Expressions, Sentences, Propositions

Abstract

The paper articulates and defends the view that paired structures of mentally ‘represented’ phonological and semantic features should, for all theoretical purposes, replace the notions of proposition and sentence. Following Chomsky, I refer to such pairs as expressions (EXP). In the first part, I elaborate the notion of an EXP and contrast it with that of sentence/proposition. The paper’s second part questions a range of considerations which putatively show that propositions are fundamental to our understanding of meaning and cognitive attitudes. I argue that while these considerations are effective against the theoretical worth of sentences, as usually understood,  they may be accommodated by the EXP-conception or otherwise eschewed.

 

1: Introduction

 

The aim of the sequel is to articulate and defend the position that, for all theoretical purposes, the notion of a proposition or sentence should give way to the notion of an expression (EXP). Following recent work in generative linguistics, we may initially think of an EXP as a pair <PHON, SEM>, where PHON = a set of phonological features, and SEM = a set of semantic features, both drawn from the inherent features of a set of lexical items selected as the input to a derivation performed by the language faculty. Prima facie, it might seem that this thesis is a mere notational variant of the familiar position that an (interpreted) language is a set of pairs <S, M>, where S = sound/marks, and M = meaning (cf. Lewis, 1975). The differences are, however, numerous and fundamental. EXPs, although abstract pairings, always give way to actual structures realised in the mind/brain. Further, a given set {EXP} does not define a language as a set of strings; rather, the set serves as an abstract specification of the output profile of the language faculty - a state of the human mind/brain integrated with other systems serving speech/motor control and conceptuality/intentionality, or, more simply, sound and meaning. The set {EXP} is not something with which the mind/brain engages, an independent object minds represent. If a definition of ‘language’ is sought, then we may say that it is the generative procedure that ‘creates’ the convergence of sound and meaning involved in EXPs. That is, the component of the mind/brain we dub the language faculty counts as linguistic just because its output codes for phonological articulations which are robustly associated with the expression (not necessarily to others) of certain ‘meanings’. EXPs specify such associations. So understood, language will vary from speaker to speaker, for no two speakers generate the same EXPs; we all use and understand language in different ways. Chomsky (e.g., 1986, 2000a) has dubbed this conception I-language: an I-language is an individual speaker/hearer’s internally realised (‘represented’) generative procedure - or function in intension - that maps from arrays of items drawn from a stored lexicon to EXPs. The claim of the sequel is that sentences and propositions should be treated as outmoded ontology, to be replaced by EXPs, that is, features drawn from lexical items and structured  so as to be useable in linguistic production and understanding.

      Firstly, I shall deal with some preliminary issues to do with the ‘truth bearer’ debate and how it should be understood. Secondly, I shall present the notion of an EXP as one which makes redundant, for all theoretical purposes, the common notion of a sentence. The argument that EXPs may profitably replace propositions will constitute the remainder of the paper. Propositions are essentially posits to carry the burden of truth and, concomitantly, meaning and the content of cognitive states. I shall consider a range of constraints on truth/meaning bearers which are often understood to militate for propositions. It will be argued that these constraints are indeed not met by a wooden notion of a sentence often presented as a straw-man, but that they are smoothly satisfied by the internalist model of EXPs.

 

2: Preliminaries: Primacy and Catholicism

In a sense, the proposal advertised above is an entry into the traditional debate over truth bearers: What kind of objects may bear the property of truth? What ontology is implied by our truth predications? These sorts of questions, while vigorously engaged by some contemporaries (e.g., Alston, 1996, Soames, 1999, and Dummett, 1999, 2002), are perhaps more often dismissed in favour of a catholic attitude under which sentences, propositions, statements, mental states, etc. all get to bear truth equally. Independent of any issue over EXPs and whether they support truth ascriptions, the catholic attitude is clearly correct inasmuch as any of the above mentioned entities support the predication of ‘is true’ in colloquial discourse as witnessed by the schema ‘The … that p is true’, where the blank may be filled by ‘statement’, proposition’, etc. The purpose of this paper is definitely not linguistic legislation. There remains, however, the theoretical  question of primacy; that is, the theories we wish to have about truth, meaning, belief, and cognitive capacity in general appear to make demands on the kind of representational structures that are core to their explanations. For example, Horwich, 1990, and Soames, 1999, contend that a deflationary theory of truth is proprietarily stated in propositional as opposed to sentential terms, for the latter construal does not secure the a priori acceptability of instances of the disquotational schema that is meant to fix the content of ‘is true’ (see §4c). We may say that if a catholic disavowal of the issue of primacy is the correct theoretical attitude, then we require some argument over and above an observation on colloquial discourse, for there are arguments which suggest a much less ecumenical approach to truth bearers. So, even though I ultimately think that the issue over ‘sentences or propositions’ is an aimless dispute that reduces to linguistic legislation over colloquial discourse (because neither notion approaches theoretical adequacy over the linguistic phenomena in question), I do think that the motor of the debate - the issue of primacy - is sound. In general, what we should seek is a general framework of human representational capacities against which questions of truth bearers (inter alia) may be understood. In this light, my claim will be that propositions in general are surplus to requirement; EXPs are independently required to account for our linguistic competence. Once the ‘sentence vs. proposition’ debate is put on the right theoretical footing, catholicism should innocently reign as a pre-theoretical ontological view; the real issue is the character of human linguistic capacity and the structured information we must posit to account for it.

      Before we get to EXPs, let us begin to see what is wrong with the intuitive notion of a sentence.

 

2.1: Sentences: The Common Form Conception

Noam Chomsky (e.g., 1986, 2000a) distinguishes between E- and I-languages. Roughly, ‘E’ stands for ‘external’ and ‘extensional’, while, as earlier remarked, ‘I’ stands for (perhaps better, suggests) ‘internal’, ‘individual’ and ‘intensional’. I shall return to I-language shortly; let us first think about the first notion. To think of English, say, as an E-language is to construe it as an (infinite) set of objects - sentences, pairs of marks with meanings - that is individuated in terms of its members, not in terms of the function or computation that generates the set. This is what makes the conception extensional.[1] What makes it external is just the thought that such a set exists outside of any mind to the extent that it can be theorised independently of any conception of human cognitive capacities: sentences are what the mind engages with or represents, they are not part of the mind itself. E-sentences (members of E-languages) have what I shall dub a common form individuation.

       E-sentence tokens are concrete entities such as sound-waves, digital displays, ink inscriptions, etc. They may or may not have meaning; those that do have it as a contingent property among others (but see n.1). This very sentence has a meaning, as well as being black and about 9” long. Type individuation of these sentences may well appeal to grammatical structure, although commonly no such appeal is made. There are two broad ways this might be done. One way is to  define categories in terms of the sets of types which realise the relations marked as grammatical in the potential utterances of, say, the average English speaker (e.g., a (count) noun is to be identified schematically as a member of an equivalence class of substitution instances in say, ‘Most…are G’. See, e.g., Brandom, 2000.[2]) In essence, this does not constitute a departure  from identifying sentences in terms of their ‘surface’ properties of shape. The other way, is to treat grammatical categories as wholly abstract entities that are realised or instantiated in physical forms. Thus, a sentence token has a grammatical structure because it partakes, as it were, of the form of a particular grammatical structure.[3] As remarked, grammatical structure is more often than not ignored by all disputants and is certainly not prominent in propositional arguments against sentences, i.e., propositionalists tend to assume that sentences are individuated independently of such structure.

        It bears emphasis that an E-language is not necessarily a social object, it may be an idiolect (the set of utterances an individual speaker understands). It remains, however, external. Thus, two tokens belong to the same sentence type given conditions of individuation that do not factor in the cognitive structures of the particular speaker in question. So, sameness of type is determined by sameness of shape, whether in terms of inscription, sound wave, grammatical form (defined in one of the two ways above), or some other medium, where this type is a member of the set BILL or MARY, say. The simple point here is that it is one thing to define languages as (potentially overlapping) sets of sentences particular speakers understand, it is quite another thing to define languages in terms of the mental structures of particular speakers.

           In philosophical discussions, the E notion of a sentence is invariably assumed. What else could a language be, other than a set of publicly realisable signs, the members of which are individuated by public properties, such as sound or shape? Well, if this is what languages and so sentences are, then they stand in stark contrast to propositions. 

         Propositions are taken to be abstract objects that perforce do not come in types or tokens; each one is unique. Propositions do not have meaning among other properties, they are meanings. Sentences express propositions, but sentence identity is neither necessary nor sufficient for proposition identity. Tokens of the same sentence type can express distinct propositions; e.g., ‘I am ill’. Tokens of distinct sentence types can express the same proposition; e.g., ‘Snow is white’ and ‘Der Schnee ist weiss’. Although abstract, potentially structured, and not 1:1 related to sentences, propositions can still be picked out by that-clauses as the complements of  cognitive verbs (believe, desire, etc.) and speech act verbs (assert, state, etc.). We speak, then, about the proposition that p, where ‘that p’ is not a proposition, but a clause that expresses a proposition.

