Available online at www.sciencedirect.

com

Journal of Pragmatics 40 (2008) 407–430

www.elsevier.com/locate/pragma

Conventional implicature revisited

Richard Vallée *
Université de Moncton, Campus de Shippagan, 218, boul. J.-D.-Gauthier, Shippagan,
New Brunswick, Canada E8S 1P6
Received 24 August 2005; received in revised form 13 April 2007; accepted 27 October 2007

Abstract
I suggest a new approach to terms famous for introducing conventional implicatures, ‘‘even’’, ‘‘but’’, and
so on. I call these expressions View on Content Devices or VCDs. First, I focus on the nature of VCD’s
semantic contribution to sentences. I then ignore the specifics of each VCD and sketch my own account,
using the Reflexive–Referential framework Perry recently introduced (2001) in semantics. Second, I
examine the semantic behaviour of ‘‘but’’ and distinguish it from ‘‘and’’. I then show the fruitfulness and
explanatory power of the Reflexive–Referential framework. Intuitively, VCDs are very simple expressions,
and my suggestion is designed to capture that prima facie simplicity.
# 2007 Elsevier B.V. All rights reserved.

Keywords: Semantics; Conventional implicature; ‘‘but’’; Reflexive semantics; Frege; Grice

1. Introduction

Frege (1892) and Grice (1975) draw attention to what I will call View on Content Devices or
VCDs

‘‘Even’’, ‘‘but’’, ‘‘already’’, ‘‘besides’’, ‘‘anyway’’, ‘‘moreover’’, ‘‘however’’, ‘‘hence’’,

‘‘nevertheless’’, ‘‘therefore’’, ‘‘although’’, ‘‘though’’, ‘‘so’’, ‘‘too’’, ‘‘still’’, ‘‘despite’’, ‘‘yet’’, etc.1

Calling them VCDs is intended to suggest that they qualify content without being part of it.
Frege (1892) put it in a nice way, saying that they ‘‘illuminate[s] (the sense) in a peculiar
fashion’’. For example, if I say ‘‘Even Paul was sea-sick’’, it is arguable that the use of ‘‘even’’
carries the qualification that Paul’s being sea-sick is surprising.2 The qualifications introduced are

* Tel.: +1 506 336 3431; fax: +1 506 336 3477.

E-mail address: [email protected].
1
See Bach (1999) for an extensive list of VCDs.
2
This is a brute intuition and is not intended as a contribution to the semantics of ‘‘even’’. For a classic examination of
the semantics of ‘‘even’’, see Karttunen and Peters (1979).

0378-2166/$ – see front matter # 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.pragma.2007.10.006
408 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

fairly simple: ‘‘But’’ means something like in contrast (Rieber, 1997) or except; ‘‘moreover’’
means something like in addition (Rieber, 1997); ‘‘so’’ means something like in consequence
(Rieber, 1997); ‘‘even’’ means something like surprisingly or amazingly, and so on. In introductory
books to logic, ‘‘but’’ is frequently semantically read as ‘‘and’’, hence ignoring the specific features
of ‘‘but’’ as a VCD and blurring the difference between ‘‘Paul is sick and Mary went to see a movie’’
and ‘‘Paul is sick, but Mary went to see a movie’’. ‘‘But’’ raises another issue in that while ‘‘Paul is
sick and Mary went to see a movie and Peter is watching television’’ is acceptable, ‘‘Paul is sick, but
Mary went to see a movie but Peter is watching television’’ does not fit well with semantic intuitions
and is cognitively dissonant.3 I will come back to that problem and explore the semantics of ‘‘but’’.
As a standard rule, VCDs apply to sentences and form new sentences.4 Some are adverbs and apply
to one single sentence (‘‘Even’’). I will focus on ‘‘even’’ in the presentation of my view. Some
connect two sentences (‘‘but’’ and ‘‘therefore’’). Other VCDs can be handled either as adverbs or as
sentence connectives (‘‘however’’). Here are examples,

She is rich; however, she is honest.

However, she is honest.

These are grammatical sentences. This being said, there is no consensus on the syntactic category
of most VCDs. VCDs are assumed not to alter the truth conditions of sentences or utterances (see
Frege, 1892, 1918; Grice, 1975).5 For example, it is taken for granted that

(1) Joan loves Paul.

is a sentence, and

(2) Even Joan loves Paul.

is a new sentence, and that both have the same truth conditions, or express the same content
(Frege, 1892:73; Grice, 1975:167; Karttunen and Peters, 1979:27; Francescotti, 1995:153;
Rieber, 1997:51). In addition

(3) However, Joan loves Paul.

(4) Nonetheless, Joan loves Paul.

and (1) have the same truth conditions. I make the same assumption, and I will say that VCDs are
truth conditionally irrelevant in a sentence or in an utterance.6 If the truth irrelevance assumption
is correct, it implies that one can understand the sentence’s truth conditions, even if one fails to
grasp the linguistic meaning of the VCD.
I want to suggest a new approach to VCDs. I first set the philosophical background and
introduce Frege’s and Grice’s view. Both views are important to understand the motivations

3
It is a grammatical sentence but, as a referee mentions, it is hard to process.
4
In ‘‘Peter is rich but happy’’, ‘‘but’’ links two adjectives. However, this sentence should be read as ‘‘Peter is rich but
Peter is happy’’.
5
That feature is argued for on ground of intuitions, and is not given a theoretical or linguistic explanation or
justification.
6
I emphasize that this does not imply that VCDs are not truth conditionally relevant with respect to a proposition
different from the one giving the truth conditions of a sentence or an utterance (see Bach, 1999) Neither does it imply their
lack of linguistic meaning or cognitive relevance.
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 409

behind my approach. Second, I focus on the nature of VCD’s semantic contribution to sentences.
I then ignore the specifics of each VCD and sketch my own account, using the Reflexive–
Referential framework Perry recently introduced (Perry, 2001) in semantics. Third, I examine
more fully the semantic behaviour of ‘‘but’’ and distinguish it from ‘‘and’’. I then show the
fruitfulness and explanatory power of the Reflexive–Referential framework. An examination of
the distinction between ‘‘but’’ and ‘‘and’’ makes clear why it is a mistake to read ‘‘but’’ as ‘‘and’’.
Intuitively, VCDs are very simple expressions, and my suggestion is designed to capture that
prima facie simplicity.
My motivations for exploring this category of expressions, lacking the philosophically
interesting aspect of propositional attitudes or modal operators, are very simple. On the one hand
these expressions are underexamined in philosophy,7 even if they motivate strong semantic
positions, like the blurring of semantics and pragmatics. On the other hand, these expressions are
interesting because of their anomalous semantic behaviour: they are clearly meaningful, but are
assumed not to contribute to the truth conditions of sentences or utterances (Frege). If the
meaning of an expression is identified with its contribution to the truth conditions of sentences,
then VCDs lack linguistic meaning. That cannot be true. We have to stop and examine certain
very fundamental principles in semantics. Finally, according to some philosophers, VCDs
introduce a proposition: a conventional implicature (Grice, 1975). However, as a rule, a sentence,
not a word, is a linguistic item fit to introduce a proposition. Once again, a very basic semantic
principle is at stake.

