The content of grammar
3.4 Content from the interface?
That a particular domain of inquiry be studied formally is obviously desirable from a scientific point of view, if that domain lends itself to a formal analysis at all—indeed it is a condition for scientific inquiry. But any such decision is independent of the domain’s particular ontology, and it implies nothing about what this ontology is: the formal analysis of a domain says nothing about the kind of entities it contains. The formal study of grammar, too, is therefore not the same thing as the study of form (if this is taken to be ontologically distinct from‘content’). In a similar way, the formal study of patterns of planetary motion implies nothing about whether planets are
‘formal’ objects or not. Grammar can be a fully material domain of inquiry, much like planetary motion, even though it exhibits formal patterns that can be studied as such, as in the case of planetary motion.‘Form’ and ‘content’ are then not names for things that exist out there, that can be studied as such, and that are separate from one another as domains of inquiry. Everything, rather, can be studied in terms of both its form and its content (insofar as it has formal properties): that is a matter of choice (and reflection), not of ontology.
This contention raises some potential problems about the study of what, in frameworks such as Minimalism, is taken to be a ‘real biological object’: namely, the‘computational system’ of language. That a particular domain is studied formally or computationally does not mean that there is a‘computational system’ that is being studied—if this means that our object of inquiry is a purely formal system consisting of arbitrary manipulations of symbols in abstraction from semantic content. Our object of inquiry, rather, in this biolinguistic domain, is grammar. The petals of the sunflower exhibit a Fibonacci pattern. But it is not ultimately this pattern, but the flower—through the pattern—that biology studies. Grammar, too, has a form as well as a content. That content is, we will ultimately argue in Chapter9, the world— namely, as it appears to us when a grammatical pattern is laid upon it.
The Minimalist Program has brought the attempt to describe grammar in abstrac- tion from its content to a peak, thereby sharpening the question we are asking. It also instantiates but also radicalizes the long-standing attempt to eliminate stipulations in the formalization of grammar. The PSGs of the1950s showed how to characterize, in a descriptively adequate fashion, the order and projection properties of contiguous relations in grammar. Non-contiguous relations were taken care of by the transform- ational component. The latter was later unified with a PSG with a recursive base, which was then in turn simplified by X-bar theory, which stated the laws of phrase structure in non-category specific terms. Transformations were restricted in their descriptive power through general conditions on transformations. Each of these steps implied a radical reduction in the options of variation that can exist in the syntactic system of language as so described, and it already exemplifies what only later came to be called a‘minimalist inquiry’.
At the end of this process, Minimalism stripped PSG down to its bare bones: Bare Phrase Structure (BPS), viewed as built by the simplest conceivable computational operation, now in charge of both contiguous and non-contiguous relations: Merge. The grammatical relations it creates are sisterhood—two syntactic objects are merged into the same set—and containment—a set generated by Merge contains another set generated by Merge. None of these come with any grammatically specific content. Nor does the operation Merge itself, which is insensitive to any grammatically specific material such as part of speech distinctions or grammatical relations in the sense above. Even if the Minimalist claim that this system gives a minimal account of both contiguous and non-contiguous relations in grammar is true, therefore,
grammar has no content on this description that goes beyond the statement of discrete infinity and the analysis of the latter through Merge.10
This is why the content has to come from elsewhere on this account: narrow syntax, which is an abstraction from content,‘interfaces’ with the semantic compon- ent, the‘conceptual-intentional’ (C-I) systems. This embedding in the rest of our cognitive architecture is what gives it a linguistic character, in other words, an interface, which divides form and content; content, perhaps in the form of a compositional semantics in the sense discussed above, is added to the Merge-based system viewed as something that is independent as an object of scientific inquiry. And perhaps, the thought is, the language faculty is ‘perfectly designed’ in light of constraints imposed by this interface, at least the semantic one, since it appears to take priority over the other, the sensory-motor one (‘interface asymmetry’: Chomsky, 2008).
The constraints imposed by the interfaces on the semantic side however are barely investigated and have tended to be of the most generic kind, like the principle of Full Interpretation. This principle maintains that every element of a syntactic representa- tion arriving at the interface must be interpreted by the C-I systems. Since these systems know nothing about grammar, however, any meaning generated by gram- mar must be unreadable by these systems and hence cannot exist. The absence of a notion of grammatical meaning in this framework is thus not a contingent aspect of it, but built into its architecture. If grammar is uninterpretable, its rationale can only be to eliminate uninterpretable (‘formal’) features contained in it before points of access to the interface, which is regarded as the basic principle driving its dynamics (Gallego,2012; Richards, 2012; Chomsky, 2007a; 2008). But since there is no point to these uninterpretable features in the first place, there is no rationale, really, to grammar at all.
