Dan Sperber (2007) Rudiments of cognitive rhetoric. Rhetoric Society Quarterly 37(4), 361-400.

First English translation of : Dan Sperber (1975) Rudiments de rhétorique cognitive, Poétique: Revue de Théorie et d’Analyse Littéraire (23) 389-415. Deirdre Wilson’s book Presuppositions and non-truth conditional semantics (1975) and this paper were the starting points of our collaboration in developing relevance theory. [PDF version]


Dan Sperber
Sarah Cummins, translator (Département de langues, linguistique et traduction, Université Laval)
Rudiments of Cognitive Rhetoric.
Rhetoric Society Quarterly, 1930-322x, Volume 37, Issue 4, 2007, Pages 361–400.

[I am honored and flattered that this old text of mine should have been deemed worth translating and publishing in the Rhetoric Society Quarterly. It was initially intended as a chapter of my book Le symbolisme en général (Hermann, 1974; translated as Rethinking Symbolism by Alice L. Morton, for Cambridge University Press, 1975). But, under the encouragement of Tzetan Todorov, it developed beyond what I had planned and was taken out of the draft of the book. In 1975, Deirdre Wilson, who had introduced me to analytic philosophy in general and to the work of Paul Grice in particular,  published her book, Presuppositions and non-truth-conditional semantics (Academic Press). She and I decided to write a joint programmatic paper covering the ground between semantics and the rhetoric of figures and we ended up collaborating for thirty years, and developing, with the help of many students and colleagues around the world, the cognitive approach to verbal communication known as Relevance Theory. In retrospect, my 1975 ‘rudiments’ were indeed quite rudimentary. Still, re-reading the article, I confess that I find it insightful. Most insights have been integrated and improved upon in later work. Little has been done however with one of the main insights of the article: that the use of figures of speech evokes ideas not just about the topic of the utterance but also about the shared background knowledge of the interlocutors. — Dan Sperber, December 2006]

Rhetoric, the study of discourse, cannot be simply an offshoot of linguistics, the study of language. [1] Not just one but in fact at least three intellectual devices are involved in the production and interpretation of discourse: grammar—that is, knowledge of a language; the encyclopaedia—knowledge of the world; and symbolism—knowledge of the encyclopaedia. This is the claim that the present article will make and develop.

Linguistics is about sentences; rhetoric is about utterances.

A sentence consists of a pair: a phonetic representation and a semantic representation. The semantic representation of a sentence is a set of senses, with th number of senses corresponding to the degree of ambiguitiy of the sentence. A sentence is an abstract object, a potentiality. An utterance is an approximate physical realisation of this potentiality. An utterance is normally used to transmit, first, a single one of the senses of the sentence and, second, a set of implicatures that are not part of the sentence’s semantic representation. An utterance consists of a pair: a phonetic representation and a conceptual representation. While the linguistic pairing of phonetic representation with semantic representation is determined by the grammar alone, independent of any external input, the rhetorical pairing of phonetic representation and conceptual representation is determined by a complex cognitive mechanism drawing on wide and varied input: the persons involved in the discourse and their situation, extralinguistic signals, previous utterances.

The conceptual representation of an utterance consists of one of the sense of the sentence uttered (completed by the assignment of values to referential expressions) plus implicatures. The semantic representation of a sentence and the conceptual representation of an utterance are thus intersecting sets with a single element in common: one of the senses of the sentence uttered.

Thus, for example, the phonetic representation in (1a)  corresponds to the French sentence in (1b). A grammar of French assigns to it the three meanings (1c), (1d), and (1e). [2] [The corresponding English structures follow, as (1Ea-e), a format followed throughout.]

(1)     a.   žeaštelžurnal

b.      J’ai acheté le journal.

c.      J’ai acheté un exemplaire du journal.

d.      J’ai acheté l’entreprise qui édite le journal.

e.      J’ai soudoyé la rédaction du journal.

(1E)  a.      aɪvbɔtðənuzpeɪpər

b.      I’ve bought the newspaper.

c.      I’ve bought a copy of the newspaper.

d.      I’ve bought the company that publishes the newspaper.

e.      I’ve bribed the editors of the newspaper.

Now, if the sentence is uttered in a typical situation by a husband speaking to his wife as she is about to go out to run her household errands, only the sense (1c) will be retained and the implicature (1f) will be added to it.

(1)     f.       Ce n’est pas la peine que tu achètes le journal aussi.

(1E)  f.       You don’t need to buy the paper too.

The linguistic component associates the phonetic representation in (1a) with the semantic representation [(1c), (1d), (1e)]. The rhetorical component, drawing not only on knowledge of language but also on knowledge of the world, will, in the typical situation described above, assign the conceptual representation [(1c), (1f)].

This account is incomplete. In many cases the conceptual representation of an utterance as a set of propositions (sense and implicatures) does not exhaust its object but leaves a residue. Even if the hearer reconstructs the set of propositions that the speaker has explicitly or implicitly expressed, the utterance, by its very formulation, suggests or evokes something more, something which cannot be logically deduced. In these cases there intervene not only the grammar and the encyclopaedia, but also symbolism. The utterance is figural.*

Under what conditions does an utterance have a figural value? How is a figural utterance interpreted?

In the second part of this paper, I sketch an answer to these questions. To do so, I must first review certain general properties of the semantic representation of sentences and of the conceptual representation of utterances. General rhetoric, and not linguistics alone, comprises the foundation of a rhetoric of figures.

2.1       The semantic structure of a sentence in one of its senses is essentially characterised by a set of entailments. For a declarative sentence, these entailments are truth conditions and are posited as true; in a yes-no question, the truth of the entailments is what is questioned; for Wh-questions, one of the entailments contains a variable and the question is about the value of this variable; an imperative asks for the entailments to be realised and an hortative expresses the wish that they be realised.

A listing of its entailments, however, is insufficient to describe the sense of a sentence. Both sentences of the pairs of declaratives (2a) and (2b), (3a) and (3b), (4a) and (4b), and (5a) and (5b) have identical entailments—that is, the same truth conditions—but clearly different semantics.

(2)        a.      Jérôme et Ursule sont mariés ensemble.

b.      C’est Jérôme qui est marié avec Ursule.

(2E)     a.      Jerome and Ursula are married to each other.

b.      It is Jerome who is married to Ursula.

(3)        a.      On m’accuse d’être en retard.

b.      On me reproche d’être en retard.

(3E)     a.      They accuse me of being late.

b.      They reproach me for being late.

(4)        a.      Martin est généreux et il est riche.

b.      Martin est généreux mais il est riche.

(4E)     a.      Martin is generous and he is rich.

b.      Martin is generous but he is rich.

(5)        a.      Dieu a crée le monde.

b.      Dieu existe et il a crée le monde.

(5E)    a.      God created the world.

b.      God exists and he created the world.

Most, if not all, facts of this kind can be accounted for by assuming that the set of entailments of a sentence (in one of its senses) is partially ordered by a linguistically determined focal structure (which, we shall see below, itself contributes to rhetorical structure).

Thus, both (2a) and (2b) entail (6a) and (6b).

(6)        a.      Jérôme est marié avec X.

b.      X est marié avec Ursule.

(6E)     a.      Jerome is married to X.

b.      X is married to Ursula.

In (2a), the two entailments are unordered. In (2b), they are ordered by syntactic means; (6b) is less focused than (6a). The same focal effect can be achieved by phonological means, by stressing Jérôme in (2a). In either case, the rhetorical effect of this focalisation is to present (6b) as an entailment already known to speaker and hearer, and (6a) as new information provided by the speaker.

Sentences (3a) and (3b) entail (7a) and (7b).

(7)        a.      On affirme que je suis en retard.

b.      On présente mon retard comme répréhensible.

(7E)     a.      They assert that I am late.

b.      They present my being late as reprehensible.

In (3a), (7a) is more focused than (7b); in (3b), the opposite holds. Here, focus is achieved through lexical means: accuser (‘accuse’) and reprocher (‘reproach’) have the same entailments but order them differently. The rhetorical effect of this focalisation is analogous to that of the preceding example: (3a) takes the reprehensibility of the lateness for granted and directs attention to the fact of being late; in (3b), the fact of being late seems to be established, and focus is on the moral judgement.

The sentences (4a) and (4b) entail (8a) and (8b).

(8)        a.      Martin est généreux.

b.      Martin est riche.

(8E)     a.      Martin is generous.

b.      Martin is rich.

In (4a), the two entailments are only weakly ordered by their order in the sentence. In (4b), (8a) is strongly unfocused, compared to (8b). This focalisation is achieved by the selection of the coordinating conjunction mais (‘but’). As for their rhetorical effect, in (4a) the hearer is encouraged to consider the two entailments equally and to consider their joint consequences; in (4b), the hearer is invited to pay less attention to (8a) and more to (8b) and to consider the consequences of (8a) lessened because of the consequences of (8b). Since nothing in the utterance makes explicit the particular relation between (8a) and (8b), the hearer is led to construct an implicature to explain it. (We will return to this example in the next section, in the discussion of how implicatures are calculated.)

Sentences (5a) and (5b) both entail (9).

(9)        Dieu existe.

(9E)     God exists.

In (5a) this entailment is at the lowest focus level, for logical reasons (it is entailed by a series of other entailments of (5a) while entailing none of them). In (5b), it is explicit and thus at the highest focal level. The rhetorical effect of this difference in focalisation is that in (5a) the existence of God is taken for granted, while in (5b) it is robustly re-asserted.

These facts concerning the relative focalisation of the entailments of a sentence (in one of its senses) have received different treatments within the theory of presupposition over the last decade. Is the notion of presupposition necessary? Is it sufficient? The answers to these questions are of little importance here.  It is enough to realise that the meaning of a sentence in one of its senses is essentially characterised by a set of entailments that are partially ordered and thus receive different focus. The difference in focus is a strictly linguistic phenomenon, as is clear from the systematic intuitions to which it gives rise, regardless of utterance context. This linguistic fact plays an important role in the conceptual interpretation of utterances—that is, in rhetoric. To account for it in our rhetorical description, we need only to observe it, even if linguists have not yet provided an explanation. Focus, the consequence of linguistic phenomena, is the cause of rhetorical phenomena and it is in this light that it should be examined.