 

3: Expressions

The common form conception of a sentence is, I think, ill-suited to enter into any useful theory of language or linguistic understanding. In this section I shall more fully set out the notion of an expression (EXP) on the basis of the notion of I-language and submit that it is fit to replace the common form conception of a sentence. As indicated, the notion of an EXP has a rich and satisfying life in linguistic theory independent of any issue to do with truth bearers. My aim here is not to stretch linguistic concepts beyond their proper reach, but to show, rather, that in their proper place they undermine the appeals to sentence and proposition.

         A central tenant of the generative program - one which has survived the many changes of theoretical architecture - is  that a theory of linguistic competence must involve, at a minimum, the identification of different interfaces between what we may call narrow syntax and independent systems of sound and meaning. This architecture captures the intuitive notion that language, at its most abstract, is simply a pairing of meaning and sound. We can think of the human mind/brain as innately realising an initial schematic structure - Universal Grammar (UG) - that comprises a uniform computational procedure. This constitutes the language faculty. The faculty grows or develops through its being constrained to make certain ‘decisions’ along a finite number of parameters, which, by current assumptions, pertain just to the morphological marking of Case and agreement. The decisions are made on the basis of the linguistic data to which the speaker/hearers are exposed. This exposure determines a steady state of the faculty (one in which all decisions have been made, one way or another), a variation on the biologically endowed theme of UG. The variation pertains crucially to the lexicon a speaker/hearer acquires (words, with their idiosyncratic semantic and phonological features) and, consequently, the structures the procedure generates at the two interfaces, for these structures can consist of nothing over and above what the computation draws from selections from the lexicon. The output structures may count as ‘representations’ only in terms of their interface relations internal to the mind/brain, they are not  about something extra-linguistic. We may say that this interface understanding is an internalist conception of representation. Thus, the faculty does not represent a language as an independent object, the  language itself just is the internal steady state in the above sense: it is a set of complex structures, individuated by sui generis concepts of syntactic (formal), phonological and semantic features, that count as units of investigation given their explanatory role in accounting for linguistic competence. Correlatively, languages are individual in the sense that  they are individuated by the representations particular to individual speaker/hearers, not in terms of communities of speakers. Finally, language is intensional in that it is a generative procedure that issues in sound/meaning pairs, it is not the set of such ‘objects’ understood in abstraction from the procedure that generates them. Thus, on this conception, our external linguistic tokens are reflections of information structures realised in that part of the brain dedicated to linguistic cognition and potentially exploited by other faculties.

       EXPs are what the language faculty generates: convergence between phonological and semantic properties as determined by the formation of structures from selected arrays of lexical items.[4] The nature of this convergence (effectively, the algorithm UG realises) is an empirical question and not to be stipulated. By recent minimalist proposals, however, there are two basic options (which admit variation, of course). At the heart of minimalism is the idea that the nature of the language faculty (narrow syntax) is determined by the requirements of the systems with which the faculty interfaces.[5] This is not to say that the syntactic ‘rules’ have access to semantic or phonological features; rather, the idea is that the external systems impose output conditions on narrow syntax in the sense that whatever it produces must be legible to the systems (useable in some minimal sense), and the faculty is ‘designed’ to meet these conditions in an optimal manner, i.e., the computation contains no redundant steps and the contribution of each step is such that all and only legible information reaches the interfaces.  Now, a natural way of satisfying these constraints is for the algorithm to generate two types of structure, PF and LF. Each is a level of representation in that they are termini of the algorithm where all and only information within the computation is made legible to the respective external systems,  is realised in a form which is useable by the systems. This gives us a simple picture of <PF, LF> directly mapping onto <PHON, SEM> = {EXP}. That is, the computation of the faculty branches, in a Y-formation, as it were, with features legible to motor systems of speech being stripped away at a point of spell out to be interpreted phonologically and terminating at PF. The computation continues covertly post spell out and terminates at LF, where only features legible to intentional-conceptual systems are ‘represented’. EXPs are thus simply the reflection of <PF, LF> pairs, which inherit their convergence from the identity of the lexical items that enter into the initial array from which <PF, LF> pairs are derived. This is the first way of understanding the architecture of the language faculty.[6]

       The second approach, which I’ll dub the Transfer Model (TM), is of recent vintage and marks a break from traditional thinking.[7] By TM, there are no levels of representation, no PF or LF, and so no (single) point of spell out. Instead, the computation cyclically selects from a lexical numeration; each cycle is a phase of the computation resulting in a syntactic object, a merging of the items selected. Each phase marks a point in the derivation of a simultaneous transfer to the external systems. Thus, rather than there being two levels at which ‘everything comes together’, there is a cycle of ‘spell outs’ or transfers which send PHON and SEM to the respective external systems, with the phases being actively stored and then somehow clipped together outside of the syntax proper.[8]

       The TM approach is a recent speculation, although it is gaining support. The initial motivation for it, at least as offered by Chomsky, 2000c, 2001a, is cut down on computational complexity. A point of spell out should strip away all and only that which is not legible at LF, but how can the computation know what is not legible at LF without some form of look-ahead, essentially a global search? Ex hypothesi, at the putative point of spell out, the computation can not see what is not legible at the semantic interface. By making spell out cyclic, the quandary may be resolved by the computation being able to ‘look’ within a phase to see which lexical items have been moved so as to acquire features (Case/agreement). This is crucial, for LF-legible features are inherent to the items; they are not acquired within the derivation. In effect, because the computation can see the affect of movement, and movement is for checking of PF legible features, then by exclusion, the computation acts as if it were able to look ahead to LF, for it strips away just that which is not LF-legible. But, if at each phase that which is not LF-legible is stripped away, then that which remains is LF-legible, and so meets the interface conditions of SEM. Consequently, if the computation transfers information to PHON in a  phase cycle, then it should do likewise with information for SEM, for, necessarily, the structure meets the SEM output conditions at the phase level, perforce further computation to arrive at a level of LF would be supererogatory.  The old distinction between overt (the pronounced) and covert (that which carries on unpronounced to LF) thus gives way: there is a simultaneous transfer of information direct to PHON and SEM respectively without the intercession of representational levels.

       How ‘big’ are phases? The very notion of a phase implies a structure that is (typically) less than the full merging of the lexical items occurring in the initial lexical numeration. Chomsky (2000c, 2001a, 2001b) understands a phase to be a minimal propositional structure, by which is meant either a CP (clausal phrase, but not finite TP) with topic and focus or a v*P (a verb phrase with all thematic roles satisfied; passives and unaccusatives are thus excluded). As indicated, the motivation for the TM model is to do with complexity, but this does not give us ‘propositional’ phases directly.[9] I think, however, that the proposal is optimally flush with the idea of simultaneous transfer to PHON and SEM. That is, while the features legible to PHON can indeed be spelt out in structures smaller than CP/v*P such as DPs or adjunct PPs, such structures appear to be too ‘small’ to be legible at the SEM interface.[10] In crude terms, the propositional (in the technical sense) is the smallest kind of object that has the requisite semantic integrity to be legible to the external operations that govern cognitive processes concerned with truth, inference, scope, etc. Of course, ‘propositional’ is here a term of art which simply means that a structure identifiable within the syntax is potentially apt to support intentional-conceptual relations; the syntactic operations themselves do not identify the structure in such terms; rather, internal to the syntax, phases count as such because of Merge (internal/external) for the checking of uninterpretable features, such as Structural Case, verbal agreement, EPP, etc. Further, a CP, say, might be legible at the interface - a coherent instruction - without in fact being useable by an external semantic system; the extent of such divergence is an empirical issue not to be a priori stipulated. So, phases of the complexity of CP/v*P are the smallest (hence most economical to the computation) that are fit to support the simultaneous transfer of features to PHON and SEM. Thus, the basic idea is that, while a phase derivation is internally required to cut search complexity to the local, the size of the phases is determined by the requirements of the SEM interface.

       The issues orbiting around the difference between these two approaches are quite complex and there is not yet any kind of stable doctrine. That said, I think the notion of EXP is more or less invariant over either model. On the first approach, an EXP (=<PHON, SEM>) is simply the reflection of the output at the convergent levels, and so can be directly read off <PF, LF>. On the second, an EXP is more a reflection of the derivation that cyclically maps to PHON and SEM. Here, the components of an EXP given some initial lexical array are clipped together external to narrow syntax. Of course, the idea of ‘clipping together’ is a mere metaphor that must be cashed for perhaps as simple a mechanism as concatenation, although some constraints must be play. The phases are not simply tipped out, as it were, but delivered in an order demanded by the basic conditions on Merge. Similarly, each phase must still feature an ‘active’ part so that further phases can clip on, a site from which movement may occur. We may say that the most embedded CP of a multi-clause structure will be such that the next delivered phase will clip onto its left to create relations of c-command for scope and binding within the complex structure (here I assume that such relations are established within the syntax only if necessary, not if optional, such as with bound variable anaphora; cf. Zwart, 2002.) So, the same EXPs will characterise the output of the faculty on either model: on one option they will be reflected at the two internal interface levels, on the other, they will be generated as a reflection of the derivation internal to the faculty.