2. Background

2.1. Frege and cognitive significance

In ‘‘Sense and Reference’’, Frege writes

Subsidiary clauses beginning with ‘‘although’’ also express complete thoughts. This
conjunction actually has no sense and does not change the sense of the clause but only
illuminates it is a peculiar fashion. We could indeed replace the concessive clause without
harm to the truth of the whole by another of the same truth value; but the light in which the
clause is placed by the conjunction might then easily appear unsuitable, as if a song with a
sad subject were to be sung in a lively fashion. (Frege, 1892:73).
As distinct from ‘‘necessary’’ for example, ‘‘although’’ does not alter the truth value of a
sentence. If in

Although the children are sleeping, Joan is not watching television.

the first clause is true, then any true sentence can be substituted for ‘‘The children are sleeping’’,
and we obtain a true clause. For example

It is raining.

can replace ‘‘The children are sleeping’’ in the previous example and we obtain

7
But see Potts’s The Logic of Conventional Implicature (2005) and Iten’s Linguistic Meaning, Truth Conditions, and
Relevance (2005).
410 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Although it is raining, Joan is not watching television.

The first clause also expresses a true proposition. What about cases in which ‘‘the light in which
the clause is placed by the conjunction might then easily appear unsuitable’’? Consider a case
where John is a student, and the sentences

All the students are sick. Even John is sick.

In the actual example, the use of ‘‘even’’ is appropriate. Consider now

All the students are sick. Nevertheless John is sick.

Why ‘‘nevertheless’’ ? Using that term seems inappropriate in the situation.

VCDs bring in a component contributing to the informational aspect or cognitive value of
the sentence or utterance. One can argue that the light shed on the clause is cognitively
significant: it is appropriate or inappropriate, and makes a difference as to what a speaker
conveys by her utterance as well as to what a hearer understands (Frege, 1892; Grice, 1975;
Rieber, 1997). In (2), the use of ‘‘even’’ intuitively conveys as part of its linguistic meaning, for
example, that the content is surprising. Perry (1986) identifies a useful condition on the
cognitive significance or cognitive value of an utterance, a condition tied to the cognitive
notion of acceptance:
If there is some aspect of meaning, by which an utterance u of S and an utterance u’ of S’
differ, so that a rational person who understood both S and S’ might accept u but not u’,
then a fully adequate theory of linguistic meaning should say what it is. (Perry,
1986:194)
That condition focuses on the cognitive value of utterances rather than sentences. In the present
case, one can accept an utterance of (1) and reject an utterance of (2) because one thinks that it is
not surprising that Joan loves Paul. One would then write (1) but refuse to write (2) in a letter. One
might also think that ‘‘even’’ is just inappropriate in the circumstances (Francescotti, 1995).
According to the standard Perry suggests, ‘‘Even’’ is cognitively significant, and makes a
difference to the cognitive significance of the utterance.
Frege does not contend that the carried suggestion is a different proposition or content. I will
call his view a One Content View. According to that picture, a hearer understanding a sentence
containing a VCD does not grasp two or more different senses or propositions, and such a
sentence does not carry two or more autonomous senses or propositions. Frege suggests that
VCDs lack sense because they do not make a difference to the truth conditions of the sentences in
which they occur, and do not contribute to the thoughts expressed by these sentences. Since
Fregean senses, are also linguistic meanings and cognitive values of terms, he is bound to say that
they lack linguistic meaning and cognitive value. However, one does not want to argue that
‘‘even’’ lacks linguistic meaning and cognitive value. That is not plausible. I will come back to
Frege.

2.2. Grice and the many propositions

Grice (1975) distinguishes, between what is meant by a VCD sentence, what is centrally said
and what is conventionally implicated by that sentence.
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 411

He proposes a picture of the semantics of VCDs assuming them to carry conventional

implicatures. A conventional implicature is a proposition carried by the meaning of the VCD
sentence, and the plausibly suggested view on the content expressed or on what is said in Grice’s
picture. For example, an utterance of (2) would be true if and only if Joan loves Paul – this is what
is said – and the VCD would conventionally implicate, or suggest, in addition to what is said that
it is surprising that Joan loves Paul or, according to Karttunen and Peters (1979), that other people
besides Joan loves Paul. Sentence (1) and sentence (2) say the same thing, but they do not mean
the same thing because of the conventional implicature carried by ‘‘even’’. What is suggested or
conventionally implicated by (2) is a proposition different from what is centrally said, or what the
speaker centrally meant, and adding to what is centrally said by the sentence. In (2) the
proposition that is conventionally implicated would be for example that Joan’s love for Paul is
surprising or that other people besides Joan love Paul. I will say that Grice proposes a Many
Propositions View. Grice writes:
If I say (smugly), ‘‘He is an Englishman; he is, therefore, brave’’, I have certainly
committed myself, by virtue of the meaning of my words, to its being the case that his
being brave is a consequence (or follows from) his being an Englishman. But while I have
said that he is an Englishman, and said that he is brave, I do not want to say that I have said
(in the favored sense) that it follows from his being an Englishman that he is brave, though
I have certainly indicated, and so implicated, that it is so. (Grice, 1975:166–167)
In a diagrammatic form, we obtain,

What is conventionally implicated is true or false and hence, strictly speaking, on that theory
the VCD is truth conditionally relevant, contributing an implicated truth valuable proposition
different from what is said. What is said can be true, and the conventional implicature false
(Grice, 1975:167). For instance, it might be true that Joan loves Paul, and false that it is surprising
that Joan loves Paul. According to Grice’s picture, only what is centrally said qualifies as the truth
conditions of the sentence or utterance. I will say that according to Grice, VCDs are proposition
introducing devices.
I am reluctant to endorse Grice’s approach to VCDs. Prima facie, nothing in a VCD sentence
backs the view that VCD sentences express two different, independent propositions, both being
412 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

truth conditions expressed by the sentence or utterance. In addition, a very basic semantic
principle underlies the Semantic enterprise, the Principle of Semantic Cohesion:

PSC: A simple sentence expresses one proposition as its truth conditions.

Grice’s framework rejects PSC. Recently, Bach (1999) offered an alternative to that framework.
According to Bach’s picture, VCD sentences express more than one proposition, both part of
what is said. In all cases, the VCD introduces a new proposition. Still, Bach ponders over the
importance of each proposition by ranking them (Bach, 1999:353). Bach (1999:350) argues that
VCDs force the rejection of the PSC. He contends, for instance, that ‘‘Cal is still on the phone’’
expresses two propositions – and does not express a conjunction:

Cal is on the phone.

Cal has been on the phone.