2.2.1    Understanding an utterance involves, among other things, recognizing it as a sentence of the language, selecting one and only one of the meanings of this sentence, assigning a value to referential expressions, and calculating implicatures. These intellectual operations rely not only on grammatical competence, but also on world knowledge: they are part of performance, involving rhetorical and not linguistic competence. We carry out such rhetorical operations in all aspects of our daily lives, without paying much attention to them. The result of these operations—the conceptual interpretation of an utterance—appears so obvious to us that it requires serious effort to realise the complexity of the work unconsciously carried out. In general, we have been content to say that the context determines the interpretation of an utterance. But how this determination is achieved has never been described—far from it.

It is possible, however, to informally describe (and formalisation, at this stage, would be bogus) some of the principles that underlie the rhetorical mechanism. To do so,  certain essential concepts must first be set forth. These are the concepts of shared knowledge, mobilised shared knowledge, field of relevance (either wide or restricted), informativeness, and relative relevance.

At a given moment in a verbal exchange, participants share certain knowledge: they live in the same universe, are members of the same culture, and perhaps of the same social group; each possesses encyclopaedic knowledge that he can assume the other also possesses. If they are in the same place, each sees what he knows the other also sees. Everything that was said previously in their conversation is also part of this shared knowledge, which is augmented by each new utterance. Shared knowledge is as important to verbal communication (or nearly so) as is a shared language. To a great extent, rhetoric is concerned with how utterances access and modify shared knowledge.

Only knowledge that is knowingly shared is pertinent to rhetoric. If each of two participants in an exchange knows that p but does not know that this is shared knowledge, it is as if p is not a part of their shared knowledge.  The purpose of a large part of verbal communication is not to introduce new knowledge about the outside world but to determine the extent of shared knowledge. For example, if I arrive late at a dinner party, I can be fairly certain that my hosts know that I am late; nonetheless, I say “I’m late” to inform them not of my lateness but of the fact that I am aware of it.

The analysis of some rhetorical data depends on the degree to which knowledge is shared mutual knowledge: does the hearer know that the speaker knows that the hearer knows that…? For example, Pierre says to Paul, in a neutral tone:

(10)      Aragon est le plus grand poète français.

(10E)   Aragon is the greatest French poet.

But both Pierre and Paul believe:

(11)      Aragon est un poète mineur.

(11)      Aragon is a minor poet.

If Paul does not know that Pierre believes (11), he may legitimately believe that Pierre, in uttering (10), has spoken sincerely and literally. If Paul does know that Pierre believes (11), but does not know that Pierre knows that Paul knows that Pierre believes (11), Paul may legitimately conclude that Pierre has spoken insincerely and literally, that he wished to mislead Paul on his opinion of Aragon. It is necessary that Paul know that Pierre knows that Paul knows that Pierre believes (11) for the only reasonable and legitimate interpretation of (10) to be an ironic one.

The notion of shared knowledge (at a particular moment between particular participants) allows for a definition of the general notion of relevance. As a first approximation, a relevant proposition is one that, when added to shared knowledge, brings about new consequences. To do this, the proposition must supply, for a previously known object, information that was not previously known. In other words, at least one of the entailments of the proposition must be part of shared knowledge and at least one other must not be. So, with regards to our shared knowledge, (12) is not relevant because we know nothing of its subject, and (13) is not relevant because it contains no new information.

(12)      Le Grand Zouzou Sacré est mort.

(12E)   The Great Holy Zouzou has died.

(13)      La capitale du Japon est Tokyo.

(13E)   The capital of Japan is Tokyo.

On the other hand, any proposition of which some but not all of the entailments are part of shared knowledge is relevant. The wide field of relevance comprises all propositions that are relevant in this sense. It can thus be described as the potential complement of shared knowledge, and any modification in shared knowledge will bring about a modification in its complement.

This definition of relevance is clearly too broad. Shared knowledge, at any given moment, is not a homogeneous whole. Relevance is a function of shared knowledge and thus of memory. At a given moment, an individual’s memory comprises at least two parts: passive memory, information that has been gathered and stored throughout a lifetime;  and active memory, information that has been acquired, or called up from passive memory, in the previous moments. Within active memory, not all information is mobilised to an equal extent at a given moment. More specifically, the conversation directs attention to some of the information only. For example, a question mobilises a small section of information linked to the entailments of the question itself.

Just as general shared knowledge defines the wide field of relevance, mobilised shared knowledge (mobilised most often by the conversation itself) defines the restricted field of relevance, which includes all propositions capable of being the complement of mobilised shared knowledge. The difference between the broad and the restricted fields of relevance has one consequence which may appear paradoxical: a proposition can be part of the restricted field without being part of the broad field. A proposition that is part of shared knowledge is by definition absent from the broad field of relevance. But if this proposition is not part of mobilised shared knowledge, it can be part of the restricted field—in other words, it may be relevant to mobilise it. This difference explains why an utterance like (13), generally not relevant, can become relevant in the context of a specific conversation, for example as the answer to the request “Name a country and its capital with the same number of letters.”

Propositions that are relevant in regard to shared knowledge (whether mobilised or not ) are not all equally relevant. It might seem at first glance that a proposition increases in relevance as its informativeness increases, but a moment’s reflection shows that this is not the case.

A proposition in an utterance increases in informativeness to the extent that, when added to shared knowledge, it has more consequences. This notion of informativeness becomes a bit less intuitive in the case of an ordered series of propositions such that the proposition n + 1has proposition n among its consequences, while the opposite does not hold. In this situation, n + 1 has all of the consequences of n, plus its own consequences, and thus is more informative than n.  For example (considering, for simplicity, only entailment relations and not implicatures), utterance (14a) is less informative than (14b), which is less informative than (14c).

(14)      a.      Isidore a mangé.

b.      Isidore a mangé des épinards.

c.      Isidore a mangé tous les épinards.

(14E)   a.      Isidore has eaten.

b.      Isidore has eaten spinach.

c.      Isidore has eaten all the spinach.

It is intuitively obvious that, in a series such as (14)—in which, by definition, informativeness increases at each stage—relevance first increases and then decreases. In other words, the additional consequences of each proposition vis-à-vis the preceding one in the series first rise then fall in number and importance. For example, imagine that only one of the following three propositions is part of shared knowledge:

(15)      On a besoin d’épinards.

(15E)   We need spinach.

(16)      Isidore est allergique aux épinards.

(16E)   Isidore is allergic to spinach.

(17)      Isidore devrait rester à jeun.

(17E)   Isidore is not supposed to eat anything.

In the situation where (15) is part of shared knowledge, (14c) has the most relevance. When (16) is part of shared knowledge, (14b) is maximally relevant, and the additional consequences of (14c) over (14b) are minimal; inversely, in these two situations, (14a) is too uninformative to be relevant. In the situation of (17), (14a) is most relevant and there is no increase in relevance with the additional information given by (14b) and (14c) but instead a decrease.

To use a mathematical metaphor, we can say that the degree of relevance of a proposition in an utterance is a linear function of the degree of mobilisation of the shared knowledge it relates to, and a parabolic function of its own informativeness. If we eliminate from this analogy the erroneous impression of precise measurement it conveys, we can say that at a given moment in a verbal exchange, there is not a point but rather a zone of maximal relevance in regards to which the relevance of each utterance can be intuitively assessed.

With these notions established, we can turn to the rules that govern verbal exchanges. These rules all relate to an obvious principle: the speaker is expected to do whatever is necessary in order to be understood. The speaker has in mind a conceptual representation that he wishes to transmit to the hearer.  It is virtually never necessary to explicate this representation, to utter it wholly and unambiguously, in order to make it understood to the hearer. In any case, to do so would be exorbitant: all referential expressions would have to be replaced by lengthy descriptions, additional clauses would be needed to avoid any ambiguity, and each implicature would have to be made exlicit at the same level of detail. In many cases, an utterance of a few words would require a corresponding explication running to several pages of text.

Precisely because speaker and hearer share knowledge, the speaker may explicate only a small part of the conceptual representation he wishes to transmit, knowing that the hearer can complete it. Once again, understanding an utterance is entirely different from understanding the meaning of the sentence uttered.

It is the job of rhetoric to explain how, on the basis of a fragment of a conceptual representation (which may, moreover, be expressed by an ambiguous sentence), the hearer manages to reconstruct the complete representation, and how the speaker can feel certain that the hearer will do so. We must assume that the speaker is supposed to follow certain rules and that the hearer takes for granted, unless shown otherwise, that the speaker is doing so. Below I sketch some of these rules and their effects.

I will mention only very briefly disambiguation and the assignment of referential values; in fact, contrary to widespread belief (due to confusing semantic ambiguity and referential ambivalence with conceptual equivocation), these play only a minor role in figural speech. When several conceptual representations within the zone of maximal relevance can be constructed from the meanings of the sentence uttered, the meaning selected will be the one corresponding to the least informative representation. This rule stems from the fact that the speaker is held to be responsible for what he says, and this responsibility is certain to hold only for the weakest interpretation of what was said—the interpretation with the fewest consequences, the least informative one in regard to shared knowledge. A similar rule applies to remove any referential ambiguity.

I will have more to say about how the fragment of conceptual representation that is an utterance comes to be completed. An utterance can be fragmentary in two ways. First, it expresses only some of the propositions of the conceptual representation and must be completed by implicatures. Second, in some cases, the best known of which is ellipsis, the utterance may contain gaps and completely express none of the propositions of the conceptual representation. I will discuss in turn the rules which allow implicatures to be calculated and those that allow gaps to be filled.

2.2.2    Leaving aside the beginnings of discourse or conversation and changes of subject (which have their own conditions), the speaker is expected to speak in such a way that Conditions I – III are met:

I.       There exists  a conceptual representation of the utterance such that the proposition expressed by the utterance is in the restricted field of relevance.

II.     The proposition uttered is neither too informative nor too uninformative; it is maximally relevant.

III.    The linguistically determined focus ranking of the entailments of the proposition corresponds to their degree of relevance.