      Hereon, let us sideline the PHON part of EXP and just think about the SEM part. We may think of SEM as dividing into two: an intentional part concerns what we might think of as worldly relations of truth, reference (binding), inference, etc., while a conceptual part concerns matters of scope, focus, topic and discursive aspects more generally (see n.11). EXPs, therefore, are in part constrained (given the nature of the external systems) to support a notion of truth conditions. It bearers emphasis, however, that the arguments to follow, and the general position they defend, neither depend on nor entail any particular approach to natural language semantics, not even a truth conditional semantics, as commonly conceived.[11] This might cause some confusion. Let me explain.

        On the internalist model sketched above, neither LF nor propositional phases (on the SEM side, as it were) are semantic representations; they are syntactic structures that encode those properties of a selection of lexical items that are relevant to intentional-conceptual interpretation (as opposed to phonological interpretation). The map of syntax onto SEM may be seen as an unfurling of the inherent features of lexical items that make up given syntactic structures into a coherent package of information - an ‘instruction’ for the employment of the truth-relevant states of external systems which give rise to our as assertions, beliefs, judgements, etc. This, in part, is what makes SEM semantic in the more familiar sense of truth/reference. It is a theoretical question, not one of stipulation, whether SEM might reduce to or be wholly explicable in terms of a notion of truth conditions. It seems not, but we should not lose sight of the fact that SEM must encode ‘instructions’ for  ‘truth-aptness’. To acknowledge this is not to usurp the internalism of EXPs. The notion of ‘truth-aptness’ does not bear on the way the world  is anyhow, but on our internal competence with such structures, our ability to offer disquotational evaluations, where the evaluation differs as a function of mood. Thus, we understand what it would be for a declarative to be true, a question to be answered, a promise to be kept, an imperative to be followed, etc. from an understanding of the structure itself, through wh/inflection movement, topic, focus, etc. We may view this competence at understanding the cognitive/discursive role a structure may play due to (in part) its inherent features as establishing a minimal propositionality on the semantic side as a reflection of the minimal ‘propositionality’ (Chomsky’s CP/v*P condition) within the syntax (or at LF). The question, then, is: Why should we make this internalist move and turn our back on the traditional E-conception of languages/sentences?

       The inadequacy of the  E-conception is not simply to do with the obscurity of the idea,  that it is difficult to  find any workable conditions of individuation for what is to count as one language as opposed to another. There is indeed massive variation between our idiolects, and there is no arbitration as to what is and is not English save for social conventions which themselves vary massively. Still, perhaps we can define a set all of whose members would be regarded as English sentences by competent speakers. Yet Chomsky’s chief claim is that, even if some such individuation were available, E-languages would remain explanatorily redundant. Chomsky (e.g., 1986) has highlighted two questions which any worthwhile account of language must address: What do individual speaker/hearers know when they know a language? How do they acquire such knowledge? The specific problem with the E-conception is that E-languages are, in effect, abstractions which make little sense in the absence of the I-conception. Thus, on the first question, how could any finite mind know an infinite object such as the set of all ‘English’ sentences? We appear to require a finite base from which the infinity might be generated, but then it is the finite base we know, not the infinite E-language; the latter is more a product of our knowledge rather than the knowledge itself. An I-language may be understood as the finite state which supports our infinite competence. Similar remarks apply to the acquisition problem. Humans reach maturity understanding (at least) one language, but they might have acquired any other language. This immediately shows that humans (as opposed, say, to chimpanzees or rabbits) have an innate capacity for language; UG is a theory of the constitution of that capacity. It is a further empirical question whether UG or something general and non-linguistic is the correct model.[12] But patently, what is innate cannot be an E-language or a Boolean function on the set of E-languages. Again, we seem to require an independent  generative procedure to so much as make sense of the notion that we acquire an E-language.  If these thoughts are anywhere near correct, then the E-conception of language is a shallow, non-explanatory abstraction, for it does not begin to touch on the fundamental questions of linguistic understanding. I shall not further develop the familiar arguments sketched, instead I wish to turn to the virtues of EXPs.[13]

        By the E-conception, sentences are individuated by common form: roughly, if two tokens look or sound sufficiently alike, then they are regarded as belonging to the same type. Yet in no sense can physical properties of sound order, shape, etc. determine linguistic properties of syntax or meaning. NPs or inflectional heads, for example, are simply not found in the world, still less are ‘empty categories’.

       Consider (1)a:

(1)a.  Bill expects to leave by himself.

     b. Mary wonders who Bill expects to leave by himself.

A simple-minded compositionality principle tells us that what a sentence means is a function of what its constituent words mean and how they are put together. In some sense, then, it is supposed that all there is to meaning, and so what determines truth conditions (here I speak neutrally, in the innocent disquotational sense), is provided in the common form we are given. This appears to work with (1)a. We know that the reflexive - himself - cannot be pronominal, i.e., it must co-refer with a local (masculine, singular) noun. Bill is the only such nominal, so we recognise that (1)a implies that it is Bill who expects Bill himself to leave alone, not someone else. But now consider (1)b. Here (1)a occurs as a complement to the interrogative verb wonder. Does (1)a - the common form - mean the same thing as it occurs in (1)b as it does when standing alone? If so, then (1)b means: Mary wonders if Bill expects to leave by himself. This is obviously incorrect: Mary is wondering about Bill’s expectations as to who is leaving, she isn’t wondering about Bill’s expectation of his leaving (although Bill well might have such an expectation, independent of Mary’s wondering). Roughly, (1)b means: Mary wonders which x is such that Bill expects x to leave by himself. This is striking: we witness a change in the understood subject of the verb leave between (1)a and b, even though (1)a exhibits no ambiguity as to who the subject is, and it appears unchanged in (1)b. It would seem that our simple-minded compositionality principle must be wrong.

        The problem here, it bears emphasis, is not that Bill might not be the understood subject of leave in (1)a; rather, it is that to know that Bill is the subject appears to require an interpretation of the ‘empty’ subject position, an interpretation, as (1)b attests, that can and does differ across constructions where the local environment of the empty position remains constant. It is not to the point to suggest that a speaker can work out the interpretation of the empty position from what is provided in the common form. In a sense, it is trivially true that speakers do just that. The rub is that the resulting structures are essentially richer than what appears phonologically; they feature elements that are provided by the speaker’s mind, rather than just those that may confront the speaker’s mind, as it were.   There might be some dispute about this, but I know of no remotely plausible explanation of the phenomenon witnessed that appeals to the ‘visible’ common form alone, mutatis mutandis for all such structural features.

         The standard explanation of the difference is to appeal to empty categories that are not in the ‘world’, out there in sound or ink to be perceived.[14] This appeal issues in structures approximately like those in (2):

 

(2)a  Billi expects [PROi to leave by himself][15]

     b. Mary wonders [who Bill expects <who> to leave by himself]

 

Here, ‘PRO’ functions like an anaphoric pronoun that is controlled by - co-referential with - Bill (marked by the index), where, contra simple-minded compositionality, PRO not Bill is the proper clause-mate antecedent of himself. Note that if we go to spell out the empty position, the result is either a different meaning or nonsense: Bill expects him/Bob/*he/*she to leave by himself. In contrast, with (2)b, the antecedent of himself is the item <who> not spelt out, but whose copy higher in the structure at CP is pronounced. Here there is no structural relationship between Bill and himself, not even one mediated via a controlled empty category. So, against all pre-theoretical intuitions of the E-conception, in a very real sense, (1)a does not occur in (1)b. At the level of syntax that is individuative of EXPs, the two structures are quite distinct, notwithstanding the fact that they ‘sound and look’ the same. Nor is this merely a syntactic matter; the difference of meaning explained above is precisely due to the unpronounced elements which determine the SEM features of the structures.

       A perhaps simpler example of the inadequacy of the E-conception is proved by our humble notion of a noun. Nouns cannot be defined in terms of phonology/morphology - properties we take to be independent of syntax/semantics - unless one were simply to list all the nouns in all of the world’s languages (mutatis mutandis for other grammatical categories). But this is absurd, for every speaker/hearer is somehow sensitive to noun-hood, as it were, but no subject has an exhaustive look-up table in their heads. Nor can nouns be identified by what they refer to; anything at all can be referred to by a noun: events, situations, concrete objects, abstract objects, fictitious objects, etc. Nor, as already pointed out, can nouns be identified schematically via some notion of frequency or substitutional class. It would seem that noun-hood is a feature that is intrinsic to our minds, which is only reflected, although essentially so, in our linguistic products. Since this feature, and all other grammatical categories, go to determine what a sentence means (why we produce and understand the language we do) any external notion of a sentence appears to be demonstrably inadequate. Similarly remarks may be made about ambiguity (syntactic and lexical), inferential relations, lexical features, etc. The richness of our linguistic competence appears to demand structures as rich and as highly articulated as EXPs.[16]

       How much of a departure is the EXP conception from the E-type conception of a sentence vis-à-vis propositions? A number of points are relevant.

