Linguistic Meaning, Communicated Meaning and Cognitive Pragmatics*

Robyn Carston

Abstract
Within the philosophy of language, pragmatics has tended to be seen as an adjunct to,
and a means of solving problems in, semantics. A cognitive-scientific conception of
pragmatics as a mental processing system responsible for interpreting ostensive
communicative stimuli (specifically, verbal utterances) has effected a transformation in
the pragmatic issues pursued and the kinds of explanation offered. Taking this latter
perspective, I compare two distinct proposals on the kinds of processes, and the
architecture of the system(s), responsible for the recovery of speaker meaning (both
explicitly and implicitly communicated meaning).

1. Pragmatics as a Cognitive System

1.1. From Philosophy of Language to Cognitive Science
Broadly speaking, there are two perspectives on pragmatics: the philosophical and the
cognitive. From the philosophical perspective, an interest in pragmatics has been largely
motivated by problems and issues in semantics. A familiar instance of this was Grices
concern to maintain a close semantic parallel between logical operators and their natural
language counterparts, such as not, and, or, if, every, a/some, and the, in the face of
what look like quite major divergences in the meaning of the linguistic elements (see Grice
1975, 1981). The explanation he provided was pragmatic, i.e. in terms of what occurs when
the logical semantics of these terms is put to rational communicative use.
Consider the case of and:
(1)

a.
b.
c.

Mary went to a movie and Sam read a novel.

She gave him her key and he opened the door.
She insulted him and he left the room.

While (a) seems to reflect the straightforward truth-functional symmetrical connection, (b) and
(c) communicate a stronger asymmetric relation: temporal sequence in (b) and a causeconsequence relation in (c). The semantic options for accounting for this are unappealing:
either a three-way ambiguity (hence three lexical items and, only one of which is
semantically identical with the logical conjunction operator), or a single item whose semantics
is considerably richer than the logical operator, in that it includes temporal and causal features.
However, we dont have to accept either of these. The Gricean approach maintains that the
natural language connective is unambiguously truth-functional and explains the richer
*

Many thanks to Richard Breheny, Sam Guttenplan, Corinne Iten, Deirdre Wilson and Vladimir
Zegarac for helpful comments and support during the writing of this paper.
Address for correspondence: Department of Phonetics & Linguistics, University College London,
Gower Street, London WC1E 6BT, UK.
Email: [email protected]

connections as a function of maxims concerning proper conversational practice; what the

words say (the semantics of the utterance) and what the speaker means diverge. So, in the
case of (1c), for instance, what is said, or the proposition expressed, is a truth-functional
conjunction while, on the basis of considerations of communicative informativeness and/or
relevance, we infer that what the speaker meant is that there is a cause-consequence relation
between the conjuncts:
(2)

what is said:
P&Q
what is meant: Q IS A CONSEQUENCE OF P

The role of pragmatics is essentially to siphon off any elements of understood meaning that
might complicate the semantics and interfere with the hoped-for parallels between logic and
natural language. The proposition meant is a conversational implicature and implicatures,
which are the result of such extra-linguistic considerations as communicative appropriateness,
have no bearing at all on the truth conditions of the utterance. Each of the and conjunction
cases in (1) is true provided just that each of the conjunct clauses is true.
On this view, the role of the communicative norms (truthfulness, informativeness,
relevance, etc) is confined to the inferential derivation of implicatures; the central truthconditional core of the utterance is given semantically. This Gricean implicature gambit has
been widely employed by semanticists in order to defend a favoured semantic analysis of
some natural language expression. For instance, Neale (1990) has preserved a Russellian
quantificational semantics for definite descriptions by treating their apparent referentiality on
particular uses as a case of conversational implicature. Others have claimed that the two
sentences in (3) (with different but co-referring names) express the same proposition and are
truth-conditionally equivalent, with the obvious difference between them being captured at the
level of implicature (for discussion, with dissociation, see Recanati 1993, 17.2):
(3)

a.
b.

Lois Lane believes Superman is valiant.

Lois Lane believes Clark Kent is valiant.

The advent of cognitive pragmatics, specifically of the relevance-theoretic approach,

has brought a rather different orientation: pragmatics is a capacity of the mind, a kind of
information-processing system, a system for interpreting a particular phenomenon in the world,
namely human communicative behaviour (see Sperber & Wilson (1986/95) and this volume).
It is a proper object of study in itself, no longer to be seen as simply an adjunct to natural
language semantics. Set within a cognitive-scientific framework, this kind of pragmatic
theorising is answerable to quite different sources of evidence and criteria of adequacy from
that of any philosophical analytical investigation. For instance, evidence from childrens
communicative development, from people with specific communicative and interpretive
difficulties or deficits and from certain psycholinguistic experiments on comprehension may
well have a bearing on an account of how the pragmatic system works, as may facts about the
functioning and architecture of other mental capacities which interact with the utterance
comprehension system, such as the language faculty and the so-called theory of mind
mechanism for interpreting peoples behaviour in terms of certain of their mental states
(beliefs, desires, intentions). Many of the papers in this issue reflect the way in which these
sorts of considerations bear on pragmatics (see, in particular, Bloom, Happe & Loth, Langdon,
Davies & Coltheart, and Papafragou).
There are (at least) three possible stances on the domain of pragmatics and so on what
sort of a cognitive system it is. In order of increasing specificity, these positions are that: (a) It
is a system for interpreting human actions/behaviour in terms of the mental states (beliefs,

