Truth-Conditional Content and Conversational Implicature∗

?
Truth-Conditional Content and
∗
Conversational Implicature
ROBYN CARSTON
1 Background and Overview
1.1 Grice on Implicature
Grice made a distinction between what is said by a speaker of a verbal ut-
terance and what is implicated. What is implicated might be either conven-
tional (that is, largely generated by the standing meaning of certain linguis-
tic expressions, such as ‘but’ and ‘moreover’) or conversational (that is,
dependent on the assumption that the speaker is following certain rational
principles of conversational exchange). What appears to have bound these
rather disparate aspects of utterance meaning together, and so motivated the
common label of implicature, was that they did not contribute to the truth-
conditional content of the utterance, that is, the proposition it expressed, or
what the speaker of the utterance said.
This truth-conditional/non-truth-conditional distinction was essential to
Grice in his concern to defeat the ‘illegitimate use’ arguments of a certain
group of ordinary language philosophers (Grice 1967, lecture 1). I will not

∗ Many thanks to Alessandro Zucchi for his lucid challenge to my main argument, and
to Richard Breheny and Deirdre Wilson for very helpful discussion of the issues.


The Semantics/Pragmatics Distinction.
Claudia Bianchi (ed.).
Copyright © 2004, CSLI Publications.
1

---

2 / ROBYN CARSTON
review those arguments here, but the utility of the distinction and the line of
argument it enabled can be demonstrated with the following example. It is
odd to produce an utterance of the sentence ‘This looks red to me’, refer-
ring with ‘this’ to a patently red pillar-box directly in front of one and in
good lighting conditions. However, this oddness need not militate against
the use of such statements in a theory or analysis (in this case, of percep-
tion), as some philosophers had argued, because the statement made (the
proposition expressed/said) by the utterance is perfectly true and that is all
that matters for the theory or analysis. The oddness or infelicity lies outside
the truth-conditional content of the utterance; it is due (merely) to the con-
versational implicature that such an utterance would be likely to convey:
that there is some doubt about the redness of the pillar-box, an implication
which, in the given circumstances, is false. A similar story can be run for a
case of conventional implicature which gives rise to some conversational
infelicity (e.g. ‘This looks red to me but it is red’). The general situation is
summarized as follows:

(1) what is said
vs.
what is implicated
truth-conditional

non-truth-conditional


if false, utterance is false
if false, utterance is merely odd

According to the standard interpretation of the Gricean account, what is
said (the truth-conditional content of the utterance) is very closely related to
the conventional meaning of the linguistic expression employed. Of course,
that linguistic expression may include ambiguous or indexical elements, so
that contextual considerations have to be brought to bear for a full determi-
nation of ‘what is said’ (Grice 1975: 44-45). However, it seems that Grice
conceived of the role of his Cooperative Principle and system of conversa-
tional maxims (quality, quantity, relevance and manner) as confined to the
determination of conversational implicatures; that is, these maxims come
into play in resolving the issue of why a speaker, who is assumed to be a
rational agent, has said what she has said, or, in other words, what she
means (intends to communicate) by having uttered a particular linguistic
expression. This then leaves open the questions of how it is that the referent
of a use of ‘she’ or ‘that’ is determined and how the intended sense of an
ambiguous form like ‘coach’, ‘bank’ or ‘bug’ is determined. On this mat-
ter, Grice was essentially silent, mentioning just a vague criterion of best
1
contextual fit.

1 However, Neale (1992) suggests that Grice may have envisaged some kind of relevance
maxim as playing a role in disambiguation and reference resolution; this hinges on the inter-

---

CONTENT AND IMPLICATURE / 3

1.2 Semantic Underdeterminacy and Grice’s circle

The issue of how context-sensitive aspects of truth-conditional content are
determined by an addressee/interpreter has become more pressing in recent
years, as more and more pragmatists have come to accept the ‘semantic
underdeterminacy’ view of verbal utterances. According to this view, the
discrepancy between the explicit content (what is said) of an utterance and
the conventional (or ‘encoded’) meaning of the linguistic expression em-
ployed is far greater than that presented by ambiguous words and overtly
indexical expressions, and pragmatic inference (that is, maxim-guided in-
ference) is required to make up the shortfall. Some of the cases discussed
in the literature as instances involving this underdeterminacy are listed in
(2)-(6):

(2)
a.
I slept well. How about you?

b.
I haven’t eaten yet.