A corollary of II and III is that the most relevant proposition of the conceptual representation that the speaker wishes to convey must be uttered and focalized.

It can happen that one of the semantic and referential interpretations of the sentence uttered is sufficient to fulfill Conditions I – III; in this case, the calculation of implicatures is not required. The only task that falls to the hearer is to disambiguate the sentence uttered and assign values to referential expressions. The hearer is guided in this task by Conditions I – III, eliminating any interpretation that does not satisfy them.

On the other hand, any time Conditions I – III are not directly satisfied by an interpretation of the utterance itself, it is necessary to calculate implicatures in order to satisfy them.

A first example:

(18)      a.      Pierre: — Irez-vous vous promener?

b.      Paul: — Il va pleuvoir.

(18E)   a.      Pierre: “Are you going for a walk?”

b.      Paul: “It’s going to rain.”

No interpretation of (18b) directly satisfies Condition I. The hearer starts from the principle that there is a conceptual representation of (18b) that does meet this condition—in other words, that it is possible, by adding certain propositions to mobilised shared knowledge, to broaden the restricted field of relevance so that (18b) will be part of it. For example, if (19) is part of shared knowledge and can be mobilised, Pierre can deduce the implicature in (20) from the conjunction of (18b) and (19):

(19)      Paul ne se promène pas quand il pleut.

(19E)   Paul does not go for walks when it is raining.

(20)      Paul n’ira pas se promener.

(20E)   Paul is not going for a walk.

With the addition of the implicature in (20), the conceptual representation of (18b) meets Condition I. (18b) is more informative than (20) because, on the basis of shared knowledge, (20) is a consequence of (18b) whereas the opposite is not true. If, moreover, Paul might legitimately think that Pierre would like to know why he isn’t going for a walk, then (18b) is not only more informative but also more relevant than (20), so Condition II is also met. Condition III is automatically met.

A second example:

(21)      a.      Le juge: —À quelle heure exacte êtes-vous rentré chez vous?

b.      L’accusé: — Entre huit et neuf heures.

(21E)   a.      Judge: “At exactly what time did you return home?”

b.      Accused: “Between eight and nine o’clock.”

The accused’s answer is not informative enough to be maximally relevant, and Condition II is thus not directly satisfied. Note that the answer is perfectly truthful, even if the accused knows that he returned home at exactly 8:47. However, if it were later proven that the accused had this knowledge, he could be considered to have misled the court. It is assumed that the accused did his best to satisfy Condition II. From this rule and (21b), the following implication can be deduced:

(22)      L’accusé ne sait pas à quel moment précis il est rentré chez lui.

(22E)   The accused does not know exactly what time he returned home.

With the implication in (22), Condition II is met; thus the accused can be held accountable for this implicature if it is later proven false.

Imagine that instead the accused had answered:

(23)      Je suis rentré à 8h47, au moment où la speakerine de la télévision a fait un lapsus et dit : «Et voici maintenant le film de Karl Marx.»

(23E)   I returned home at 8:47, just as the announcer on the television made a blooper by saying, “And now here is the film by Karl Marx.”

Once again, Condition II is apparently violated, but this time by too much information. The relevance of this additional information can be established and Condition II restored when the court’s wish to have proof of the accused’s statement is added to shared knowledge, when (23) is understood as implicating that the accused heard the announcer’s blooper while watching television at home at 8:47. If it was later proven that the accused was indeed at home at 8:47 but did not turn on the TV and only learned of the blooper from a friend the following day, he could again be held to have misled the court, even though (23) in no way asserts that the accused actually heard the announcer’s blooper.

A third example (repeating (4b)):

(24)      Martin est généreux mais il est riche.

(24E)   Martin is generous but he is rich.

As we saw above, the use of mais ‘but’ focuses the second proposition. Suppose that (24) is said in reply to (25):

(25)      On m’a dit que Martin est pauvre et généreux.

(25E)   I was told that Martin is poor and generous.

In this case, Condition III is directly satisfied because the second proposition, in contradicting an opinion known to both speaker and hearer, is more relevant than the first proposition, which merely provides confirmation. In this case, no implicatures arise from (25).

If, however, we suppose that nothing in shared knowledge immediately justifies the focal structure of (24), Condition III is not met and an implicature must be calculated. Imagine then that (26) – (28) are part of the speaker and hearer’s shared knowledge:

(26)      La générosité est une grande vertu.

(26E)   Generosity is a great virtue.

(27)      Une vertu est d’autant plus grande qu’elle est difficile.

(27E)   The more difficult a virtue is, the greater it is.

(28)      La générosité est facile aux riches.

(28E)   It’s easy for rich people to be generous.

Based on the proposition Martin est généreux ‘Martin is generous’ and (26), one might be tempted to conclude that Martin is very virtuous. But if Martin est riche ‘Martin is rich’ and (27) – (28) are added to these premises, this first conclusion is invalidated. If Martin’s degree of virtue is relevant, then his wealth is more relevant than his generosity, and with the implicature in (29), the utterance in (24) now meets Condition III.

(29)      Martin n’est pas aussi vertueux que sa générosité pourrait le faire croire.

(29E)   Martin is not as virtuous as his generosity might lead one to believe.

These examples illustrate the general principle followed in calculating implicatures: when the interpretation of an utterance does not directly satisfy Conditions I – III, the hearer looks for propositions that can be deduced from the conjunction of the utterance and shared knowledge and which, when added to the conceptual representation of the utterance, will satisfy these conditions.

By definition, an implicature is not more informative in regard to shared knowledge than the utterance it is entailed by. But Condition II has a corollary whose importance is such that we will state it as a separate condition:

(IV)     An implicature is not more relevant than the utterance that implicates it.

This condition is particularly important here because, as we will see, when the calculation of implicatures does not satisfy it, it may lead to a figural interpretation of the utterance.

2.2.3.   An utterance expresses only a fragment of the conceptual representation that the speaker wishes the hearer to construct. Moreover, this fragment may itself be fragmentary: as well as the propositional implicatures discussed above, the utterance may have lexical implicatures—or more accurately, sub-propositional implicatures—which I will call gaps.

One type of gap—ellipsis—is well known and has been fairly well studied. An ellipsis is a gap revealed by the syntactic analysis of the sentence uttered; certain underlying syntactic components have no surface lexical realisation. Less attention has been paid to other kinds of gaps, whose presence may be suggested by semantic anomalies or contradictions but which ultimately can only be definitely established by the conceptual representation of the utterance. Here I am entering into a new domain and the following hypotheses, even more than the previous ones, should be taken as exploratory.

To say that an utterance contains gaps means that the hearer can, in certain cases, add constituents to those that have been made explicit, in order either to construct a sentence, when the utterance does not constitute a sentence on its own, or to modify an uttered sentence when it has no acceptable conceptual representation. It is obvious that this possibility will be compatible with the principle that the speaker does what is necessary to be understood only if the omission of constituents by the speaker and their restitution by the hearer are greatly constrained.

An utterance consists not only of a sequence of lexical elements, but also the syntactic relationship among them. Syntactic relations play as great a role in semantic interpretation as does the meaning of lexical elements: the semantic interpretation of a syntactic relation is a logical function. The hearer must start from the principle that the lexical elements, along with the logical functions the speaker expresses in syntax, contribute to the relevance of the utterance; thus the adjunction of a new constituent should not eliminate any of these functions. Hence Condition V:

V.        The complete interpretation of an utterance with a gap must maintain the logical functions expressed syntactically in the utterance.

For example, if  the utterance in (30a) is interpreted as having a gap, then (30b) is a possible interpretation and (30c) is not a possible interpretation.

(30)      a.      La Neuvième plaît aux amateurs.

b.      La Neuvième Symphonie de Beethoven plaît aux amateurs de concerts.

c.      Le début de la Neuvième plaît à l’élite des amateurs.

(30E)   a.      The Ninth pleases music lovers.

b.      Beethoven’s Ninth Symphony pleases music lovers who go to concerts.

c.      The openning of the Ninth pleases the elite among music lovers.

In (30c), la Neuvième and les amateurs no longer hold the logical functions of subject and complement of the verb plaire [in English, music lovers and the Ninth no longer hold the functions of subject and object of like] and are instead complements of a noun. If such changes in function were allowed, how could the hearer ever manage to reconstruct the conceptual representation intended by the speaker? The possibilities would be far too numerous.

Condition V has an interesting corollary. Either the syntactic analysis reveals functions that are unfilled in the utterance—this is an ellipsis and the position of additional constituents to supply is clearly indicated; or there is a non-elliptical gap, in which case additional constituents can be inserted into only two types of position: they may either dominate none of the constituents that are expressed, or they must dominate all of them. In other words, in non-elliptical gaps, additional constituents are either complements of expressed constituents, or else the set of expressed constituents, along with their logical-syntactic relationships, is a complement of the additional constituents.

Note that most kinds of elliptical gaps also fall into these same position types; (31) shows a dominated ellipsis and (32) a dominating one.

(31)      On a déjà donné (de l’argent).

(31E)   We already gave (money).

(32) (Passez-moi) le sel et le poivre.

(32E)   (Pass me) the salt and pepper.

In (33), on the other hand, the elided constituent is in a position of partial domination within the sentence.

(33)      Isidore a mangé des épinards et Théodule (a mangé) des salsifis.

(33E)   Isidore ate spinach and Theodule (ate) salsify.

The elided element in (33) repeats an element appearing earlier in the sentence; this phenomenon is quite different from (31) and (32), in that syntactic analysis reveals not only the unfilled position but also the constituent that must fill it. This is not a gap that must be filled by recourse to memory and reasoning.

We can thus propose that, for all genuine gaps, whether elliptical or not, the omitted constituents cannot both dominate and be dominated by the expressed constituents; thus gaps are either entirely embedding or entirely embedded. Condition V therefore severely restricts the range of functions that may be filled by omitted constituents.

Non-elliptical embedded gaps are common, such as the example in (34):

(34)      Richard se pique (à l’héroïne).

(34E)   Richard shoots up (heroin).