intentions) underlying them (i.e. it is identical to the general theory of mind system); (b) It is a
system for the understanding of communicative behaviour, that is, for figuring out what the
producer of the ostensive behaviour is trying to communicate; (c) It is dedicated to the
understanding of specifically linguistic communicative behaviour. Obviously, no matter which
of these one takes as the domain of pragmatics, linguistic communication is included, so there
must be an interface with natural language semantics, but on this cognitively-oriented approach
to pragmatics, natural language semantics is not taken to be the point of investigative departure.
The relevance-theoretic account advocates the second position: the domain of pragmatics is a
natural class of environmental phenomena, that of ostensive (= communicative) stimuli; verbal
utterances are the central case, but not the only one, and they themselves are frequently
accompanied by other ostensive gestures of the face, hands, voice, etc, all of which have to be
interpreted together if one is to correctly infer what is being communicated. For the most part
in this paper, I will focus on linguistic ostensive stimuli (i.e. utterances).
The move from the semantic adjunct view to the cognitive system view of relevance
theory brings a range of changes with it. The components of the theory are quite different from
those of Gricean and other philosophical descriptions; they include on-line cognitive
processes, input and output representations, processing effort and cognitive effects (see
Sperber & Wilson, this volume). The phenomenon of conversational implicature is no longer
thought of as a useful tool for philosophical analysis, but rather as a representational level,
derived in a particular way and playing a particular role in the process of understanding. The
semantics of the linguistic expression type employed in an utterance, while clearly crucial to
comprehension, is seen as having just an evidential, rather than a fully determining, role in the
identification of what a speaker has explicitly communicated (what is said). This is most
obvious in the case of subsentential utterances, which abound in actual communication. These
are generally blatantly subpropositional, so have no determinate truth conditions as a matter of
their intrinsic linguistic meaning (or even linguistic meaning topped up by contextual
disambiguation and reference determination).
Consider the following very ordinary situation: its breakfast time and, coming into the
kitchen, I see my companion searching around in the lower reaches of a cupboard; knowing his
breakfast habits, I guess that hes looking for a jar of marmalade and I utter:
(4)

On the top shelf.

Although the proposition I have expressed here is something like The marmalade is on the top
shelf, the linguistic semantic input to the pragmatic processor is, arguably, just whatever
meaning the language confers on that prepositional phrase, that is, a far from fully propositional
logical form, one which consists of just a location constituent (which denotes a property).
Given that, on the particular cognitive conception of pragmatics adopted here, the
content of a communicative intention may be inferred in the complete absence of any coded
material (say, on the basis of just an ostensive facial or hand movement), it is not surprising
that when a code is involved it need do no more than provide whatever clues, whatever piece
of evidence, the speaker judges necessary to channel the inferential process in the right
direction. The linguistically encoded element of an utterance is not generally geared towards
achieving as high a degree of explicitness as possible, but rather towards keeping processing
effort down (no more than is necessary for the recovery of the intended cognitive effects), so
information that is clearly already highly activated in the addressees mind (The marmalade is
here somewhere, for instance) is often not given linguistic expression.
In the next subsection, I outline a distinction between two kinds of explanation of
mental activity, with a view to considering the kind appropriate for an account of pragmatics

construed as a cognitive system as opposed to that more characteristic of philosophical

accounts in the Gricean tradition.
1.2. Levels of Explanation: the Personal and the Sub-Personal
The distinction between the personal and the sub-personal levels of explanation of human
behaviour was first introduced by Daniel Dennett (1969). Persons are conscious thinking
agents, who engage in actions (voluntary behaviours) which can be explained in terms of
reasons, that is, in terms of commonsense psychological attributions of beliefs, desires and
practical inferences that would normally lead to such actions. A mundane example of such
personal-level explanation is the following: X picked up her umbrella before she went out of
doors because she believed it was going to rain and she wanted to stay dry. This Intentional
(belief/desire) explanation makes her action reasonable or justified, makes it an intelligible
behaviour. The hallmark of this sort of explanation is that it is normative, it is given in terms of
what ought to be the case; we find someone intelligible as a person by interpreting her
behaviour as embedded in a wider pattern of rational activity (see Elton 2000, 2). For
instance, we might explain what a speaker meant by her utterance in terms of what it would be
rational for her to have meant given the words she used in the particular context.
Sub-personal explanation, on the other hand, deals in entities and properties that can be
shown to play a causal role in the action or behaviour, without necessarily standing in rational
or normative relations to it. A physiological account in terms of the neuronal activity in the
brain which accompanies the production, or the understanding, of an utterance would have
nothing to do with considerations of people as agents with reasons and would be an obvious
instance of a sub-personal explanation. However, if the current cognitive-scientific case for an
autonomous level of unconscious syntactically-driven mental computation holds, there would
seem to be another level of sub-personal explanation, a psychological level of informationprocessing mechanisms, which is, arguably, not reducible to the neurological (see Davies
2000). Talk of the sub-personal level of description and explanation in this paper is directed
solely at this assumed level of psychological mechanisms.
On the assumption that this is a distinction which has useful application to all areas of
mental theorising (something that might be questioned), let us consider how it stands for the
case of pragmatics. At which level is an account of utterance interpretation (to be) conducted,
the personal or the sub-personal (or both)? Discussion of pragmatics within the philosophy of
language is most often conducted at the level of the person (the hearer/interpreter as person
reasoning about the speaker/actor as person). For instance, Recanati (this volume, p.00)
presents the Gricean view of pragmatics, one which he largely endorses, as follows: It
[pragmatic interpretation] is not concerned with language per se, but with human action. When
someone acts, whether linguistically or otherwise, there is a reason why she does what she
does. To provide an interpretation for that action is to find that reason, that is, to ascribe the
agent a particular intention in terms of which we can make sense of the action. ... Pragmatic
interpretation is possible only if we presuppose that the agent is rational. ... On this view of
pragmatics, understanding an utterance is one instance of a more general personal-level activity
of interpreting other peoples purposeful behaviour: the hearers interpretation of the speakers
linguistic behaviour rests on the assumption that the speaker is a rational agent acting in
accordance with certain norms (truthfulness, an appropriate degree of informativeness, etc) and
he attributes to her beliefs and intentions that provide reasons for her to have spoken as she
did. The Gricean schema for figuring out a speakers conversational implicature(s) from what
he or she has said is a clear case of such personal-level practical belief/desire reasoning; it is
conscious, rational and normative: He has said that p; there is no reason to suppose that he is
not observing the maxims, or at least the CP [Cooperative Principle]; he could not be doing this