        EXPs, from one view, are abstract structures of information, whose individuation does not factor in properties of the human brain or any other physical system. In this regard, EXPs might appear similar to propositions. It is, however, constitutive of the notion of EXPs in general that they give way to the PHON and SEM features  realisable by the human brain; for they are nothing other than those structures of information whose physical realisation explains (in part) human linguistic ability (cf., Chomsky, 2001a, pp.41-2). In this sense, a physical structure, an array of neurones say, counts as an EXP or, to be precise,  a physical structure realising a set of high-level properties through its role in the language faculty.[17] We can thus differentiate between EXP types - the abstract information structure - and tokens - the instantiations of such information structures - in terms of the sameness of the information being realised at different times. I think that this idea can be extended to sentences on the E-type model. That is, all public tokens - inscriptions, sound waves, etc. - may be viewed as projections from the cognitive structures that lead to token production or are called upon to understand (consume) such products.

      A question arises as to how equal consumption and production are. Take, for example, a particularly peculiar gust of wind that forms an inscription on a beach with the following shape: Paris is the capital of France. Should we view the inscription as a sentence token? In one sense, clearly not, for it has not been produced by a cognitive agent that invested the marks with any structure or meaning. On the other hand, it can easily be taken to be a sentence; one can consume the marks as a sentence - project structure onto it. After all, if one is ignorant of its curious origin, then why should one suspect it is merely marks in the sand. The appropriate answer to this quandary, I think, is that production does matter: external marks count as sentences only if they are produced by a cognitive system that assigns syntactic and semantic properties to the marks. Sentences, in an extended sense, are structures, in various media - sound, ink, hand gestures in ASL - that are generated by human beings. The basis for this generation is the species specific procedure that generates mental structures which encode the essentially linguistic information we invest in the external structures. 

         If EXPs are individuated in part by semantic properties (not truth and reference, mind, but the inherent features of the lexical items), should we then think of them as meanings, and so as closer to propositions than to sentences? Well, consider No Entry road signs; are we tempted to call them meanings? Of course not; they are typically made of metal adorned with appropriate paint-work. Even so, something counts as such a sign just if it receives a specific interpretation or is produced with that interpretation in mind; weight or shape are not individuative of the signs. Just so with EXPs. They are meaningful qua realised at the interface of syntax with interpretive structures of sound and meaning of particular speaker/hearers. As such, they are realised by a range of divergent physical properties, and so are not meanings in the usual abstract sense, but these properties do not constitute them as belonging to the same EXP type. EXPs are essentially meaningful in virtue of being those structures that support human intentionality-conceptuality, but they are not meanings. Indeed, from the optic of the I-conception, meanings, and so propositions, have no role whatsoever to play in our understanding of the mind. The EXP-conception is a way of doing without propositions, it is not a way of understanding propositions, still less making them naturalistically respectable.

        The conditions for the realisation of I-languages are a function of the computational processes underlying linguistic competence. Just what these processes are is an empirical question - glanced at above - whose answer we do not yet know; but the guiding assumption of current linguistics is that we do know enough to hypothesise the information structures such processes are defined over, and these structures (again, at a certain level of processing) generate EXPs. In this sense, we should view EXPs as structures visible from a computational or informational level that, in the first instance, are realised by neuronal structures, whose salient features we have yet to discover. Sentences in the external sense may be seen as projections from the EXPs called on in the production of the sentences.

      Much more, of course, could be said about the notion of an EXP. Still, enough has been said to support the claim that the model has greater coherence and clarity than the conception of a sentence as arbitrary concreta with a certain form. The best way of showing that EXPs are indeed fit to replace our ontology of sentences and propositions is to see how they avoid the major criticisms of traditional sententialism and may usurp the familiar role of propositions. This is the brief of the following section.

 

4: In Defence of Expressions

Propositions appear so indispensable, I think, because the only other option entertained is a common-form notion of a sentence. A host of complaints have been made against traditional sententialism, some to the effect that sentences simply cannot bear meaning or truth, others more modestly claiming that sentential truth is derivative of propositional truth. There is not a general criticism underlying this motley assault. However, the complaints ride on the back of certain intuitive constraints imposed on candidate truth bearers. My method, therefore, will be to go through a number of criticisms and show that the constraints they presuppose are either inapplicable in themselves or may be smoothly satisfied by the EXP model. So, propositions will be surplus to requirements, not because they necessarily have something of the night about them, but that even if they were to be transparent, we need make no recourse to abstract, essential meanings, independent of the employment of human linguistic capacity.

 

a)   The common practice objection

It is often claimed that common usage is at odds with sententialism. Alston, 1996, p. 9, for example, avers that “there is no ordinary, nontechnical practice of applying ‘true’ and ‘false’ to sentences, and no sense of these terms in ordinary use in which they apply to sentences”. Such claims are echoed by many (see, e.g., Barwise and Etchemendy, 1987, pp. 9-11 and Soames, 1999, Chp. 1).

       A number of issues arise with this objection to do with ambiguity and context dependence that I shall tackle below; for the moment I want to deal specifically with the topic of common practice. As should be evident, common practise does not sanction EXPs - internal mental structures of abstract features. Here it will be argued that the current objection offers no support whatsoever to propositions against either sentences or EXPs.

         The thought that ordinary usage does not sanction predication of truth to sentences is, I think, based upon two reflections. Firstly, the paradigm of traditional sententialism is the context (SP):

 

(SP) ‘…’ is true,

 

where the blank is filled with sentence tokens to form quotation-names of the sentences. Such contexts, let us agree, principally occur in philosophy or logic books. It thus appears that (SP) is an artifice that does not reflect the behaviour of our truth predicate in ordinary use. Secondly, the things whose truth or falsity we are generally concerned with are people’s assertions, claims, beliefs, sayings, etc., which are not sentences as such, but that which is asserted, claimed, etc., i.e., propositions.

         Let us begin with the first reflection. It is true that (SP) contexts are not the norm, but this is of glancing interest. It certainly does not demonstrate that there is anything incoherent about the contexts; we understand them perfectly well. Moreover, there is a practice, if not an ‘ordinary’ one, of using (SP). Consider marking, revising, translating, or otherwise working on a text. Here it seems that we are concerned with sentences. Perhaps we are in fact concerned with the propositions they express, but the issue is at best tendentious and has nothing to do with ordinary usage, for that is what is in dispute, it does not decide the issue. People just do not have intuitions on such matters. As I argued in §2, the question of truth bearers ought to be a theoretical one concerning the kinds of assumptions we should make in a general account of human representational capacities - in particular, accounts of truth and meaning. In this light, any attempt to deduce primary truth bearers from ‘ordinary usage’ is wholly misconceived.

       It might still be that common practice does sanction propositional truth bearers as opposed to sentential ones, even if such a licence is of no particular interest. Here we come to the second reflection that the things whose truth we are interested in are assertions, beliefs, etc., and so propositions. This idea depends on a particular view of how to understand mental state terms, not a neutral survey of the folk’s ontology. For instance, the propositional analogue of (SP) - The proposition that p is true - is as  rare as instances of (SP). By parity of reasoning, if sententialism is dismissed on the basis of the reconditeness of (SP), then propositionalism ought also to be dismissed. It is, of course, often claimed that that-clauses name propositions, but they do not, obviously, name sentences. I find it curious that such a thought retains its pull. Clauses are not nouns: there is no syntactic basis at all for thinking that that-clauses name anything at all, let alone propositions. That a clause can be picked up by a discursive pronoun it matters little: It is true that p is an extraposed version of That p is true; the pronoun is pleonastic, it doesn’t refer to anything. It might be that we require propositions in order to account for the semantics of cognitive verbs (see below), but there is no independent reason to think so; the syntax leaves it open.

      All this stand-off shows is that truth bearers cannot be explicitly read off the practices of the folk; or rather, if we follow the folk, then nigh-on anything can bear truth and there is no question of primacy. This being so, the mere appeal to our interest in assertion, belief, etc. militates not a jot for propositionalism without an argument that the objects of such cognitive states and speech acts are propositions. The assumption that propositions are the objects is not a reason for propositionalism, it is a prejudice against sententialism. I shall return to this issue below and consider some of the arguments for the supposed requirement of propositions to account for beliefs and other cognitive states. For the moment, I only hope to have established that a mere appeal to common practice or vernacular is otiose: argument and theory are required.

 

b)  The ambiguity and context dependence objection

This is perhaps the main objection to sententialism and I think that it does cause real problems for what I have called the common form conception. It does not, however, challenge the EXP model.

           We saw in §3 that EXPs are individuated such that  syntactically distinct structures make for distinct EXPs; this is so whether we assume LF or a phasal transfer model.  Thus, tokens of, say, ‘They are flying planes’ are not ambiguous; they fall under one or the other of two homonymous EXP types; that is, given the difference between the merged structures v + [flying planes] and [DP flying planes], the respective EXPs will differ with regard to SEM, but, if we ignore issues of stress pattern, be equivalent with regard to PHON. Likewise for semantic ambiguity: tokens of bank are not ambiguous; they fall under one of (at least) two homonymous lexical types, one marked to refer to financial institutions, the other marked to refer to sides of rivers.

