reordering in Swiss German. In M. Niepokuj, M. Van
Clay, V. Nikiforidou & M. Van Clay (Eds.), Proceedings of the Eleventh Annual Meeting of
the Berkeley Linguistics Society. Berkeley, CA: Berkeley Linguistics Society.
Ohala, J. J. (1981). The listener as a source of sound change. In C. S. Masek, R. A. Hendrick, and
M. F. Miller (Eds.), Papers from the Parasession on Language and Behavior (pp. 178-203).
Chicago: Chicago Linguistic Society.
_______. (1983). The direction of sound change. In M. P. R. Van den Broecke and A. Cohen (Eds.),
Abstracts of the Tenth International Congress of Phonetic Sciences (pp. 253-258). Dordrecht:
Foris.
Paul, H. (1880). Prinzipien der Sprachgeschichte [The Principles of Language]. Halle: M. Niemeyer.
Petruck, M. R. L. (1986). Body part terminology in Hebrew: A study in lexical semantics. Unpub-
lished doctoral dissertation, University of California, Berkeley.
Porzig, W. (1967). Das Wunder der Sprache. Bern: A. Francke.

Rosch, E. H. (1978). Principles of categorization. In E. Rosch & B. B. Lloyd (Eds.), Cognition and
Categorization (pp. 27-48). Hillsdale, NJ: Lawrence Erlbaum.
Saussure, F. (1959). A course in general linguistics. In C. Bally, A. Sechehaye & A. Riedlinger
(Eds. and Trans.), New York: McGraw-Hill. (Original work published 1915)
Shepherd, S. C. (1981). Modals in Antiguan Creole, child language, and history. Unpublished doc-
toral dissertation, Stanford University, Stanford, CA.
_______ (1982). From deontic to epistemic: An analysis of modals in the history of English,
Creoles and language acquisition. In A. Ahlqvist (Ed.), Papers from the 5th International
Conference on Historical Linguistics (pp 316-323). Amsterdam: John Benjamins.
Stern, G. (1931). Meaning and change of meaning. Bloomington, IN: Indiana University Press.
Sweetser, E. E. (1986). Polysemy vs. abstraction: Mutually exclusive or complementary? In V. Niki-
foridou and M. Van Clay (Eds.), Proceedings of the Twelfth Annual Meeting of the Berkeley
Linguistics Society. Berkeley, CA: Berkeley Linguistics Society.
_______. (1987a). Metaphorical models of thought and speech: A comparison of historical direc-
tions and metaphorical mappings in the two domains. In J. Aske, N. Beery, L. Michaelis & H.
Filip (Eds.), Proceedings of the Twelfth Annual Meeting of the Berkeley Linguistics Society.
_______. (1987b). The definition of lie: an examination of the folk models underlying a semantic
prototype. In D. Holland and N. Quinn (Eds.), Cultural models in language and thought.
Cambridge: Cambridge University Press.
_______. (1988). Grammaticalization and semantic bleaching. In S. Axmaker, A. Jaisser & H.
Singmaster (Eds.), Proceedings of the Fourteenth Annual Meeting of the Berkeley Linguistics
Society. Berkeley, CA: Berkeley Linguistics Society.
_______. (in press). From etymology to pragmatics: the mind-as-body metaphor in semantic struc-
ture and semantic change. Cambridge: Cambridge University Press.
Talmy, L. (1982). Borrowing semantic space: Yiddish verb prefixes between Germanic and Slavic.
In O. Gensler & M. Macaulay (Eds.), Proceedings of the Eighth Annual Meeting of the Berke-
ley Linguistics Society. Berkeley, CA: Berkeley Linguistics Society.
_______. (1988). Force dynamics in language and cognition. Cognitive Science, 12(1), 49-100.
Traugott, E. C. (1982). From propositional to textual and expressive meanings: Some semantic-
pragmatic aspects of grammaticalization. In W. Lehmann & Y. Malkiel (Eds.), Current Issues
of Linguistic Theory: Vol. 24. Perspectives on Historical Linguistics. Amsterdam: John Benja-
mins.
_______. (1986a). From polysemy to internal semantic reconstruction. In D. Feder, M. Niepokuj, V.
Nikiforidou, and M. Van Clay (Eds.), Proceedings of the Twelfth Annual Meeting of the Berke-
ley Linguistics Society. Berkeley, CA: Berkeley Linguistics Society.
_______. (1986b). On regularity in semantic change. In T. Eaton (Ed.), Journal of literary seman-
tics (pp. 155-173). University of Kent at Canterbury, Great Britain: Julius Groos Verlag.
_______. (1987). From less to more situated in language: The unidirectionality of semantic change.
Proceedings of the International Conference on English Historical Linguistics V. Cambridge:
Cambridge University Press.
_______. (1988). Pragmatic strengthening and grammaticalization. In S. Axmaker, A. Jaisser & H.
Singmaster (Eds.), Proceedings of the Fourteenth Annual Meeting of the Berkeley Linguistics
Society. Berkeley, CA: Berkeley Linguistics Society.
_______. (1989). On the rise of epistemic meanings in English: An example of subjectification in
semantic change. Language, 65(1), 31-55.
Trier, J. (1931). Der deutsche Wortschatz im Sinnbezirk des Verstandes. Die Geschichte eines spra-
..
chlichen Feldes. I: Von den Anfangen bis zum Beginn des 13. Jh. Heidelberg: C. Winter.
Turner, M. (1987). Death is the mother of beauty. Chicago: University of Chicago Press.
Ullmann, S. 1967. The principles of semantics (Glasgow University Publications, No. 84). Oxford:
Basil Blackwell.
_______. (1974). Words and their meanings. University Lectures. Canberra: The Australian
National University Press.
Van Oosten, J. (1984). The natures of subjects, topics, and agents: A cognitive explanation. Unpub-
lished doctoral dissertation, University of California, Berkeley.
Wang, W. S. Y. (1969). Competing changes as a cause of residue. Language, 55(1), 1-36.
Watkins, C. (1970). Language of gods and language of men: remarks on some Indo-European
metalinguistic traditions. In J. Puhvel (Ed.), Myth and Law among the Indo-Europeans. Berke-
ley: University of California Press.
Weinreich, U., Labov, W. & Herzog, M. (1968). Empirical foundations for a theory of language
change. In W. Lehmann & Y. Malkiel (Eds.), Directions for historical linguistics: A sympo-
sium. Austin: University of Texas Press.
Wierzbicka, A. (1980). Lingua mentalis: the semantics of natural language. New York: Academic
Press.
Wilkins, D. (1981). Towards a theory of semantic change. Unpublished doctoral dissertation,
A.N.U.