Barker (2003) states a principle very similar to PSC

SCM: The semantic content of a sentence S – the content S expresses by virtue of linguistic
rules and context, and upon which logical particles may potentially operate – is to be
identified with S’s truth conditional content (Barker, 2003:2)
Barker rejects SCM by invoking the semantic behaviour of VCDs. He also draws dramatic
consequences for truth conditional semantics like Davidson’s. I think that the PCS is sound, and
that a semantic theory preserving it would have much in its favour.8 Finally, Grice’s approach is
mute on why ‘‘Paul is sick, but Mary went to see a movie but Peter is watching television’’ is at
best hard to process or, more appropriately, cognitively dissonant. I will come back to this issue at
length.
The literature on the expressions we are concerned with relies predominantly on the notions of
what is meant, what is said and what is implicated. These notions are notoriously controversial
and I dispense with them in trying to capture the phenomenon we are interested in. I set aside
Grice’s semantic notions and cast a Fregean net. I argue that VCDs by virtue of their linguistic
meaning introduce an element into the cognitive value of the utterance, one illuminating the
content or truth conditions in a certain way, but one not part of the truth conditions of the
utterance. I will say that it introduces a cognitively relevant view on the content expressed, or the
truth conditions, and no truth conditionally relevant content constituent. My suggestion has many
advantages. First, it handles conventional implicatures in purely semantic terms. Semantics used
to be defined in terms of truth conditions, and pragmatics in terms of the non-truth conditional
aspects of sentences. Because VCDs do not alter the truth conditions of sentences, their
contribution to sentences is sometimes taken to be pragmatic in nature even if it depends on the
linguistic meaning of the expression. I align them to the side of meaning, and do not rely
on pragmatics. I use a very conservative view of semantics. The idea that semantics is
concerned with the truth conditions of sentences has recently been challenged (see especially

8
It is important to emphasize that VCDs do not alter the illocutionary force of the speech act then performed. If they
did, they would not throw a specific light on the content or illuminate it in a peculiar fashion, but add a new burden to the
success condition of the speech act performed. For example, ‘‘frankly’’ is not truth conditionally relevant, but adds
sincerity conditions to the successful utterance of (2). ‘‘Even’’, and other VCDs do not. An utterance of ‘‘Frankly, Joan
loves Paul’’ does not throw a peculiar light on the expressed content, Joan loves Paul.
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 413

Récanati, 2004). I can ignore that debate, which is irrelevant to my main point. Second, my
suggestion is not ad hoc: it uses notions already available, and as distinct from Grice’s view, it
does not require major changes in our picture of how some sentences behave. Finally, on Grice’s
view, the close relationship between what is said and what is conventionally implicated is not
adequately captured. My view exhibits that close relationship.

2.3. Features of conventional implicatures

Some of Grice’s basic ideas shape our view and educate our intuitions on VCDs. I think that
these ideas are not controversial and should be accounted for by any theory on VCDs, including
more Fregean approaches like the one I propose. Conventional implicatures have characteristic
features.

2.3.1. Conventional
A conventional implicature depends on the conventional meaning of a sentence, not on any
maxim (Grice, 1989:41) or on the context of utterance. In Grice’s view, relying on extralinguistic
data is not required to identify or ‘‘build’’ an implicated proposition. According to other authors,
it is required (See especially Bennett, Francescotti and Bach). I follow Grice’s intuition. This
feature draws attention to the fact that what a VCD suggests depends on its linguistic meaning,
and not on any contextual aspect of the utterance. This feature is independent from Grice’s
framework, and it fits within different frameworks.

2.3.2. Detachable
According to the Gricean paradigm, one can say the same thing without carrying the
conventional implicature. In so far as what is said is defined by the truth conditions of the
sentence and that VCDs do not introduce truth conditional components, this seems to be a trivial
feature, resulting from how VCDs are characterized. In a Fregean framework, one will simply say
that they are not truth conditionally relevant, and that is exactly what Frege wrote.

2.3.3. Not calculable

In a plausible understanding of that feature, it simply means that one does not need an argument
or extra premises, going beyond the linguistic meaning of the VCD, to identify a conventional
implicature or what is suggested, and that one’s understanding of the meaning of the words is
necessary and sufficient to get the speaker’s view on the content. This feature goes hand in hand with
the conventional aspect of conventional implicatures. I base my view around that feature.

2.3.4. No cancellation without anomaly

I cannot utter (2) and add ‘‘it is not surprising that Joan loves Paul’’, hence cancelling what is
suggested, without being incoherent. I want to provide strong reasons for that being the case.
Let me add two more features.

2.3.5. No VCD without justification

If one expresses an amazing attitude, or illuminates the content in an unexpected way, then a
hearer will tend to question the use of that specific VCD. One can follow the maxim ‘‘no VCD
without justification’’. For example, I can say ‘‘Snow is white, but 2 + 2 = 4’’. Why ‘‘but’’? If
‘‘but’’ introduces a contrast between ‘‘snow is white’’ and ‘‘2 + 2 = 4’’, why is a contrast
414 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

introduced? However, the meaning of the VCD does not carry information on the justification for
the idea of the contrast introduced.

2.3.6. Context insensitive

I want to emphasize that VCDs are context insensitive items. Prima facie, and in contrast with
indexicals, nothing in their meaning makes them sensitive to a component of the context of
utterance or to the speaker’s background belief.
It is important to underline the fact that VCDs do not explain why the speaker uses a specific term
and throws a specific light on the content expressed. To repeat, the justification for the use of a
specific VCD is not to be found in the meaning of that VCD. Suppose that ‘‘even’’ carries the idea of
surprise. This term does not explain why the speaker is surprised or expresses/suggests surprise.
Consider (2). Why does the speaker express surprise? There are many options: because an amazing
number of people love Paul, because Joan loves no one, and so on.9 In any case, the reason is not to
be found in the meaning of the sentence. Philosophers and linguists are prone to read very rich
suggestions conveyed by the VCD. Bennett (1982) and Lycan (1991) are good examples (See also
Francescotti, 1995). But this view requires going well beyond the meaning of the VCD.

2.4. Characterizing VCDs

As previously mentioned, syntactically VCDs are either sentence connectives (‘‘but’’, ‘‘so’’),
adverbs (‘‘even’’, ‘‘still’’, ‘‘moreover’’, ‘‘nevertheless’’,) or both (‘‘however’’). Semantically,
however, they all behave in a very similar way: VCDs suggest a relation between two contents or
propositions: a contrast or an opposition, or an addition. For example, in

Peter will come to the party, but Joan has to study for an exam.
My flight is scheduled to leave at 2; however, I am asked to be at the gate at 10.
My flight is scheduled to leave at 2; nevertheless, I am asked to be at the gate at 10.

‘‘but’’, ‘‘however’’ and ‘‘nevertheless’’ suggest a contrast between the content expressed by the
first sentence and the content expressed by the second sentence. Contents contrasted are not
confined to those expressed by a simple sentence. For example, ‘‘Peter is playing football and
Mary is studying, but you can stay’’ is correct. What is expressed by complex sentences and what
is expressed by a simple sentence can be contrasted. In some cases, one content involved in the
relation is not part of any sentence used, as in

However Peter will not come to the party.

The content with which ‘‘Peter will not come to the party’’ is contrasted is not part of the
sentence and it must be recovered. It can be, for example, that many people will come to the party
(see Rieber, 1997). VCDs are called discourse markers. Most authors take VCDs to be
connectives on grounds of their semantic behaviour – they connect propositions. Rieber (1997)
and Blakemore (1989) call them discourse connectives. So will I. Now, in so far as these
expressions are blind to truth value and make no contribution to the truth conditions of sentences,
they are not truth functional connectives.