          The context dependency of the semantic values of deictic items is similarly accommodated. Indexicals and the rest have what Kaplan, 1978, has called character: a meaning associated with the lexical type that determines, given context, content (what, in regard to SEM, we would call intentional features). So, roughly, the character of ‘I’ is a rule, or constant function, which determines that a token of ‘I‘ refers to the speaker of ‘I’. Now it is straightforward to adopt this notion of character to EXPs: their SEM components are so marked as to take designated semantic values in context of production or interpretation (processing lexical input). The details, for sure, will not be straightforward, but the principle is clear enough (see Larson and Segal, 1995, Chp.6).

      As is widely rehearsed in the literature, traditional sententialism (i.e., E-sententialism) has difficulties with ambiguity and context dependency. The root of the problem is that the principle of individuation employed is non-linguistic: the common form of either type or token does not include syntactic or semantic features. The EXP model suffers no such problems, for its principle of individuation factors in both kinds of features without appeal to any notion of a proposition.

 

c)   Truth-Meaning objection

Scott Soames, 1999, pp.102-7; 243-4, has argued that only propositional truth bearers establish a basic conceptual link between the concepts of truth and meaning. It would seem, then, that “our ordinary notion of truth” carries “an implicit commitment to propositions” (Ibid, p. 106). I think that Soames is right about the link, but not about the implicit commitment, for EXPs serve equally to support the link. This, it should be noted, they must do, insofar as SEM codes for the minimal assertoric feature required for intentional use of a structure.

       Truth and meaning are conceptually linked in that one way of expressing the meaning of a sentence is to state its truth conditions. This principle may be codified schematically as (TM):

 

(TM) If s means in L that P, then s is true in L iff P,

 

where the substitution instances of ‘s’ are singular terms denoting/describing sentences, and the instances of ‘P’ are the sentences so picked out. The instances of (TM), let us agree, are a priori available to us. It thus seems that knowledge of truth conditions should conceptually double as knowledge of meaning. That is, the instances of the disquotational schema (DS) should a priori express the contents of the sentential substitutes of ‘s’ from which the instances of (TM) may inherit their obviousness:

 

(DS) s is true iff P.

 

Now this appearance, Soames, Ibid, p. 105, avers, is real only if the “primacy of propositions as bearers of truth” is assumed; that is, only if (DS) is not primary. What is wrong with (DS)? Well the link between meaning and truth conditions is a conceptual one, a priori, yet instances of (DS) are not knowable a priori, they are contingent extensional equivalencies which do not a priori express the content of their designated sentences; they would be false if evaluated in those possible worlds, say,  where our sentences mean something different. The language index does not alter this, for we do not a priori know that we speak L, as it were. Nor would indexing the truth predicate to ‘my language’ establish the required connection, for there is nothing about (DS) or (TM) which establishes that one understands the quoted sentences, even if they do belong to one’s language. Thus, the apparent a priori status of the connection enshrined in (TM) is not supported by instances of (DS), for they tell us nothing at all about meaning unless we already know what the designated sentences mean. Thus the link between meaning and truth conditions enshrined in (TM) cannot be established as purely conceptual. If, however, the truth bearers are propositions, then the appropriate schema is (PT):

 

(PT) The proposition that P is true iff P.

 

The instances of (PT) are knowable a priori because to know that an instance of (PT) is true is to know a priori the proposition expressed; as such, to deliver truth  conditions via (PT) is necessarily to deliver meaning. Thus, if we wish to ground the conceptual connection between our ordinary concepts of truth and meaning, then the propositional (PTM) is the proper expression of the connection:

 

(PTM) If s expresses the proposition that P, then the proposition P expressed by s is

            true iff P,

 

and so (DS) must be seen as derivative of (PT).[18]

        This argument works fine against E-sentences, on the assumption that they  carry their meaning contingently, but the EXP model is perfectly flush with Soames’s connection. Let us use ‘SEM(s)’ as a dummy for the structure encoded by the SEM features of a given EXP. In LF terms, one could think of SEM(s) as the product a given LF structure instructs the intentional-conceptual systems to generate. Now, it seems that the following schema is available:

 

(SEM-T) SEM(s) is true in L iff p,

 

where ‘p’ expresses the assertoric content of SEM(s) as expressed in say the ‘surface form’ of a speaker. So understood, the instances will be extensionally correct. Further, the instances will be intensionally correct, for since ‘SEM(s)’ is individuated by semantic properties, it will, we may presume, express the same content in every possible world. But are the instances knowable a priori as are those of (PT)? If we take given instances of (SEM-T), then, in an obvious sense they are not a priori knowable by competent speakers; after all, no speaker ‘knows’ the SEM structure that underlies (in part) the sound-meaning association of ‘English’. Yet this is quite beside the point. By assumption, the very fact of being linguistically competent (being in a certain cognitive state), is constituted by the internal generation of EXPs. Instances of (SEM-T), therefore, should be read as statements about speakers’ judgements rather than about particular inscription types or tokens. Consequently, we shouldn’t construe the instances of ‘SEM(s)’ as the subjects of truth predications. The relation we want  might be roughly depicted as

(SEM-TS) SEM(s) takes the value true for speaker S (at t, in C, etc.) iff SEM(s)

                  supports S’s judgement that p (at t, in C, etc.)

Whether or not, then, a speaker would even so much as understand a formal representation of an EXP  is not at issue. The instances of (SEM-TS) may be viewed as expressing a relation between the publicly expressed thought on the right-hand side and that structure which supports the expression. This relation, in a sense, is a priori, for its instances express precisely the information that is necessary and sufficient for a speaker to mean what she means according to our (and Soames’s) intuitive disquotational criterion of truth conditions. Thus, each relevant instance of (SEM-TS) counts as a priori relative to speaker/hearers as a fact about their competence as opposed to what they would consciously judge to be correct.[19]

        The notion of knowledge here used is theoretically loaded, not our ordinary concept. I am perfectly happy, therefore, to accept Soames’s point in favour of propositions against sentences as a remark upon the shape of our colloquial concepts. Even so, my aim in this paper is to seek a deeper and theoretically richer conception of our representational capacities that leaves behind sentences and propositions alike.

 

d) The propositional attitude objection[20]

Perhaps the chief argument for propositions is that they capture a sameness of meaning relation over sentential differences in the intensional contexts of mental state reportage. Indeed, as reflected in the vernacular, propositional attitude, propositions are often defined just as what serve as the ‘objects’ of our cognitive states. If one were to claim that EXPs, qua being able to fill the role of propositions here, are propositions, then I would take the issue to be merely semantic, for EXPs, as detailed, depart markedly from what is usually meant by ‘proposition’.[21] An ‘eliminativism’ about propositions, however, does not amount to an eliminativism about cognitive states such as beliefs, judgements, etc. (inclusive of speech acts: assertions, statements, et al.) - all those things which we are normally interested in per the first objection above.[22] Let us agree that it is a desideratum on any candidate substitute for sentence/proposition that it provide representations for cognitive states.

        The first quandary a non-propositional account faces is what we may call the absent structure problem. Let first consider sentences, for their falling foul of this problem is taken to militate quite strongly for propositions. It is prima facie coherent to think of sentences, however defined, as the objects of sayings, assertions, and other speech acts because there is some sentence in use. For sure, the mere tokening of a sentence is not sufficient to make it fit to serve as such an object. All the same, a sentence is there as a candidate. When we turn to cognitive states - I shall restrict myself to belief - we typically do not even have a sentence as a candidate. Indeed, a sentence is available to bear the truth of a belief just if we linguistically express the belief. Yet our intuitions on this matter suggest that most of our beliefs go unsaid (e.g., Dennett, 1978, distinguishes between beliefs - states that govern actions - and opinions - linguistic expressions of beliefs.) The question, then, is: What bears the truth of all those beliefs we do not express?

         The sententialist, of course, is entitled to claim that for every potential belief, there is a potential sentence token that articulates its content, which can thus be  understood to bear its truth. This position appears to be the one favoured by Field, 1978, where the sentential objects of S’s beliefs are understood (roughly) to be those sentences S would be disposed to assent to.[23] This claim, though, seems to be beside the point. In line with Dennett’s distinction, we take people to have beliefs whether or not they express their ‘opinions’; indeed, people who are incapable of uttering anything whatsoever are still treated as believers. So, while the truth conditions of each potential belief attribution may be analysable on the basis of sentence tokens, the truth of the beliefs themselves is not supported by the truth of sentence tokens.[24]

        There is no clear route by which the traditional sententialist may escape from this problem: sentence tokens just do not march in step with beliefs. On the other hand, the EXP model is faced with no such problem. Sentence tokens, I have suggested, count as EXPs via projection from cognitive structures realised in the mind/brain of the producer/consumer of the token; it is the internally represented structure that supports truth ascription, not that onto which it is projected. This model, then, does not entail that a sentence token  must accompany each belief, nor even that an agent be disposed to utter anything whatsoever. What it does entail is that the truth values of our belief attributions are inherited from the semantic structure realised in EXPs, which may or may not find expression in utterances. Therefore, the absent sentence problem has no purchase here, for the complaint is wholly concerned with the truism that beliefs are tokened without a tokening of a sentence. This truism is quite consistent with the EXP model.

