The previous chapter dealt with the imposition of afirst grammatical layer on words, which thereby become‘parts of speech’. The emergence of part of speech distinctions cannot be the key to grammar, however. An adjective (e.g.‘male’) may function as a predicate (‘[is] male’), but need not do so (‘the male and the female’). A verb (e.g. ‘fly’) can be a predicate (‘flies’), but need not be (‘Flying is beautiful’). A nominal like ‘a man’ can be used to refer to an object, but can be a predicate (‘[is] a man’), resulting in a truth value when applied to a subject, much as a predicate realized through an adjective or verb. Predication, therefore, is a grammatical category independent of the parts of speech, though it can be realized by them. A verb is also statistically more often a predicate than a nominal, but this does not show that the notion of‘predicate’ can be defined as a ‘verb’, or as ‘either an adjective or a verb’, or as‘either an adjective or a verb or sometimes a noun’. On the contrary, under- standing what an adjective or verb is will in part involve understanding what grammatical roles they play in relation to other parts of speech. ‘Predicate’, we argue, is a term indicating a grammatical relation or function, and it cannot be accounted for by any morpholexical distinctions.
We stress that, in what follows, we will use the term ‘predication’ in a purely grammatical sense, not a broader logical one: namely, to denote the sentential predicate that is applied to a sentential subject, as‘flies’ is applied to ‘Socrates’ in ‘Socrates flies’, yielding a truth value as a result. In this sense of predication, a sentence only ever has one predicate (which again can be realized through different parts of speech, with language-specific differences in which parts of speech can play which roles), exactly as it only ever has one truth value. In the phrase kill Bill, the verb ‘kill’ is also predicated of the internal argument ‘Bill’, but the result is crucially not a truth value: a verb phrase is not evaluated for truth, in the way a sentence is.
Predication, in this sense of a grammatical relation between the subject and predicate of a sentence, is not a part of speech: it relates different parts of speech playing distinct grammatical roles, and hence is not present in either of them. Predication in this sense is not found in morphology, let alone the lexicon. In
morphology, as Di Sciullo (2005) points out, a complex word can, say, encode thematic structure (bank-robber, shop-lifter, etc.), but neither predication, nor quan- tification, reference, and truth can be found within a single word, encoded in its internal structure. Relatedly, a sentence is not a part of speech, although clauses can be arguments, a different matter, and although the temptation has arisen to turn the sentence into a part of speech, say an ‘inflectional’ or ‘extended’ projection of the verb. Of course, predication is not a lexical concept either. There is a word‘predica- tion’, which we can use to refer to the topic under discussion, namely predication: but this topic is the grammatical relation of predication, not a lexical concept.
This chapter, then, is about the transition from parts of speech to grammar proper, which is a far better candidate for a genuine invariance in the species than morph- ology-based distinctions. But what is grammar, so that it can be invariant? To answer this question, we will ask about its content. What is the ‘content of grammar’, as distinct from the content encoded in lexical concepts or parts of speech distinctions? To explain this question, consider how one might answer it after looking at a contemporary textbook on Minimalist syntax. The content of grammar, one would then conclude, consists of a‘computational system’ made up of the following: A. The local relation of (recursive)Merge, where Merge(A,B) = {A,B} and A can be
contained in B or vice versa. B. The long-distance relationAgree. C. Constraints oncomputation. D. Interface constraints.
(A) gives an arguably minimal account of the combinatoriality of language. (B) accounts for the fact that elements in a grammar can enter long-distance relations with other such elements elsewhere in a given structure. (C) prominently includes principles of computational efficiency that are argued in Minimalism to be expected to be operative in any well-designed computational system in nature, and to be indeed factually operative in language (Chomsky,1995). (D) encodes the assumption that what computations take place is constrained by demands imposed by interfacing performance systems in the mind that access the computational system.
As we would conclude from this list, the content of grammar is purely formal. Merge in particular, as noted, is a generic operation that will not tell us anything about what is specific to language. Language-specific content will thus rather come from what Merge applies to or from what enters Agree relations: lexical features, said to make up lexical items, which are conceived as sets of such features in standard Minimalism. These features are morphologically visible in language-specific ways, for example in the form of Case andç-features (Person, Number, Gender), which are taken to play a crucial role in driving the grammatical derivation forward in this framework. But some of these features, too, in some or all of their occurrences, are
commonly taken to be‘uninterpretable’, and grammar is said to be designed so as to ‘check’ and eliminate them through Agree and Merge, before the semantic compon- ent is accessed which is said not to be able to‘read’ them. Constraints on computa- tion (C),finally, are commonly regarded as ‘third factor’ conditions: constraints not due to either experience or UG (Chomsky,2005). Hence they aren’t language-specific and certainly not semantic in nature either. In short, interpretable content and language-specific information comes either from the lexicon or the interfaces—it is external to grammar, and grammar is completely content-free.