unless he thought that q; he knows (and knows that I know that he knows) that I can see that the
supposition that he thinks that q is required; he has done nothing to stop me thinking that q; he
intends me to think, or is at least willing to allow me to think, that q; and so he has implicated
that q (Grice 1975, 50).
The relevance-theoretic approach, on the other hand, embedded as it is within the
assumptions and methods of current cognitive science, aims at a causal mechanistic account, an
account in terms of the processes of interacting sub-personal systems. In recent years, this
orientation has become particularly clearly established with the proposal that the
comprehension system is a mental module: it is fast and automatic, and, more crucial to the
position, it is domain-specific, in that it is activated exclusively by ostensive stimuli and
employs its own proprietary concepts and processing strategies and routines (see Sperber
1994b and Sperber & Wilson, this volume). This move constitutes a leap across the Fodorian
modular/nonmodular divide (Fodor 1983, 2000). Fodors persistent claim is that while input
and output systems are domain-specific, encapsulated systems, that is, modules, the central
conceptual systems are architecturally unstructured and holistic, that is, nonmodular. Alan
Leslie and others are currently making the modularity claim for another central interpretive
system, the theory of mind mechanism (ToMM) (see, for instance, Scholl & Leslie 1999,
Leslie 2000a, 2000b). The two systems are closely related (if not one and the same, as some
have claimed): the theory of mind system interprets the behaviour of others by attributing to
them such Intentional (that is, world-representing) mental states as beliefs, desires and
intentions, and the pragmatic comprehension system interprets communicative behaviour in
terms of an intention on the part of the speaker to bring about a certain belief state in the
addressee. Currently, the idea is being developed that the latter is a sub-system of the former,
that is, that the relevance-based comprehension module may be a sub-module of the more
general mental-state attributing module (see Sperber 2000 and Sperber & Wilson, this
volume).
The explanatory vocabulary in these (sub-personal-level) accounts of how we interpret
each others behaviour (whether it is communicative or noncommunicative) includes the
propositional attitude terminology (intention, belief, etc) which is typical of explanation at
the level of the person. In effect, the theory of mind mechanism is an information-processing
system which, in a presumably limited, unconscious and automatic way, computes
interpretations which are a counterpart to conscious, rational and reflective personal-level
explanations of human actions. Some of the mental states which might be cited as reasons for a
particular action by a personal-level thinker, intent on making a persons behaviour
intelligible, are given a sub-personal causal status in the workings of the theory of mind
mechanism. So, for instance, an explanation along the lines of he believes the bus is about to
arrive, and he wants to get on it and ... for someones behaviour of running towards a bus-stop
might occur as part of a rationalising personal-level explanation or as an interpretive output of
the theory of mind mechanism. Much the same convergence of personal-level explanation and
output of a sub-personal mechanism appears to hold for utterance interpretation; a personallevel explanation might have the form her reason for saying that it is late is that she wants her
addressee to believe that it is time to leave and this might be matched in the sub-personal
comprehension mechanism by an input representation, she has said it is late, and an output
representation, she intends me to believe that (she wants me to believe that) it is time to leave
(see Sperber 1994a for discussion of the multi-level metarepresentation here). Of course, the
unconscious inferential processes internal to the modular mental systems, which mediate input
and output representations, are very likely to be quite distinct from the conscious, normative
rationalisations of personal-level thinking.

I assume that a cognitive-scientific account of pragmatics is, or at least aims at, subpersonal description and explanation1. However, this assumption does not go unchallenged, as
will be seen in section 4, where Francois Recanatis account of pragmatic processes is
compared with the relevance-theoretic account. In the next section, I outline some of the
pragmatic processes that, according to both accounts, mediate the transition from linguistic
meaning to explicit utterance content.
2. Pragmatic Processes of Explicature Derivation
2.1. Linguistic Input and Pragmatic Output
A major development in pragmatics since Grices work is the recognition that linguistically
decoded information is usually very incomplete and that pragmatic inference plays an essential
role in the derivation of the proposition explicitly communicated. This is especially clear in
the case of subsentential utterances, such as that discussed above, but it holds also for the vast
majority of fully sentential cases. Various terms for this are used in the literature; the linguistic
expression employed is described as providing an incomplete logical form, a semantic
skeleton, semantic scaffolding, a semantic template, a proposition/assumption schema (see,
for instance, Sperber & Wilson 1986/95, Recanati 1993, Bach 1994, Taylor 2001). What all
of these different locutions entail is that the linguistic contribution is not propositional, it is not
a complete semantic entity, not truth-evaluable.
On the other hand, what is communicated, that is, the output of the pragmatic processor,
is usually a set of fully propositional thoughts or assumptions, which are either true or false of
an external state of affairs in the world. There are two kinds of communicated propositions,
those that are explicitly communicated and those that are implicitly communicated There is
some debate about the precise nature of this explicit/implicit distinction, how it is to be drawn,
and whether any such two-way distinction can do justice to the levels and kinds of meaning
involved in utterance interpretation. However, it is generally agreed that while implicatures
are wholly external to, and distinct from, the linguistic meaning, the proposition explicitly
communicated is, in some sense, built out of the semantic template contributed by the linguistic
expression used. There are several different, and somewhat confusing, terms in currency for
the propositions explicitly communicated, including explicature (in Sperber & Wilsons
relevance theory), what is said (in Recanatis reconstrual of the Gricean term) and
impliciture (used by Kent Bach (1994), who takes this communicated proposition to be
implicit in what is actually said). Very much the same sort of entity is denoted by these three
terms, though there are some major differences in the wider semantic/pragmatic frameworks
they inhabit, one of which is discussed in section 4.
It is uncontentious that processes of disambiguation and indexical reference assignment
play a crucial role in identifying the explicature of an utterance, but there is some disagreement
about how they are effected, about what guides or drives them, specifically about whether or
not the speakers communicative intention plays a role (hence whether or not pragmatic maxims
1

It has been suggested that the personal level, at least as conceived here, is not
primarily a level of scientific description, since its explanations are not concerned with
subsuming events under covering laws about how the world works. The description of
persons as experiencing, thinking subjects and agents is characteristic of explanation in the
philosophy of mind, though we do not rule out the possibility that these descriptions may also
figure in scientific theories. (Davies 2000, 93). For more on this contentious issue and on the
relation between the personal and the sub-personal levels, see the special issue of Philosophical
Explorations, volume 3 (1), January 2000.