(3)
a.
It’ll take time for your knee to heal.

b.
Your application requires some processing.

(4) a. Everyone
left
early.

b.
There’s nothing on.

(5)
a.
You’re not going to die.



[uttered by mother to small child wailing over a


scratched
elbow]

b.
She gave him the key and he opened the door.

c.
The road layout had changed and she lost her way.

(6)
a.
Only 22,000 miles. Like new.



[uttered by a used car salesman]

b.
Ann:
Mary is refusing to answer my emails.

Bob: Typical.

Let me briefly indicate some of the elements of the propositional content of
these utterances which seem not to be linguistically specified: in (2), the
relevant temporal spans of the sleeping and the not eating are considerably
narrower than that encoded in either the simple past or the past perfect (I

pretation of a passage in Grice’s early paper on meaning, reprinted in Grice (1989b, 222).
For discussion, see Neale (1992: 530) and Carston (2002: 105-6).

---

4 / ROBYN CARSTON
slept well last night, I haven’t eaten dinner yet this evening); similarly in
(3), the ‘taking time’ and the ‘some processing’ are not understood as in-
volving just any quantity but as an amount relevant to mention in that con-
text (perhaps more time/processing than the addressee appears to expect in
each case); in (4), the domain of the quantifiers, ‘everyone’ and ‘nothing’
has to be determined (perhaps: everyone at such and such a party, nothing
worth watching on television); in (5a) (example due to Bach 1994) and
(5b), meaning expressible using prepositional phrases seems to be recov-
ered: from that scratch and with that key; (5b) is further enriched with a
temporal ordering relation so the event in the first conjunct is understood as
preceding that in the second, and in (5c), as well as the temporal order, a
cause-consequence relation is understood as holding between the first and
second conjuncts, though there is no linguistic element encoding either of
these relations; finally, the (nonelliptical) subsentential utterances in (6)
require substantial recovery of contextually available material in determin-
ing the proposition expressed.
Among those who support the underdeterminacy view are relevance
theorists, such as Sperber & Wilson (1986/95) and Carston (1988, 2002),
and philosophers and linguists, who follow Grice to varying degrees, in-
cluding Recanati (1989, 1993, forthcoming/2003), Bach (1994, 2000),
Stainton (1994), Levinson (1988, 2000) and Neale (forthcoming), though
some of them might not agree that all the examples given in (2)-(6) are per-
tinent cases. There is a fair amount of variation in the proposals that differ-
ent theorists make for a semantic/pragmatic account of the underdetermi-
nacy phenomena and in their analyses of particular cases. In this paper, I
shall start by looking at some of Stephen Levinson’s recent observations
about the problem underdeterminacy poses for the classical Gricean account
and go on to consider the direction in which he looks for a solution, com-
paring it with the approach pursued within relevance theory.
According to Levinson, this situation gives rise to a kind of circularity,
an untenable interdependence, between saying and implicating (as Grice
conceived of them):

Grice’s account makes implicature dependent on a prior determina-
tion of “the said”. The said in turn depends [on implicature: it depends] on
disambiguation, indexical resolution, reference fixing, not to mention el-
lipsis unpacking and generality narrowing. But each of these processes,
which are prerequisites to determining the proposition expressed, [may]
themselves depend crucially on [processes that look indistinguishable
from] implicatures. Thus what is said seems both to determine and to be
determined by implicature. Let us call this Grice’s circle.

---

CONTENT AND IMPLICATURE / 5
.... Then truth-conditional content depends on most, perhaps all, of the
known species of pragmatic inference; or the theory of linguistic meaning
is dependent on, not independent of, the theory of communication.



(Levinson 1988: 17-18; 2000: 186-87)
[Square brackets in the quote indicate disparities between
the 1988 and 2000 versions.]