Cogsci 89 60 PDF

Uploaded by

Copyright:

Available Formats

Cogsci 89 60 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

Cogsci 89 60 PDF

Uploaded by

Copyright:

Available Formats

DIACHRONIC REGULARITY AND IRREGULARITY:

STRUCTURAL PARALLELS BETWEEN

Kiki Nikiforidou and Eve Sweetser

Cognitive Science Program

Institute of Cognitive Studies

University of California at Berkeley

The production of the Berkeley Cognitive Science Report Series is supported by a

Recent work in historical semantics challenges traditional views of meaning-change as

We examine some of the evidence about the structure of psychological categories in

DIACHRONIC REGULARITY AND IRREGULARITY:

Kiki Nikiforidou and Eve Sweetser

1. Feature analysis and cognitive structure in semantics and phonology.2

1.1. Problems with binarity.

1.2. Problems for objectivist semantic analysis.

1.3. Cognitive category structure in phonology.

explanation in terms of non-classical psychological categories.6 Nor does classical feature

2. Structural parallels between sound change and meaning change.

2.1. Two case studies in parallel.

2.2. Phonemes and morpheme-senses.

(1) Proto-Hellenic /a:/ Attic-Ionic /e:/

2.3. Paradigmatic structuring in sound and meaning.

2.4. Syntagmatic factors influencing change.

Our perception of phonological category-membership is strongly influenced by the

2.5. Abstraction in sound and meaning.

2.6. Syntagmatic and paradigmatic influences on regularity and

3. General trends and spatio-temporal unities: semantic and

Although it is definitely restricted by spatio-temporal boundaries, it is also regular by virtue

4. Directions of metaphor, metonymy and analogy.

4.1. Metonymic relations in phonology and semantics.

Because of the unidirectional tendency to view abstract domains in terms of concrete

4.3. Analogy in sound, meaning, and grammatical paradigms.

4.4. Directions of analogical spread.

5. Category structure, universals and directions of change.

Porzig, W. (1967). Das Wunder der Sprache. Bern: A. Francke.

You might also like