9
Francescotti (1995) introduces very sophisticated conventional implicatures for ‘‘even’’ because he confuses the
communication of the attitude of surprise and the communication of an explanation for that attitude.
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 415

3. A Reflexive–Referential proposal

3.1. Sense and suggestion

Frege notes that VCDs lack sense10 because they are truth conditionally irrelevant. This
remark is perplexing. Senses have different roles, among which are linguistic meaning, cognitive
value, truth bearer, thought content and referent in oblique context. We have

Sense (i) Linguistic meaning

(ii) Cognitive value
(iii) Truth bearer
(iv) Thought Content
(v) Referent in oblique context

Frege’s remarks plausibly convey that VCDs are simply truth conditionally irrelevant, not that
they lack linguistic meaning or cognitive value. One does not want to argue that ‘‘but’’ has no
linguistic meaning. Finally, it is not plausible to believe that VCDs are cognitively irrelevant.
The use of a VCD can be appropriate or not, and this fact cannot be explained if VCDs lack
cognitive relevance. So, I will preserve these functions of senses for these expressions. (iii) is
dropped as far as VCDs are concerned. (i)–(ii), and the dropping of (iii) are easily captured in
Perry’s Reflexive–Referential approach to natural languages. Frege also mentions that the sense
of a sentence, a proposition, is a thought content, (iv), and the referent of the sentence in a
propositional attitude context, (v). Perry’s framework is silent on this topic. I will return briefly to
this aspect of Fregean senses at the end of the paper.

3.2. The Manifold View of Propositions

Let me introduce the framework that supports my suggestion. There a different versions of
that theory, and that might be confusing. My version is grounded on Perry’s ‘‘Individuals in
Informational and Intentional Content’’ (Perry, 1990) and Perry’s Reference and Reflexivity
(2001). I slightly alter Perry’s formulation to make my point clearer. Perry’s view focuses on
utterances, rather than sentences. An utterance is the use of a sentence by a speaker at a place at a
moment of time. He (1997, 2001) introduces a view according to which

Manifold View of Propositions:

an utterance of a sentence expresses many propositions

Consider an utterance of (5)

(5) Birds have wings.

and suppose that you understand the linguistic meaning of the sentence as type but know nothing
about the utterance. Knowing the meaning, you know linguistic rules determining propositions.
The linguistic meaning is not a proposition. However, it can determine propositions, and it
might determine different propositions for different utterances of the same sentence.

10
I remind you that ‘‘sense’’ must be understood as Fregean sense here.
416 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Propositions are contents. To make the theory clearer, let me introduce a type of proposition, the
proposition S, not found in Perry’s but allowing me to emphasize the focus on utterances.
Following Perry’s picture (Perry, 1990, 1997, 2001), and eliminating possible ambiguities,
knowing the linguistic meaning of (5) you know that

Content S
<For any x, if x is a bird, then x has wings; u>

The first component echoes the linguistic meaning of the sentence. It provides the truth
conditions of the sentence. The second component is a variable for an utterance. I add this
variable at that stage to emphasize Perry’s focus on utterances. (5) is utterance insensitive, so on
understanding the sentence one understands the truth conditions of any utterance of that sentence.
Consider the specific utterance u of (5). Focussing on that specific utterance and the linguistic
meaning, we can drop the reference to the utterance and get
Content M
<For any x, if x is a bird, then x has wings>

Since that specific Content M is not sensitive to the utterance, and does not contain an utterance as a
constituent, all utterances of (5) will have the same Content M. Content M is the cognitive value of
the utterance. The first component of the Content S of (5) is then the cognitive value of the utterance,
and all utterances of (5) have the same cognitive value. Under what conditions are they true? The
truth conditions of this utterance are given by Content D or the official truth conditions of the
utterance

Content D
<For any x, if x is a bird, then x has wings>

All utterances of (5) have the same truth conditions that one can identify with their cognitive
value. Things do not always work this way.

Consider my utterance of (6)

(6) I am sick.

Thanks to your semantic competence, you know that

Content S
<The speaker of the utterance is sick; u>

The first part of the Content S is given by the linguistic meaning of (6) and it refers to an
utterance. The second part is a variable for an utterance. Suppose that you focus on the specific
utterance, u. You know that this utterance, u, is true if and only if the speaker of the utterance u is
sick.

Content M
<the speaker of u is sick>

Here the Content M contains the specific utterance. Perry calls it the reflexive content of the
utterance because it contains the utterance as a constituent. The content M is arguably the cognitive
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 417

value of the specific utterance. Different utterances of (6) will have different cognitive values
because the Content M will contain a different utterance. Suppose now that you know enough about
the context to identify the speaker of the utterance. You have the content D of the utterance

Content D
<ME, being sick>

The content D is what we usually take to be the truth conditions of utterance. My utterance of
(6) is true if and only if I am sick. Perry calls the Content D the official truth conditions of the
utterance. The content D is obtained once what indexicals, definite descriptions and proper names
designate is given.11
I emphasize that according to this picture, each proposition is determined by the linguistic
meaning of the sentence as type. When features of the utterance are required to obtain truth
conditions, they can be introduced into the propositions because utterances are primary in the
framework. The overall picture can be summarized in the following diagram

11
I blur the distinction between Content C and Content D found in Perry’s Reference and Reflexivity, since this
distinction has no use in my paper.
418 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Sometimes, the meaning of a sentence does not fully determine the truth conditions of an
utterance – this is the case of non-articulated constituents (Perry, 2001), or as I prefer to say,
unarticulation.
A constituent is articulated if it is provided by a lexical element, and if it is part of the truth
conditions of an utterance. In my understanding, articulation is about content D. My utterance of
(6) expresses a proposition D having all its constituents articulated. Unarticulation recently
entered the philosophical landscape (Perry, 2001; Récanati, 2002). A constituent is unarticulated
if it is not provided by a lexical item in the sentence but it is required to have a truth valuable
entity or proposition D. Unarticulation is about content D. For example, the content D expressed
by an utterance of ‘‘It is raining’’ contains an unarticulated constituent, a place, because there is
no term for a location in this sentence (Perry (2001)) and a place is required to have a truth
valuable content D. Rain is always located in space. Following Frege, VCDs lack sense. Under
one plausible interpretation, it can be read as the idea that it does not contribute to the truth
conditions of utterances. Exegetical evidence backs this interpretation. However, it cannot imply
that VCDs lack linguistic meaning or cognitive significance. A semantically competent speaker
knows the meaning of VCDs, witness the fact that he/she can select an appropriate VCD, and its
cognitive value, witness the fact that one can appreciate the suitability of the ‘‘light’’ thrown by a
VCD. So, a VCD sentence does have a content S and a content M echoing the linguistic meaning
of the VCD. However, the latter is not echoed in content D. I will say that VCDs do not articulate
any constituent in Content D (see also Vallée, 2005).
The Manifold View of Propositions is designed to make semantics sensitive to the reflexivity
of indexicals and demonstratives. One feature of this picture is that it distinguishes the linguistic
meaning of an utterance, the cognitive value of that utterance and the truth conditions of that
utterance. It can also distinguish the contribution of the VCD to the meaning and to the cognitive
value of the utterance from its contribution to the truth conditions of the utterance. Let us go back
to an utterance of (2). To simplify the presentation, I take ‘‘even’’ to linguistically mean what
‘‘surprisingly’’ means. From the meaning of the sentence we obtain the first component of
Content S, and I add the usual second component, a variable for an utterance.
Content S
<Surprisingly Joan loves Paul; u>
We move to the specific utterance and can drop the second component, the variable for an
utterance. Because that content, the first component of Content S, is not utterance sensitive, all
utterances of (2) have the same content M:

Content M
<Surprisingly Joan loves Paul.>

All utterances of (2) have the same Content M, and the latter is identical with the first component
of Content S. So, in virtue of her understanding of the sentence, and assuming that the meaning of
‘‘even’’ is carried within the meaning of ‘‘surprisingly’’, a speaker understands what Grice takes
to be the conventional implicature.12 Note that the contribution of the VCD is part of the Content
S and Content M of the VCD sentence. ‘‘Even’’ does not contribute to the truth conditions of the
utterance. Content M is sufficient to get the official truth conditions of the utterance or what Grice
takes to be what is said.