Non-elliptical embedding gaps have received less attention. Most are implicit modalisations. I use the term modalisation in a fairly broad sense here, encompassing constituents that may take an entire clause as their complement, for example il faut que, il est possible que, on dit que, on suppose que, c’est comme si, on fait comme si (‘it is required that, it is possible that, people say that, it is assumed that, it is as if, people act as if’). The function of  these modalisers is to limit the scope of the complement clause, as the entailments of the complement clause of a modaliser are not the truth conditions of the complete modal proposition. As we shall see, this option is used whenever shared knowledge or the logical-semantic properties of the utterance itself allow exclusion of the possibility that the speaker wished an uttered proposition to be understood as true. Consider, for example, the utterances of (35) and (36).

(35)      Vous tournez à droite au carrefour.

(35E)   You turn right at the corner.

(36)      Vous avez gagné le gros lot. Qu’est-ce que vous allez faire?

(36E)   You’ve won the jackpot. What are you going to do?

Each of these utterances is equivocal. (35) is either an assertion or an order. If it is an order, it has a gap and must be embedded under il faut que (‘it is necessary that’). (26) is either an assertion or a supposition. If it is a supposition, it has a gap and must be embedded under on suppose que (‘suppose that’).

Condition V restricts the range of functions that can be filled by a constituent added to an utterance containing a gap. But it says nothing about the meaning of such constituents.

Logically, a function could be filled by any one of an infinite number of conceivable constituents. Therefore, there must be strict constraints on the choice of constituents that can fill a given function.

It is not essential that the function be able to be filled by a single lexical item; however, if the speaker is to be understood, it is necessary that all of the constituents that the hearer might choose entail the same encyclopaedic consequences—in other words, that they be conceptually equivalent. Two things will guide the hearer in making this choice: first, mobilised shared knowledge and second, the hypothesis that the speaker has not violated Conditions I – III. A speaker may licitly produce a gap only if a unique solution can be determined by these two elements.

Hence Condition VI:

VI.       In a gapped utterance, the omitted element fills a function which, on the basis of mobilised shared knowledge, can be filled only by conceptually equivalent constituents, so that Conditions I – III are met.

Thus when an utterance has a gap, the hearer must seek, among the range of choices allowed by Condition V, a function for which there exists a class of constituents that satisfy Condition VI. The presence of a gap, which initially seemed to leave the door wide open for all kinds of misunderstandings, in fact allows the speaker (provided he follows Conditions V and VI) to make himself understood economically.

It should be noted that, just as Condition IV is a corollary of Condition II as regards implicatures, Condition VI is a corollary of Condition II as regards gaps. If  mobilised shared knowledge does not allow gaps to be immediately and unequivocally filled, then the speaker has been too uninformative and has not uttered, even lacunally, the most relevant proposition of the conceptual representation that he wished to convey.

A few examples will show how, when semantic interpretation and the calculation of implicatures do not produce a conceptual representation of the utterance satisfying Conditions I – IV, the hearer is led to hypothesise the presence of a gap and to fill it, in accordance with Conditions V and VI. These examples also show that certain problems which would cause serious difficulties and even paradoxes if treated in a semantic framework can receive a relatively simple rhetorical solution.

First example:

(37)      a.         — L’accusé avait avoué le vol, mais je ne sais pas si maintenant il a avoué le viol et le meurtre.

b.         — Il a avoué.

(37E)   a.         “The accused confessed to the theft, but I don’t know if he has now confessed to the rape and the murder.”

b.         “He has confessed.”

In (37b) the complement of avouer (‘confess’) is absent. The hearer can immediately eliminate le vol (‘the theft’), because if it were the content of the gap, (37b) would not be relevant. Shared knowledge, mobilised by (37a), allows for four other possibilities:

(38)      a.         Il a avoué le viol.

b.         Il a avoué le meurtre.

c.         Il a avoué le viol ou le meurtre.

d.         Il a avoué le viol et le meurtre.

(38E)   a.         He confessed to the rape.

b.         He confessed to the murder.

c.         He confessed to the rape or the murder.

d.         He confessed to the rape and the murder.

Intuitively (and unless very unusual circumstances hold), (37b) will be completed by (38d). For example, if it later proved that only (38a) was true and that the speaker knew this, he could be accused of having misled the hearer. Why is this so, given that the utterance does not make explicit what exactly the accused confessed to?

Since there are four non-equivalent possibilities, it would seem that Condition VI is not met. But the hearer starts from the principle that the speaker has obeyed Condition VI and that there is therefore a way of eliminating three of the four possibilities. The first three options do not directly meet Condition II; they are not maximally relevant and thus would lead to the following implicatures: (39a) for (38a), (39b for 38b), and (39c) for (38c).

(39)      a.   Le locuteur ne sait pas si l’accusé a avoué le meurtre.

b.   Le locuteur ne sait pas si l’accusé a avoué le viol.

c.   Le locuteur ne sait pas si l’accusé a avoué le meurtre ou bien le viol

(39E)  a.   The speaker does not know whether the accused confessed to the murder.

b.   The speaker does not know whether the accused confessed to the rape.

c.     The speaker does not know whether the accused confessed to the murder or to the rape.

If one of the propositions in (39) was already a proposition of mobilised shared knowledge instead of being a new implicature, it would allow for the corresponding proposition of (38) to be selected, and (37b) would thereby have a single interpretation. But since none of the propositions of (39) is known, it is impossible to choose, as an interpretation of (37b), among the interpretations (38a), (38b),  and (38c), which do not directly satisfy Condition II. The speaker thus could not use the gapped utterance of (37b) to convey one of these interpretations, as there is no implicature which would select it. Only (38d) (or, more accurately, only the set of propositions equivalent to (38d)) directly satisfies Condition II and thereby Condition VI.

Second example:

(40)      Martin n’est pas riche, il est extrêmement riche.

(40E)   Martin is not rich, he is extremely rich.

Logically, (40) should be a contradiction but intuitively, it is not. The first clause of the utterance is understood not as meaning that Martin is poor, but that the term riche (‘rich’) does not suffice to describe him. To describe this kind of utterance in semantic terms, we would have to invent an ad hoc ambiguity for negation. In one sense, negation would mean that the proposition it applies to is not true. In another sense, it would mean that the proposition it applies to is not appropriate. Such a device would mean that all negative sentences would be ambiguous in this way. In fact, the second “meaning” is only found in special cases—almost always in utterances that echo a previous utterance. An utterance like (40), for example, normally comes in response to an utterance asserting that Martin is rich. The device would also be fairly costly: to handle a few special cases, the meanings of negative sentences are doubled in number, with all the problems that entails for logical calculations and disambiguation. The solution is moreover unnecessary, since the problem posed by utterances like (40) has an easy rhetorical solution—one requiring no apparatus that is not independently motivated and preserving the sole truth-functional meaning of negation.

A negative proposition is true if one of the entailments of the corresponding positive sentence is false. In this sense, every negative proposition is equivocal and the hearer must determine which of the entailments the speaker is negating. This ambiguity is partially resolved by Condition III, which entails that the most focused entailments are the ones negated and the least focused ones are presented as true. Consider, for example:

(41)      a.      Martin est extrêmement riche.

b.      Martin n’est pas extrêmement riche.

c.      Martin est riche.

d.      Martin n’est pas riche.

(41E)   a.      Martin is extremely rich.

b.      Martin is not extremely rich.

c.      Martin is rich.

d.      Martin is not rich.

(41a) entails (41c) and thus if (41c) is false or, equivalently, if (41d) is true, then (41b) is true. However, negation reverses the order of entailments: (41d) entails (41b), rather than the contrary. Therefore (41b) is less informative than (41d). However, the focal order of the entailments of the positive sentence still holds, and thus (41d) is less focalized than the overall proposition expressed by (41b). Therefore, if the speaker uttered (41b) in order to convey (41d), he would not be obeying Condition III. This is why, barring any special implicatures, (41b) is normally understood as meaning that Martin is rich, but not extremely rich.

Let us return now to (40). If the utterance is complete, there will always be a contradiction, whatever implicatures are negated. The problem is resolved by assigning  to (40) a gapped interpretation and by introducing into the field of negation an additional constituent of the same type as extrêmement, ‘extremely,’ so that negation now applies only to the specific entailments of this constituent and that Martin’s being rich, as in (41b), is not negated. Extrêmement itself suggests the class of possible constituents—adverbs of the same class but incompatible with it. Hence the gapped interpretation of (40) is:

(42)      Martin n’est pas ordinairement riche, il est extrêmement riche.

(42E)   Martin is not ordinarily rich, he is extremely rich.

Rather than ordinairement ‘ordinarily,’ the hearer could choose simplement ‘merely,’ comme tout le monde ‘like everyone else,’ etc. In any case, there is no ambiguity and Condition VI is met, as is, transparently, Condition V.

A third example:

(43)      La maison que Pierre habite, c’est celle de Paul, mais il a un garage en plus.

(43E)   The house where Pierre lives is Paul’s house, but he also has a garage.

This utterance is equivocal. On one interpretation, Pierre lives in Paul’s house and has a garage in addition. On a second interpretation, Pierre lives in a house identical to Paul’s but with a garage in addition. To explain the second interpretation semantically, we would have to say for example that the verb être, ‘be,’ is ambiguous and one of its meanings is être semblable à, ‘be similar to.’ As with the previous example, we would be doubling the number of meanings of all sentences containing the verb être in order to account for a few special cases. And here again, there is a simple rhetorical solution: modalising (43) and assigning to it a complete interpretation as in (44):

(44)      C’est comme si la maison que Pierre habite c’était celle de Paul, mais il a un garage en plus.

(44E)   It’s as if the house where Pierre lives is Paul’s house, but he also has a garage.

To meet Condition VI, c’est comme si, ‘it’s as if,’ must be posited; an utterance like (43) will typically come in response to a question like Comment est-ce la maison que Pierre habite? ‘What’s the house that Pierre lives in like?’

However, in the absence of Condition V, the complete interpretation of (45) would have been preferred:

(45)      La maison que Pierre habite, c’est comme la maison de Paul, mais il y a un garage en plus.

