This book "supplies a vocabulary of English words and idiomatic phrases 'arranged … according to the ideas which they express'. The thesaurus, continually expanded and updated, has always remained in print, but this reissued first edition shows the impressive breadth of Roget's own knowledge and interests."
Francois Recanati, (2000) Oratio Obliqua, Oratio Recta: An Essay on Metarepresentation, MIT Press, Cambridge, Massachusetts & London, England. xviii + 360p, paperback US$24.95.
Reviewed by: Anne Bezuidenhout, Department of Philosophy, University of South Carolina, Columbia, SC 29208, U.S.A.
1: Introductory Remarks
A metarepresentation is a representation of another representation. Call what is represented by a metarepresentation an object-representation. This object- representation could be either a linguistic representation, such as a sentence, or a mental representation, such as a belief or a desire. It could also be a painting, a film, a sculpture, a tapestry, a Morse code signal or some other such thing. An object-representation represents a fact or state-of-affairs in some possible world, and a metarepresentation represents the fact that an object- representation represents such a worldly fact or state-of- affairs. Metarepresentations themselves can be either linguistic, mental, pictorial or whatever. One can believe or say something about a belief or a picture, a picture can depict another picture, and so on.
The notion of metarepresentation currently plays a large role in cognitive science. For instance, many cognitive psychologists and philosophers of mind subscribe to the view that humans are endowed with a special metarepresentational faculty. This faculty is what enables people to ascribe representational mental states (e.g., beliefs, wants, fears, hopes etc.) to others, and hence to explain and predict the behavior of others. Many linguists and philosophers of language hold that metarepresentational capacities play a crucial role in verbal communication and particularly in the pragmatic processes involved in utterance interpretation. In developmental psychology there is a great deal of current research on the question as to when children acquire the capacity for metarepresentation. Some researchers in neuroscience and psychopathology hold that autistic individuals have a damaged capacity for metarepresentation. Finally, there is an on-going debate in cognitive science as to whether metarepresentational capacities are a matter of being able to simulate what others are thinking, feeling, wanting etc., or whether it is a matter of having internalized a "theory" about what sorts of mental causes are likely to produce what sorts of behavioral effects.
I agree with Recanati when he says in his preface that "not much theoretical progress can be made in this area until we know more about the very structure of metarepresentations" (p. xii). What cognitive science badly needs is a careful philosophical account of the notion of metarepresentation, but one that clearly has some relevance to the many empirical debates that presuppose the concept.
There have been seemingly endless philosophical discussions of one special class of metarepresentations, namely propositional attitude ascriptions such as 'John believes that he is being persecuted'. However, a researcher from another field who ventures into this territory is likely to find it pretty inhospitable. Although there are definite philosophical camps, each advocating some general sort of approach to the problem, it is hard to find any real consensus, and it sometimes seems that every new paper on the topic has a different account to offer. There is also a lot of terminological confusion even with regards to some of the core distinctions in the debate, such as the de re/de dicto, relational/notional, and transparent/opaque distinctions. These are often conflated with one another. Moreover, it is unclear whether all parties to the debate are using these distinctions in the same sense.
Philosophical discussions also tend to revolve around a few hackneyed examples, such as 'Ralph believes that Ortcutt is a spy', 'Ralph believes that Smith's murderer is insane', 'Lois Lane believes that Superman can fly'. (This last example is usually treated as though it was a belief ascription to a real person, when in fact it is a meta- meta-representation, for strictly speaking it is only in the Superman comic series that Lois Lane believes that Superman can fly). Hence it is unclear whether and to what degree current accounts can be generalized to other sorts of metarepresentations. Perhaps theories of propositional attitude reports can be readily extended to account for indirect speech reports, such as 'Ralph said that Ortcutt is a spy'. But rarely are there discussions of other sorts of metarepresentations, such as 'In the film, a 3rd World War is declared', 'In Monet's painting, the water is lilac- colored', 'According to the Bible, God created the universe in six days'. Moreover, all the examples given so far are linguistic metarepresentations. One would want an account that generalizes to other sorts of metarepresentations too, such as mental and pictorial ones.
Recanati's book provides an excellent and sure-footed guide through this philosophical terrain. For a cognitive scientist interested in philosophical explorations of the notion of metarepresentation, Recanati's book would be a good place to start. Recanati's aim is to set out his own highly original account of metarepresentation. But along the way he gives very clear sketches of some of the alternative accounts on offer. He also does an excellent job at clarifying some of the important terminology in this area. He discusses a range of different types of metarepresentations, and it is clear that he is aiming for an account that can be generalized. But most importantly, his account seems tailor-made to fit into current cognitive science debates about metarepresentation, because it sees a connection between metarepresentation and simulation. Also, at several points (p.57, pp.79-84) Recanati makes interesting connections to discussions in psychology about the development of children's metarepresentational capacities.
Recanati's book is divided into six parts. In section 2 I lay out Recanati's arguments in some detail. For those not interested in the details, it is possible to skip to section 3, in which I give a much briefer summary of Recanati's book. In section 4 I give a brief critical assessment of the book.
2. Detailed summary of book
In Part I, Recanati lays out and discusses some of the principles that should constrain any account of metarepresentation. From Recanati's perspective, one of the most important constraints is that our account should preserve "semantic innocence". In his famous paper 'On sense and reference', Frege argued that the sentences embedded in the complement-clauses of belief and speech reports do not have their customary references (namely, truth-values). Instead these embedded sentences refer to propositions (their customary senses). In other words, Frege accepted a thesis of semantic deviance. Words inside attitude contexts behave differently from words outside such contexts. Davidson (1984:108) challenged philosophers to return to a pre-Fregean semantic innocence and recognize that it is "plainly incredible that the words 'The Earth moves', uttered after the words 'Galileo said that', mean anything different, or refer to anything else, than is their wont when they come in different environments."
