The content of grammar
3.3 Content from composition?
The basic formula of modern compositional semantics might be put as (7): (7) Lexical meaning + syntax = propositional meaning.
But doesn’t this express the very message above, or indeed of this book at large, namely that it is grammar that mediates between the lexicon and propositional meaning? If so, the framework for addressing these issues is already in place and we need to look no further. Addressing this objection will help to bring the claim of this chapter into better focus. The problem, we will now argue, is that how compositional semantics implements the suggestion formulated in (7) recapitulates the original problem, by misidentifying the locus of its solution.
What syntax does on this conception is to assign a part–whole (constituent) structure to an expression. In contemporary syntax, this structure is modelled by Merge, hence through a binary set potentially containing other such sets. Take the sentence (8) and its (vastly simplified) formal generation in (9):
(8) John swims
(9) Merge (John, swims) = {John, swims}
There is nothing in the set {John, swims} that makes it in any way intelligible why the meaning of (8) arises from the independent meanings assigned to ‘John’ and ‘swims’—except that we are now told that the meaning of ‘swims’ is such that when combined with the meaning of‘John’ it yields the meaning of (8). This is not false but it is the problem we started with, and compositional semantics is in this sense explanatorily vacuous. There is not necessarily an ontology of‘properties’ now, of course, but only of‘semantic values’, but this does not change the basic explana- tory problem. If no independent meaning or ontology of the parts is assumed, on the other hand, and their meanings are taken to be abstracted from the meaning of (8) taken as a whole, the explanatory problem becomes even more obvious: now we are deprived, in principle, of an explanation of how the meaning of (8) is strictly a function of its independently meaningful parts (Collins,2011: ch. 3).
Either way we run into the foundational problem noted above: the distinction between an object and a Fregean‘concept’, or an object and a property, we argued, is not an ontological or semantic but a grammatical and relational one. It depends on how a concept or lexical item grammatically functions, which no concept or word as such can predict. We have to wait for grammar to rule the matter one way or the other. There is therefore no semantic value that‘horse’ as such has and that we could insert into the compositional computation of grammatical content. And when it is decided which of these things the word does stand for, this cannot be explained non-relationally by going back to ontology or semantics and positing some ontological or semantic entity that it stands for, in order to then explain what meaning results.
This explanatory problem, which compositional semantics faces when added to a formal syntax as an independent component, is general. Zimmermann (2012) gives the following general account of the practice of compositionality in linguistic semantics. Given a linguistic expression C, its syntactic construction type ˜ is identified, of which two immediate constituents A and B are parts (possibly after changing C’s surface appearance into what is taken to be its ‘logical form’). With that construction type a functionˆ is then uniquely associated that assigns the semantic value of C to the pair consisting of the semantic values of its immediate parts. One of these values will usually be regarded as unproblematic—as when ‘John’ is simply assigned John as its semantic value, an object (a problem, if object-reference is grammatically configured: see Chapter 4). The other value is then defined to be the mapping of this value to the value of the whole. In the case of (8), ˆ will assign a meaning to‘swims’ defined to be the very meaning that maps John to the meaning of (8). But if this function is viewed as a set, say—the set of those objects that are mapped by the function to True if they swim—the definition of the set requires an
understanding of the very predicate in question, depriving the move of independent explanatory value (cf. Davidson,2005).
It is true, then, on both the present and the standard compositional view, that grammar mediates between lexicon/concepts and propositional meaning. But as compositional semantics spells out this intuition, the idea becomes dependent on a notion of‘syntax’ that precisely deprives it of playing a constitutive role in the genesis of propositional meaning, with the consequence that the explication of how propos- itional meanings are generated compositionally is explanatorily vacuous, and formal- ontological distinctions are misidentified as lexical-semantic ones.
Consider, in this regard, again the semantic difference between (10) and (11). As described in Chapter2, taking inspiration from the Modists, this is a difference in the perspective with which we refer to an object, not in what it is that we refer to (the external significatum):
(10) John laughs (11) John’s laugh
The difference between (10) and (11) is not lexical, in the sense that it cannot be explained by what content the lexeme laugh contributes to the expression, since that root is identically present in both. The function of the lexicon, on the account of Chapter2, is to collect perceptual features of environmental stimuli into equivalence classes: but in terms of perception, nothing needs to change when we move between using (10) and (11) on a given occasion of language use. So the difference between them is not lexical. One could always make it lexical, by ‘composing’ the lexeme laugh with an added lexical ‘action-feature’ in (10) and an ‘object-feature’ to (11). But this seems non-explanatory, especially as the process leading to the duality in (10)–(11) is productive, and we would have to distribute such new ‘lexical features’ to an endless set of items in the substantive lexicon, missing the systematicity of the process and that it rather seems to depend on how a given lexeme is grammaticalized on an occasion, in the course of a given syntactic derivation.
Similarly, we observe a difference between the clause Superman left, when it is used in isolation to make an assertion of truth, as in (12), and when it occurs embedded, as in (13), in which case no assertion of truth is made about it by the speaker:
(12) Superman left
(13) Lois believes Superman left.