12
Grice never mentions ‘‘even’’, and here I am just applying my own view.
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 419

Content D
<Joan, Paul, loves>

The relation ‘‘loves’’ is satisfied by Joan and Paul. The cognitive significance or Content M is the
mode of presentation of the Content D. The cognitive significance of ‘‘even’’ is part of the mode of
presentation of D, even if it does not introduce any component in Content D. Now, according to this
picture we go from the cognitive value of the utterance, including the contribution of the VCD, to
the truth conditions of the utterance. In Grice’s vocabulary, it would mean that we go from the
conventional implicature to what is said. I take my picture to be more natural, and to exhibit strong
and simple relationships. In addition, according to my picture, one’s knowledge of content D does
not require one to understand the specifics of the VCD. Since the latter is truth conditionally
irrelevant, one can lack understanding of the meaning of the VCD with no consequence on one’s
knowledge of the content D. This is a welcome consequence. To summarize, we have the following
determination chain

3.3. Features of conventional implicatures

Let us go back to the celebrated features of conventional implicatures.

3.3.1. Conventional
The suggestion is carried by the conventional meaning of the VCD and is semantically echoed
in a proposition. This is captured by a Reflexive–Referential view on VCD sentences, since the
suggestion is determined by the meaning of the VCD and is part of a proposition, the Content M,
semantically expressed by a VCD sentence.

3.3.2. Detachable
The suggestion is detachable, since the VCD can be removed from the sentence and the
content D will stay the same.

3.3.3. Not calculable

No calculation is required to obtain the suggestion, since the latter depends only on the
meaning of the VCD.
420 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

3.3.4. No cancellation without anomaly

Suppose that one utters (2), conveys

Content M
<Surprisingly Joan loves Paul>

and adds that it is not surprising that Joan loves Paul. That person is incoherent, conveying
through the use of ‘‘Even’’ that the it is surprising that Joan loves Paul – this is the Content M –
and adding explicitly that it is not – hereby negating the Content M. On my view, the content M of
‘‘Even Joan loves Paul’’, namely, ‘‘Surprisingly Joan loves Paul’’ or ‘‘It is surprising that Joan
loves Paul’’, and the content M of ‘‘It is not surprising that Joan loves Paul’’ cannot be true at the
same time as one is the negation of the other. The problem with cancellation is a cognitive value
problem, a problem having to do with the cognitive value of the utterances, not a problem with the
truth conditions or Content D of the utterances. The cognitive value of the second utterance is the
negation of the cognitive value of the utterance of (2), hence the anomaly. Needless to say that the
speaker makes incoherent utterances. My view clearly captures that fact. Notice also that the
content M of an utterance of (2) might be false – it is not surprising that Joan loves Paul – and the
Content D of that utterance might be true – because Joan loves Paul. This fits Grice’s intuitions.
The last two features of conventional implicature ‘‘No VCD without justification’’ and context
insensitivity are also captured. Notice that the use of the VCD must be justified in the context, and
that the justification is not given by the VCD itself.

3.4. Presuppositional analysis

Glanzberg (2005) proposes a presuppositional analysis of ‘‘even’’. That picture is not Gricean
in that VCDs are not assumed to introduce a proposition as conventional implicature. Let me
sketch that view and compare it to my suggestion. In Glanzberg’s picture, sentences and context
are distinguished, the context being the information forming the common ground of speakers
(Stalnaker, 1974). Let us represent the sentence S and the context C as follows

S+C

Some sentences have presuppositions, the latter being part of the context. If presuppositions fail,
that is, when this information is not part of C, then the sentence is infelicitous in the context.
Glanzberg contends that syntactic constructions and lexical items trigger presuppositions, that is
generate a presuppositional requirement. Sentences containing triggers require the presence of
the presupposition. Let us go back to ‘‘even’’. Consider ‘‘Even John solved the problem’’. That
sentence’s presupposition is something like ‘‘Someone other than John solved the problem, and it
was unlikely that John might have done so’’, and ‘‘even’’ triggers that presupposition. In case the
presupposition is not part of the context, the sentence is infelicitous.13
I tend to prefer my suggestion. First, it is more economical than Glanzberg’s, dispensing with
the notions of background information and infelicity. In addition, my view does not introduce
anything in the common background, and VCD utterances can convey new information,

13
The presupposition can, of course, be introduced in the common ground by accomodation: if an utterance u
presupposes a proposition P not part of the context, then the utterance by itself introduces P in the context. (see Stalnaker,
2002 for a recent discussion of accomodation). My view not requiring the notion of common ground, it need not invoke
accomodation.
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 421

information not part of the background. VCD sentences can build common ground. Second, it
approaches ‘‘even’’ in a more standard, Fregean, way in assigning it meaning and cognitive
significance, rather than taking it to be a trigger. Thirdly, it does not link ‘‘even’’ to a rich
proposition – something Grice’s view does – and assigns it the modest role it intuitively has. The
presuppositional content Glanzberg provides – ‘‘Someone other than John solved the problem,
and it was unlikely that John did’’ – probably motivates the use of ‘‘even’’. However, one does not
want what motivates that use to be captured by the semantics of ‘‘even’’.

4. ‘‘But’’

Let us move to the semantic contribution of specific VCDs. I take unary VCDs to be adverbs,
and I rely on adverbs to make explicit what they semantically contribute to both Content S and
Content M. Binary VCDs are conjunctions. I take this to mean that they semantically contribute a
conjunction. But that cannot be the whole story. I want to examine ‘‘but’’, because there is an
important literature on that word. My view suggest a new angle on its semantics. I will not pay
attention to other binary VCDs. What is characteristic of each VCD is of interest in lexical
semantics, and that goes beyond the point of my paper.

4.1. ‘‘And’’ vs ‘‘But’’

In logic textbooks, ‘‘but’’ is semantically identified with ‘‘and’’. For example, Klenk (1994)
asks the reader to symbolize ‘‘but’’ and ‘‘however’’ as ‘‘and’’, since, she contends, they have the
same ‘‘logical force’’. I think that this is a mistake.14 Philosophers of language are wrong in
accepting this identification and in neglecting the characteristic features of ‘‘but’’. Consider the
following

Paul will come to the movie and Mary will stay home.
Mary will stay home and Paul will come to the movie.