Another constraint that Recanati argues for is one that he calls the Principle of Iconicity (p.10). Syntactically, metarepresentations (at least of the linguistic variety) are sentences that contain other sentences. Thus 'John believes that he is being persecuted' contains the sentence 'He is being persecuted'. One can think of such metarepresentations as consisting of a metarepresentational prefix 'd' and an object-representation 'S'. The Principle of Iconicity says:
"Attitude reports and other metarepresentations contain the object-representation not only syntactically (in the sense that dS contains S), but also semantically: the proposition Q expressed by dS 'contains' as a part the proposition P expressed by S - and that's why one cannot entertain Q without entertaining P."
Many contemporary accounts of attitude reports treat them as expressing a relation between a person and a proposition. On this view, the that-clause of an attitude report refers to a proposition, and the report as a whole says that the ascribee stands in some relation (e.g., the relation of believing) to that proposition. In Part I, Recanati argues that there can be no semantically innocent analysis of this general sort. This is an important argument, as recently Crimmins & Perry (1989), Pietroski (1996) and others have claimed that one can treat that- clauses as referring terms without loss of semantic innocence. Their claim is that the sentence embedded in the that-clause still has its normal semantic function - i.e. it expresses a proposition and refers to a truth-value. It is the complete that-clause, not the embedded sentence, which refers to a proposition. In response to this proposal Recanati (p. 18) says:
"if, in order to protect innocence, we draw a sharp distinction between the embedded sentence (which expresses a proposition) and the 'that'-clause (which names it), we run the risk of making the former disappear from the logical scene. For the relevant semantic unit is the complete 'that'-clause. At the level of logical form the sentence 'John believes that S' has the form aRb - it consists of a two-place predicate and two singular terms. The embedded sentence plays a role only via the 'that'- clause in which it occurs. What role? Arguably a pre- semantic role analogous to that of the demonstration which accompanies a demonstrative... If that is right, then semantically the complexity of the 'that'-clause matters no more that the pragmatic complexity of a demonstrative-cum- demonstration or the pictorial complexity of a quotation."
Recanati argues that the best way to respect the Principle of Iconicity is to treat 'John believes that' and other such metarepresentational prefixes as sentence forming operators that attach to a sentence S to form a compound sentence. Then it is clear that the object-representation S is both syntactically and semantically a part of the metarepresentation.
One might think that it is possible to respect Iconicity while maintaining the standard view that that-clauses refer to propositions. The claim would be that this is possible if one treats that-clauses as directly referring expressions. Since the content of a directly referring expression just is its referent, what the that-clause contributes to the content of the attitude report is simply the proposition it refers to, namely the proposition expressed by the content sentence. Thus it looks as though one can hold that the proposition expressed by the content sentence will be semantically a part of the proposition expressed by the attitude report, just as Iconicity requires.
Recanati argues that this strategy will not work, as it requires us to hold that the content of a term (namely, the that-clause) is the same as the content of a sentence (namely, the embedded content sentence). But terms and sentences belong to different semantic categories. According to what Recanati calls the Heterogeneity Principle (p.19), if two expressions are of different types, their contents should be of different types also. The direct reference approach violates this principle.
Recanati also considers the paratactic account offered by Davidson (1984). This account appears to be both semantically innocent and to respect Iconicity. However, Recanati rejects it on the grounds that it violates the Grammatical Constraint (p. 28), according to which we should minimize discrepancies between syntactic and semantic structure. Syntactically attitude reports of the form 'John believes that S' are single sentences, but on Davidson's semantic account they actually consist of two sentences, each with its own semantic contribution. As Recanati (p.27) puts it: "The embedded sentence S expresses the proposition P; the prefixed sentence 'John believes that' expresses the proposition Q which does not contain P as a part, but refers to P (via the demonstrative 'that') and says that John believes it."
In the remainder of his book, Recanati develops an account of metarepresentation that satisfies the various constraints he has laid down in Part I.
In part II he considers two kinds of theories that satisfy the Principle of Iconicity, which he calls modal theories and simulation theories. Modal theories regard metarepresentations of the form 'dS' as consisting of an object-representation 'S' and a circumstance-shifting prefix 'd'. According to such accounts, an attitude report of the form 'John believes that p' is true in a circumstance c iff p holds in all circumstances c' compatible with what John believes in c. Similarly, 'In the film, a 3rd World War is declared' is true in circumstance c iff a 3rd World War is declared in the circumstance c' depicted in the film.
Modal accounts regard the content of the object- representation as being a proper part of the content of the metarepresentation. Simulation theories on the other hand regard the content of the object-representation and the content of the metarepresentation as coinciding. The metarepresentational prefix contributes no content of its own. It simply signals that the content of the object- representation is not being asserted. The speaker is merely pretending. On this view, when we say that John believes that S, we simulate John's mental state and assert S within this simulation or pretense. Metarepresentations are seen as pretend assertions.
There follows a very interesting discussion of conditionals and the sense in which they can be said to involve simulation. The conclusion is that both with respect to conditionals and metarepresentations a pure simulation account is too simplistic. We need to distinguish between metarepresentations and pretend assertions. Metarepresentations do involve simulation in some sense. We see things from the point of view of the believer or the film or whatever. But metarepresentations are also serious assertions. They require us to step outside the pretense and to assert something about what holds in the pretense.