         The second and more fundamental problem that is taken to cripple sentences and so, therefore, support propositions, is that any sententialist individuation of cognitive states will be too fine. Sentences, under whatever construal, are finely individuated in the sense that two tokens count as belonging to distinct types even though they may ‘say’ the same thing. Thus, sentences are more finely individuated than propositions. Indeed, the EXP model has a hyper-fine individuation; it differentiates between lexically and phonetically identical structures - it is as fine as syntax, which can include elements that have no phonological properties at all (witness the discussion of (1).) Such fineness of cut, however, does not sit comfortably with the claim that EXPs may serve as the structures our cognitive attitudes cover. Identity of belief appears to be comparatively coarse in that sameness of that-clause meaning appears to entail belief-type identity (Church, 1950; Scheffler, 1955, Schiffer, 1987). For example, John asserts ‘Snow is white’ and Johann asserts ‘Der Schnee ist weiss’. Do they have the same belief? Well, since translation preserves meaning and ‘Snow is white’ translates ‘Der Schnee ist weiss’, it seems that they do. We, after all, would report Johann as asserting/believing that snow is white (in those terms). But if belief (or assertion) were sententially individuated, then it would come out that John and Johann  believe different things. It follows that our report on Johann would be false, surely a wrong result.

       Now this observation might cause problems for the view that belief is a relation between believer and E-sentence (but see Field, 2001, Chp.5). Yet even if it were to, it would not follow that propositions are individuative of cognitive states; in particular, a refutation of traditional sententialism is not perforce a refutation of the EXP model.

      I said above that belief individuation appears to be coarser than sentence individuation. The judgement that this is so, though, is perhaps based more on examples than on a thorough survey of the factors to which belief attribution is sensitive. Content is what individuates beliefs, it thus seems reasonable to say that content just is the proposition (or meaning) expressed by the that-clauses in belief attributions. What this thought  ignores are the many dimensions along which belief attribution, and so identity, is sensitive. Larson and Ludlow, 1993, and Larson and Segal, 1995, Chp. 11, suggest that belief attribution is sensitive along dimensions we would not readily classify as determinants of the meanings of the sentences in the clausal positions of belief attributions, e.g., spelling, pronunciation, intonation contour, as well as lexical, syntactic and semantic properties. For example, if we report on someone’s ironic utterance without ourselves marking the irony, we would be justly accused of false reporting (just so for typical interrogative, declarative and imperative intonation, which, familiarly, can all be carried by the same structure). We might also predict differences of belief between English and American English speakers upon their reading, ‘The sky was the color of coal’. An English speaker might read the sentence under poetic licence and believe that ‘color’ refers to texture, say. Thus, we may say that the individuation of belief is at least as fine as the individuation of the clausal complements, inclusive of their syntactic, semantic, lexical and phonetic properties. Now a structure realising such a cluster of properties is precisely what an EXP is. It thus seems, ceteris paribus, that EXPs can be individuative of belief, for they are not too fine. Quite the contrary, such fineness is precisely what is required to account for the hypersensitivity of belief attribution.[25]

          The problem remains of how to account for those cases where, far from being sensitive to linguistic factors, belief attribution appears to be wholly insensitive to linguistic form, as in the example of John and Johann. Such examples appear to show conclusively that the EXP model is not sufficient to type identify beliefs. In one sense, this appearance is veridical, for we clearly do want to say that, in some sense, ‘Snow is white’ and ‘Der Schnee ist weiss’ say the ‘same thing’; such is why content individuation is only at least as fine as EXP individuation. Yet I submit that this sense of insufficiency does not have a crucial bearing on the question of individuation. Examples of the John/Johann form reveal that our reporting of what another believes or says, at least in indirect speech, is not wholly sensitive to the linguistic form (if any) of that  on which we are reporting. This fact is familiar from our use of pronouns in discourse changes of person: A reports on B’s utterance, ‘I am ill’, by saying that he is ill; similarly, A reports that B believes that he (= B) is under the weather today, the next day A’s report is that B believes that he was under the weather yesterday. In these examples the reporter understands herself to be expressing (or saying) the same content as the reportee. Let us agree that it is this relation, what Davidson, 1968, dubs samesaying, that must be preserved by an account of content; this is the desideratum revealed by the examples.

          The Church/Scheffler/Schiffer observation we are considering holds that this desideratum may be satisfied only if there is a single representation between samesayers to bear the truth of their beliefs, assertions, etc. Traditionally, this kind of constraint is understood to license extra-linguistic entities - propositions - to which sentences stand in some meaning constituting relation. Yet the desideratum manifestly does not demand the identity of the representations; in particular, the mere fact that our mental reportage is in some respects insensitive to linguistic form, does not mean that structures as fine as EXPs are not the structures of belief. Following this lead, we can say that two EXPs realise a samesaying relation just if the first can be employed to report accurately on the second tokening. The constraint of accuracy is crucial: the truth of our reports on another depend upon a range of variable factors, those very factors that individuate beliefs. In some cases it may be that extensional equivalence is sufficient to establish an accurate report, in other cases it may be that we have to use the reportee’s very words, such, after all, is partly why we adopt foreign words. Consider, then, John and Johann who both simply state facts; here our report on what John says can double as a report on Johann’s assertion, because a semantic interpretation is shared, that is, what both speakers say is true just if snow is white (mutatis mutandis for examples involving indexicals). Sameness of semantic interpretation is not always sufficient to realise a samesaying relation; the relation depends upon the accuracy of one’s reporting, which in turn depends upon the sensitivity of the individuation of the state reported. For example, if I say that Berlioz was a restrained composer with a rising intonation contour marking irony, it would be inaccurate to report, ‘J.C. believes that Berlioz was a restrained composer’ without, in some sense, marking the irony yourself; only then would we be samesayers. Indeed, we may count as samesayers if you report, ‘J.C. believes that Berlioz was not a restrained composer.’ Far from scuttling the EXP model, the shape of our indirect speech idioms lends it weight, for the model offers an explanation of the sensitivity of mental reporting which a simple appeal to propositions leaves untouched. Such is another virtue of the EXP model.[26] 

       In the preceding discussion of the propositional attitude  objection, I have assumed much and been equally prodigal with promissory notes. My brief, however, is not comprehensively to justify the EXP account of cognitive states, but only to demonstrate that the considerations taken to weigh in favour of propositionalism have no purchase on the EXP model. The model, for sure, faces a host of problems and counter hypotheses, some of which I have considered elsewhere; yet such competition is the fate of any theory.

6: Conclusion

I have argued that internal structures paired by the output of the language faculty   should replace our familiar notions of sentence and proposition for all theoretical purposes. This position depends upon a cognitive conception of language. This conception is certainly not a priori mandatory, but it is not supposed to be. It is an empirical approach to the real phenomenon of linguistic competence. Compared to it, the common E-conception of language is an otiose, shallow abstraction, and propositions are a needless reification. If, then, we grant the primacy of the cognitive approach, the EXP model naturally follows from it, for EXPs are the bearers of linguistic meaning.[27]

 

 

Notes



[1] An anonymous referee suggested that ‘language’ construed externally need not be extensional, for languages might be understood to contain their sentences (plus their meanings) essentially. Such a position is adopted by, e.g., Soames, 1984. I think that the point is correct: externality and extensionality may come apart. That said, the salient point is that an E-language is individuated independently of the cognitive capacities of individual speakers, whereas, as we shall see, an I-language just is the generative function that produces linguistic structures. Suffice it to say that none of the arguments to follow rest upon a necessary link between externality and extensionality.

[2] Although such approaches are still attractive to many, they were, to my mind, demolished in Chomsky, 1955/75.

[3] For such Platonism, see Katz, 1981, and Higginbotham, 1989, 2001 (among others). Unlike the substitutional model, Platonism has not been refuted; indeed, it is somewhat difficult to see what would falsify it. Even so, the motivation for it is not easy to appreciate. One requires a psychological conception of language to explain its acquisition and its retention in mature competence, but not only does Platonism offer no enlightenment here, it makes it wholly obscure how our utterances may partake of the grammatical forms; that is, it threatens to make the facts of language acquisition unrealisable. Certainly for Higginbotham, it seems that grammar is Platonic just because the language faculty is representational, and so requires an independently constituted represented object, viz. a grammar.  Per internalism, language is an internal structure, it is not an independent thing that is represented (see below).