or principles are involved). I wont attempt to argue it here, but given the plainly highly
context-sensitive nature of sense selection and reference assignment, I take it that they are
matters of speaker meaning, not determinable by any linguistic rule or procedure for mapping a
linguistic element to a contextual value, and so just as dependent on pragmatic principles as the
processes of implicature derivation. For more detail and argument, see Carston (2000,
forthcoming b) and Recanati (2001, this volume).
Identifying the intended sense of an ambiguous word or structure and giving values to
indexicals (also known as saturation) are mandatory processes, that is, they must be carried
out in all contexts in which the ambiguous or indexical form is used. In the next two
subsections, I consider two other kinds of contribution that, according to relevance theorists,
and certain other cognitively-oriented theorists, pragmatics can make to the derivation of the
explicature(s) of an utterance. These are optional or free processes in the sense that they
need not occur in every context in which the linguistic expression at issue is used.
2.2. Unarticulated Constituents and Free Pragmatic Enrichment
There is a wide range of cases where it seems that pragmatics contributes a component to the
explicitly communicated content of an utterance although there is no linguistic element
indicating that such a component is required. That is, there is no overt indexical, nor is there
any compelling reason to suppose there is a covert element in the logical form of the utterance,
and yet a contextually supplied constituent appears in the explicature. Consider utterances of
the following sentences, whose interpretation, in many contexts, would include the bracketed
element which is provided on pragmatic grounds alone.
(5)

a.
b.
c.
d.
e.
f.

Sally has a brain.

[VERY GOOD BRAIN ]
Something has happened.
[SOMETHING IMPORTANT/TERRIBLE]
Ive had a shower.
[TODAY]
Its snowing. [IN LOCATION X]
Mary gave John a pen and he wrote down her address. [AND THEN] [WITH
THE PEN MARY GAVE HIM ]
Sam left Jane and she became very depressed.
[AND AS A RESULT]

Given disambiguation and saturation of indexicals, each of these would, arguably,

express a proposition (hence be truth-evaluable) without the addition of the bracketed
constituent, but in most contexts that minimal proposition would not be what is communicated
(speaker meant). One class of cases, represented here by (5a) and (5b), would express a
trivial truth (after all, every person has a brain, and, at any given moment, something or other
has happened), and it is easy to set up cases of obvious falsehoods (the negations of (5a) and
(5b), for instance). Others, such as (5c) and (5d), are so vague and general as to be very
seldom what a speaker would intend to communicate (they would not yield sufficient cognitive
effects). Across most contexts in which these sentences might be uttered, obvious implicatures
of the utterance would depend on the enriched proposition: in (5a), for instance, the implicated
proposition that Sally will make an intelligent contribution to a debate; in (5c), the implicature
that the speaker doesnt need to take a shower at the time of utterance. The relevance-theoretic
position, then, is that, in the vast majority of contexts, it is the enriched propositions that are
communicated as explicatures, with the uninformative, irrelevant, and, sometimes, truistic or
patently false, minimal propositions playing no role in the process of utterance understanding.
While the issue with disambiguation and saturation processes is how they are brought
about, whether with or without pragmatic principles geared to uncovering the speakers
meaning, the issue with free enrichment is more fundamental. It is whether or not there really is

any such process, so whether or not there are such things as constituents of the explicit content
of the utterance which do not occur in any shape or form in the linguistic representation.
Philosophers of language who insist on the psychological reality of the process include
Recanati (1993, 2001) and Bach (1994, 2000). However, a current school of semantic
thinking, represented by Stanley (2000, this volume), Stanley & Szabo (2000) and Taylor
(2001), holds that if a contextually supplied constituent appears in the explicit content of an
utterance then it must have been articulated in the logical form of the utterance, whether by an
overt indexical or by a phonologically unrealised element (a covert indexical). In other words,
the only pragmatic processes at work at this level are disambiguation and saturation, and there
is a lot more saturation going on than the surface syntactic form reveals; any other process of
pragmatic inference involved in understanding an utterance results in an implicated
proposition.
What lies behind this denial of free enrichment is a particular view of natural
language semantics and its relation to linguistic communication (see Stanley, this volume). The
claim is that the truth-conditional content of an utterance is entirely determined by (a) its
logical form, and (b) the (context-relative) meanings of its most basic components (words or
covert elements), that is, it satisfies a strict principle of semantic compositionality. This is
essentially the Gricean position on what is said but with a great many more indexical
elements requiring contextual instantiation. Relevance theorists have a rather different view of
linguistic semantics, one which also complies with strict compositionality but which is not
truth-conditional. Linguistic semantics is a system of mappings between elements of linguistic
form and certain kinds of cognitive information and, as already discussed, the result of these
mappings is standardly a subpropositional schema for the (pragmatic) construction of fully
propositional representations. Truth-conditional semantics is a distinct enterprise and its
proper object is not linguistic expressions but fully propositional entities, such as thoughts and
communicated assumptions (semantic/pragmatic hybrids). On this view of linguistic semantics,
then, the possibility of constituents in the proposition explicitly communicated which have not
been articulated in the logical form of the linguistic expression does not raise any semantic
problems.2
2.3. Pragmatic Adjustments of Conceptual Encodings
Free enrichment is a process which involves the addition of a conceptual constituent to the
decoded logical form; for example, its snowing [IN ABERDEEN]. There are other cases
where it seems that a better way of construing what is going on is that a lexical concept
appearing in the logical form is pragmatically adjusted, so that the concept understood as
communicated by the particular occurrence of the lexical item is different from, and replaces,
the concept it encodes; it is narrower or wider (or some combination of the two) than the

2
There is now in the semantic/pragmatic literature quite a complex array of arguments
for and against unarticulated constituents. Stanley (2000) and Stanley & Szabo (2000) have
argued that, whenever there is thought to be a constituent of explicitly communicated content
which has been recovered on wholly pragmatic grounds, it is really the value of a hidden
indexical element in the linguistic logical form. In different ways, Bach (2000), Carston (2000),
Breheny (this volume) and Recanati (forthcoming) defend the existence of unarticulated
constituents and the pragmatic process of free enrichment. Stanley (this volume) claims to have
found a new problem for advocates of pragmatic enrichment. He argues that, as presented by
relevance theorists and by Bach (2000), it is a process that over-generates, making false
predictions about possible interpretations of utterances. This allegation remains to be addressed.

lexical concept from which it was derived. Consider an utterance of (6a) by a witness at the
trial of X who is accused of having murdered his wife; the utterance is a response to a question
about Xs state of mind at the time leading up to the murder:
(6)

a.
b.

He was upset but he wasnt upset.