In Part II Recanati also presents his 'Austinian semantics' for metarepresentations. This semantic machinery relies on the notion of a situation and the facts supported by that situation in a particular world. What Recanati calls a fact is a structured entity. Facts are essentially what others call Russellian propositions. Utterances state or are about facts, but they also concern situations. For example, suppose Recanati utters 'It is raining' in Paris at some time t. This states the fact that it is raining at t, and it concerns the situation in Paris. This can be represented by the Austinian proposition:
(1)[Paris] |= @ << At t, it is raining >>
This says that the situation in Paris in the actual world @ supports the fact that it is raining at time t. Now if Recanati were to reflect on his situation and make it explicit, he might utter the sentence 'It is raining in Paris'. He is now stating a fact that concerns a wider situation, in the sense that implicitly he is contrasting what is happening in Paris with what is happening elsewhere, say in Europe. This new fact and its different, wider situation might be represented by the following Austinian proposition:
(2)[Europe] |= @ << Paris |= @ << At t, it is raining >> >>
The cognitive process of reflection is the converse of a process of projection. This latter process is involved in cases where a situation is first mentioned, and then the speaker projects herself into the situation, and states something with respect to that situation. An example would be 'Berkeley is a nice place. There are bookstores and coffee shops everywhere.' The following pair of Austinian propositions give the analysis of these sentences:
(3) [USA] |= @ << Berkeley is a nice place >>
(4) [Berkeley] |= @ << There are bookstores and coffee shops everywhere >>
This distinction between reflection and projection can also be used to distinguish between metarepresentations and pretend assertions. Pretend assertions are cases in which a speaker projects herself into someone else's situation and makes assertions under that pretense. Cases of free indirect speech provide examples of this phenomenon. For example: 'John is totally paranoid. Everybody spies on him or wants to kill him, including his own mother.' This would be analyzed as follows, where S and S' are situations, @ is the actual world, and w is John's paranoid world:
(5) [S] |= @ << John is paranoid >>
(6) [S'] |= w << Everybody spies on John etc. >>
Metarepresentations on the other hand involve taking a reflective stance on some imaginary situation. For example: 'John believes that he is being persecuted.' This could be analyzed as follows:
(7) [S]|= @ << S' |= w << John is being persecuted >> >>
Metarepresentations depend on the dual nature of situations. Situations can be viewed both from without as objects and from within as supporting certain facts. In metarepresentations we view situations in both ways simultaneously. Although metarepresentations are not pretend assertions they do involve an element of pretense. As Recanati says: "Even though in metarepresentational talk we take a reflective stance toward the imaginary situation we are mentioning, still the content of the simulation is displayed, and to that extent metarepresentations involve simulation." (p.87)
Actually, Recanati's analysis of metarepresentations is more complex than I have indicated. He introduces the notion of a delta-structure (d-structure for short). In such structures we simultaneously view a situation from the outside and from within. Recanati then distinguishes between homogeneous d-structures (such as (2) above) and heterogeneous d-structures (such as (7) above). Heterogeneous d-structures involve a world shift, whereas homogeneous ones do not. It turns out that both conditionals and metarepresentations involve a world shift and hence both are heterogeneous d-structures. Recanati considers various ways in which one might modify his initial analysis in order to distinguish conditionals from metarepresentations. His reasoning is subtle and rather intricate, and so I will not summarize it here. Several other considerations are in play as well, such as the need to provide an analysis that distinguishes 'In the book, there is a 3 page chapter' and 'In the book, there is a winged horse'. Despite their superficial grammatical similarity, the latter, but not the former, is a metarepresentation.
In Part III Recanati aims to give an account of the sort of opacity that is peculiar to metarepresentations, but in a way that respects Iconicity and is semantically innocent - a seemingly tall order! Recanati notes that the fact that extensionally equivalent expressions fail to be substitutable in attitudinals and other metarepresentations does not establish the opacity of metarepresentations. Such failures of substitutivity occur whenever there is a circumstance-shifting prefix, even in cases such as 'Some years ago, the president of the U.S.A. was an actor'. To be freely substitutable in the embedded sentence, the expressions must have the same extension in the shifted circumstance introduced by the circumstance-shifting prefix 'Some time ago'. However, metarepresentations also exhibit intensional substitution failures. Terms that have the same intension/content may fail to be substitutable in attitudinals such as 'According to John, lawyers are crooks'. John may not realize that an attorney is a lawyer (in American English), and so he may fail to believe that attorneys are crooks. This says Recanati, is a hint that some sort of mention takes place in attitudinals.
Before exploring the use/mention issue, Recanati tries to bring some order to the terminological confusion surrounding discussion of the relational/notional and the transparent/opaque distinctions. These are often conflated and said to be instances of the so-called de re/de dicto ambiguity of sentences of the form 'John believes that S'. However, Recanati argues that this is a mistake. The relational/notional distinction should be drawn in terms of the type of belief that is ascribed, namely whether a singular or a general belief is being ascribed. Recanati assumes that when the embedded sentence S contains a genuine singular term, the only possible reading is a relational one, since the ascribed belief must be a singular one. However, when S contains a quantified or descriptive term, the ascription will be ambiguous between a relational and a notional reading. When the description is understood to take wide scope over the metarepresentational prefix, then we have a relational reading, because the ascribed belief will be understood to be singular. On the other hand, when the description has narrow scope and falls with the scope of the metarepresentational prefix, we will have a notional reading, because the ascribed belief will be understood to be general.
The transparent/opaque distinction is orthogonal to the relational/notional one. When the embedded sentence of a belief ascription contains a singular term the ascription must be understood relationally, but even such an ascription can be given either a transparent or an opaque reading. On the opaque reading the singular term has a dual role. It serves not only to refer to the object the ascribed belief is about, but it also indicates how the believer thinks about that object. Opaquely understood a belief ascription can be thought of as saying that the believer so-believes that S. This is reminiscent of Quine's famous example 'Giorgione was so-called because of his size.' So Recanati embarks on a discussion of this example.