[4] The semantic properties at issue are inherent features of the lexical items and will minimally include thematic structure (not necessarily grids, see Chomsky, 2001a) of the familiar kind: cause, function, patient, etc. We may also assume a far richer qualia structure of the kind investigated by Pustejovsky, 1995. The merging of items will also produce structural features of scope, focus, topic, etc. We may speculate that these two divide into intentional/conceptual, see below.

[5] The minimalist program originates from Chomsky, 1995 (especially see chps. 3, 4). For overviews of the program, see Chomsky, 2000a, Chp.1, 2000b, 2002. For a deep but accessible overview in relation to other fields, see Uriagereka, 1998.

[6] This model might be called first-wave minimalism, and is best represented by Chomsky, 1995.

[7] This model is articulated in Chomsky, 2001b, and intimated in Chomsky, 2000c, 2001a. For similar proposals, which differ from Chomsky’s in matters of detail, see Epstein, et al., 1998, Uriagereka, 1999, Epstein1999, and Epstein and Seely, 2002.  

[8] Here I assume what Uriagereka, 1999, dubs the “radical” thesis which contrasts with a “conservative” one, under which a phase post transfer remains “frozen” in the derivation for further merge, but cannot project, take further lexically dependent arguments. (Uriagereka speaks of a “cascade” instead of phase transfer, but they mean approximately the same thing) See n.9.

[9] Chomsky does not insist that phases must be CP or v*P (see Chomsky, 2001a, p.12; 2001b, p.5, n.16), but thinks it reasonable that only such objects of dual phonological/semantic “integrity” are apt to undergo simultaneous transfer. For the authors mentioned in n.7, any merged structure can count as a phase, certainly by the standards of search complexity. While this position makes sense for a phase spell out of PHON, I can’t see how it would work for SEM (see below).  Wolfram Hinzen is sceptical of my semantic assumptions here. I agree with him that what should fall under ‘semantics’ is an empirical issue not to be stipulated. I’m prepared to settle for: the extent to which external systems impose ‘sentential/propositional’ conditions on the faculty is the extent to which phases will be propositional in the technical sense. Either way, we are obliged to make certain working assumptions about the semantics side of the interface.  

[10] Here I follow the proposal of Chomsky, 2001b, 2002, that SEM divides into intentional and conceptual parts. We may take these two aspects to determine interface conditions which the computation must meet. Per minimalist principles, we take the computation to reflect the conditions in terms of two species of Merge: external, which pairs two objects to determine intentional theta-theoretic configurations, and internal, which merges a part of an object to itself (‘move’), determining conceptual properties of scope, discourse effects, etc.

[11] For a truth-conditional semantics for LF, see Higginbotham, 1985, and Larson and Segal, 1995. Higginbotham explicitly understands semantic properties to relate LF structures with external, extra-linguistic objects. This contravenes the internalist perspective, as understood above. Larson and Segal are much less explicit and may be read in a more internalist manner. See below.

[12] Although the bare idea of UG is an empirical hypothesis, it is fair to say that it has no serious competitor;  the interesting empirical questions are internal to the UG model. There is, of course, much more to linguistic theory than generative grammar. Insofar as linguistics is concerned with cognitive capacity, however, it is certainly arguable that, from a distance, the major theories are really different ways of saying much the same thing (see Newmeyer, 1987.)

[13] For I-conceptions of human mentality in general, see Jackendoff, 1992,  and Chomsky, 2000a.

[14] Empty categories are items of a structure that are not phonologically realised; their nature and conditions of occurrence are a matter of controversy, but the general thought that not all items of a structure are mapped to PHON is not. Ontologically speaking, the only place for them to reside is in the mind of the speaker/hearer. See Haegeman, 1991/94, for an overview of the traditional taxonomy of empty categories within the GB framework. See, e.g., Hornstein, 1995, for a discussion of empty categories in recent minimalist theory; I keep with the traditional ‘PRO’ below for purposes of exposition. In terms of a general defence of I-languages, the difference is unimportant.

[15] The indexes indicate co-reference. They are a notational convenience; they are not part of the syntax.

[16] As described above, the E-conception may afford a notion of grammatical structure that can serve to disambiguate common forms such as They are flying planes. For the reasons given, however, it is difficult to see these proposals as competitors to the I-conception.

[17] This somewhat tortuous way of putting things is intended to capture the idea that EXPs are not represented in the mind, as if syntactic structure were out there in the world; rather, they are part of the structure of the mind, yet individuated in terms of sui generis properties. 

[18] Field, 1994, 2001, Chp.5, proposes an idiolect-relative conception of truth, which, he argues, supports an a priori status for DS and, correlatively, TM. I think that this model of truth faces serious difficulties (Collins, 2002), although I do agree with Field that propositions are not necessary to capture the proposed connection between truth and meaning.

[19] I am unsure how Soames would respond to this proposal. Soames, 1999, pp. 258-9, n. 21, rejects, rightly in my judgement, the view that truth bearers are pairs of syntactic objects and meanings. EXPs, however, are not such pairs. Moreover, the pair proposal does not embody the cognitive notion of knowledge that is carried by the EXP conception. 

[20] The points to follow have some similarity with those made by Harman, 1973, and Field, 1978. There are differences, though. Most importantly, Harman and Field posit a ‘language of thought’ to accompany, but in contrast to, public language (E-language). On the understanding of this paper, language just is the representational structure of the language faculty; public signs or utterances are theoretically otiose.

[21] King, 1995, 1996, for one, takes LF (of some description, presumably GB) to provide the structure for propositions, where the proposition is the composed semantic value of the lexical items of an LF representation. Here, full syntactic structures (including lexical items) serve as the “vehicles” of propositions, although ones which essentially contribute  to proposition individuation without exhausting it (LF or its like is a necessary condition for proposition individuation). The main motivation for this position is to establish that the ‘objects’ of thought are structured as opposed, say, to sets of possible worlds. King does not consider the view that we can simply let EXPs serve in place of propositions. That is, it remains unclear on King’s view why propositions are not merely redundant reifications of aspects of EXPs. Why should we reify when EXPs themselves provide the required structured ‘objects’? Why should we think of the mental structure as a mere vehicle when it essentially contributes to the ‘proposition’? Indeed, the individuation conditions on ‘propositional attitudes’ suggests that the reification is a retrograde step as the object is far too course, i.e., they can serve as the ‘objects’ of thoughts only if they essentially include the full features of an EXP, including PHON.  Ludlow, 1999,  has also argued, on the basis of an A-theory semantics of tense, that the objects of the attitudes must have values which change over time, i.e., they cannot be eternal or immutable, properties which characterise propositions  (especially see p.50). In sum, King’s reification of propositions appears to be simply a way of saving the notion of a proposition, a needless abstraction once EXPs are acknowledged (cf., Higginbotham’s position on language; see n.3).

[22] Thanks go to an anonymous referee for suggesting that my position is best expressed in terms of a propositional eliminativism. 

[23] By Field’s, 1994, 2001, chp.5, later deflationary view, content is a ‘deflated’ (non-truth-involving) notion, based (principally) on idiolectic inferential role. I agree with Field that content attributions do not answer to linguistically independent propositions, but I do not share his deflationary inferential role account of meaning. It seems clear that the semantic properties mapped onto syntax that enter into EXPs are not exhausted by inference, however broadly conceived (wide or narrow). Indeed, the semantic properties explain our inferential competence, they are not constituted by it.

[24] See Burge, 1986, and Schiffer, 1987, Chp. 5, for this form of argument against an extension of Davidson’s, 1968, utterance based account of indirect speech to mental states in general. Lepore and Loewer, 1990, p.99, correctly point out that Davidson’s general paratactic proposal does not rely on the demonstrated element being an utterance. (See below for a Davidson-style proposal without the parataxis.) 

[25] The theory that the objects of cognitive states are interpreted logical forms (a cousin, as it were, of the current account) has been argued for by Higginbotham, 1986, Larson and Ludlow, 1993, Larson and Segal, 1995, Segal, 1998, and Collins, 2000. Not all of these positions are explicit on whether the syntactic structures are to replace propositions or be the ‘vehicles’ for propositions, as in King’s approach - here I include myself (see n.20). Larson and Segal, 1995, pp.432-437, are certainly explicit that propositional theories are to be replaced, but the other authors, to my mind, leave the question hanging. Given Higginbotham’s general Platonism (see n.3), he might see propositions as abstract linguistically structured entities that are represented by the mind, with LF-like structures being their internal vehicles. For the reasons already provided, I think we should reject such a position.

[26] For a discussion of the cognitive, semantic and pragmatic issues raised by the notion of samesaying see the references in n.25. One problem with the way the interpreted LF notion of samesaying is often presented is that phonological features are (inadvertently) precluded, even though attitude ascription is clearly sensitive to phonology. By definition, LF is a level of syntax independent of PF, the level that codes for sound; consequently, phonological features cannot occur in the interpretation of LF. EXPs smoothly avoid this pitfall, for, even if we operate with a LF level (as opposed to phase transfers), an EXP is not simply an interpreted LF representation, a pair <LF, SEM>, but is, rather <PHON, SEM>. Alternatively, we could express the pair as <PHON(PF), SEM(LF)>, where the respective members designate the mapping of properties onto the interfaces of the faculty (cf. Segal’s, 1999, notion of a Total Form as a replacement for ILFs). On this model, PHON(PF) may not be relevant to all ‘samesaying’ relations, just as the difference between an active and passive construction is only sometimes relevant to whether speakers count as samesayers (e.g., note the difference between, The Japanese build good TVs and Good TVs are built by the Japanese.)