The truth functional connective’s only semantic function is to operate on truth values, and the two
sentences can be inverted without altering the truth conditions of the complex sentence.
Semantically, these two complex sentences are equivalent.15 In addition, they are also cognitively
on a par. The cognitive value of the first one is identical with the cognitive value of the second
one: one accepting an utterance of the former and rejecting an utterance of the latter is irrational.
One willing to utter the former and refusing to utter the latter, for any reason, is also irrational.
Things are different in the following sentences:

Paul will come to the movie but Mary will stay home.
Mary will stay home but Paul will come to the movie.

14
I am not convinced that Frege made that mistake (Frege, 1892:73). Frege writes ‘‘Subsidiary clauses beginning with
‘Although’ also express complete thoughts. This conjunction actually has no sense and does not change the sense of the
clause but only illuminates it in a peculiar fashion*.’’ In the note he adds ‘‘similarly in the case of ‘but’, ‘yet’’’. If ‘‘but’’
illuminates a clause in a peculiar fashion, it contributes more than ‘‘and’’. In ‘‘Thoughts’’, he clearly writes that ‘‘but’’
differ from ‘‘and’’ (Frege, 1918:39).
15
I reject the idea that ‘‘and’’ carries tense, and that ‘‘and’’ is responsible for our intuition that ‘‘They got married and
had children’’ and ‘‘They had children and got married’’ are not equivalent because intuitively the order of the sequence
of events in time is different in each case.
422 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Paul will come to the movie; however, Mary will stay home.
Mary will stay home; however, Paul will come to the movie.
Paul will come to the movie; nonetheless, Mary will stay home.
Mary will stay home; nonetheless, Paul will come to the movie.

These sentences are prima facie not equivalent. For example, prima facie ‘‘Paul will come to the
movie but Mary will stay home’’ is different from ‘‘Mary will stay home but Paul will come to the
movie’’. ‘‘But’’, as well as ‘‘however’’ and ‘‘nonetheless’’, each introduce a contrast (see Wilson,
1975), and introduce it whatever the truth value of the sentences it connects. Let us say that in
virtue of its linguistic meaning, ‘‘but’’ introduces a cognitively significant contrast between
propositions. Now I don’t want to suggest that ‘‘but’’ is unequivocal. That term might well
semantically be ambiguous. However, that has no impact on my main point. Let me distinguish
the base proposition and the contrasted proposition. The contrasted proposition is the one
following ‘‘but’’ (see also Rieber, 1997). Consider the following

Paul will come to the movie but (in contrast) Mary will stay home.
Mary will stay home but (in contrast) Paul will come to the movie.

Bach (1999:327) suggests that in the previous examples a contrast is indicated between
coming to the movie and staying home. If so, the two previous sentences are cognitively
equivalent, because in both cases the same contrast is indicated. If there is a contrast between
coming to the movie and staying home, then there is one between staying home and going to the
movie. However, the cognitive impact of the first ‘‘but’’ sentence differs from the cognitive
impact of the second ‘‘but’’ sentence. Each introduces a proposition meant to be the contrasted
proposition, and which proposition is the contrasted proposition matters. Because the contrasted
proposition follows ‘‘but’’, inverting the sentences alters the cognitive value of the complex
sentence. ‘‘But’’ is both cognitively significant, contrasting cognitive contents, and truth
conditionally irrelevant. Consider the following sentences

She is honest, but she is rich.

She is rich, but she is honest.

In both cases, the contrast is clear. These sentences are not equivalent in all respects, even if
sharing the same truth conditions, and they are not appropriate in the same circumstances. One
might utter the former, and refuse, for any reason, to utter the latter, or vice-versa, depending on
the contrast one wants to emphasize: richness, as opposed to honesty, or honesty, as opposed to
richness. These sentences differ in cognitive significance: what is contrasted, the contrasted
content, differs in both cases. In that respect, theses sentences differ from

Paul will come to the movie and Mary will stay home.
Mary will stay home and Paul will come to the movie.

These last sentences are equivalent in all respects, and they are appropriate in the same
circumstances.
I want to draw attention to the fact that a ‘‘but’’ sentence contrasts propositions, but does not
explicitly state why there is a contrast. To explain the asymmetry between ‘‘A but B’’ and ‘‘B but
A’’, one could invoke a conversational maxim like ‘‘Be orderly!’’. This strategy is sometimes
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 423

used to explain in non semantic terms the intuition that ‘‘They got married and had many
children’’ and ‘‘They had many children and got married’’ are different without arguing
that ‘‘and’’ carries the idea of time. However, arguing that that maxim, or a similar maxim,
establishes a contrast or a priority, or make natural a certain order, is not plausible. There is
nothing beyond the contrast introduced by the meaning of ‘‘but’’, contrasting the second
component with the first. In addition, were a maxim like ‘‘Be Orderly’’ invoked to make sense of
the contrast or emphasis, it would turn a prima facie semantic issue into a pragmatic one,
connected to conversational maxims, while a semantic explanation is already available. Another
issue should be addressed.
Grice (1975) writes that a ‘‘but’’ sentence has the truth conditions of an ‘‘and’’ sentence, and in
addition suggests a contrast. It is not infrequently argued that

A +: ‘‘but’’ = ‘‘and’’ + something else

Wilson (1975), for instance, writes: ‘‘It seems to me that sentences containing but are
assigned a truth-value as a function of the values of their constituent conjuncts, and that in
addition hearers attempt to construe a contrast of some sort between the conjuncts.’’ (p. 119).
Gazdar (1979:38) echoes the same intuition and Bach (1999) mentions it when writing, with respect
to the following:

(7) Shaq is huge but he is agile.

(8) Shaq is huge and he is agile.

‘‘According to common wisdom, the truth of (7) (my numbering) requires nothing more than the
truth of (8), although in uttering (7) rather that (8) one is indicating that there is some sort of
contrast between being huge and being agile.’’ (p. 327). (8) is true if and only if the two conjuncts
are true, and it is false otherwise. ‘‘But’’ is then to be read as containing a hidden conjunction.
Should a conjunction be introduced between the sentences? If so, in addition to being a truth
functional connective, ‘‘but’’ has a cognitive value ‘‘and’’ does not have. Rieber (1997) offers a
similar suggestion. He (Rieber, 1997:58) writes

‘‘For ‘‘but’’ is equivalent in meaning to ‘‘and’’ plus a parenthetical verb.’’

If so, then ‘‘but’’ is a truth functional connective, in addition to introducing a parenthetical verb.
Following Frege, I take ‘‘but’’ to be a truth conditionally irrelevant discourse connective, and for
that reason it cannot be a truth functional connective, or contain a hidden truth functional
connective in addition to something else. It is blind to truth values. I do not adopt an approach
dictated by A+. Let me insist. ‘‘But’’ alters the cognitive value of the sentence, and contrasts
contents. By the same token, it does not ‘‘add’’ or ‘‘put together’’ contents by way of the logical
connective ‘‘and’’. There is more.