As given here, there seem to be two distinct charges of circularity: (a) that
between saying and implicating; (b) that between semantics and pragmatics.
Of course, for those who equate linguistic meaning with what is said, as
Levinson seems to do here, there is no such distinction to be made. How-
ever, there is a perfectly respectable sort of theory of linguistic meaning for
which this equation does not hold and which does not depend on a theory of
communication, at least not synchronically. This is the account of the
meaning or information encoded in linguistic expression types, which pro-
vides the scaffolding on which processes geared to the recovery of commu-
nicated meaning (speaker meaning) build. The linguistic meaning of a
phrase or lexical item is obviously not propositional, and the linguistic
meaning of a sentence is also not generally, if ever, fully propositional.
What it provides is a template or schema, that is, clues to, or constraints on,
the process of recovering the proposition the speaker intended to express. It
is, plausibly, the output of an encapsulated language processor, hence free
from the modifications that come with access to extra-linguistic context and
speaker intentions. This kind of semantics is, by definition, independent of
the account of communication; thus there is no circularity between seman-
tics so construed and pragmatics.
Be that as it may, I shall concentrate here on the alleged say-
ing/implicating circle, leaving aside for now the issue of whether that dis-
tinction is to be usefully equated with a semantics/pragmatics distinction.
This interdependence is, I think, an inescapable issue once one accepts that
the proposition expressed (what is said) is heavily dependent on pragmatics
(the underdeterminacy thesis) and puts this together with the standard
Gricean assumptions given in (7a) and (7b):

(7) a. All pragmatically-derived (maxim-dependent) meaning

constitutes conversational implicature.
b. Conversational implicatures arise from the application of

conversational maxims to ‘the saying of what is said’ and

so require the prior determination of what is said.

---

6 / ROBYN CARSTON
The point is that these Gricean assumptions are not compatible with there
being pragmatic (maxim-driven) input to what is said, and as the truth of
the latter seems indisputable (anyway, it is not disputed by any of the theo-
rists I am discussing here), something has to give in the Gricean account.

1.3 Outline of Some Possible Solutions

The solution favored by semanticists such as Jason Stanley, Zoltan Szabo
and Jeffrey King involves a revision of (7a): ‘broadly Gricean mechanisms’
do play a crucial role in determining what is said (Stanley & Szabo 2000:
236). However, this role is kept clearly distinct from their role in determin-
ing implicatures. The maxims have only a ‘weak’ pragmatic effect on
‘what is said’, that is, they merely supply values to indexicals that occur in
the linguistic form of the utterance (a process known as ‘saturation’). Im-
plicature derivation is a ‘strong’ pragmatic effect in that it is ‘free’ from
linguistic mandate or control. The Gricean assumption in (7b) can be main-
tained on this view; the maxims can first perform their role of determining
the semantic content of the utterance (what is said), which is then input to
the next phase, that of implicature derivation.
Among the various solutions to the problem at hand, this one is the
most ‘semantic’ and the most preservatory of the original Gricean story. As
Stanley (2000: 391) puts it: ‘all truth-conditional effects of extra-linguistic
context can be traced to logical form’. The cost of the approach, however,
is the positing of a great many covert indexical elements in linguistic form,
one for every instance of a pragmatically derived contribution to explicit
utterance content, each of which requires independent justification. I will
not address this account in any detail in the rest of the chapter, though cer-
tain details of King & Stanley’s (forthcoming) work along these lines will
be mentioned when they bear on the analyses I do want to discuss; see Bre-
2
heny (forthcoming) for an assessment of their account.
Levinson, on the other hand, accepts that there are ‘strong’ (as well as
‘weak’) pragmatic effects on truth-conditional content. He does not offer

2 On the face of it, indexical saturation accounts seem more plausible for certain cases of
pragmatic effects on propositional content than for others, for instance, the case of quantifier
domain restriction as opposed to, say, the case of causal connections between the conjuncts in
cases of ‘and’-coordination. However, on the one hand, both Bach (2000) and Neale (2000)
argue against Stanley & Szabo’s (2000) indexical account of quantifier domains, and, on the
other, King & Stanley (forthcoming) argue for an indexical saturation account of the prag-
matic contribution to ‘and’-conjunctions in particular linguistic contexts (see section 3 be-
low). For more general arguments against the indexicalist approach and in favor of ‘strong’
pragmatic effects on propositional content, see Carston (2000) and Recanati (2002).

---

CONTENT AND IMPLICATURE / 7
any sort of overall solution to the circularity problem that he has empha-
sized as arising from this, but suggests that his independently motivated
theory of default (or generalized) conversational implicature can make sub-
stantial inroads on it, reducing its dimensions.
He draws a theoretical and empirical distinction between two kinds of
conversational implicature: generalized (GCI) and particularized (PCI).
The following exchange exemplifies the two kinds of implicature, each of
which, in his view, is derived by a distinct kind of inferential process, gov-
erned by distinct pragmatic maxims or heuristics:

(8)
A:
Did the children’s summer camp go well?