X WAS UPSET* BUT X WASNT UPSET**

As far as its linguistically supplied information goes, this is a contradiction, but it was not
intended as, nor understood as, a contradiction. The two instances of the word upset were
interpreted as communicating two different concepts of upsetness (as indicated in (6b) by the
asterisks), at least one, but most likely both, involving a pragmatic narrowing of the encoded
lexical concept UPSET; the second of the two concepts carries certain implications (e.g. that he
was in a murdering state of mind) that the first one does not, implications whose applicability
to X the witness is denying.
There are many other cases where any one of a wide range of related concepts might be
communicated by a single lexical item; for instance, think of all the different kinds, degrees and
qualities of feeling that can be communicated by each of tired, anxious, frightened,
depressed, well, happy, satisfied, sweet, etc. In one context, an utterance of Im
happy could communicate that the speaker feels herself to be in a steady state of low-key
well-being, in another that she is experiencing a moment of intense joy, in yet another that she
is satisfied with the outcome of some negotiation, and so on. The general concept HAPPY
encoded by the lexical item happy gives access to an indefinite number of more specific
concepts, recoverable in particular contexts by relevance-driven pragmatic inference.
The examples considered so far have involved a narrowing or strengthening of the
encoded concept, but there are others that seem to require some degree of widening or
loosening. Consider what is most likely communicated by the highlighted lexical item in
utterances of the following sentences:
(7)

a.
b.
c.
d.

There is a rectangle of lawn at the back.

This steak is raw.
On Classic FM, we play continuous classics.
Mary is a bulldozer.

The area of lawn referred to in (7a) is very unlikely to be truly a rectangle (with four right
angles, opposite sides equal in length); rather it is approximately rectangular, and this holds for
many other uses of geometrical terms: a round lake, a square cake, a triangular face, etc.
In (7b), the steak, perhaps served in a restaurant, is not really raw but is much less cooked than
the speaker wishes; in (7c), the classical music played on the radio station is interspersed with
advertisements and other announcements, so not strictly continuous, and so on. In each case,
a logical or defining feature of the lexically encoded concept is dropped in the process of
arriving at the intended interpretation: EQUAL SIDES in the case of rectangle, UNCOOKED for
raw, UNINTERRUPTED for continuous, MACHINERY for bulldozer.
While the existence of a pragmatic process of free enrichment, as discussed in the
previous subsection, is disputed by some truth-conditional semanticists, the process of
pragmatic concept construction has not (yet) been challenged by semanticists or Griceanoriented pragmatists, perhaps because it is a relatively new player on the scene.3 Although this

3
For further discussion of the role of ad hoc concept construction within the
relevance-theoretic view of utterance understanding and its implications for the account of
metaphor, see Carston (1997) and (forthcoming a, chapter 5), Sperber & Wilson (1998),

process does not bring about a structural change in the transition from linguistic logical form to
proposition explicitly communicated, as does free enrichment (expansion), it clearly does take
us well away from encoded linguistic meaning and has no linguistic mandate, so marks yet
another considerable departure from the Gricean semantic notion of what is said.4
3. Relevance Theory and the Mutual Adjustment of Explicatures and Implicatures
According to relevance theory, the pragmatic inferential system employs the following strategy
in order to arrive at the intended interpretation of the utterance:
(8)

Consider interpretations (disambiguations, saturations, enrichments, implicatures, etc)

in order of accessibility (i.e. follow a path of least effort in computing cognitive
effects); stop when the expected level of relevance is achieved.

Interpretive hypotheses are made rapidly, on-line, and in parallel. The mechanism that
mediates the inferences from logical form to communicated propositions is one of mutual
parallel adjustment of explicatures and implicatures, constrained by the comprehension
strategy. The result should consist of (sets of) premises and conclusions making up valid
arguments, but the important point is that the process need not progress strictly logically from
the accessing of premises to the drawing of conclusions. For instance, a particular conclusion,
or type of conclusion, might be expected on the basis of considerations of relevance and, via a
backwards inference process, premises constructed (explicatures and implicatures) which will
make for a sound inference to the conclusion. The process may involve several backwards and
forwards adjustments of content before an equilibrium is achieved which meets the systems
current expectation of relevance.
Ill illustrate the process with an example which involves free enrichment. See
Sperber & Wilson (1998) and, in particular, Wilson & Sperber (2000) for examples in which
pragmatic concept construction plays a central role. Bobs utterance in (9) is a response to
Anns immediately preceding question. In such cases, expectations of relevance are quite
constrained and specific since the question has indicated the sort of information that would be
relevant (would have cognitive effects).
(9)

Ann: Shall we play tennis?

Bob: Its raining.
Explicature: ITS RAINING AT LOCATIONA/B
Implicated premise: IF ITS RAINING IN LOCATIONX THEN IT IS UNLIKELY THAT
PEOPLE WILL PLAY TENNIS AT LOCATIONX
ANN AND BOB WONT PLAY TENNIS AT LOCATIONA/B

Implicated conclusion:

Wilson & Sperber (2000), and Breheny (1999) and (forthcoming). For his related notions of
analogical transfer and metonymical transfer, pragmatic processes which contribute to the
proposition explicitly communicated, see Recanati (1993, section 14.4) and (1995).
4

Many semanticists and pragmatists follow Grice in preserving a conception of what

is said which is minimally distinct from the semantics of the linguistic expression used. Bach
(1994) aims for a wholly semantic notion, one which is free from any consideration of speaker
intentions and allows for the contextual fixing of only pure indexicals. In Carston
(forthcoming a, chapter 2) and (forthcoming b), I have argued in some detail against there being
any role for such a notion (intermediate as it is between linguistic expression type meaning and
communicated propositions) in an account of the cognitive processes and representations
involved in utterance interpretation.

In understanding Bobs utterance, the explicature constructed from the logical form has to be
enriched with a location constituent in order that the implicated conclusion is properly
warranted. In this case, the location is anchored to the place of utterance, though in a different
context it might not be, so this is a matter of pragmatic inference.
The following step by step description of the pragmatic processes involved in
understanding Bobs utterance in (10) is closely modelled on analyses given in Wilson &
Sperber (2000):
(10)

a.
b.
c

d.

e.
f.

g.

h.