4.2. Blakemore hypothesis

Diane Blakemore (1989:32) offers the following observation:

‘‘whereas and can conjoin any number of propositions, but can only be used to connect two.’’
424 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

If Blakemore is right, then A+ is not defensible: her observation implies that it is false that
‘‘but’’ can replace any occurrence of ‘‘and’’, or any occurrence of ‘‘and’’ plus something else.
She is right in an important way. I want to take a few pages to explore her view. It tells us a lot
about the very limited capacities of VCDs as conjunctions. I will conclude by formulating a
generalization concerning binary VCDs based on Blakemore’s observation. Let me rephrase her
view in a slightly different language. The idea that ‘‘and’’ can connect many propositions is
ambiguous. It can mean that it can take many simple propositions (or a complex proposition) on
its right side and many simple propositions (or a complex proposition) on its left side. It can also
mean that it can take one simple proposition on the right side and one complex proposition on the
left side, or one complex proposition on the right side and one simple proposition on the left side.
Consider the example:
P and Q and R and S and T and U and V.
That sentence is multiply ambiguous, and must be disambiguated, one way or another as in

((P and Q) and ((R and S) and T)) and (U and V).
Different disambiguation will result in different assignments of truth value to the complex
proposition if the truth values of the simple sentences differ. In long conjunctions, brackets
should be introduced to get a clearly truth valuable conjunction with no impact on the linguistic
meaning and cognitive value of the sentence. However, things are different with
P but Q but R but S but T but U but V.
or
Peter plays violin but Gary is learning piano, but Mary plays basketball but John swims in his
college team but Diane cooks but Alan teaches in a high school but Bob rides horses.

That example is grammatical, but barely understandable. If one introduces brackets as before,
one gets the also grammatical but barely understandable

((P but Q) but ((R but S) but T)) but (U but V).
or
((Peter plays violin but Gary is learning piano), but ((Mary plays basketball but John swims in his
college team) but Diane cooks)) but (Alan teaches in a high school but Bob rides horses).

What is the problem with these long ‘‘but’’ sentences? Why does ‘‘but’’ have these limited
capacities as a conjunction? What limits free bracketing? How can these sentences be made
readable? Two options are on the table. Consider again
P but Q, but R, but S, but T, but U, but V.
Q, R, S, T, U, V can each be compared to P, or every proposition can be compared to the preceding
simple proposition. These are the only plausible intelligible options. I will call them forced
options. Others options are not plausible. For instance in

Peter plays violin but (Gary is learning piano), but Mary plays basketball)
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 425

the last sentence (‘‘Mary plays basketball’’) is contrasted with the complex sentence (‘‘Peter
plays violin but Gary is learning piano’’).
This way of reading the long sentences is not plausible. One may wonder how a contrast is
introduced between the complex sentence and the simple sentence, or how one is supposed to
contrast ‘‘Peter plays violin but Gary is learning piano’’ and ‘‘Mary plays basketball’’.
Remember that in the long sentence ‘‘but’’s are introduced, and contrasts identified. I will come
back to this dictated by the forced option and the second. A better reading of the sentence is given
by

Peter plays violin but Gary is learning piano.

Gary is learning piano but Mary plays basketball.

‘‘P but Q, but R, but S, but T, but U, but V.’’, in the interpretations we examined, are predicted as
semantically incorrect by Blakemore Hypothesis. I will replace ‘‘proposition’’ with ‘‘sentence’’
and restate

Blakemore Hypothesis 1:
‘‘But’’ can only connect two sentences.

For example,

((P but Q) but ((R but S) but T)) but (U but V).

or

P but (Q but (R but (S but (T but (U (but V)))))).

where simple and complex sentences are contrasted. However, these options are not
plausible – they are barely understandable – and are excluded. Blakemore’s hypothesis is
more specific than she states. ‘‘But’’ can connect a simple sentence to a simple sentence, a
simple sentence to a complex sentence containing logical connectives or a complex sentence
containing logical connectives (like ‘‘and’’) to a complex sentence containing logical con-
nectives (like ‘‘and’’). In these cases, the result is intelligible. For instance

Peter plays violin and John is learning piano, but Mary plays basketball.
Peter plays violin but Mary plays basketball and John swims in his college team.
Peter is rich but he is unhappy and he is depressive.
Peter plays violin and Gary is learning piano, but Mary plays basketball and John swims in his
college team.

‘‘But’’ can connect simple and complex sentences, but it can not connect a simple sentence to
a complex sentence containing many ‘‘but’’s, or a complex sentence containing ‘‘but’’ to a
complex sentence containing ‘‘but’’. The following are cognitively dissonant and not acceptable,
unless the third simple sentences is contrasted with the first simple sentence or contrasted with
the simple sentence immediately preceding it.

Peter plays violin but John is learning piano, but Mary plays basketball.
Peter plays violin but Mary plays basketball but John swims in his college team
426 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Peter is rich but he is unhappy but he is depressive.

Peter plays violin but Gary is learning piano, but John swims in his college team.

On the other hand, the following are perfectly acceptable

Peter plays violin but John is learning piano and Mary plays basketball.
Peter is rich and he is unhappy but he is depressive.

Let me rephrase the hypothesis.

Blakemore Hypothesis 2:
‘‘But’’ can only connect simple sentences to simple sentences, simple sentences to complex
sentences not containing ‘‘but’’, or complex sentences not containing ‘‘but’’ to complex
sentences not containing ‘‘but’’.

The Hypothesis can be generalized, since it applies to

Peter plays violin; however, John is learning piano, but Mary plays basketball.
Peter plays violin; but Mary plays basketball; however, John swims in his college team.

These sentences are cognitively dissonant and not acceptable, unless the second and third simple
sentences are compared to the first simple sentence or to the sentence immediately preceding it.
BH 2 also applies to ‘‘however’’. For instance it predicts that

Peter plays violin and John is learning piano; however, Mary plays basketball.
Peter plays violin; however, Mary plays basketball and John swims in his college team.
Peter is rich; however, he is unhappy and he is depressive.
Peter plays violin and Gary is learning piano; however, Mary plays basketball and John swims in
his college team.

are acceptable and that

Peter plays violin; however, Mary plays basketball but John swims in his college team.
Peter is rich; however, he is unhappy but he is depressive.

are hard to process and cognitively dissonant. So, I suggest

A binary VCD can only connect simple sentences, simple sentences to complex sentences
not containing binary VCDs, or complex sentences not containing binary VCDs to complex
sentences not containing binary VCDs.

Let us go back to the initial example:

Mary will stay home, but Paul will come to the movie.

We have the following contents:

 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 427

I contend that ‘‘but’’, like ‘‘however’’ and ‘‘nonetheless’’, does not introduce a truth functional
conjunction. I also contend that it semantically introduces contrasts between contents in Content
M. The semantic value of these terms separate contents and operate on cognitive contents, not on
truth values. Neither ‘‘but’’ nor ‘‘however’’ leaves a trace in content D. Blakemore offers no
explanation for the principle she offers. Let me suggest one for BH2: binary VCDs do not operate
on truth values but on cognitive contents, and they contrast propositions in Content M.
What is wrong with ‘‘Peter plays violin but Gary is learning piano, but Mary plays
basketball’’? The model predicts the following:

Content M, the cognitive content, contrasts a proposition and two contrasted propositions, not a
proposition and a proposition. Contrasting with contrasted propositions does not make much
sense. For example, using brackets, ‘‘Peter plays violin in contrast (Gary learns piano, in contrast
428 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Mary plays basketball)’’ does not make much sense. Of course, the forced options are still
available. My view captures the intuition that there is something wrong with the example. It also
provides an appropriate explanation, identifying the problem as having to do with the cognitive
value of the utterance and as having nothing to do with the truth conditions of the sentence or the
utterance.
Now remember that Content M is the mode of presentation of Content D. In the mode of
presentation of Content D, the propositions are contrasted. Someone saying ‘‘Mary will stay
home, but Paul will come to the movie’’ and adding ‘‘Mary’s staying home is not in contrast with
Paul’s coming to the movie’’ would contradict herself, even if expressing the same Content D.
The Content M expressed by her utterance of the ‘‘but’’ sentence and the Content M of her
utterance of ‘‘Mary’s staying home is not in contrast with Paul’s coming to the movie’’ cannot be
true at the same time.