(45E)   The house where Pierre lives is like Paul’s house, but there is a garage as well.

While c’est comme si, ‘it’s as if,’ in (44) compares the known world to an imaginary world in which the house Pierre lives in is Paul’s house, comme, ‘like,’ in (45) compares two objects in the known world: Pierre’s house and Paul’s house. (45) therefore has a more restricted and more precise meaning. Intuitively, it is also part of what the speaker wishes to convey to the hearer.

Is the only reason for preferring (44) to (45) the wish to preserve Condition V, which is so useful for understanding gapped utterances but which is violated by (45)? No, not entirely. In its gapped interpretation, (43) is slightly hyperbolic, which the two-world comparison of (44) accounts for but the two-object comparison of (45) does not. Moreover, if (44) is to meet Condition II and thereby Condition VI, shared knowledge must determine the relevance of c’est comme si; in other words, (44) must implicate a proposition like (45). Hence the intuition that (45) enters into the conceptual representation of (43) is confirmed; but it does so as an implicature and not as a completed interpretation. We will see below how the figurativeness of certain tropes is brought about by the insufficiency of shared knowledge in determining the scope—and thus maximizing the relevance—of a modalisation like c’est comme si.

The fourth and final example:

(46)      J’avais aperçu Jules au meeting mais il m’a juré qu’il n’y était pas.

(46E)   I glimpsed Jules at the meeting but he swore he wasn’t there.

If (46) is a complete utterance, it implies that Jules is either a liar or an amnesiac. In many situations, such an implicature would be more relevant than the utterance itself and thus (46) would violate Condition IV. If this is the case, (46) will be interpreted as having a gap and modalised as in (47):

(47)      J’avais cru apercevoir Jules au meeting, mais il m’a juré qu’il n’y était pas.

(47E)   I thought I glimpsed Jules at the meeting, but he swore he wasn’t there.

Note—and this is the reason for the example, which is otherwise unproblematic—that while the superficial syntactic relations in (47) are changed, the deep syntactic relations, the ones corresponding to logical functions, are maintained: the deep subject of apercevoir ‘glimpse’ is indeed je, ‘I,’ as in (46). Condition V is thus met.

Similar cases to the examples given here are common in ordinary speech and require the notion of a gap which, we will see below, also plays a crucial role in figural speech. Among gaps, only ellipsis is part of the linguistic analysis of the sentence; other gaps appear in the conceptual representation of the utterance and thus in the rhetorical analysis. The widespread confusion in the rhetorical literature between a sentence and an utterance has led certain authors to an unmotivated extension of the notion of ellipsis, and others (sometimes the same ones) to almost entirely ignore non-elliptical gapping phenomena, especially modalisations. The somewhat sterile debate on the nature of metaphor—an elliptical comparison for some, a figure with meaning substitution for others—stems from this confusion. We return to the matter below.

3.1       When none of the meanings of an uttered sentence directly satisfies Conditions I – III, the hearer has two resources: he must seek, on the basis of mobilised shared knowledge and with the guidance of Conditions IV – VI, which the speaker is expected to follow, either an implicature or a gap filler that will, when added to the utterance, allow a canonical conceptual representation to be constructed. At the end of this process, the hearer’s object of attention, an utterance “in quotes,” is transformed into a set of analysed propositions that describe the information that the speaker wished to convey.

What happens when mobilised shared knowledge does not allow the hearer to construct a conceptual representation meeting Conditions I – IV and the utterance, without a complete analysis, remains in some sense still in quotes? The hearer has several possible hypotheses. He may think that the speaker has not managed to express himself or has overestimated the extent of shared knowledge. It is also possible that the speaker has deliberately violated the conditions of verbal communication out of hostility towards the hearer. Thus, in example (21): the accused, when asked the exact time he returned home, answers “between eight and nine o’clock;” if the fact that the accused knows the exact time he returned home is part of shared knowledge, the implicature in (22) is ruled out, and the accused’s answer also constitutes a refusal to answer. More often, when mobilised shared knowledge does not lead to a canonical conceptual representation, the speaker is displaying neither incompetence nor recalcitrance but merely inviting the hearer to seek a figural interpretation. What could not be achieved on the basis of mobilised shared knowledge can be accomplished through symbolic evocation.

Encyclopaedic memory has a two-fold organisation: on one hand a relatively stable classification of information based on numerous conceptual hierarchies, and on the other a network of associations that are constantly replenished from occasional analogies and juxtapositions made outside of the classificatory system. We might say that the encyclopaedia has both a rational and a symbolic organisation (eliminating from these terms any connotation of value judgements); or rather, that the encyclopaedia, rationally organized knowledge of the world, is itself the object of symbolic knowledge, symbolism  being a meta-encyclopaedia within the encyclopaedia. While the rational organisation of the encyclopaedia allows information to be summoned up directly—invoked—on the basis of the concept it is attached to, symbolic organisation allows information to be evoked on the basis of other information it is associated with.

In Rethinking Symbolism I suggested that, when mobilised knowledge and rational invocation are insufficient to account for the object of attention by a fully analysed conceptual representation, appeal is made to symbolic evocation in the following way. First, attention is directed to the particular conceptual condition whose non-fulfilment caused the failure of the conceptual representation. This focalisation defines a field of evocation in passive memory from within which the missing information can be reconstructed. Secondly, evocation scans this field in order to satisfy the unmet condition. Thirdly, if evocation is successful, the defective conceptual representation can be completed and the initial object of attention thus receives its symbolic interpretation. Now it is symbolically associated with all the information that had to be evoked in order to assign to it a conceptual representation.

Invocation is a sequential process, evocation a parallel process, in the sense these notions are used in cognitive psychology. Invocation is a process of reasoning in which each step is determined by the outcome of the previous step. Evocation in a process of sorting: different objects are examined in turn or in parallel and each operation is logically independent of the others. In invocation, therefore, operations are ordered and their order is logically determined. For evocation, no order is necessary, and if there is one, it is determined by factors relating to energy. In other words, impulses and desires, which can only hinder invocation, are on the contrary an engine for evocation. The goal of invocation is a single pre-existing object; evocation constructs its object. In symbolic interpretation, only focalisation is determined by the object of interpretation; evocation depends on the idiosyncrasies of the interpreter. It is thus pointless to search for a meaning that will be systematically associated with a symbolic phenomenon. No such meaning exists.

With these notions established, it is possible to sketch an answer to the two questions posed at the beginning of this article: Under what Conditions does an utterance have a figural value? How is a figural utterance interpreted? I propose:

VII.     An utterance takes on a figural value when mobilised shared knowledge is insufficient to assign to it a conceptual representation in accordance with Conditions I – VI and this deficiency is not attributed to the speaker’s incompetence or recalcitrance.

VIII.    When an utterance takes on a figural value, the unsatisfied condition responsible for its figural character is focalized; evocation is used to restore the condition and thereby correct the initial conceptual representation.

This is not all. Even if focalisation and evocation lead to an acceptable interpretation of the utterance, the fact remains that the speaker has acted as if the information evoked was part of mobilised shared knowledge or could be invoked. If the hearer, having rejected the idea of speaker incompetence or recalcitrance, wishes to interpret the speaker’s behaviour as obeying the principles of conversation, he can only do so symbolically. Only a second evocation, bearing on the relationship between speaker and hearer, no longer about the utterance [l’énoncé] but about the utterance act [l’énonciation], will allow the hearer to conceive how the knowledge that was originally evoked could have been invoked. Hence:

IX:      When an utterance has received a symbolic interpretation under Conditions VII and VIII, the utterance act itself is symbolically interpreted. Attention is focused on the inadequacy of shared knowledge and a second evocation attempts to reconstruct the conditions under which the first evocation would have been superfluous—that is, the conditions under which the information that was first evoked could have instead been invoked.

I suggest that this two-stage evocation, about the utterance and about the utterance act, is a characteristic of figural speech which distinguishes it from all other kinds of verbal symbolism. [3]

When Conditions I – III are not met, the hearer looks for an implicature or a gap filler. If none is found in shared knowledge, or one is found but it does not meet Condition IV in the case of an implicature or Conditions V – VI in the case of a gap, then the hearer must turn to evocation. There are therefore two major types of figures, depending on whether the evocation is intended to establish or correct an implicature or to establish or correct a gap-filler. These two types can of course be freely combined. I will give examples of figures by implicature and of figures by gap without attempting to cover, even schematically, the full range of possibilities. My purpose here is not to present a taxonomy, but to study certain general mechanisms of figural interpretation.

3.2       Figures from implicature


(48)      Ma femme, m’invitant a goûter son tout premier soufflé a par inadvertance laissé tomber une cuillerée sur mon pied, fracturant ainsi plusieurs petits os. (Woody Allen)

(48E)   My wife, inviting me to sample her very first soufflé, accidentally dropped a spoonful of it on my foot, fracturing several small bones. (Woody Allen)

This utterance implies:

(49)      Le soufflé avait une densité de l’ordre de celle du plomb.

(49E)   The soufflé was as heavy as lead.

This implicature is a gross violation of Condition IV, whereby an implicature must be less relevant than the utterance which provides it. To restore the condition, it is necessary to first evoke a paradoxical soufflé, an unprecedented culinary disaster. Now, under what circumstances would it have been possible to invoke this image rather than having to evoke it? If among the knowledge shared by Woody Allen and his readers was the view that his wife, or young wives in general, are prone to culinary disasters. Thus (48) evokes in the second stage a complicity of attitude among male chauvinists towards women who fail miserably, however hard they try. In short, the first evocation, about the utterance, is quite astonishing;  and the second evocation, about the utterance act, comes down to saying “What else would you expect?”

A second example: When Sieyès was asked what he had done during the Reign of Terror, he replied:

(50)      J’ai vécu.

(50E)   I lived.

If he had not lived, he would not have been around to talk about it; therefore the least that can be said is that Condition I is not directly satisfied. But ‘to live’ implies ‘to not be killed,’ and in shared knowledge is the fact that it was not easy for a member of the National Convention to escape being guillotined during the Terror. Hence, the implicature in (51), which gives a minimum of relevance to (50):

(51)      Sieyès a fait ce qu’il fallait pour ne pas être guillotiné.