Feature-specifications in the lexicon, however, do not seem the right place to account for grammaticality. Moreover, a theory is lacking of which ‘features’ the lexicon contains or fails to contain. Against the above picture, therefore, and consistent with Boeckx’s (2010; 2012) critique of the role of features in Minimalist syntax, we will argue in this chapter against a lexico-centric focus in the theory of grammar, and the implicit contention in Minimalism that grammar is not—or not directly—involved in the generation of content. Note that if semantics is distinguished from syntax as ‘content’ is from ‘form’, we may expect that the programme of a universal grammar will become harder to formulate, not easier. This is because semantics is commonly regarded—in typology as well as (and essentially uncontroversially) in the philosophy of language, where language-specific variation in general is barely discussed—as the invariant part of the organization of human language. So if semantics is independent of syntax in the way that content is of form, then we have no particular reason to expect invariance in syntax: for syntax will now be arbitrary and ungrounded in semantics. Why should such an arbitrary system be invariant? It would seem like an unmotivated evolutionary accident (see also Christiansen and Chater,2008). We could maximize the explanatory force of third factor conditions, so as to make the syntactic system less arbitrary, but this, as noted, will not lead to insight specific to language, and will divorce us from the functions of language.
Thefield of comparative syntax has sought to model variation in syntax through a system of (arbitrary)‘Principles and Parameters’, but as we will discuss in Chapter 5, this programme faces significant challenges today (Newmeyer, 2005; Boeckx, 2011), which has resulted in the suggestion to shift the locus of variation solely to the lexicon and morphophonology (Berwick and Chomsky,2011). Yet, the claim that the essence of universal syntax is recursion or hierarchical organization (or A and perhaps C above) is too weak to be an inroad into the language-specific content of grammar; and it also leaves the principles of semantic content unaddressed, giving grammar no intrinsic semantic rationale and semantics no intrinsic grammatical rationale.
A foundational—even if initially maybe unsurprising—claim in what follows will be that the content of grammar is essentially relational. If grammar is relational, its content cannot be accounted for by any content that is intrinsic to any word or ‘concept’. Indeed, we stress throughout this book that how words function is—apart
from their substantive lexical content or the concept they encode—itself a matter of what grammatical relations they enter into: these change how a given word signifies. Grammar is about relations between words, and about what novel content arises from this: this is grammatical meaning, in our sense. We claim that this content cannot be reconstructed from the contents of words. It does not reside in‘features’, e.g.‘Person’, or ‘ACC’ (for Accusative) either. Instead, theoretical descriptive terms of traditional grammar such as‘Person’ or ‘Case’ refer to grammatical relations or dependencies into which nominal arguments enter. These hold in the case of Person between determiner phrases (DPs) and T/C in the left periphery of the clause, and they show up on the surface of languages as a morphological feature marked on (pro-)nouns, verbs, both, or neither. Similarly, we will argue in Chapter6 that ‘ACC’ is a morphological feature that, where it exists in a language, encodes syntac- tic dependencies between v and an argument DP. The dependencies are part of grammar; the feature itself is not. Grammar appears to contain no uninterpretable features at all (cf. also Sigurðsson,2009; 2012; Kayne, 2008). Consider an example. It makes no difference to the interpretation of German (1) that the 1st Person singular is also marked on the verb, which it isn’t in its English translation (‘I go’):
(1) Ich geh-e
I go-1p
Hence the Person feature on the verb might be said to be‘uninterpreted’. However, if grammar is relational and what is interpreted is not words or features but relations, it makes perfect sense for there to be a relation of ‘Person’ expressing that a given referent (a person) carrying a certain theta-role is the subject of a given sentential predicate, resulting in a proposition involving an event, whose only participant is also the speaker in the speech event, and hence the one asserting the proposition that this event obtains. This (highly complex) relation holding between the DP in subject position and the verbal predicate then shows up in the form of morphological features on the subject, the predicate, both, or neither, in language-specific ways that make no difference to the interpretation just depicted. While there can be no question that Person, as a relation, is interpretable, the same question, however, arises for other syntactic dependencies such as Case-relations, where, as we shall see in Chapter6, the answer is more complex.