Many people following Quine would conflate the notion of an occurrence of a singular term being referential with its being transparent, and both in turn with its occupying a position that is open to substitution and that can be quantified into. But Recanati skillfully argues that all these notions must be kept distinct. The occurrence of a singular term t in an embedded sentence S is purely referential if its narrow semantic contribution to the truth-conditions of S is simply its referent. Such an occurrence will be "innocent". But a term t may make a wider semantic contribution to the truth-conditions of S if the semantic contribution of some other expression in S depends on the identity of t. In such a case the occurrence of the term will not be transparent. Hence transparency entails pure referentially, but not conversely. This is what happens in Quine's example. The occurrence of 'Giorgione' is purely referential, since its narrow semantic contribution is simply its referent. However, it is not transparent, since the semantic contribution of 'so- called' depends on the identity of the term. If one substitutes the term 'Barbarelli' one alters the truth- conditions and hence (potentially) the truth-value of the sentence.
Recanati introduces the concept of a reflecting context, namely a linguistic context containing an expression whose semantic value depends on the identity of some other term in the context. Quine's 'Giorgione' sentence is a reflecting context. So too claims Recanati are belief ascriptions opaquely understood, as they have the form 'a so-believes that S'. On the other hand, when transparently read, belief sentences are not reflecting contexts. Such a transparent reading Recanati calls the minimal reading, whereas the opaque reading is an enriched one. It is a "cumulative" reading (p.133), because the embedded sentence expresses a proposition, but in addition says something about the believer's way of believing.
One may think that when a belief ascription is read transparently, a singular term in the embedded sentence will be substitutable. That is, it can be replaced by a co- referring expression without altering the truth-conditions of the belief sentence. However, Recanati argues that transparency does not entail substitutability. He says: "in the same way in which a purely referential occurrence may not be transparent if it occurs in a reflecting context, a transparent occurrence may not be substitutable if it occurs in an unstable context." (p.150) An unstable context is one that might be turned into a reflecting one by the substitution. Belief sentences are like this, Recanati claims. Whether transparently read or not, a singular term is never substitutable into the embedded portion of a belief report.
Finally, Recanati argues contra Quine that it is always possible to quantify into the position occupied by a singular term in the embedded portion of a belief report. This is because he thinks that any opacity is filtered out in the process of existential generalization. Thus it is always legitimate to infer from 'John believes that a is F' to 'There is something that John believes to be F'. In the process of generalizing one filters out all readings but the minimal (transparent) one.
In Part IV Recanati worries that by relying on a quotational paradigm (the 'Giogione' example) for understanding the opaque readings of metarepresentations he may have violated semantic innocence. Quotation can be thought of as a context-shifting device. The terms inside the quotes are to be interpreted with respect to a shifted context. For instance, in direct speech reports such as:
(8) John said 'I am here now.'
the indexicals are not to be interpreted with respect to the speaker's context (the external context) but with respect to John's context when he uttered the quoted words (the internal context). Do metarepresentations shift context? Recanati says that semantic innocence can be maintained only if they do not, so in the remainder to the book he is concerned to explore this issue. His final conclusion is going to be that metarepresentational prefixes are semantically innocent. All they do is shift the circumstance for evaluating the internal sentence (the object-representation). However, metarepresentations do provide an environment for the operation of certain pragmatic processes that give rise to semantic deviance. The primary cause of such deviance will be attributed to a pragmatic process of deference, which is extensively discussed in Part VI.
Part IV begins to set the stage for this discussion of deference and deviance by discussing direct speech reports, such as:
(9) John said 'The slithy toves did gyre and gimble in the wabe.'
He also discusses some varieties of semi-quotation. Recanati is particularly interested in cases of what he calls mixed quotation, such as:
(10) Quine said that quotation 'has a certain anomalous feature.'
In (10) the quoted material plays a dual role. It is not simply being displayed, as in (9), but is also in active use. Mixed quotation is a sub-class of what Sperber & Wilson (1986) have called echoic uses. These are cases in which one uses a word while at the same time implicitly ascribing its use to another person or group of persons. Cases of mixed quotation are echoic uses occurring in the sentential complement of an indirect speech report or attitude report.
Recanati distinguishes between cumulative and non- cumulative instances of mixed quotation. Cumulative cases entail the sentence with the quotation marks removed, but not so with non-cumulative uses such as:
(11) Mr. Greenspan said he agreed with Labor secretary R. B. Reich 'on quite a lot of things.' Their accord on this issue, he said, has proved 'quite a surprise to both of us.'
In non-cumulative cases it appears that there is a context- shift. Not all the quoted material can be interpreted relative to the speaker's context, but must be interpreted relative to the 'internal' context. Thus the final personal pronoun 'us' in (11) refers to Greenspan and Reich, and not to a group of which the reporter is a member. To retain its intended referent when unquoted, the 'us' would have to be transposed to 'them'.
If such context-shifting can occur in metarepresentations, then we potentially have a threat to semantic innocence.
In Part V Recanati discusses metafictional statements such as:
(12) My children think that Santa Claus can't come until we're all asleep.
Here there does seem to be a context-shift. The speaker of (12) doesn't believe that 'Santa Claus' refers to anything. He is taking up his children's perspective and pretending that there is such a person. But Recanati doesn't want to say that (12) is a pretend assertion, for then it couldn't be true, whereas (12) clearly could be. Recanati considers the idea that pretense can sometimes be used for serious purposes. To account for attributions such as (12) Recanati invokes Walton's (1990) idea of Meinongian pretense. The children in thinking about Santa Claus are entertaining a pseudo-singular proposition, since there is no such individual. In order to ascribe such a pseudo-singular belief the ascriber exploits the Meinongian pretense and makes as if there were such an individual. Since it is clear that this pretense is a pretense, this fictive ascription amounts to the factive ascription of a pseudo- singular belief to the children. It is only possible to directly ascribe either a singular or a general belief to a subject. So the only way to ascribe a pseudo-singular belief is indirectly, by adopting the Meinongian pretense.