Bob has uttered sentence with logical form: [it is raining] (Output of
linguistic decoding.)
Bobs utterance is optimally relevant to Ann. (Presumption of relevance.)
Bobs utterance will achieve relevance by providing an affirmative or negative
answer to Anns question. (Standard expectation created by the asking of a
yes-no question.)
If it is raining in a particular location then it is not likely that one can play tennis
in that location. (Highly accessible assumption which might help to answer
Anns question.)
It is raining at Ann and Bobs location. (First accessible enrichment of Bobs
utterance which could combine with (d) to yield an answer to Anns question.)
Ann and Bob cant play tennis at their location.
(Inferred from (d) and (e); satisfies (c); accepted as an implicature of Bobs
utterance.)
They cant play tennis at their location because it is raining at their location.
(Further highly accessible implicature inferred from (d) and (e), which,
together with (f) and various other (weaker) implicatures, such as (h),
satisfies (b), the general expectation of relevance.)
Ann and Bob will have to find some other entertainment.
They could go to the cinema, etc.

Bob has not given a direct yes/no answer to Anns question; rather, Ann has to infer an
implicated answer. The extra inferential effort required by Bobs indirect reply to Anns
question is offset by extra effects, specifically, the strongly communicated implicature in (10g)
which supplies a reason for the negative answer to her question, and perhaps other weakly
communicated implicatures, such as those in (10h).
Two caveats are in order here. First, I have given natural language paraphrases of
explicatures and implicatures here which, as always, are merely suggestive of the actual
conceptual representations involved. Second, as the comments above about the mutual
adjustment process indicate, the steps in the derivation are not to be thought of as sequential.
Interpretive hypotheses about aspects of explicit and implicit content are made on-line and
adjusted in parallel until both the hearers expectation of relevance is met and a final stable
state of sound inference is achieved.
It is clear from just this one example and the general comments about the relevancetheoretic derivation process, that we have here a considerable departure from the widely held
Gricean view of how conversational implicatures are derived and, so, of their derivational
relation to the explicit content of the utterance. According to that view, they are inferentially
derived on the basis of the antecedently determined what is said and arise as a response to a
consideration of why the speaker is saying what she said, what she means (communicatively
intends) by saying it. As will be seen in the next (and final) section, this difference of
conception is central to two opposing views on the cognitive architecture of the pragmatic

capacity.
4. How Many Pragmatic Systems?
As we have seen, there is a variety of conceptually distinct pragmatic tasks. These may or may
not involve distinct kinds of process, and distinct kinds of process may or may not involve
distinct mechanisms (or architectural units). The following three positions on these
relationships have actually been taken up by different pragmatists:
[1]

The various different pragmatic tasks are performed by processes that comprise a single
system, which takes decoded linguistic meaning as its input and delivers the
propositions communicated (explicatures and implicatures).

[2]

There is a crucial split between the processes involved in deriving explicit utterance
content, on the one hand, and the processes of implicature derivation, on the other, with
the two sets of processes each belonging to a distinct cognitive system, the output of the
first (explicature or what is said) being the input to the second.

[3]

There are distinct processes for at least some of the (conceptually) distinct pragmatic
tasks (disambiguation, indexical reference assignment, recovery of unarticulated
constituents, speech act assignment, etc) and each of these distinct processes is
performed by a distinct cognitive system.

The third position, which I wont explore here, has been adopted for purely practical
reasons by some computationalists in attempts to provide an implementation of particular
pragmatic tasks, and, on more theoretical grounds, by Asa Kasher (1991a, 1991b). The second
position is the standard one (one system for the pragmatic processes involved in the recovery
of the proposition expressed or explicature, and the other for implicature derivation). It is held
by a range of people, whose outlooks otherwise diverge considerably: Grice, for whom
conversational maxims were responsible for the derivation of implicatures, but not, it seems,
for the pragmatic processes of disambiguation and indexical reference fixing required for a full
identification of what is said; the semantic theorists, Larson & Segal (1995, chapter 1), who
assume there is a system for identifying the referents of indexicals which is distinct from a
pragmatics system (for implicature generation); the post-Gricean pragmatist, Stephen Levinson
(2000), who distinguishes a system of default rules for generating what he calls generalised
conversational implicature, which can contribute to the truth-conditional content of an
utterance, hence to what is said, and a system of general communicative principles (probably
relevance-based) for inferring particularised conversational implicatures; Recanati (1993,
1995, 2001, this volume), who makes a fundamental distinction between primary pragmatic
processes and secondary pragmatic processes.
I will focus on position [2] as it is developed in Recanatis work. Primary pragmatic
processes are all those that contribute to explicature (or what is said, in his nonGricean,
nonminimalist sense of the term), whether obligatory processes like saturation or optional ones
like free enrichment; secondary pragmatic processes are responsible for implicature
derivation. Although both kinds of pragmatic processes are wholly dependent on context (in
the widest sense), they are very different in other respects: the primary ones are associative
and free from considerations of the speakers intention, while the secondary ones are properly
inferential and require representation of the speakers intention; the two kinds of processes are
also governed by distinct principles (a principle of highest accessibility for the primary
processes, Gricean type norms for the secondary processes) and the primary ones are prior,

both logically and temporally, to the secondary ones. Indeed they are so fundamentally
different as to belong to different levels of description: The determination of what is said
takes place at a sub-personal level, ... But the determination of what the speaker implies takes
place at the personal level, ... (Recanati, this volume, p.00).
In all these distinguishing respects, this view is at odds with that of the relevancetheoretic account which, as should be evident from the preceding section, takes the first of the
three positions laid out above. There is a single pragmatic comprehension system, informed by
a single overarching principle: Every utterance (more broadly, ostensive stimulus) carries a
presumption of its own relevance (see Sperber & Wilson, this volume). The system operates
in accordance with a comprehension procedure (given in (8) above) which is dedicated to the
processing of communicative stimuli and distinct from the procedures of other systems (such as
the general theory of mind system). Recovery of the two kinds of communicated assumption,
explicatures and implicatures, proceeds in parallel and is effected by a process of mutual
adjustment which may involve processes of inference from implicature to explicature as well
as from explicature to implicature. The function of this single pragmatics system is to recover
speaker meaning, that is, what the speaker communicatively intended, but this is a sub-personal
system and so does not require conscious reflection on what that intention is.
In the little space remaining, I will consider just two of the many differences between
Recanatis binary position and the relevance-theoretic unitary position: (a) the question of the
temporal order of processes of explicature and implicature derivation, and (b) the issue of
kinds and levels of processing.
There is a reasonably clear sense in which explicatures are logically prior to
implicatures: explicatures function as premises in sound patterns of inference in which (some,
at least) implicatures are conclusions. But Recanati is also claiming that explicatures are
temporally prior; they are the output of a system of primary pragmatic processes and the input
to a system of secondary pragmatic processes which result in implicatures. If the relevancetheoretic view is right, however, there is no generalization to be made about which of the two
kinds of communicated assumption is recovered first and functions as input to the recovery of
the other; the parallel adjustment process entails that neither is wholly temporally prior to the
other. An addressee may have quite specific expectations of relevance that, as it turns out,
pertain to information which the speaker implicates, so that the pragmatic development of the
linguistic logical form is, at least partly, made in order to provide inferential grounding for that
implicature.
Setting aside the specifics of the relevance-theoretic view, there is a class of widely
recognized implicatures, known as bridging implicatures, which have to precede the full
derivation of an explicature (see Clark 1977, Levinson 2000, Matsui 2000). These are
contextual assumptions that must be accessed, whether or not they are already known to the
addressee, in order to identify a referent. So, for instance, in order to identify the referent of
the beer in (11a), the addressee has to access the implicature in (11b):
(11)