Results in comparative cognition have not, on our view, come to lend support to an architecture where a grammar, although minimal in formal terms, is arbitrary and unprincipled in substantive terms, and interfaces with a semantic component, which is non-minimal in formal terms but principled in substantive ones.11This predica- ment is what we predict if even positing‘concepts’, let alone propositions, on the non-linguistic side of the interface, is subject to severe doubts, if‘concepts’ means
10
That language is‘discretely infinite’ is a property it shares with myriad other systems in nature that can be modelled in this fashion, including arithmetic, music, phyllotaxis, and DNA (see e.g. Katz and Pesetsky,2009; Mukherji, 2009: ch. 6; Gimona, 2006; Searls, 2002). Merge, which amounts to discrete infinity, therefore, gives us a minimalist analysis of what we knew before, that human language is recursive.
11 Chomsky (2008: 141) in particular suggests that ‘C-I incorporates a dual semantics, with generalized
argument structure as one component, the other one being discourse-related and scopal properties. Language seeks to satisfy the duality in the optimal way ( . . . )’. No evidence for this hypothesis is provided, and prospects forfinding evidence that such a semantics—in essence, all of thought as it is manifest in the semantics of human languages—is present in non-linguistic systems of thought appears slim.
anything like the meanings of human words. Even the most optimistic accounts of ‘animal thoughts’ acknowledge that a gap with the kind of thought expressed in language remains: a chimpanzee, or a zebrafinch, do not think like us. There is no common metric by which we can compare non-linguistic thought with linguistic thought in any systematic fashion, however, so as to assess how much of the latter there is in the former. No one has studied whether non-human animals can process the contents of verb phrases, say, but not sentences; or of non-finite clauses, but not finite ones (which would mean to use a linguistic metric, given the absence of a non- linguistic one). For all we can tell, there appears to be no phrasal organization of meaning at all, which also is arguably not found in the communications of the ‘language-trained’ apes (Tomasello, 2008). Nor is there evidence for a semantics of this nature in animal thought rather than communication (Fitch,2005).
This is again not to deny that the formal-ontological notion of an ‘object’ is partially given in perception—and again not to affirm that, thereby, a creature with such perception has thoughts about objects involving the concept of‘object’. Con- sider in this regard the processing of a voice, an auditory object that is the perceptual correlate of some acoustical pattern that reaches our ears. This pattern is cognitively analysed and different perceptual categories are imposed on it, which might involve the recognition of the voice as human, or as that of a particular person (a source), in which case certain structural invariants would be used for categorization corres- ponding to characteristics of the speaker’s vocal tract. It might also involve the recognition of a vowel as associated with a particular pattern of change that can be produced by different possible sources and representing an invariance across differ- ent auditory events and sources. What comes to be perceived in each of these cases is an auditory object involving an extraction of abstract properties from a given auditory input. These are meaningful for us and required for the perception of the ‘same’ voice at different times and in different states of vocal health, or of the same vowel or melody produced by different persons at different times. It will also typically involve sensory integration, as when the perception of a voice is integrated with the visual processing of a face. The emergence of a unified auditory object, from low-level temporal and spectral (frequency) analyses, is eventually helped by the attribution of ‘meaning’ in the sense of triggering a particular concept of what is being perceived.12Perceptual objects in this sense, however, even as analysed by meaningful categories, never involve a relation of reference, since none is required, given how the representations generated in this domain are rooted ipso facto in the sensory context. Moreover, before we apply the notion of object to
12 Stages before and after attribution of meaning appear to be neuro-anatomically distinct (Griffiths and
perception naively, we should be aware that the very notion of‘perceptual object’ is also quite problematic, with some frameworks explicitly rejecting this ontology.13
We will continue to assume, therefore, that perceptually experiencing the world as structured into objects falling into various more abstract classes is very different from an ability to deliberately refer to such objects symbolically. As we suggested in Chapter2, the reason that we can refer to objects to which we are not exposed in ‘online’ perception is that we have concepts—the meaningful categories that we apply to perceptual objects—which as such characterize possible objects of reference. So acts of reference are anchored in concepts, and such acts can succeed even when they are based on little more than that, as when I refer generically to kings of France: nothing, in this case, anchors the act of reference except the descriptive content of the phrase kings of France (and how it functions grammatically). Reference can be less generic and more specific than that, yet some residue of descriptive content is never completely absent, as noted even for the case of pronouns and indexicals above. This appears to be the crucial difference to functional reference in animal alarm calls, which do not tap into abstract conceptual resources in addition to perceptual classifications. No concept is needed.
When an amodal concept evolves from a modality-specific or multi-modal per- cept, the concept is not‘indexed’ to the context of immediate sensory-motor pro- cessing anymore. The connection thus has to be reinstated (cf. p. 47 above). But perception cannot do this, since grammar, when accessing concepts and grammati- calizing them, can establish acts of reference that cannot be linked to any percept, even in principle. Grammar, therefore, needs to contain elements and procedures that re-indexicalize the novel contents that it generates. This is a sensible rationale for the existence of traditionally so-called ‘indexical’ expressions in grammar, such as pronouns and demonstratives. It also rationalizes the grammatical processes that license these items in appropriate positions where they take on interpretable gram- matical functions (for example movement into subject position to become the subject of a predication in the present sense, i.e. a truth-value bearing expression), which involves Case or Person relations. Much of the machinery of grammar, in short, serves to implement the re-indexicalization that is required in acts of reference based on concepts as distinct from percepts.