This account of metafictional statements treats the pretense as global. The pretense does not occur within the scope of an attitudinal. Rather, the attitudinal is within the scope of the pretense, so that we don't have to see the metarepresentational prefix as inducing the context-shift, and hence as semantically deviant.
We've already seen instances of context-shifting in cases of mixed quotation, such as the Greenspan example above. Uses of untransposed indexicals are also common in free indirect speech, such as in:
(13) After a while she gave her response. Tomorrow she would meet me with pleasure; but she was too busy now.
At least on one possible reading, the orientational indexicals 'tomorrow' and 'now' are untransposed. They are to be interpreted with respect to a shifted context, namely the context in which the woman whose response is being reported first gave her response. They are not to be interpreted with respect to the reporter's current context. Although such untransposed indexicals seem quite common in free indirect speech, they appear to be less common in indirect speech and attitude reports, although Recanati does give the following example from E. Szuyska's Textes Choisis (I give an English translation):
(14) I knew that today was the day.
It is clear from the context that 'today' is not to be interpreted with respect to the context of utterance (the 'external' context), but with respect to the context of the remembered event that is being narrated.
Recanati argues that although metarepresentational prefixes signal a world shift, the context shifting that goes on in the sorts of examples given above is an additional feature, which extends the ongoing pretense signaled by the world shift. Thus although the context shifting introduces an element of semantic deviance, it is not attributable to the metarepresentational prefix per se.
Part V also contains a discussion of cases of partial pretense, such as:
(15) 'Quine' hasn't finished his paper yet.
This is supposed to be a case in which it is clear from the context that 'Quine' is not being used in its normal sense, but is being used mockingly, to refer to a man McPherson who James (the person being mocked) mistakenly thinks is Quine. Recanati posits the idea of a context shifting function, which is an unarticulated constituent of the utterance of (15). It maps the character of the name 'Quine' onto a distinct character PERSON NAMED 'QUINE' BY JAMES. What this distinct character picks out in the actual context is just what the normal character would pick out inthe shifted context.
This sort of context shifting function looks as though it won't work for cases such as:
(16) My 3-year-old thinks I'm a 'philtosopher'.
Here the echoed word does not have a normal character. So it cannot be mapped into a distinct character that picks out in the actual context what is picked out by the normal character in the shifted context.
In Part VI Recanati proposes a way of dealing with such cases, which he calls strongly deviant non-cumulative echoes. (Cases such as (15) are weakly deviant non- cumulative echoes). The strongly deviant cases can be handled by appeal to a deferential operator. Ultimately though Recanati wants a unified account of all cases of mixed quotation (i.e., of echoic uses inside the object- representation of a metarepresentation). He finally offers an account of all such echoes (cumulative and non- cumulative, strongly deviant and weakly deviant) in terms of a deferential operator D x,m [S], where 'S' is an expression of a language, 'x' the person to whom we defer and 'm' is a way of using that expression.
I will not summarize the discussion here as it is rather intricate. It takes a detour through an examination of the notion of deference, and of Sperber's (1975) notion of quasi-belief. There are also interesting connections made to work by Burge, Millikan and others. The discussion also takes a detour through a consideration of the notion of context. Recanati defends a pragmatic conception of context. Here he cites work by Travis, Searle and others in support of his views, and contrasts his views with those of Kaplan.
The final chapter of the book (chapter 20) ties together the discussion of the distinction between transparent and opaque readings in Part III with the discussion of cumulative mixed quotation that is offered in chapter 19 section 19.5. Earlier, opaque readings were said to involve a pragmatic enrichment of the minimal (transparent) reading. To say that John believes that S understood opaquely is to say that John so-believes S. The metarepresentation has as part of its truth-conditional content the content of the object-representation, but in addition it contributes information about the way in which the subject believes that content. A similar account could be given of cases of cumulative mixed quotation, such as:
(10)Quine said that quotation 'has a certain anomalous feature.'
This expresses the same content as its unquoted version, but in addition it expresses information about how Quine made his point.
However, the account of cumulative mixed quotation given in chapter 19 in terms of the deferential operator suggests a different picture. A sentence like (10) expresses the same content as its unquoted version, but under a different, metalinguistic, character. This is the character introduced by the deferential operator. A similar account can be given for belief sentences interpreted opaquely. They express the same content as their transparent counterparts and differ only in that this content is presented via a metalinguistic character.
Can these two versions be reconciled? The earlier version seems to say that transparent and opaque ascriptions differ with respect to their truth-conditional content, whereas the later version says they have the same truth-conditional content and differ only at the level of character. Despite this apparent contradiction, Recanati suggests that the two version can and should both be accepted. The normal content expressed by a belief sentence is just its minimal content, and yet at the same time it is contextually enriched (p.308).
Recanati's explorations in his book lead him to the overall conclusion that metarepresentational prefixes are in and of themselves semantically innocent. All they do is shift the circumstance for the evaluation of the object- representation. However, metarepresentation provides an environment for the operation of a couple of pragmatic processes. One of these is contextual enrichment, which is semantically innocent. However, another is the process of deference, and this does introduce deviance. Opaque readings are associated with deviant, metalinguistic characters.