a.
b.

The picnic was awful. The beer was warm.

The beer was part of the picnic.

In the absence of any argument that denies the status of implicature to assumptions like those in
(11b), they seem to present strong evidence in favour of a system of pragmatic interpretation
which derives explicatures and implicatures in parallel.
Turning now to the issue of kinds of process and explanatory levels, consider the
following statement by Recanati (this volume, p.00):

As Grice emphasized, implicature-determination in the strict sense is a reflective

process. Instead of merely retrieving what is said through the operation of unconscious,
primary pragmatic processes, we reflect on the fact that the speaker says what he says
and use that fact, together with background knowledge, to infer what the speaker means
without saying it. As Millikan writes, the true communicator is in a position to tinker
with the mechanisms of normal language flow, is sensitive to symptoms that the other is
tinkering with these mechanisms, and can rise above these automatic mechanisms if
necessary (Millikan, 1984: 69). That is what happens in special cases. The retrieving
of conversational implicatures, in particular, involves reflective capacities that are not
exercised in what Millikan calls normal language flow.
Leaving aside Millikans own concerns and focussing just on Recanatis use of her
views for his own purpose, there is a strong (and, to me, highly implausible) claim here that
linguistic communication involving implicatures is special and abnormal, in some sense, that
implicatures are only derived when something has gone awry with the normal automatic smooth
processes of linguistic communication. With this claim in mind, lets consider some examples
of utterances which clearly communicate an implicature, starting with the first example Grice
gave when illustrating the role of his conversational maxims (Grice 1975, 51):
(12)

A:
I am out of petrol
B:
There is a garage round the corner
(Gloss: B would be infringing the maxim Be relevant unless he thinks, or thinks it
possible that the garage is open, and has petrol to sell; so he implicates that the garage
is, or at least may be, open, etc.)

Now, B could have given a more explicit response to A, one in which the information that
petrol is currently being sold at a garage round the corner is part of what is said by the
utterance. For instance, she could have uttered the sentence in (13):
(13)

B:

There is a garage round the corner which sells petrol and is open now.

According to the view just given, this utterance would have maintained the normal language
flow while the one B actually gave, in (12), disrupts that normal flow. It is perhaps difficult to
have a sure sense of what is meant by the notion of normal language flow, but the exchange in
(12) seems to be about as natural, normal and flowing a conversation as there is, while,
arguably, the implicature-less one in (13) is somewhat awkward, being quite unnecessarily
explicit (in the absence of any doubt about the functioning of the garage).
Furthermore, if there is any statistical basis to the use of the word normal, then
Recanatis claim cannot be right, since the majority of our exchanges are implicature-laden.
Note in this regard that, in (12), B has taken A to have implicated that she wants some petrol
and his utterance is a response to that implicature. According to the relevance-theoretic single
system view, the processes of explicature and implicature derivation proceed on-line, in
parallel and in response to each other, without any major switch of processing mode, thereby
reflecting what seems to be the normal communication flow of exchanges such as that in (12).
This is what we would expect from a system which has evolved to solve the adaptive problem
of figuring out a speakers meaning, which may consist of just an explicature but, more often
than not, consists of implicatures and explicature.
In the next example, the first part of Bs utterance is a direct explicit response to As
question, but it raises a further (implicit) question what does B want?, which the second
segment of Bs utterance answers indirectly; B implicates that she wants some paper:

(14)

A:
B:

Do you want something?

Yes. # Ive run out of paper.

Having processed Bs utterance up to the point marked by #, A is very likely to have formed an
anticipatory assumption schema [B wants __ ] for which the next section of Bs utterance
provides a completion. It so happens that the resulting proposition B wants some paper is an
implicature of the next part of Bs utterance. Had the second unit been Some paper or I want
some paper, the answer would have been direct and the completed assumption schema would
have been an explicature of the utterance. According to Recanatis view quoted above, the
understood answer, that is the completed schema, B wants some paper, must be achieved by
fundamentally different kinds and levels of processes in the two cases, and the one in which it
is an implicature is in some way special, is brought about by a disruption in the linguistic
communicative flow. However, this is not supported by intuitions about the two possibilities,
and, in the absence of compelling arguments, it is difficult to see why we should adopt this
view.
Moreover, if the idea that utterance comprehension processes include the formation of
anticipatory hypotheses is correct, the binary view imposes a very odd requirement: particular
hypotheses, such as the schema above, would have to be categorised as a feature of one kind of
processing system (unconscious, subpersonal-level) in the case where it turns out to be
completed as an explicature, but part of a distinct one (conscious, personal-level) when it is an
implicature. According to the relevance-theoretic view, such anticipatory hypotheses about
where the relevance of an utterance is going to lie are a common occurrence and, given the
single system of interlocking processes of explicature and implicature derivation, the problem
of their having to switch from one sort of status to another does not arise.
There clearly are times at which the normal communicative flow is disrupted: certain
instances of garden-path utterances, especially when exploited by speakers for particular, often
humorous, effects; some cases of complex figurative use which require an effortful conscious
search for an interpretation; other cases where there is some apprehended difficulty in
satisfying oneself that the intended interpretation has been reached (it doesnt seem sufficiently
relevant, for instance). The appropriate distinction between modes of processing and levels of
explanation would seem to be between, on the one hand, a modular (sub-personal) pragmatic
processor which, when all goes well, quickly and automatically delivers speaker meaning
(explicatures and implicatures), and, on the other hand, processes of a conscious reflective
(personal-level) sort which occur only when the results of the former system are found wanting
in some way.
To conclude, these interesting issues in the study of linguistic communication have
arisen only since pragmatics has moved from its place of origin in philosophy to its new
location within cognitive science. Clearly, there is a long way to go before they are fully
resolved and it seems very likely that empirical evidence from experiments on the time course
of processing, from child development and from people with communicative deficits will play
an important part in their resolution.
Department of Phonetics and Linguistics
University College London
References
Bach, K. 1994: Conversational impliciture. Mind and Language, 9, 124-162.
Bach, K. 2000: Quantification, qualification, and context: a reply to Stanley and Szabo. Mind
and Language, 15, 262-283.