[27] My thanks go to two anonymous referees for useful criticisms of an earlier draft. I am especially grateful to Wolfram Hinzen for general insight and his forcing me to be much clearer about my syntactic assumptions; the article owes much to his persistence. My thanks also go to Noam Chomsky for his intellectual and material contribution.

References

Alston, W.: 1996, A Realist Conception of Truth, Ithaca, Cornell University Press.

Barwise, J. and Etchemendy, J.: 1987, The Liar: An Essay on Truth and Circularity,

      Oxford, Oxford University Press.

Brandom, R.: 2000, Articulating Reasons: An Introduction to Inferentialism,

      Cambridge, MA., Harvard University Press.

Burge, T.: 1986, ‘On Davidson’s “Saying That”’, in E. Lepore (ed.), Truth and

      Interpretation: Perspectives on the Philosophy of Donald Davidson,

     Oxford, Basil Blackwell, pp.190-208.

Chomsky, N.: 1955/75, The Logical Structure of Linguistic Theory, Chicago,

     Chicago University Press.

Chomsky, N.: 1986, Knowledge of Language: Is Nature, Origin, and Use, Westport,

      Praeger.

Chomsky, N.: 1995, The Minimalist Program, Cambridge, MA, MIT Press.

Chomsky, N.: 2000a, New Horizons in the Study of Language and Mind, Cambridge,

      Cambridge University Press.

Chomsky, N.: 2000b, The Architecture of Language, New Delhi, Oxford University

      Press.

Chomsky, N.: 2000c, ‘Minimalist Inquiries: The Framework’, in R. Martin, D.

      Michaels, and J. Uriagerka (eds.), Step by Step: Essays on Minimalist Syntax in

      Honor of Howard Lasnik, Cambridge, MA, MIT Press, pp.89-155.

Chomsky, N.: 2001a, ‘Derivation by Phase’, in M. Kenstowicz (ed.), Ken Hale: A

      Life in Language, Cambridge, MA, MIT Press, pp.1-52.

Chomsky, N.: 2001b, ‘Beyond Explanatory Adequacy’, MIT Occasional Papers in

       Linguistics, No.20, pp.1-28.

Chomsky, N.: 2002, ‘An Interview on Minimalism’, in On Nature and Language,

       Cambridge, Cambridge University Press, pp.92-161.

Church, A.: 1950, ‘On Carnap’s Analysis of Statements of Belief and Assertion’,

       Analysis 10, 97-99.

Collins, J.: 2000, ‘Theory of Mind, Logical Form and Eliminativism’, Philosophical

       Psychology, 13, 465-90.

Collins, J.: 2002, ‘Truth: An Elevation’, American Philosophical Quarterly, 39, 325-

       342

Davidson, D.: 1968, ‘On Saying That’, Synthèse 19, 130-46.

Dennett, D.: 1978, Brainstorms: Philosophical Essays on Mind and Psychology,

       Cambridge, MA., MIT Press.

Devitt, M., 1984/91, Realism and Truth (second edition), Oxford, Blackwell.

Dummett, M.: 1999, ‘Of What Kind of Thing is Truth a property?’, in S. Blackburn

       and K. Simmons (eds.), Truth, Oxford, Oxford University Press, pp. 264-82.

Dummett, M.: 2002, ‘Two Faces of the Concept of Truth’, in R. Schantz (ed.), What

       is Truth?, Berlin, de Gruyter, pp.249-263.

Epstein, S.: 1999, ‘Un-principled Syntax and the Derivation of Syntactic Relations’,

       in S. Epstein and N. Hornstein (eds.), Working Minimalism, Cambridge, MA,

       MIT Press, pp.317-345.

Epstein, S., Groat, E., Kawashima, R., and Kitahara, H.: 1998, A Derivational

       Approach to Syntactic Relations, Oxford, Oxford University Press.

Epstein, S., and Seely, T. D.: 2002, ‘Rule Applications as Cycles in a Level-Free

       Syntax’, in S. Epstein and T. D. Seely (ed.), Derivation and Explanation in the

       Minimalist Program, Oxford, Blackwell, pp.63-89.

Field, H.: 1978, ‘Mental Representation’, Erkenntnis 13, 9-61.

Field, H.: 1994, ‘Deflationist Views of Meaning and Content’, Mind 103, 249-285.

Field, H.: 2001, Truth in the Absence of Fact, Oxford, Oxford University Press.

Haegeman, L.: 1991/94, Introduction to Government and Binding Theory (second

      edition), Oxford, Blackwell.

Harman, G.: 1973, Thought, Princeton, Princeton University Press.

Higginbotham, J.: 1985, ‘On Semantics’, Linguistic Inquiry 16, 547-593.

Higginbotham, J.: 1986, ‘Linguistic Theory and Davidson’s Program in Semantics’,

      in E. Lepore (ed.), Truth and Interpretation: Perspectives on the Philosophy of

     Donald Davidson, Oxford, Basil Blackwell, pp. 29-48.

Higginbotham, J.: 1989, ‘Elucidations of Meaning’, Linguistics and Philosophy 12,

      465-517.

Higginbotham, J.: 2001, ‘On Referential Semantics and Cognitive Science’, in J.

      Branquinho (ed.), Foundations of Cognitive Science, Oxford University

      Press, Oxford, pp.145-156.

Hornstein, N.: 1995, Logical Form: From GB to Minimalism, Oxford, Blackwell.

Horwich, P.: 1990, Truth, Oxford, Blackwell.

Jackendoff, R.: 1992, Languages of the Mind: Essays on Mental Representation,

       Cambridge, MA, MIT Press.

Kaplan, D.: 1978, ‘On the Logic of Demonstratives’, Journal of Philosophical Logic

       8, 81-98.

Katz, J.: 1981, Language and Other Abstract Objects, Totowa, NJ, Rowman and

       Littlefield,

King, J.: 1995, ‘Structured Propositions and Complex Predicates’, Noûs 29, 516-535.

King, J.: 1996, ‘Structured Propositions and Sentence Structure’, Journal of

       Philosophical Logic 25, 496-521.

Larson, R. and Ludlow, P.: 1993, ‘Interpreted Logical Forms’, Synthèse 95, 305-355.

Larson, R. and Segal, G.: 1995, Knowledge of Meaning: An Introduction to Semantic

     Theory, Cambridge, MA, MIT Press.

Lepore, E. and Loewer, B.: 1990, ‘A Study in Comparative Semantics’, in C. A.

     Anderson and J. Owens (eds.), Propositional Attitudes: The Role of Content in

     Logic, Language, and Mind, Stanford, CSLI., pp. 91-111.

Lewis, D.: 1975, ‘Languages and Language’, in K. Gunderson (ed.), Minnesota

      Studies in the Philosophy of Science, Vol. VII: Language, Mind and Knowledge,

      Minneapolis, Minnesota University Press, pp.3-35.

Ludlow, P.: 1999, Semantics, Tense, and Time: An Essay in the Metaphysics of

      Natural Language, Cambridge, MA, MIT Press.

Newmeyer, F.: 1987, ‘The Current Convergence in Linguistic Theory: Some

      Implications for Second Language Acquisition Research’, Second Language

      Research 3, 1-19.

Pustejovsky, J.: 1995, The Generative Lexicon, Cambridge, MA, MIT Press.

Scheffler, I.: 1955, ‘On Synonymy and Indirect Discourse’, Philosophy of Science 22,

      39-44.

Schiffer, S.: 1987, Remnants of Meaning, Cambridge, MA, MIT Press.

Segal, G.: 1998, ‘Representing Representations’, in P. Carruthers and J. Boucher

      (eds.), Language and Thought: Interdisciplinary Themes,

      Cambridge, Cambridge University Press, pp. 146-161.

Soames, S.: 1984, ‘What is a Theory of Truth?’, Journal of Philosophy 81, 411-429.

Soames, S.: 1999, Understanding Truth, Oxford, Oxford University Press.

Uriagereka, J.: 1998, Rhyme and Reason: An Introduction to Minimalist Syntax,

      Cambridge, MA, MIT Press.

Uriagereka, J.: 1999, ‘Multiple Spell-Out’, in S. Epstein and N. Hornstein (eds.),

      Working Minimalism, Cambridge, MA, MIT Press, pp.251-282.

Zwart, J-W.: 2002, ‘Issues Relating to a Derivational Theory of Binding’, in S.

      Epstein and T. D. Seely (eds.), Derivation and Explanation in the Minimalist

      Program, Oxford, Blackwell, pp.269-304.