3. Brief summary of book
(If you've read the detailed summary in section 2, you can skip to section 4)
Part I: This lays out various principles that Recanati thinks should constrain any account of metarepresentation. The two most important of these are that the account be semantically innocent and that it obey the Principle of Iconicity. An account is semantically innocent if expressions occurring in the complement clauses of metarepresentations of the form 'John believes that S' are assigned their normal (as opposed to deviant) semantic roles. Iconicity is the principle that the sentence S (the object-representation) should be not only syntactically but also semantically a part of the metarepresentation. This means that the content of the object representation should be contained in the proposition expressed by the metarepresentation. Recanati also gives an argument to the effect that no account that treats that-clauses as referring terms can be semantically innocent. This is an important argument, since most contemporary accounts assume that belief and other attitude reports express a relation (e.g. of believing) between a subject and a proposition, which, according to these accounts, is picked out by the that-clause.
Part II: Think of a metarepresentation such as 'In the film, a 3rd World War is declared' or 'John believes that a 3rd World War has been declared' as having the form 'dS'. Here 'd' is a metarepresentational prefix (e.g., 'In the film' or 'John believes that') and 'S' is an object- representation (e.g., 'A 3rd World War has been declared'). Recanati considers two kinds of accounts of metarepresentations that satisfy the Principle of Iconicity. He calls these modal theories and simulation theories. According to the former, the content of the object-representation is a proper part of the content of the metarepresentation. The metarepresentational prefix is an operator that shifts the circumstance with respect to which the object-representation is to be evaluated. For example, 'In the film, a 3rd World War is declared' is true in a circumstance c iff a 3rd World War is declared in the circumstance c' depicted in the film. Simulation theories on the other hand regard the metarepresentational prefix as contributing no content of its own. It simply signals that the speaker is projecting himself into the world of the film or the believer or whatever, and making assertions within that pretense. Metarepresentations on this view are pretend assertions rather than serious assertions about what holds within some pretense. Recanati rejects pure simulation accounts as too simplistic. This does not mean that the idea of simulation is completely rejected. Even in modal accounts there is an element of simulation, since the object-representation that is reported on is simultaneously displayed. Recanati also introduces his 'Austinian semantics'. This semantic apparatus invokes the notion of a situation and the facts it supports. It allows Recanati to represent the idea that metarepresentations involve a world/circumstance shift. It also shows that in metarepresentations a situation is viewed both from within (as one that supports certain facts) and as an entity that can be viewed from an external perspective.
Part III: This contains some important clarifications of terminology. The relational/notional and transparent/opaque distinctions are often conflated and said to be instances of the so-called de re/de dicto ambiguity of sentences of the form 'John believes that S'. Recanati argues that these are in fact orthogonal to one another. The relational/notional distinction should be made in terms of the sort of belief ascribed. We have a relational reading if the belief ascribed is singular, and notional if the ascribed belief is general. Recanati assumes that when there is a singular term in the embedded sentence S, then only a relational reading will be available. Yet even such an ascription can be given either a transparent or an opaque reading. On the opaque reading, which has the form 'John so-believes that S', a singular term in S has a dual role. It refers to the object the ascribed belief is about, but it also indicates how the believer thinks of the object. This analysis is reminiscent of Quine's famous example 'Giorgione was so-called because of his size'. So Recanati embarks on a discussion of this example. Here too he offers terminological reforms. Many following Quine conflate the notions of a singular term being referential with its being transparent, and both in turn with its occupying a position that is open to substitution and that can be quantified into. Recanati argues that all those notions must be kept distinct. The occurrence of a singular term can be purely referential yet not be transparent, because it occurs in a reflecting context, i.e., a linguistic context containing an expression whose semantic value depends on the identity of some other term in the context. 'Giorgione' in Quine's example is a purely referential occurrence. However, it is not transparent, because Quine's sentence is a reflecting context. Replacing the name 'Giorgione', even with a co-referring name, alters the semantic value of 'so-called'. Similarly, opaque readings of attitude reports are reflecting contexts. One might think that if a belief report is given a transparent reading, then any singular terms occurring in the embedded sentence must be replaceable by co-referring terms. However, Recanati denies this. Belief reports are unstable contexts, and substitution can change a context from a non- reflecting one into a reflecting one. Thus substitution is never possible into the embedded sentences of attitude reports and other metarepresentations. On the other hand, and contra Quine, "quantifying in" is always possible. That is, from 'John believes that a is F' one can infer 'There is something of which John believes that it is F'. This is always possible because in the act of existentially generalizing one filters out all readings but the transparent one.
Part IV: Recanati worries that by relying on a quotational paradigm (the 'Giorgione' example) for understanding opaque readings of metarepresentations, he may have violated semantic innocence. Quotation can be thought of as a context-shifting device. For example, in a direct speech report such as "John said 'I am sick today'", the indexicals are not to be interpreted with respect to the speaker's context (the 'external' context), but with respect to John's context when he uttered the quoted words (the 'internal' context). A better model for the sort of mention that goes on in opaque readings of metarepresentations is mixed quotation. For example:
(1) Quine said that quotation 'has a certain anomalous feature.'
Here the quoted material plays a dual role. It is not simply being displayed (as is the case in direct quotation), but it is also in active use. Recanati distinguishes between cumulative and non-cumulative cases of mixed quotation. Cumulative cases entail the sentence with the quotes removed, but not so for non-cumulative cases. These involve a context-shift, so that not all the quoted material can be interpreted relative to the speaker's context. Consider:
(2) Mr. Greenspan said he agreed with Labor Secretary R. B. Reich 'on quite a lot of things.' Their accord on this issue, he said, has proved 'quite a surprise to both of us.'
Here the final 'us' refers to Greenspan and Reich, not to a group of which the speaker is a member. If the quotes were removed, the 'us' would have to be transposed to 'they'. If opaque readings of metarepresentations involve such context shifting, we potentially have a threat to semantic innocence.