A long tradition resists the contention that the content of grammar is relational and that it cannot be accounted for in terms of the contents of words. For example, if the relations in question are denoted by (or the ‘referents’ of ) words, as has been widely assumed since Russell (1903), all content can be reconstructed lexically, and no content needs to be grammatical—if there are relational aspects of grammar that matter to meaning, there are lexical items to denote them. They exhibit these relational aspects as part of their meaning. For example, instead of saying that
‘being a predicate of a subject’ is a grammatical relation, we could say that this is an intrinsic property of certain lexical items, say verbs, or even certain‘concepts’, say blue.
We think this is a mistake and will rather follow Frege in this regard, who, as we will argue, did importantly not hold such a lexicalist view. He rightly saw that terms denoting the grammatical functioning of a given word, and associated formal-onto- logical distinctions in their denotation, co-vary with grammar and its intrinsic modes of signification.1 In particular, the distinction between ‘object’ and ‘concept’ is, at
heart, a grammatical one—part of the intrinsic content of grammar—not a meta- physical one. It arises on occasions of language use, where acts of reference are configured. We develop this claim in Section 3.2, which contains our basic case for grammar having a content of its own.
Even Frege himself, however, sticks to a word-based conception of semantics, as his puzzlement about so-called identity sentences (and intensionality more generally) illustrates. Thus, (2) is supposed to have the logical form ‘a = b’, where ‘a’ and ‘b’ are names, whose referent is an object:
(2) Hesperus is Phosphorus.
If the content of such a sentence is‘composed’ of the meanings of its parts (ultim- ately, words), then of course the content should be the same as that of‘Hesperus is Hesperus’—which it intuitively isn’t. Hence the puzzle that Frege’s theory of ‘senses’—which are equally word-based as well as non-linguistic entities—was meant to solve. But the puzzle disappears if what a word contributes to the meaning of a sentence is its grammatical meaning as well, not merely its lexical meaning: in particular, whether it functions as a predicate or subject, properties that make the relation of the two nominals in a clause always and necessarily asymmetric (see Moro,1997; and Woodard, in preparation).
If the content of grammar cannot be reconstructed lexically, or by somehow putting together (‘composing’) the intrinsic contents that the lexical items have that it contains, the standard claim that meaning in language is‘compositional’ has to be qualified, and we turn to this issue in Section 3.3. In essence, the principle of compositionality as standardly viewed deprives grammar of a content again, at the cost of having to misinterpret it as lexical: for compositional meaning is ultimately solely word-based again. But neither reference, predication, nor intensionality belong to either lexical or compositional semantics. Section3.4 revisits another claim, that the content of grammar could come from an‘interface’ with non-linguistic systems of thought, which of course we are committed to reject here: such systems could never interpret the outputs of grammar, if grammar has a content of its own. If
grammar is productive of a new kind of meaning—reference in its various forms, we contend—then the architecture assuming an ‘interface’ is not correct.
The crucial grammatical relations transpire within grammatical templates that we will here identify with the phases of contemporary Minimalist syntax, reinterpreted in Section 3.5 as the smallest units of grammatical organization and as units of (intentional) reference, which can take a number of different forms in human language, ordered on a scale. The root meaning (and earliest developmental mani- festation) of this notion of reference we take to be (e.g. indexfinger) pointing, which is deictic. By vastly expanding the range of possible reference beyond where our index finger can point, grammar becomes a device for ‘extended’ deixis. Its organization reflects this function. This notion of deictic reference has to be carefully distinguished from a notion of (i) lexical reference or content, which content words and concepts have intrinsically to their lexical specifications and without mediation by grammar; and (ii) functional reference that wefind in other species. Crucially, the invariance of such a system of (extended) deictic reference in the human species is not in doubt, and hence it is a candidate for grounding the invariance that grammar represents.
Finally, this relational reconstruction of grammar is, in Sections3.5 and 3.6, the basis of reconstructing the recursivity of grammar, which is a derived notion rather than a primitive here (Arsenijević and Hinzen, 2012). This reconstruction will explain why grammatical recursion cross-linguistically takes on a specific form and is subject to restrictions, which we argue current Merge-based accounts of ‘narrow syntax’, which leave the matter to the‘interfaces’, neither explain nor predict. The restrictions also do not seem to be due to lexical specifications of the lexical items merged. We suggest that they need to follow from the organization of the grammar itself, or its intrinsic content, and that they do indeed fall out freely from the phasal dynamics as interpreted here.