B:
Some of them got stomach ‘flu.


GCI:
Not all the children got stomach ‘flu.


PCI:
The summer camp didn’t go as well as hoped.

While the PCI of B’s utterance depends on the context provided by A’s
question and would not arise in a different context (e.g. a context in which
the issue is whether all the children were able to sit their exams), the GCI
arises quite generally across contexts and only drops out if it encounters a
context with which it is inconsistent.
The claim, then, is that this sort of generalized pragmatic inference can
contribute to the propositional content of certain kinds of utterance, such as
that in (9). The second part of the utterance, the comparative, makes a per-
fectly coherent statement, but this is only possible if we interpret ‘some’ (of
your exams) as some but not all (of your exams).

(9)
You shouldn’t be too upset about failing some of your exams; it’s

much better than failing the whole lot.

On this approach to the phenomenon of pragmatic contributions to the
proposition expressed by an utterance, the Gricean assumption in (7a) ap-
pears to be preserved, though, in fact, only by virtue of an extreme loosen-
ing of the sense of the term ‘implicature’ (this point will be pursued later),
while (7b) is modified so that a restricted kind of pragmatic inference (lo-
cal, default) can apply without the prior determination of ‘what is said’.
This sort of inference is generated by a system of default rules which are
attached to particular lexical items and constructions, the word ‘some’ in
the case of (8) and (9). These rules are implemented in parallel with the
process of linguistic decoding, so in (8), for instance, the inference to not
all the children will have been made before the predicate ‘got stomach ‘flu’
is processed.

---

8 / ROBYN CARSTON
However, as Levinson acknowledges (2000: 236-39) some PCIs also
seem to affect truth-conditional content and, as I’ll claim later, a substantial
subset of cases that fall in his class of GCIs do not, so in fact this approach
makes quite limited headway with the task of accounting for the underde-
terminacy phenomena and the issue of Grice’s circle.
A third approach also recognizes strong (as well as weak) pragmatic ef-
fects on truth-conditional content, that is, pragmatic contributions which are
not triggered by the requirement of indexical saturation but are entirely
pragmatically motivated, but it does not construe these as a kind of implica-
ture. Rather, it makes a distinction between these ‘free’ pragmatic contribu-
tions to content, on the one hand, and conversational implicatures, on the
other, which as Grice maintained, lie outside truth-conditional content.
This position, which is known as ‘truth-conditional pragmatics’, is held by
pragmatists across a range of otherwise different frameworks, including
Francois Recanati (1989, 1993, 2003), Kent Bach (1994, 2000), Anne Be-
zuidenhout (1997, 2002b), and Stephen Neale (2000, forthcoming). It has
also been central to the relevance-theoretic framework since its inception
(Sperber & Wilson (1986/95), Carston (1988, 2002)), and it is this particu-
lar manifestation of the truth-conditional pragmatic position that I’ll call on
in the rest of the paper. The relevance-theoretic term for the pragmatically
imbued level of truth-conditional content is ‘explicature’, while Recanati,
Levinson, King & Stanley and others continue to use the term ‘what is
said’; in what follows I shall use the terms interchangeably depending on
whose work I am discussing, but it is worth bearing in mind that this con-
ception of ‘what is said’ is quite different from Grice’s original conception
(which was independent of considerations of speaker intentions, hence of
maxim-driven inference).
Relevance theorists make no distinction of any theoretical import be-
tween generalized and particularized implicatures. Of course, implicatures
vary in their generality, some being very general, others less so, and some
being essentially one-off (nonce), but this is a continuum situation. No im-
plicatures are a matter of default inference; rather, all must be warranted by
contextual relevance. The Gricean assumption in (7a) is dropped, since one
and the same pragmatic principle (based on the concept of ‘optimal rele-
vance’) is responsible for both all cases of conversational implicature and
all pragmatic contributions to truth-conditional content. The assumption in
(7b), that ‘what is said’ is determined prior to the derivation of conversa-
tional implicatures, is also relaxed and the two levels of communicated con-
tent are taken to be derived in parallel via a mechanism of ‘mutual adjust-
ment’, so that, for instance, an interpretive hypothesis about an implicature
might lead, through a step of backwards inference, to a particular adjust-
ment of explicit content. For detailed justification and exemplification of

---

CONTENT AND IMPLICATURE / 9
this account, see Wilson & Sperber (2002) and Carston (2002: section
2.3.4).
By way of brief illustration, omitting all technical detail, let us consider
B’s utterance in each of the following exchanges:

(10)
A: Will John get any support from accident compensation?