The transition to grammar is thus the transition from sensory exposure to the world and conceptualization to reference to this world, itself an indication of thought. Our thoughts can now be wrong. That again is an indication of thought, as opposed to dealing with the world perceptually and practically. This possibility of falsehood, however, only arises at the end of the grammatical process: it is not an
13 In the psychophysical model of Bregman (1990), for example, the fundamental elements of the
auditory world are auditory‘streams’, rather than objects, and the concept of an auditory object is explicitly rejected. For discussion, see Griffiths and Warren (2004).
option yet at the level of object reference or the nominal phrase, where all such reference comes with a presupposition of existence, as crucially distinct from asser- tions of existence (as in‘there exists a king of France’), which of course can be wrong. Acts of reference to objects as existing (‘the king of France’), by contrast, cannot be: no truth value can be assigned to‘the king of France’. In this sense there is, in the case of reference to objects in grammar, ‘immunity from error’: error, like falsehood, requires the sentence. Someone who knew that there wasn’t a king of France couldn’t point this out to a person who didn’t know it by using a noun phrase.
So questions of existence are not decided or resolved at the level of nominal reference to objects, and no such resolution is a precondition for symbolic reference in grammar. But that Superman, by some non-grammatical (such as physical) criterion, does not exist, is a consideration irrelevant to the Superman aficionado who is infatuated with his (or her) hero and refers to him, as something existing, by
the name of ‘Superman’. In a less affected mood, he may agree with us that
‘Superman does not exist’, which would be contradictory if using this proper name implied existence or a claim to existence. Language moulds our sense of reality, creates a world, rather than merely responding to one.‘Reference to non-existents’ is no anomaly of language, but its hallmark, and no motive for re-interpreting expres- sions as non-referential ones when a presupposition of existence fails.
The absence of the phenomena of both reference and truth in perceptual systems or animal cognition confirms our stance, assuming the essential absence of substan- tive constraints on grammar as coming from postulated C-I systems, and the absence of independent predictions from a ‘semantic component’ for what we will find in grammar. Given the emergence, with grammar, of a novel referential-deictic system distinct from both perception and classification, the conclusion should be that there is no system there on the C-I side of the interface, equivalent in semantic power to a grammar-based system, which could then constrain the grammatical system. In fact, as noted we predict that whatever non-linguistic system exists on the other side of the posited interface, couldn’t even systematically interpret the gener- ations that arrive on the grammatical side (Hinzen,2006; 2008; 2009). C-I systems should look at the deliverances of grammatical derivations with bafflement or stupor, failing to interpret them altogether, much as we would be when confronted with communications from aliens in a never-encountered code.
This in turn engenders another conclusion: that the form-content dichotomy must be given up. Content has to lie on the side where the form is: the grammatical one. This, of course, will hold exactly insofar as meaning is grammatical. Insofar, that is, as there is a difference that grammar makes to the organization of meaning. The specific content of grammar, we here suggested, lies in grammatical relations, since we know it does not lie in the atoms of the lexicon. Yet Merge applies to lexical atoms generating binary sets, and the members of a set stand in no inherent relations to one another. Lexical items in turn are collections of lexical features, and some of these
(Case, ç-features) will trigger relations of Agree and potentially applications of ‘internal Merge’ (IM) on the standard model: the merging of a syntactic object A contained in another one, B, to that object B. IM creates a discontinuous constitu- ent, and hence grammatical relations, with interpretive consequences as in the case of operator-variable dependencies or binding. But there is no theory of grammatical relations in grammar so construed, viewed as primitives or fundamental elements of grammar. If they exist, they are coded featurally. Feature-based coding does not of course exclude grammatical relations, but by coding these uniformly through features and generic operations such as Agree and Merge, a restrictive and principled theory of grammatical relations and hence of the content of grammar does not come into view. Since formal semantics, in turn, tends to code relations by assigning them as denotations of predicative expressions, hence effectively lexically, the very term ‘grammatical relation’ lacks a status within the Minimalist framework. It will be a theoretical primitive in the present one, and our basis for reconstructing the recur- sivity of grammar, following Arsenijević and Hinzen (2012), in the rest of this chapter. This position contrasts with the standard Minimalist position, but is not, on our view, inconsistent with Minimalism as such. The difference is that recursion as described through Merge is a purely formal universal of human language that does not say anything about grammar specifically. It is the culmination of a formal tradition in linguistic theory that abstracts from semantic interpretation and is, given its bare minimality, the peak to which such a perspective can be driven. That is the strength of the Merge-based viewpoint—but also, as we will argue, its weakness. For as we may already conclude from the above, Merge leaves us without a handle or substantive insight on grammatical relations, or on the content of grammar. Gram- mar is recursive; but its substantive content is not recursion.14