(51E)   Sieyès did what was necessary to avoid being guillotined.

But (51) is more relevant than (50), so Condition IV is not met. However, (51) is not yet sufficiently informative to answer the question asked of Sieyès with optimal relevance, as required by Condition II. Everything which would make (51) more relevant to the question asked and less relevant to the answer given must be evoked. This can only be done by suggesting that the question itself was of little relevance and that the answer is self-evident.

The Terror is evoked, then, as a period when all politics, options, factions came down to a simple choice: either die for one’s ideas or conceal them in order to survive. Depending on the extent of his knowledge, the hearer can invoke in greater or lesser detail the manoeuvres, alliances, and betrayals that Sieyès must have engaged in so as to escape the guillotine; the hearer’s preferences will determine whether he sees these as a sign of cowardice or of shrewdness. With this kind of evocation, the implicature in (51) answers the question, which has lost quite a bit of relevance; and the utterance in (50) explicates to a degree what is most relevant in the answer. Conditions II and IV are, to the extent possible, restored.

Nonetheless, the image of the Terror that is evoked in this way is not only not a part of shared knowledge but actually contradicts the common opinion. Under what circumstances would it be immediately present in the mind of speaker and hearer? Only if they had a completely jaded, cynical view of political life. Thus, in the second stage, a connivance in cynicism is evoked not by the content of the utterance but by the act of having uttered it.

A third example:

(52)      Il faut manger pour vivre et non pas vivre pour manger.

(52E)   One must eat to live and not live to eat.

This utterance, a classic example of antimetabole (the repetition of two words or phrases, in reverse sequence), also comprises an antanaclasis (the repetition of a single word in different senses), less noticeable because it is hidden by ellipsis: et non pas, ‘and not,’ is elliptical for et il ne faut pas, ‘and one must not.’ The verb falloir, ‘must,’ has two meanings: material necessity and moral obligation, and both are used here. Moreover, the inverted symmetry suggested by the antimetabole is only superficial. The two propositions composing (52) are ambiguous not only because of the two meanings of falloir but also for syntactic reasons: pour vivre, ‘to live,’ can be the complement of manger, ‘eat,’ or of il faut manger, ‘one must eat,’ and pour manger, ‘to eat,’ can be the complement of vivre, ‘eat,’ or of il ne faut pas vivre, ‘one must not live.’ Therefore each proposition has four meanings, corresponding to those in (53) and (54).

(53)      aa.     [Il est matériellement nécessaire de manger] pour vivre.

ab.    [Il est moralement obligatoire de manger] pour vivre.

ba.    Il est matériellement nécessaire de [manger pour vivre].

bb.    Il est moralement obligatoire de [manger pour vivre].

(53E)   aa.     [It is materially necessary to eat] in order to live.

ab.    [It is morally obligatory to eat] in order to live.

ba.    It is materially necessary to [eat to live].

bb.    It is morally obligatory to [eat to live].

(54)      aa.     [Il est matériellement nécessaire de ne pas vivre] pour manger.

ab.    [Il est moralement obligatoire de ne pas vivre] pour manger.

ba.    Il est matériellement nécessaire de ne pas [vivre pour manger].

bb.    Il est moralement obligatoire de ne pas [vivre pour manger].

(54E)   aa.     [It is materially necessary to not live] in order to eat.

ab.    [It is morally obligatory to not live] in order to eat.

ba.    It is materially necessary to not [live to eat].

bb.    It is morally obligatory to not [live to eat].

Hence there are 16 meanings [4] resulting from the possible combinations of (53) and (54). Intuitively, the meaning of (53aa) + (54bb) is the one selected and, incidentally, this meaning obeys the principle of disambiguation suggested above. For the second proposition, no hesitation is possible, because the three eliminated meanings are completely paradoxical. For the first proposition, however, the typical reaction of an informant will be to choose first (53aa) and then, as if seized with remorse, hesitate and consider the possibility of (53bb). We shall see that, while only (53aa) is asserted by (52), (53bb) is evoked by the symbolic interpretation, which accounts for the informant’s hesitation.

The two propositions of (52) conjoined by et, ‘and,’ are equally focused, although they do not have equal relevance with regards to common shared knowledge. It goes without saying that in order to live, one must eat; the first proposition has only the minimal relevance of a reminder. The second proposition also seems too uninformative, since even among gluttons, few would claim that eating is the purpose of life. But the hearer starts from the premise that the speaker has maximised relevance, and when Condition II is not directly satisfied by a semantic interpretation of the utterance, it can be restored by an implicature. Hence the following implicature arises:

(55)      Manger est le but de la vie pour un gourmand.

(55E)   Eating is the purpose of life for a glutton.

Since gluttony is fairly common, if the implicature in (55) is associated with the proposition (54bb), the latter expresses a strong—even severe—judgement, and the relevance of the second conjunct of utterance (52) is restored. Once this is accomplished, two problems remain to resolve—one optionally and the other obligatorily. First, the difference in relevance of the two equally focused propositions in (52) is increased; this does not directly violate Condition II, which requires only that when one proposition is less focused than another it must also be less relevant, but not the contrary. Nonetheless, the speaker said et, ‘and,’ when he could have said mais, ‘but,’ and it would be preferable to find an interpretation of the first conjunct of (52) that increases its relevance (note that (52) with mais rather than et intuitively has less figural import, and the present analysis accounts for this). Moreover, the implicature in (53) is more relevant (if only because it is questionable) than the proposition that implicates it, and thus Condition IV is violated. It must be restored and this can only be done through symbolic evocation.

Attention is thus shifted from the utterance itself to its implicature, (15), whose over-relevance defines as field of evocation anything in memory or imagination that could make it less paradoxical. Realistic knowledge of the world shows that, between the pure ascetic and the pure hedonist, all gradations are possible: the pleasure of food and the necessity of sustenance are two motives that combine in variable proportions. But, if (55) is to go without saying, this continuum must be reconstructed as a clear opposition: the slightest hint of gluttony is taken as a complete inversion of the means (eating) and the end (living), and the glutton is seen as having crossed an absolute boundary, all the more daunting in that nothing indicates its place. Simultaneously, what lies before this boundary—alimentary virtue which consists in ingesting food only because it is necessary for survival, with no consideration of the tempting pleasures of eating—is also evoked. This complementary evocation solves the minor problem of the first proposition’s weak relevance. This proposition does indeed have the meaning of (53aa) but it evokes the more relevant meaning of (53bb). To accept, albeit reluctantly, the physical necessity expressed by (53aa) is to accept the moral obligation expressed by (53bb).

Here, then, is an utterance that seems trite, weak, and insufficiently relevant, but which is balanced and enriched by symbolic evocation. Moreover—and as always—the knowledge reconstructed from passive memory, the conception evoked by the figural utterance, is itself subject to a second evocation: it is presented as shared knowledge, mobilised or invokable, common knowledge to speaker and hearer; its moral sense, unbeknownst to them, is reshaped by an utterance whose symbolic effectiveness is only enhanced by its apparent conceptual banality and inoffensiveness. Beware of proverbs.

It will be noted that the superficial syntactic inversion of utterance (52), disproven by the syntactic and semantic analysis, is restored by the symbolic evocation. The antimetabole and the antanaclasis appear as catalysts, putting additional focus on a figure  based on the non-correspondence between the relative focus of the two propositions in the conceptual representation, and their relevance with regards to shared knowledge.

These three examples, provided to illustrate how an overly relevant implicature becomes the focal point of a figural interpretation, suggest two other incidental observations.

First, in the three examples, the initial evocation, about the utterance, led to nuances in encyclopaedic knowledge being recast as stark contrasts. While not all symbolic evocations work like this, the similarity is not fortuitous. Lévi-Strauss’s analysis of myths showed how the symbolic organisation of the encyclopaedia creates contrasts out of gradation, draws on differences as much as similarities, highlights distance as much as contiguity. Symbolism is unitary, and one should not be surprised to find these processes in figural interpretation.

But while these three examples may initially seem similar, they differ greatly in tonality. Woody Allen is funny, Sieyès is witty, and the proverb is sententious. These differences, it seems to me, stem less from the evocation of the utterance than from the evocation of the utterance act. The complicity evoked by the act of uttering (48) is entirely imaginary:  there exists no male chauvinist who would actually find it normal for a young wife’s first soufflé to be as heavy as lead; thus the uttering of (48) gives speaker and hearer an image of themselves that they enjoy all the more in recognizing that it is false and that the opprobrium attached to it will not fall upon them. A minor desire receives symbolic, risk-free satisfaction.

When Sieyès says “j’ai vécu,” the connivance in cynicism evoked by the utterance is equivocal—neither entirely realistic nor entirely imaginary. In its imaginary aspect, it flatters the intelligence, containing an element of “you and I know…which so many others, the poor things, didn’t know, and look where they wound up.” In its realistic aspect, it challenges moral vanity: “You who are listening to me, you would not have died for your ideas either.” This is wit, but it is not humour.

As for the proverb in (52), it reawakens in us a censorious voice: “Be careful. Sometimes a virtuous person is merely unaware of his own vices.” “Yes, daddy, I know; yes, my father, I know.” Neither complicity nor connivance, but rather shared submission, is evoked by the act of uttering (52).

3.3.1    I will now turn to gapped figures. Gaps must obey Conditions V and VI. Condition VI arises from logical analysis and not recourse to memory; thus, if it is violated, it cannot be restored by evocation. We will see nonetheless that it plays an indirect part in the figural interpretation of tropes. Condition V does involve memory and, in simple gapped figures, evocation aims to restore this condition alone. I will give only two examples, before turning to tropes, the more complex and more interesting examples of figures.

At a party in Paris in 1975, all the young intellectuals there are a bit bored. Someone  proposes:

(56)      Et si on fumait?

(56E)   Why don’t we smoke?