Part V: Recanati begins with a discussion of metafictional statements such as 'My children think that Santa Claus can't come until we're all asleep.' Here there appears to be a context-shift. The speaker doesn't believe that 'Santa Claus' refers to anything. He is taking up his children's perspective and pretending that there is such a person. Recanati doesn't want to say that metafictional statements are pretend assertions. Rather, we exploit pretense for a serious purpose. By pretending that there is a person called 'Santa Claus' we are able indirectly to ascribe a belief to the children that couldn't be ascribed directly. In this case the pretense is global. It does not occur within the scope of a metarepresentational prefix. On the contrary, the latter is within the scope of the pretense, so we don't have to see the prefix as inducing the context- shift and hence as semantically deviant. This section of Recanati's book also contains a discussion of context shifting that occurs in free indirect speech and in cases of partial pretense. An example of the latter is:
(3) 'Quine' hasn't finished his paper yet,
where it is clear from context that 'Quine' is being used mockingly, to refer to a man McPherson who the person being mocked mistakenly thinks is Quine. This is accounted for by positing a context-shifting function which maps the normal character of 'Quine' onto a distinct character PERSON NAMED 'QUINE' BY THE ADDRESSEE. In the actual context this picks out just what the normal character would pick out in the shifted context determined by taking up the addressee's perspective.
Part VI: Context shifting functions of the sort just posited won't help with cases such as:
(4) My 3-year-old believes I'm a 'philtospher'.
Here the quoted term doesn't have any normal meaning. Recanati calls cases such as (4) strongly deviant non- cumulative echoes. Cases such as (3) he calls weakly deviant non-cumulative echoes. Cases such as (1) are cumulative echoes. All these cases are now accounted for in a unified framework by introducing the idea of deference in language use. In this framework, cumulative echoes, which are the closest models for opaque uses of metarepresentations, are said to express the same content as their unquoted versions, but under a different, metalinguistic character. This metalinguistic character is determined by applying the deferential operator to the quoted expression(s). This account gives rise to a slightly different account of the transparent/opaque distinction than was given in Part III, where the opaque reading was seen as a contextual enrichment of the minimal, transparent reading. Thus Recanati discusses how these two accounts are to be reconciled. Recanati's overall conclusion is that metarepresentational prefixes are in and of themselves semantically innocent. All they do is shift the circumstance of evaluation of the object-representation. However, metarepresentation provides an environment that is hospitable to the operation of various pragmatic processes. One of these is the process of enrichment, which is semantically innocent. But another pragmatic process is the one of deference, and this does indeed introduce semantic deviance.
4. Critical assessment
Some of Recanati's ideas have already been presented in a series of papers. However, having these ideas organized as they are in this book makes one appreciate that Recanati is in the process of developing a highly innovative and interesting account of metarepresentation and related phenomena, such as quotation, direct and indirect speech reports and metafictional statements. One of the strong points of the book is how wide-ranging it is. Topics that may at first sight have seemed remote from one another are linked in interesting ways. For example, a discussion of conditionals proves to throw light on the structure of metarepresentations, and a discussion of the idea of deference helps to unify various sorts of echoic uses of language, including opaque uses of metarepresentations.
Other highlights are the conceptual clarifications Recanati proposes in Part III. His proposals for understanding the relational/notional and transparent/opaque distinctions, and his treatment of the notions of pure referentiality, transparency, substitutability and existential generalizability are likely to change the way people talk about these topics in the future.
As I said, the discussion is wide ranging, and sometimes it is not obvious how the discussions in the various parts hang together. Sometimes Recanati explicitly attempts to hook the parts together, as when in his final chapter he connects the discussion of opacity in Part III with the later discussions about echoic uses and the deferential operator in Parts IV - VI. However, there are other connections that remain unclear, or at any rate unstated.
For example, how is the semantic machinery presented in Part II (Recanati's 'Austinian' semantics) to be applied to opaque uses of metarepresentations? Opaque uses involve pragmatic enrichment. Recanati admits that the process of enrichment is one that affects the truth-conditional content of an utterance. He agrees with Levinson (2000) that there can be pragmatic intrusion into truth- conditional content. Presumably d-structures (see detailed summary) are meant to capture the truth-conditional content of metarepresentations. So enrichments should be reflected in d-structures. But how? The following is the d-structure Recanati suggests for a belief report such as 'John believes that a 3rd World War has been declared':
(1) [S] |= @ << John's belief state |= @ << Situation in John's belief world |= w <<3rd WW has been declared >> >> >>
I assume that (1) corresponds to the transparent interpretation of the belief report. How is the enrichment (viz. that John so-believes that a 3rd World War has been declared) to be incorporated? It is not a fact supported by the situation in John's belief world, but a fact about John's mode of believing. So it is external to John's belief world and hence a fact (if it is a fact) supported by some actual situation S'. One possibility is that enrichments are given by independent Austinian propositions, such as:
(2) [S'] |= @ << The content of John's belief is presented under the mode 'A 3rd WW has been declared' >>
(This is just an ordinary Austinian proposition, not a d- structure, because it doesn't involve a situation seen both from within and from without). Thus the opaque reading of the belief report is given by the conjunction of (1) and (2). Recanati claims that opaque readings entail transparent ones. If opaque readings were conjunctions of this sort, this would support his entailment claim.
It is also unclear how context shifting is to be handled in terms of Recanati's Austinian semantics. Recanati will presumably say that context-shifting all takes place at the level of character not the level of content, and thus needn't be reflected in d-structures, unless we are dealing with cases where we have both a context shift and pragmatic enrichment. Suppose we want to represent the following belief ascription:
(3) John believes that 'Quine' has finished his paper.