Instead, (12) as occurring in (13) is interpreted intensionally, namely as referring, in terms of intensional semantics, to a function from possible worlds to truth values, rather than to a truth in the actual world. This semantic difference again does not seem to be due to the words involved, nor to how this lexical material is‘composed’: the string of words in (12) and of the same clause as occurring in (13) at least seems to
be exactly the same. It follows that the difference in meaning just noted must either be due to differences in hidden lexical material (a covert functional head, say: cf. de Cuba,2007), or else a difference in grammatical meaning in our sense, reflecting the content of grammar, which is what we propose here.7The grammar of‘Superman left’ clearly is different in the two cases: for the clause is a grammatical argument in (13), but not in (12). The challenge is to give an explanatory account of this grammat- ical difference that can yield the intensionality effect. We sketch such an account in Section3.5 (see further Hinzen, Sheehan, and Reichard, 2013).
The claim, then, is that grammar and the lexicon contribute to meaning in different ways, inducing different kinds of effects. Intensionality falls on the grammatical side: it is not an effect of the lexicon, nor can it be explained by composing the contents of lexical items—which would explain why it has caused puzzlement and consternation in over100 years of compositional semantics. The same is true for reference as well (as we would expect, if the claim applies to intensional interpretations, which in the terms of Chapter4 is one weak way in which clauses can refer). Thus, consider that no (substantive) lexical item is as such referential. As noted, a common noun such as‘man’ cannot be used to refer to a particular object, such as a particular man, unless it is uttered together with an act of pointing or a demonstrative, in which case the object-reference is contributed by the gesture or the deictic element, not the content word, which rather contributes the NP-restriction:
(14) [Pointing:] man (15) this man
Where it is clear enough which man is meant,‘this’ in (15) can be reduced to the weaker form‘the’, which lacks the locative/deictic element contained in ‘this’ or ‘that’, and referentiality to an individual object will then still be maintained. Referentiality in this case, too, is an aspect of the phrase as a whole, and the act of (either stronger or weaker) deixis (or grammatical pointing) that it encodes within its grammatical context. Reference thus comes from grammar, not the lexical item or its substantive content, which doesn’t refer, but merely supplies concepts that become part of an identifying description. Reference to particular objects, as in the paradigmatic use of (15), is thus not due to composing the referents of lexical items:‘man’, as such, lacks such reference (taken in isolation, it doesn’t tell us even whether it will be used as a noun or a verb); and the word‘this’ has no substantive lexical content to begin with, and equally lacks object-specific reference when used in isolation (as when uttered out of the blue when entering a room, with no identifying description).
7 We argue extensively against the former approach in Sheehan and Hinzen (2011), and return to such
Grammar provides what‘man’ and ‘this’, each taken in isolation, lack: reference to objects, based on a description. It thus doesn’t add to the contents of lexical items, but to their grammatical semantics: how they can be used to refer. Grammar maps lexical concepts/contents, which as such lack reference, to objects of reference. Even where it combines two content-words, say table and top, so as to obtain table-top, we see reference decided in some initial way: whatever a person uses this compound to refer to, it will be a top (of a table), not a table (with a top). This crucial difference is nothing that is specified in either ‘top’ or ‘table’: it comes from grammar (namely, headedness). Even the grammatical operation of modification is thus not symmetric, and it is certainly not free of grammatical meaning. That meaning, however, adds nothing to the lexical features or content of‘table’ or ‘top’. Reference is not more lexical content, it is grammar.
Even a nominal compound, though, still lacks individual-specific reference: table- top, as such, cannot refer to any particular table-top. Modification is not enough. The compound must enter the grammatical template illustrated in (15), and combine with a lexically meaningless item, such as‘the’, rather than any other content word. This remains true for kill Bill, where again two contentful lexical items seem to be combined directly, as in compounds, and again without the mediation of a lexically meaningless word, but where the grammar now generates an argument instead of a modifier. Here, however, Bill stands in for the determiner in the nominal projection, and this is from where its object-referentiality derives (see Chapter4). This also illustrates that the referential effect generated from adding ‘the’ is not due to whatever residual lexical content‘the’ might have, for this putative content is absent in the case of Bill, when it stands in for the determiner.
As a lexical item, ‘man’ also does not refer to a particular set of men which is rather what‘these men’ in (16) refers to:
(16) I don’t like these men.
Clearly, we cannot take the less complex item ‘man’ to denote something that a grammatically more complex item denotes, namely ‘these men’, when the latter inherently contains the former as a proper part. If‘these’ is dropped, ‘men’ comes to stand on its own; but then it still does not pick out a set of men, for which the determiner ‘the’ or ‘these’ would have to be added. Instead it refers generically to anything qualifying as a man as in (17–18):
(17) I don’t like men. (18) Men are brutes.
In (19), in turn, ‘men’ is quantificationally interpreted (meaning whether you have seen some men):
We see, then, that in (14–19), formally different entities, all constrained by the same identifying description, are being referred to: a single given man, a set of men, arbitrary instances of the kind man, and an indefinite number of such instances. The meaning of the lexical item‘men’ cannot account for any of these facts about reference, which depend on how this item functions grammatically: whether it is inflected for Number, the complement of a determiner, the determiner is overt or covert, and the noun is in a referential or predicative position.