The utterance in (56) is elliptical: the object of the verb fumer, ‘smoke,’ is absent. In many utterances, the object of fumer is elided and the hearer can easily supply the object: du tabac, ‘tobacco.’ But if this completion is supplied for (56) it results in a non-relevant interpretation. In our society, anyone can smoke as the spirit moves him [remember, the context here is France, in 1975—editor]; but (56) proposes a collective activity. And what is smoked collectively? Marijuana. However, in the milieu described, smoking pot is not the general or regular custom. Therefore, (56) will cause most hearers a fleeting instant of puzzlement: the missing concept is not permanently mobilised, as it would be for genuine potheads, and it must be evoked. The speaker has violated Condition VI by not observing the limits of mobilised shared knowledge. The first evocation is fairly brief and easy but is nonetheless sufficient to trigger the second evocation, not about the utterance this time but about the utterance act. Under what circumstances would the gapped utterance in (56) be interpretable without appealing to evocation? If those present were not occasional but instead regular, daily smokers of cannabis. This second evocation, of the complicity of smokers genuinely addicted to marijuana and whose presumed behaviour the group is about to imitate, passing the joint around for deep tokes, is delightful because it is imaginary.

A second example: two friends are confiding in one another. One sighs deeply and says to the other:

(57)      Ah, Julie! tu sais…Julie…

(57E)   Ah, Julie! You know…Julie…

In fact, the hearer does not know. The speaker is expressing, in gapped form, a feeling or an opinion about Julie—but what exactly? Does he love Julie? Is she causing him heartbreak? Does he think she’s wonderful? Is she not as affectionate as he would like? Only by a lengthy, uncertain evocation can the hearer manage—if  indeed he can manage—to complete the gapped utterance of (57), which grossly violates Condition VI. On the other hand, the second evocation, about the utterance act, is not so difficult: “We understand one another; with just a word or two, you know what I’m feeling….” In this way an utterance that cannot be understood in exact terms creates a marvelous feeling of mutual understanding.

3.3.2    Tropes. There are two major views of tropes: one holds that a trope is a combination of periphrasis and ellipsis (in particular, a metaphor is an elliptical comparison). The second, more widespread, is that a trope is a figure  in which a figurative meaning must be substituted for the literal one. Under the first view, the metaphor of (58a) would receive the figural interpretation of (58b); under the second, that of (58c).

(58)      a.      Léon a épousé un rossignol.

b.      Léon a épousé une femme qui chante comme un rossignol.

c.      Léon a épousé une excellent chanteuse.

(58E)   a.      Leon has married a nightingale.

b.      Leon has married a woman who sings like a nightingale.

c.      Leon has married an excellent singer.

The first view confuses ellipsis and gap: strictly speaking, (58a) contains no ellipsis. It is, moreover, incompatible with Condition V, which would be violated by an interpretation like (58b). Thus in solving a problem of the rhetoric of figures, this interpretation creates another problem of general rhetoric. And what exactly does it explain of the mental processes involved in the figural interpretation of tropes?

The second view does not explain the obvious connection between metaphor and comparison, but instead classifies these two figures in radically opposite categories: one with and one without a change in meaning. The very notion of a change in meaning implies that it is possible that the speaker did not mean what he said, an idea that not only displeased Breton, but also poses almost insurmountable problems to a theory of general rhetoric. Moreover, in order to explain the mental processes of the figural interpretation of tropes, those who argue for the second conception propose that the figurative meaning is justified by the semantic features it shares with the literal meaning. For example, rossignol ‘nightingale’ would have the “seme” bon chanteur, ‘good singer.’ But if this were the case, then (59) should be the same kind of analytical contradiction as (60) is.

(59)      Les rossignols ne chantent pas bien.

(59E)   Nightingales don’t sing well.

(60)      Les rossignols ne sont pas des oiseaux.

(60E)   Nightingales are not birds.

Thus the solution to a problem of the rhetoric of figures raises countless problems of semantics. Finally, the second conception predicts that (61a) has the figural interpretation (61b), which is absurd.

(61)      a.      C’est presque un rossignol que Léon a épousé, tant sa femme chante à ravir.

b.      C’est presque une excellente chanteuse que Léon a épousé, tant sa femme chante à ravir.

(61E)   a.      The woman Leon married is almost a nightingale, so delightfully does she sing.

b.      The woman Leon married is almost an excellent singer, so delightfully does she sing.

Under the interpretation I propose, propositions like (58b) or (58c) are not interpretations but rather implicatures of the interpretation of (58a). This obviates the previous objections.

When the utterance of a complete sentence has no acceptable semantic interpretation, the utterance itself is considered incomplete, gapped, and in most cases, it is modalised. Modalisation weakens the conceptual force of the utterance. But in some cases, the utterance may suggest a proposition that contains all the same terms but in different functions; this proposition, therefore, cannot be treated as a completed interpretation and is thus an implicature. We then have an implicature that is more relevant than the interpretation completed by modalisation, violating Condition III in its two corollaries V and VI. This situation corresponds to tropes, which are figures by virtue of implicatures and gaps.

Consider the metonymy in (62), the synecdoche in (63), and the metaphor in (63), uttered about a man who has married a singer at the Paris Opera.

(62)      Léon a épousé un abonnement gratuit à l’Opéra.

(62E)   Leon has married a free subscription to the Opera.

(63)      Léon a épousé une voix sublime.

(63E)   Leon has married a sublime voice.

(64)      Léon a épousé une fauvette.

(64E)   Leon has married a warbler.

One would be tempted to assign them the following interpretations:

(65)      Léon a épousé une femme qui lui procurera un abonnement gratuit à l’Opéra.

(65E)   Leon has married a woman who will get him a free subscription to the Opera.

(66)      Léon a épousé une femme qui a une voix sublime.

(66E)   Leon has married a woman who has a sublime voice.

(67)      Léon a épousé une femme qui chante comme une fauvette.

(67E)   Leon has married a woman who sings like a warbler.

These interpretations remove the semantic anomalies or the encyclopaedic paradoxes of (62) – (64). They accomplish this easily because they eliminate the logical function—as object of the verb épouser, ‘marry’— of abonnement gratuit à l’Opéra, ‘free subscription to the Opera,’ voix sublime, ‘sublime voice,’ and fauvette, ‘warbler;’ this function is the source of the anomalies and paradoxes. But by the same token (65) – (67) cannot serve as completed interpretations of (65)-(67), since they violate Condition V, and symbolic evocation can do nothing to correct the situation. Therefore, (65) – (67) can only be implicatures of the completed interpretations of (62) – (64). The completed interpretations could be modalisations as in (68) – (70):

(68)      C’est comme si Léon a épousé un abonnement gratuit à l’Opéra.

(68E)   It’s as if Leon has married a free subscription to the Opera.

(69)      C’est comme si Léon a épousé une voix sublime.

(69E)   It’s as if Leon has married a sublime voice.

(70)      C’est comme si Léon a épousé une fauvette.

(70E)   It’s as if Leon has married a warbler.

These modalisations with c’est comme si, ‘it’s as if,’ eliminate the encyclopaedic paradoxes of (62) – (64). If one adopts a semantic theory that considers (62) – (64) to be semantic anomalies rather than paradoxes, then the modalisation could be on pourrait dire que, ‘it could be said that,’ which would eliminate the anomalies. The distinction is of little importance here: the consequences of the two kinds of modalisation are largely identical: if it is as if then it could be said that, and if it could be said that, then it is as if—or better yet, it could be said that it is as if.

With this one reservation, shared knowledge assigns to utterances (62) – (64) the completed interpretations of (68) – (70) or equivalent interpretations, for it is clear that the speaker does not wish to present (62) – (64) as either true or possible but is instead describing an imaginary world and inviting the hearer to compare it to the real world. But mobilised shared knowledge does not determine the scope of this comparison, which is not maximally relevant. If the speaker thinks the imaginary world is comparable to the real world, he should have said in what way this is so, rather than leaving it up to the hearer to guess. Thus, the necessary univocality of the completed interpretation is attained only by violation of Condition II, and Condition VI is not fully met.

But, one might say, the scope of the comparison is made clear by the implicatures of (65) – (67), which can be calculated on the basis of mobilised shared knowledge. This is probably so, but the scope is determined for only one aspect. A complete comparison contains three elements in addition to the comparative: two terms that are compared to one another and a theme of comparison. In the complete comparison in (71), the two terms are la femme de Léon, ‘Leon’s wife,’ and une fauvette, ‘a warbler;’ the theme is chanter, ‘sing.’

(71)      La femme de Léon chante comme une fauvette.

(71E)   Leon’s wife sings like a warbler.

In the comparisons (68) – (70), only the second term—the imaginary world of (62) – (64)—is made explicit; the first term and the theme are implicit. The implicatures in (65) – (67) merely make explicit the first term of the comparisons: that particular aspect of the real world to which the imaginary world is to be compared. The implicatures do not say in what way the two worlds are comparable, and the theme, as it is neither present in the utterance or invokable, has yet to be evoked. Moreover, the implicatures (65) – (67), descriptions of the real world, are more relevant in terms of mobilised shared knowledge that the uncertain comparisons in (68) – (67) which implicate them, and thus Condition IV is violated.

The task of symbolic evocation is to discover additional implicatures that could not be calculated on the basis of mobilised shared knowledge and, since no invokable theme is available, can provide the evokable themes of (68) – (70) and maximise their scope, thereby restoring Conditions IV and VI. (68) – (70) would then have maximal relevance, which restores Condition VI, and they would be more relevant than the implicatures (68) – (70), restoring Condition IV.

In what way can the real-world fact that Léon has married a woman who will get him a free subscription to the Opera be compared to the imaginary “fact” described by saying that “he has married a free subscription to the Opera”? The only way is to imagine that the subscription is the sole consequence of the marriage. But even if Léon desired or obtained no more than that, common knowledge tells us that a marriage always has other, less paltry, consequences because it creates, if not in the eyes of the spouses at least in the eyes of society and the law, a contractual bond, permanent in principle, sometimes sanctioned by religion, which forbids any other similar bond—quite different from the bond between a subscription to the Opera and its subscriber. To justify the metonymy of (62), one has to imagine an authority for whom all the significant and necessary consequences of marriage do not count, one for whom the only important consequence is the trivial contingency of obtaining a free subscription to the Opera. In this regard (62) is equivocal: the authority in question could be Léon, if he neither sought nor found anything else in his marriage; it could be his wife, if she will never offer him anything more than the subscription. Or it could be the speaker alone, if he wishes to suggest that the spouses unknowingly share a marriage which will never have other consequences. The equivocation can be eliminated by the utterances (72) – (74):

(72)      Léon a épousé un abonnement gratuit à l’Opéra; c’est tout ce qu’il attend de sa femme.