B: Someone left a manhole cover off and John broke his leg.

Implicature:
John will get accident compensation payments.

(11)
A: I was planning to climb Mt Snowdon next week.
B: Your knee needs time to heal properly.

Implicature:
A should not go mountain-climbing next week.

In (10), the answer to A’s question is indirect; B conversationally implicates
that John will get financial compensation. We know that such financial
compensation depends on the cause of the incapacity to work being negli-
gence in the workplace, hence the statement here is not just that the two
events expressed by the conjuncts took place, nor just that they took place
in the given order, but the richer proposition that there is a cause-
consequence relation between them. Given that A’s question narrowly cir-
cumscribes the expected relevance of B’s response (essentially to a ‘yes’ or
‘no’), it is fairly easy to see that this expectation warrants the derivation of
the implicature, for which the explicature (what is said) has to be appropri-
ately enriched, thereby ensuring an inferentially sound interpretation. A
similar explanation can be given for B’s utterance in (11); the quantity of
time involved must be understood as extending some way into the next few
weeks if the proposition expressed is to provide a proper inferential basis
for the implicature. Note that, in both cases, the contextual contribution to
the explicature (what is said) has been motivated by pragmatic considera-
tions alone.
My main concern in this paper is to consider the relative merits of the
second and third approaches to the underdeterminacy issue (Levinson’s and
relevance theory’s). In particular, I will take a close and critical look at
Levinson’s account of what he describes as the ‘pragmatic intrusion’ of
generalized implicatures into certain complex constructions, including
conditionals, comparatives, negations and disjunctions. Before that,
though, I will consider a question that arises for the ‘truth-conditional
pragmatics’ approach, including relevance theory, and a response to it
which involves this same set of complex constructions.
2 Truth-Conditional Pragmatics and the Scope Criterion

---

10 / ROBYN CARSTON
Suppose we have an element of utterance meaning which is clearly prag-
matically derived (i.e. we can show that it is not encoded in the linguistic
expression used but depends on pragmatic principles/maxims geared to the
recovery of the speaker’s communicative intention). This raises the follow-
ing question(s): is it an implicature or does it, rather, contribute to what is
said (explicature)? How do we know? What distinguishes the two?
Various criteria for distinguishing the two kinds of pragmatic meaning
have been proposed in the literature. In the end, they all rest, I think, on
speaker/hearer intuitions. However, there is one that has been used to
sharpen up intuitions and which has a bearing on the assessment of posi-
tions in the following sections, so I want to spend a little time looking at it.
Within relevance theory, this is known as the ‘embedding test’ and was
introduced by Deirdre Wilson as a useful tool for deciding whether some
element of utterance meaning is or is not a component of the truth-
conditional content of an utterance; Recanati (1989, 1993) called it the
Scope Principle (or Scope Criterion), and formulated it explicitly as fol-
lows:
A pragmatically determined aspect of meaning is part of what is said (and,
therefore, not a conversational implicature) if - and, perhaps, only if - it
falls within the scope of logical operators such as negation and condition-
als.
The general idea seems to have begun with Cohen’s (1971) use of an
embedding procedure in order to demonstrate that Grice could not simulta-
neously maintain the truth-functionality of ‘and’ and of ‘if’. On a Gricean
account, the meaning of ‘and’ is identical to its truth-functional logical
counterpart ‘&’, so that the two conjunctive utterances in (12) have the
same truth-conditional content (they ‘say’ the same thing). The difference
in what they communicate, concerning the order in which the events de-
scribed took place, arises at the level of conversational implicature (based
on the manner maxim of ‘orderliness’):

(12) a.
The old king has died of a heart attack and a republic has
been declared.
b.
A republic has been declared and the old king has died of
a heart attack.

The problem with this analysis that Cohen pointed out becomes apparent
when the conjunctions are embedded in the antecedent of a conditional as in
the following:

---