This is an instance of weakly deviant non-cumulative mixed quotation. Here the (minimal) truth-conditional content is given by the following d-structure:
(4) [S] |= @ << John's belief state |= @ << Situation in John's belief world |= w << McPherson has finished his paper >> >> >>
Recanati thinks that pragmatic enrichment may operate in the case of (3). This would yield the claim (understood to be a part of the truth-conditional content of (3)) that John uses the name 'Quine' as a name for McPherson. If (3) is also understood opaquely, this yields the following two Austinian propositions:
(5) [S'] |= @ << The content of John's belief is presented under the mode 'Quine has finished his paper' >>
(6) [S"] |= << John uses 'Quine' as a name for McPherson >>
Thus the enriched content of (3) is represented as the conjunction of (4), (5), and (6). Actually, it is crucial that (5) and (6) do not become decoupled, as then one may fail to realize that 'Quine' is being used deviantly in (5). So perhaps (5) and (6) need to be incorporated into a single Austinian proposition.
A further question concerns how to analyze cases of strongly deviant non-cumulative mixed quotation in Austinian terms. Recanati allows that these too can be used with enrichment. An example would be "My 3-year-old son thinks I'm a 'philtosopher'". What is unclear here is whether we are to think that the child has a deviant word for our concept of a philosopher, or whether the child has a deviant concept, which we have access to only through the character associated with the expression that is formed by attaching the deferential operator to the expression 'philtosopher'. In the example above it is plausible to think that what is being echoed is simply the child's strange pronunciation of our word 'philosopher'. But there are presumably cases where we echo the child's word because we have no idea what concept is being expressed. E.g., "My 3-year-old thinks I'm a 'pooti'." In such cases we apparently do not know the truth-conditional contents of our utterances.
Another quibble I have is that in several places Recanati simply gestures at something without developing it, whereas it would be nice to get some more details. For instance there are hints given that pragmatic factors other than those already mentioned play a role in the interpretation of metarepresentations. Firstly, consider the d-structure (1) above. One of its elements is a situation in John's belief world that supports the fact that a 3rd World War has been declared. Recanati says that in metarepresentations a situation that plays this role "is either contextually determined or left unspecified (quantified over)." (p.108) But Recanati says nothing about the pragmatic processes that might play a role in contextually determining such a situation. Secondly, when talking about opaque readings of belief reports, Recanati says that one can think of such reports as having the form 'a believes that S like that'. The adverbial phrase 'like that' denotes a manner of believing. But what manner of believing is at issue will "depend upon the dimensions of similarity which are contextually relevant." (p. 158) Presumably is it possible to use English words to ascribe a belief to someone who speaks no English. Thus an ascription of the forms 'a believes that S like that' needn't mean that the believer thinks of the content of 'S' by means of that very sentence-type, but rather by means of a relevantly similar one. Here again pragmatic factors come into play in interpreting opaque belief reports, but Recanati doesn't spell out any details.
There are parts of the book that are likely to be more controversial than other parts. I have in mind particularly those sections in Part VI where Recanati defends a pragmatic conception of context against a more traditional Kaplanian conception. As one who has been persuaded by Recanati's earlier work (e.g., Recanati 1991, 1993) of the ways in which pragmatic factors intrude into what others would regard as a purely semantic domain, I personally liked those sections very much.
I end with one caveat. I earlier said that I thought this book would make an excellent introduction for cognitive scientists interested in exploring this philosophical territory. I still believe this, but would like to stress that this is a philosophical work and thus has a style that some cognitive scientists may find hard to take. Recanati (like all good philosophers) lays out all his reasoning, including the sometimes-circuitous means by which he arrives at his conclusions. This means exploring ideas only to abandon them when they seem to be leading nowhere, and even sometimes returning to ideas that had previously been abandoned because a widened perspective can make them seem viable again. So if you're one who prefers "just the facts, please ma'am", this may not be the book for you!
Crimmins, M. and J. Perry (1989). 'The prince and the phone booth.' Journal of Philosophy 86:685-711.
Davidson, D. (1984) 'On saying that.' In Inquiries Into Truth and Interpretation, 93-108. Oxford: Clarendon Press.
Frege, G. (1970). 'On sense and reference.' In P. Geach & M. Black (eds.) Translations from the Philosophical Writings of Gottlob Frege. Oxford: Blackwell.
Levinson, S. (2000). Presumptive Meanings. Cambridge, MA: MIT Press.
Pietroski, P. M. (1996). 'Fregean innocence.' Mind & Language 11:338-370.
Recanati, F. (1991). 'The pragmatics of what is said.' In S. Davis (ed.) Pragmatics: A Reader. New York: Oxford University Press.
Recanati, F. (1993). Direct Reference: From Language to Thought. Oxford: Blackwell.
Sperber, D. (1975). Rethinking Symbolism. Cambridge: Cambridge University Press.
Sperber, D. & Wilson, D. (1986). Relevance: Communication and Cognition. Oxford: Blackwell.
Walton, K. (1990). Mimesis as Make-Believe: On the Foundations of the Representational Arts. Cambridge, MA: Harvard University Press.
Anne Bezuidenhout teaches in the Philosophy Department at the University of South Carolina. Her research is focused on issues in the philosophy of language and pragmatics. She has written on topics such as the referential/attributive distinction, the pragmatics of propositional attitude reports, generalized conversational implicatures, and metaphor. She is also currently engaged in a project in experimental pragmatics with two psychologist colleagues. The project seeks a better understanding of the factors influencing on-line processing of sentences giving rise to generalized conversational implicatures. Her e-mail address is: firstname.lastname@example.org
Anne Bezuidenhout Department of Philosophy University of South Carolina Columbia, SC 29208 U.S.A Tel (803) 777-4166 Fax (803) 777-9178 email@example.com