(72E)   Leon has married a free subscription to the Opera; that’s all he expects from his wife.

(73)      Léon a épousé un abonnement gratuit à l’Opéra; sa femme est bien décidée à ne rien lui accorder de plus.

(73E)   Leon has married a free subscription to the Opera; his wife is determined to give him nothing more.

(74)      Léon a épousé un abonnement gratuit à l’Opéra; quoiqu’aujourd’hui il s’aiment, il est nonchalant, elle est volage, et bientôt il ne s’apercevront plus que lui, de la salle, et elle, de la scène.

(74E)   Leon has married a free subscription to the Opera. They’re in love now, but he  can’t commit and she’s fickle. Sooner or later, the only time they’ll see each other is when she’s on stage and he’s in the audience.

Whether (62) is interpreted as (72), (73), or (74), the image evoked by the utterance is the image of an image:  the image of the marriage held by the speaker, by Léon, or by his wife—the reduction of a marriage to a contingent consequence, the subscription—and thus the elimination (not real but imaginary) of all the other necessary or probable effects of the matrimonial bond, to the extent that it is as if, or it could be said that Leon has married a free subscription to the Opera.

This image, which was not that of the hearer but that of the speaker, who may or may not have derived it from one of the spouses, is evoked by the utterance act as a shared image. This is a second-stage evocation and thus evokes not knowledge but a shared imaginary world that short-circuits knowledge and identifies cause with effect. Moreover, in this particular case of metonymy, it is a cynical imaginary world which, if the speaker intends it to be understood as deriving from Léon or his wife, make them either fascinating or despicable, depending on the shared moral knowledge of speaker and hearer; if it derives from the speaker alone, it makes Léon and his wife ridiculous.

In what way can the real-world fact of marrying a woman with a sublime voice be compared to the imaginary fact depicted by saying “Leon has married a sublime voice”? The relationship between the part and the whole must be conceived of as a relationship of identity. Knowledge does not allow acceptance of this identity, but one could imagine someone imagining it and for that someone, all other characteristics fade, leaving only the voice—just as in Alice’s dream, all of the Cheshire cat disappeared, leaving only its smile. This evoked image is the speaker’s and possibly, but not necessarily, Léon’s. For example, in (75), it is definitely not Léon’s image.

(75)      Léon a épousé cette voix sublime pour son argent.

(75E)   Leon married that sublime voice for her money.

Inversely, the image can be that of Léon alone, without being adopted by the speaker, in which case (63) directly implicates not (66) but (76).

(76)      Léon a épousé une femme pour sa voix sublime.

(76E)   Leon has married a woman for her sublime voice.

In this case, (63) is not a synecdoche but rather a metonymy of the end for the means and receives an account similar to that for (62) above.

When (63) is a genuine synecdoche, the utterance act evokes in the second stage not knowledge nor even an imaginary world, but rather shared symbolism, because the image evoked in the first stage by the utterance is not only not real but cannot even be imagined as real. Symbolic thought weaves onto the encyclopaedia a network of points, places their common background in the shadow, turning a cat into a smile and a woman into a voice. The utterance act—sometimes a metonymy, sometimes a synecdoche, but always a metaphor—evokes the shared nature of this symbolic thought.

Typically, a metaphor, as in (64), poses an additional problem: unlike (62) and (63), it has in (67) an implicature that is itself figural. For what is it exactly to sing like a warbler? What shared knowledge has to tell us on this subject has little to do with the way an opera singer sings. At the very most, we can say that both a warbler and an opera singer are thought to sing well. If someone had uttered (67) only to say that Léon had married a woman who sings well, he would be violating Condition II by giving excess information that did not contribute to relevance. The hearer, on the principle that Condition II can be restored, must therefore evoke other points of similarity between the woman’s singing and the warbler’s: high notes, trills, scales, solos, an impression of both delicacy and virtuosity.

This, then, is the scope of the comparison in (67); it is implicated by (70), the completed interpretation of (64), and established by evocation. But the metaphor takes evocation one step further. In fact, it is not so much a question of conceiving how Léon’s wife’s singing is comparable a warbler’s but of conceiving, on the basis of this initial evocation, how his marriage with this woman is comparable to a marriage with a warbler. While the comparison underlying any metaphor bears on two objects in the real world, the metaphor itself is modalised into a comparison between this known world and an imaginary world. The theme that sufficed for one does not suffice for the other, and not only must the points of resemblance be increased but the points of dissimilarity must be erased, which is never required in simple comparisons. For example, the woman is fragile and ethereal as a warbler. If the hearer knows little about warblers, he may take guidance from their name in French: fauvette seems to be a diminutive of fauve, thus a wild being but not a ferocious one, feline but not dangerous, with the colour but not the smell of a fauve, a tawny wild beast. Unlike the comparison of (67), the metaphor in (64) would hardly be appropriate if Léon’s wife, though she might sing marvelously, were an obese layabout. Next, everything that is not comparable must be erased: for example, that Léon’s wife does not have a beak and does not lay eggs, that a warbler does not have hands and is not paid to sing. And finally, in the fleeting seconds of the evocation, one must forget that species are endogamous. Then, yes, it is as if Léon had married a warbler.

Here again, in this necessarily solitary evocation to which the hearer has abandoned himself, too quickly to even become aware of it—this evocation founded on reminiscences, guided by his desire and merely triggered and given focus by the speaker, is presented by a second evocation as a path taken by both of them, like a dream dreamt by both. The more unusual the metaphor, the deeper and more individual the evocation and the greater the feeling of communion in symbolism. Say that Léon’s wife is a nightingale? The metaphor is banal, the shared symbolism evoked by the utterance act is indeed shared but not very symbolic, and the sentiment of communion is derisory. But say that she is a warbler, and then something has happened between speaker and hearer.

With ever greater subtility, the classical rhetoricians identified increasingly diversified figures, which they classified and then reclassified. The best among them, such as William Empson, even explicated, to the extent possible, what the intuitions of speaker and hearer might be. But to certain questions—When does an utterance take on a figural value? How is a figural utterance interpreted? —no better answer has been proposed but a theory of departure. Figural speech departs from…what exactly? Grammatical speech? But figures are frequent in the most clearly grammatical utterances. Ordinary speech? It also teems with figures. Perhaps from the “degree zero” of discourse, as found in the instructions for serving canned food and which is only defined, tautologically, by the absence of figures.

I have attempted to show that if there is a difference, it is not between different types of discourse but between different levels of conceptual representation. The figure is not in the text and is not a function of the text alone. It resides in the conceptual representation of the text and is a function of both the text and shared knowledge. Rhetoricians may debate whether, alongside phonological, syntactic, and semantic figures, there also exist figures of thought. I have tried to suggest that there are only figures of thought, for which phonological, syntactic, and semantic properties may play the role of additional focalisers, neither sufficient nor necessary, that trigger the mechanism of figural interpretation.

I have tried to put forward fairly specific predictions (fewer than I would have wished but more than is usually the case in rhetoric) concerning the conditions under which an utterance will take on a figural value and the way the figure will be interpreted. Unlike taxonomic rhetorics, cognitive rhetoric, whose rudiments I have proposed here, makes predictions; because it does so, it runs the risk of being refuted by facts. But if one does not run this risk, one can talk and talk and still wind up saying nothing.


* The original article appeared in 1975 as “Rudiments de rhétorique cognitive,” Poétique: Revue de Théorie et d’Analyse Littéraire (23) 389-415, under the joint editorship of Tzvetan Todorov and Gérard Genette. The academic dialect of this paper, by the way, might require some acclimatization by 21st century readers of RSQ. It is the dialect of generative linguistics, in a tone that, as translator Sarah Cummins phrases it, “is so Paris 1975” (witness, for instance,  the discussion of sentence 56 in terms of implicatures related to tobacco and marijuana use).

[1] No bibliographic references are given in the text; therefore I must first of all acknowledge my indebtedness. I have gratefully undergone the influence of N. Chomsky, directly and via other linguists and philosophers in the Chomskyan vein—in particular R. Jackendoff, J. J. Katz, Nicolas Ruwet, and Deirdre Wilson. A few problems were suggested to me by the work of philosophers of language O. Ducrot and J. Searle. The concept of implicatures proposed here was inspired entirely by the unpublished lectures of H. P. Grice, Logic and Conversation (1968). [note: Grice’s lectures were published in 1975, in Speech Acts, Syntax and Semantics vol. 3, edited by P. Cole and J. Morgan, New York: Academic Press. —editor] I will name no rhetoriticians here, for it was precisely to free myself from their influence that I undertook this work. I was prodded and assisted in this enterprise by the friendly provocation of Tzvetan Todorov,  but any blame must fall on me alone.

[2] Sentences and utterances are shown in Roman type, while meanings and propositions are in italics. The two types of representations must not be confused.

* Sperber and Cummins have upheld in this translation a French distinction between figural and figurative—the former relating to figures generally (i.e., roughly the same as the English figurative), the latter relating more narrowly to figures of thought only (i.e., to tropes).

[3] Thus an utterance relating a catastrophe or a passion may be evocative without the utterance act being so. Inversely, when social or professional jargons are used, the utterance act be evocative without the utterance being so. Only in figural speech are both types of evocation necessary.

[4] Under one possible semantic analysis, the verb falloir is not ambiguous but merely vague. In this case there would only be four meanings; but there would still be 16 interpretations, the only point at issue here. Note also that if the second occurrence of falloir was not elided and thereby defocused,  it could fall under the scope of negation and the number of interpretations would be doubled.