Now, what about

If P but Q, then R

Where P and Q seems to be joined by a conjunction?

If P. Q then R

It is easy to obtain a conjunction. Take ‘‘P but Q’’. You get P and Q as Content D, and then use

Conjunction Rule
P
Q
P. Q

and then go back

If P. Q then R

I do not think that there are deep semantic differences between ‘‘however’’, ‘‘nonetheless’’
and ‘‘but’’. Piling them up creates redundancy, as in ‘‘Peter is sick, however, nonetheless Mary
went to the movie’’. I will not explore that aspect of binary VCDs I let the reader put together
unary and binary VCDs, as is ‘‘Peter is sick, but even Mary doesn’t care’’.

5. Conclusion

Is there a range of views on contents? Do different VCDs express different views on contents?
I leave this question open. It requires investigations in the lexical semantics of specific languages,
and this is beyond the scope of this paper (but see Iten, 2005). Let me go back to thought content
and content reported in oblique context. Frege assumes that ‘‘but’’ makes no difference to the
thought. He writes
‘‘The way that ‘‘but’’ differs from ‘‘and’’ is that we use it to intimate that what follows it
contrasts with was was to be expected from what preceded it. Such conversational
suggestions make no difference to the thought. (Frege, ‘‘Thoughts’’, p. 39)
 R. Vallée / Journal of Pragmatics 40 (2008) 407–430 429

Assume that ‘‘thought’’ here is not read as truth conditions but thought content, reported by a
propositional attitude sentence. Frege’s view on the relationship between VCDs and thought is
then negative: VCDs make no contribution to thoughts. Frege fails to provide an explanation for
the role of what VCDs express in the cognitive life of the speaker. Consider now

Paul believes that even Mary is kind.

Paul believes that Mary will stay home but Paul will come to the movie.

What is reported in propositional attitudes? If propositional attitudes report thoughts, then

Frege suggests that VCDs make no difference to what is reported. The thought reported cannot be
the Content M, since that latter contains the semantic contribution of ‘‘even’’ and ‘‘but’’.
However, ‘‘even’’ seems to make a difference to thought. The Reflexive–Referential framework
does not take into account thought contents. If we follow Frege, that framework implies that
VCDs do not contribute to thought contents and reported thought. These are very difficult
questions. I leave this topic for another paper.

Acknowledgements

I want to thank Catherine Wearing and Arthur Sullivan, and the audiences at the SEP meeting
in Toronto and the Principia symposium in Florianopolis. The referees made very useful
suggestions and, thanks to them, the paper is better than the previous version. One of them made
very detailed comments, and made me see many problems I did not notice. Suzanne Deschênes
shared with me her linguistic intuitions on ‘‘but’’.

References

Bach, Kent, 1999. The myth of conventional implicature. Linguistics and Philosophy 22, 327–366.
Barker, Stephen, 2003. Truth and conventional implicature. Mind 112, 1–33.
Bennett, Jonathan, 1982. Even if. Linguistics and Philosophy 5, 403–418.
Blakemore, Diane, 1989. Denial and contrast: a relevance theoretic analysis of But. Linguistics and Philosophy 12, 15–37.
Francescotti, Robert M., 1995. Even: the conventional implicature approach reconsidered. Linguistics and Philosophy 18,
153–173.
Frege, Gottlob, 1892. On sense and reference. Philosophical Writings of G. Frege, vol. 1952. Basil Blackwell, pp. 56–78.
Frege, Gottlob, 1918. Thoughts. In: Salmon, N., Soames, S. (Eds.), Propositional Attitudes, Oxford University Press,
1988, pp. 33–55. There is an alternative translation: The Thought; A Logical Inquiry. In: Strawson, P.F. (Ed.),
Philosophical Logic, Oxford University press, 1976, pp. 17–38.
Gazdar, Gerald, 1979. Pragmatics. Implicature, Presupposition and Logical Form. Academic Press, New York.
Glanzberg, Michael, 2005. Presuppositions, truth values, and expressing propositions. In: Preyer, G., Peter, G. (Eds.),
Contextualism in Philosophy: On Epistemology, Language, and Truth. Oxford University Press, Oxford, pp. 349–
396.
Grice, Herbert Paul, 1975. Logic and Conversation. In: Cole, P., Morgan, J. (Eds.), Syntax and Semantics 3: Speech Acts,
Academic Press, New York, pp. 43–58. Also in Grice, 1989. Studies in the Ways of Words, Harvard University Press,
Cambridge.
Grice, Herbert Paul, 1989. Studies in the Ways of Words. Harvard University Press, Cambridge.
Iten, Corinne, 2005. Linguistic Meaning, Truth Conditions and Relevance. Palsgrave MacMillan, Houndsmill.
Karttunen, Lauri, Peters, Stanley, 1979. Conventional implicature. In: Oh, C.-K., Dinnen, D. (Eds.), Syntax and Semantic
11: Presupposition. Academic Press, New York, pp. 1–56.
Klenk, Virginia, 1994. Understanding Symbolic Logic, third ed. Prentice Hall, Englewood Cliffs, New Jersey.
Lycan, William, 1991. Even and even if. Linguistics and Philosophy 14, 115–150.
Perry, John, 1986. Cognitive Significance and New Theories of Reference. In Perry, J. 2001, 189–206.
430 R. Vallée / Journal of Pragmatics 40 (2008) 407–430

Perry, John, 1990. Individuals in informational and intentional content. In: Perry, J. 2000. The Problem of the Essential
Indexicals and Other Essays, CSLI, Stanford, California.
Perry, John, 1997. Indexicals and demonstratives. In: Hale, B., Wright, C. (Eds.), Companion to the Philosophy of
Language. Blackwell, Oxford, pp. 586–612.
Perry, John, 2001. Reference and Reflexivity. CSLI, Stanford, California.
Potts, Christopher, 2005. The Logic of Conventional Implicature. Oxford University Press, Oxford.
Récanati, François, 2002. Unarticulated constituents. Linguistics and Philosophy 25, 299–334.
Récanati, François, 2004. Literal Meaning. Cambridge University Press, Cambridge.
Rieber, Steven, 1997. Conventional implicatures as tacit performatives. Linguistics and Philosophy 20, 51–72.
Stalnaker, Robert, 1974. Presupposition. In: Munitz, M., Unger, P. (Eds.), Semantics and Philosophy. University Press,
New York, pp. 197–213.
Stalnaker, Robert, 2002. Common ground. Linguistics and Philosophy 25, 701–721.
Vallée, Richard, 2005. Complex demonstratives, articulation and overarticulation. Dialogue 97–121.
Wilson, Deirdre, 1975. Presupposition and non-truth-conditional semantics. Academic Press, London.