It follows that reference, too, like intensionality, is an instance of grammatical semantics, and that combining the lexical contents of words does not account for these differences in reference. The difference between a property and an object, which arises in the mapping of a lexical content, which as such provides an identify- ing description, to an object of reference, is not a lexical difference, any more than the difference between an object and an action is in (10–11). The difference is the contribution of grammar. Content does not only come from the lexicon, but the grammar, too, though it is one of a very different nature in the latter case. This is the second problem we see with semantic compositionality: assuming a meaningless syntax, it has all content coming from the lexicon. But the lexicon as such never entails reference. It only entails what we have called classification in Chapter 2. And just as parts of speech distinctions reflect the functioning of words in grammar, some items in the lexicon, mostly small (‘the’, ‘a’, ‘that’, etc.) reflect grammar as well, and its inherent task: regulating reference. But these items do not account for reference, which depends on whole phrases and how these function in relation to others.
The same applies to even the putatively most ‘directly’ object-referring lexical items, namely proper names, pronouns, indexicals, and clitics, as anticipated in Section2.2. (20) illustrates that proper names can function as predicates:
(20) Was the already centralized French state more likely to produce a Napoleon than the decentralized United States?8
The same is easily possible even for personal pronouns when they bear Accusative case as in (21), or for 3rd-person pronouns, which can marginally take a determiner: (21) I am me. ‘I am the person I am.’
(22) He is a she.
In short, forced in the role of predicate by grammar, commonly referential elements such as proper names and pronouns resort to whatever little descriptive content they have, leaving it to the interpreter tofigure out which salient property precisely ‘being me’, or a pronoun of feminine Gender, evokes. The same is even true, though now
8 N. Ferguson, Civilization, Penguin:2011 (157), speculating about better prospects for democracy in
much more marginally, for demonstratives. Thus we can just about interpret (23), where a person might be commenting on what is going on in the here and now (or perhaps a Hegelian philosopher is creating abstractions from acts of deixis): (23) the this and the that
And we get (24), which might be said by a patient who lost his deictic sense through brain damage but then recovered:
(24) There was no this and no that.
Clitics, too, can stand for predicative nouns, as long as they are not specified for grammatical Person, such as in the Italian (25) (from Longobardi, 2008):
(25) a. Gianni è monarchico e anticlericale; se anche Maria lo/*la/*li/*le fosse . . . Gianni is monarchical and anticlerical; if also Maria CL(3SgM)/*CL(other forms)-were . . .
b. Se Gianni fossete o anche solo se Maria lo/*ti fosse . . .
If Gianni were you or even just if Maria CL(3SgM)/*CL(2Sg)-were . . . Literally every lexical item, then, including deictic elements or traditional‘indexicals’, canfind itself in the role of a predicate. The same applies to the more grammatical elements such as Tense, which are arguably referential and deictic elements as well, and which are also inflectional and not proper lexical items to start with.
The overall conclusion must be that the lexicon has no control over the forms of reference. Rather, the process of reference begins when lexically common nouns, which are always endowed with a descriptive content, become the complement of grammatical elements without descriptive content, creating a unit of grammar that can be referential depending on its grammatical position and the closed-class items involved. With some nominals—demonstratives, pronouns, clitics—the descriptive content is next to nil but not zero, and hence it can be exploited when the grammar is such that descriptive readings are forced. In such cases these items become comple- ments of a determiner. Where they are referentially used, the determiner must remain absent.
The compositional model is thus based on a problematic idea: that the basic primitive of semantics is word meaning, and that word meaning in turn consists in reference to external objects. In the case of such reference, the idea is, nothing really is to be explained—this kind of semantics, as it were, comes for free, and can be assigned to lexical items directly. Words are simply names for things, and names are labels.9 Confronted with the question of where sentence meanings come from, which do not
9 Thus, Frege (1892) effectively defined the class of ‘proper names’ in purely semantic terms, comprising
seem to be objects like tables and chairs, this lexicalist bias, in the absence of any notion of grammatical meaning, then makes a prediction: sentence meaning must somehow arise from word meanings, by ‘composing’ the latter. Since the base notion of meaning is that of ‘standing for’ or externally referring, sentences will now be taken to also ‘stand for’ or refer to external objects—just to objects of a different kind: a Russellian compound, say, thought of as a‘chunk of the world’, consisting of the referents of the words involved, one mapped to a property, the other to an object, say, which are both making up the compound. Meaning, then, is totally non-linguistic: words are simply symbols, and sentences are symbols of other kinds. Their meanings are external objects. A puzzle with the pervasive intensionality of reference in natural language now ensues, which has defined the philosophy of language for more than100 years.
But if we are right, the alleged external meanings simply recapitulate grammatical distinctions, misinterpreting them lexically. Compositional semantics, given its lex- ical bias, bypasses grammar, failing to recognize the categorically different kind of meaning it generates. This different kind of meaning has nothing to do with external objects, as non-linguistically described. Indeed, if the lexicalist model worked, then