Breheny, R. 1999: Context-dependence and procedural meaning: the semantics of definites.

University College London PhD thesis.
Breheny, Richard. forthcoming: Maximality, negation and plural definites. Ms. Research
Centre of English and Applied Linguistics, University of Cambridge.
Carston, R. 1997: Enrichment and loosening: complementary processes in deriving the
proposition expressed? Linguistische Berichte, 8, Special Issue on Pragmatics, 103127.
Carston, R. 2000: Explicature and semantics. UCL Working Papers in Linguistics, 12, 1-44.
To appear in S. Davis & B. Gillon, (eds.), forthcoming. Semantics: A Reader. Oxford:
Oxford University Press.
Carston, R. forthcoming a: Thoughts and Utterances: the Pragmatics of Explicit
Communication. Oxford: Blackwells.
Carston, R. forthcoming b: Relevance theory and the saying/implicating distinction. In L. Horn
& G. Ward (eds.), Handbook of Pragmatics. Oxford: Blackwells.
Clark, H. 1977: Bridging. In P. Wason & P. Johnson-Laird (eds.), Thinking: Readings in
Cognitive Science, 411-420. Cambridge: Cambridge University Press.
Davies, M. 2000: Interaction without reduction: The relationship between personal and subpersonal levels of description. Mind and Society, 1, 87-105.
Dennett, D. 1969: Content and Consciousness. London: Routledge & Kegan Paul.
Elton, M. 2000: The personal/sub-personal distinction: an introduction. Philosophical
Explorations 3, 2-5.
Fodor, J. 1983: The Modularity of Mind. Cambridge, Mass.: MIT Press.
Fodor, J. 2000: The Mind Doesnt Work That Way. Cambridge, Mass.: MIT Press.
Grice, H.P. 1975: Logic and conversation. In P. Cole & J. Morgan (eds.), Syntax and
Semantics 3: Speech Acts, 41-58. New York: Academic Press. Reprinted in H.P.
Grice 1989, 22-40.
Grice, H.P. 1981: Presupposition and conversational implicature. In P. Cole. (ed.), Radical
Pragmatics, 183-198. New York: Academic Press.
Grice, H.P. 1989: Studies in the Way of Words. Cambridge, Mass.: Harvard University Press.
Kasher, A. 1991a: Pragmatics and the modularity of mind. In S. Davis (ed.), 1991.
Pragmatics: A Reader, 567-582. Oxford University Press.
Kasher, A. 1991b: On the pragmatic modules: A lecture. Journal of Pragmatics 16, 381-397.
Larson, R. & Segal, G: 1995. Knowledge of Meaning: An Introduction to Semantic Theory.
Cambridge, Mass.: MIT Press.
Leslie, A. 2000a: How to acquire a representational theory of mind. In D. Sperber (ed.),
Metarepresentations: a Multidisciplinary Perspective, 197-223. Oxford: Oxford
University Press.
Leslie, A. 2000 b: Theory of mind as a mechanism of selective attention. In M. Gazzaniga
(ed.), The New Cognitive Neurosciences, 1235-1247. Cambridge, Mass.: MIT Press.
Levinson, S. 2000: Presumptive Meanings: The Theory of Generalized Conversational
Implicature. Cambridge, Mass.: MIT Press.
Matsui, T. 2000: Bridging and Relevance. Amsterdam: John Benjamin.
Millikan, R. 1984. 1984: Language, thought, and Other Biological Categories. Cambridge,
MA: MIT Press.
Neale, S. 1990: Descriptions. Mass.: MIT Press.
Recanati, F. 1993: Direct Reference: From Language to Thought. Oxford: Blackwell.
Recanati, F. 1995: The alleged priority of literal interpretation. Cognitive Science, 19, 207232.
Recanati, F. 2001: What is said. Synthese, 128, 75-91.

Recanati, F. forthcoming: Unarticulated constituents. Linguistics and Philosophy

Scholl, B. & Leslie, A. 1999: Modularity, development and theory of mind. Mind and
Language 14, 131-153.
Sperber, D. 1994a: Understanding verbal understanding. In J. Khalfa (ed.), What is
Intelligence?, 179-198. Cambridge: Cambridge University Press.
Sperber, D. 1994b: The modularity of thought and the epidemiology of representations. In L.
Hirschfeld & S. Gelman (eds.), Mapping the Mind: Domain Specificity in Cognition
and Culture, 39-67. Cambridge: Cambridge University Press.
Sperber, D. & Wilson, D. 1986/95: Relevance: Communication and Cognition. Oxford:
Blackwells.
Sperber, D. & Wilson, D.1998: The mapping between the mental and the public lexicon. In P.
Carruthers & J. Boucher (eds.), Language and Thought: Interdisciplinary Themes,
184-200. Cambridge: Cambridge University Press.
Stanley, J. 2000: Context and logical form. Linguistics and Philosophy, 23, 391-434.
Stanley, J. & Szabo, Z. G. 2000: On quantifier domain restriction. Mind and Language, 15,
219-61.
Taylor, K. 2001: Sex, breakfast, and descriptus interruptus. Synthese, 128, 45-61.
Wilson, D. & Sperber, D. 2000: Truthfulness and relevance. UCL Working Papers in
Linguistics, 12, 215-254.