Abstract for Phoevos Panagiotidis's talk

An (almost) new perspective on grammatical categorisation
Phoevos Panagiotidis (University of Cyprus)
May 1, 2008 (Thursday)
4:15 PM - ; Room 6417, The CUNY Graduate Center

In the generative literature on grammatical category – with Chomsky 1970 and Jackendoff 1977 acting as pivots, the categorial features [N] and [V] are usually perceived as syntax-internal classificatory ones without an LF interpretation (see also the discussion in Haeberli 2002: Ch. 2). This is a serious problem for Full Interpretation and, more generally, for a conception of grammar as a system that manipulates interface-interpretable features (Chomsky 1995; 2001):
(1) What is grammatical category; what does it do in syntax and at LF?
Two paths have recently been opened towards answering this question:

A first one seeks to rid syntax from categories in the lexicon altogether, employing the mechanisms of Distributed Morphology: Halle & Marantz (1993), Marantz (1997; 2000; 2006a), Harley & Noyer (1998), Embick (2000), Arad (2003; 2007); Embick & Marantz (2006). In this programme, syntactic category is not the result of a categorial feature in a pre-syntactic lexicon: there is no pre-syntactic lexicon. Category is recast as the upshot of the existence of categorisers – a nominaliser (n), a verbaliser (v) and an adjectiviser (a), which combine with roots and whose projections constitute “word-internal” phases.

A second path is that of trying to reconceive categorial features as ordinary LF-interpretable features with a well-defined syntactic behaviour and a particular LF interpretation. This is what Baker (2003) sets out to do. He advances a theory of category where [V] and [N] are privative features. Verbs are specified as [V] and are interpreted at LF as predicates, while nouns are specified as [N] and are interpreted at LF as bearing a criterion of identity, making them referential. Although Baker’s is a straightforward and coherent theory, it is fraught with problems which prevent us from accepting it wholesale.

In this talk I wish to show that the two paths may converge, after all. The starting point will be the observation that, even if we get rid of nouns and verbs in the lexicon (I will largely ignore adjectives), we cannot eliminate the syntactic difference between the nominaliser n and the verbaliser v – a difference hopefully expressible in terms of features. Moreover, these distinctive features had better be interpretable, enabling us to recast the categorial features [N] and [V] of old as the LF-interpretable distinctive features on n and v:
(2) What differentiates between the nominaliser and the verbaliser at LF?

Keeping the above in mind, we turn to the observation that category distinctions should be taken to correspond not to inherent ontological properties of concepts but, rather, to perspectives on (concepts about) the world (Baker 2003: 293-294). Having said that, concepts of particular types are canonically mapped onto particular categories in language after language, with objects getting mapped onto ‘prototypical’ nouns (rock, tree, child) and with dynamic events onto ‘prototypical’ verbs (buy, hit, walk, fall). In other words, grammatical categorisation cannot be a completely arbitrary, grammar-internal affair, although grammatical categories are hardly coextensive to conceptual ones: thus, ‘nominality’ and ‘verbality’ (or ‘nouniness’ and ‘verbiness’) must possess an interpretive import, reducible to the interpretation of [N] and [V] on n and v. Exploiting the ‘perspective’ aspect of grammatical category, I propose the following as the LF import of [N] and [V]:
(3) An [N] feature forces an extending-into-space perspective at LF; a [V] feature forces an extending-into-time perspective at LF.

Once we conceive [N] as forcing an extending-into-space perspective on concepts and [V] as forcing an extending-into-time one, we can capture the fact that nouns are better suited to have concepts denoting objects as well as substances mapped onto them: objects and substances saliently occupy space. Similarly, events typically contain verbs because a verb is a perspective over a concept as something extending into time. Moreover, the [V] feature’s interpretive import becomes evident in verbs’ biunique relation with Tense cross-linguistically.

Understanding categorial features as imposing interpretive perspectives, we can explain why we cannot have bare roots inserted in the syntax, i.e. without them being in the complement of a categoriser – a point of criticism made by Baker (2003: 268). Note that roots are not syntactically exceptional: they merge with a number of elements and they project their own phrases (Marantz 2006a; 2006b; Harley 2006; 2007). However, if Arad (2007: Ch. 3) is correct, free roots are semantically impoverished / underspecified; so, their presence in a derivation would lead it to crash at LF. This is where categorial features become important: they close off material associated with the root, by imposing an interpretive perspective for the conceptual systems to view it in. In this respect, categorial features have a dual role, fitting for phase-edge features: a) phase-internally, they contribute the interpretive perspective and b) they identify the whole phase externally (as ‘nominal’ or ‘verbal’). Consequently, underspecified roots and their related material cannot be assigned a ‘meaning’ until they have combined with a categorial-feature bearing head: a categoriser.

In effect, if [N] and [V], interpretable categorial features, are marked on n and v, then categorisers are not functional elements and grammars possess only one ‘lexical noun’, the nominaliser, and one ‘lexical verb’, the verbaliser. The remainder of the analysis in Marantz (2000; 2006a; 2006b) applies in its entirety.

If what we call lexical elements are syntactic structures composed of at least a nominaliser or a verbaliser and a root, then
(4) What is the category of functional elements?
If we do not want features [N] and [V] to come prepackaged from the lexicon with the roots they co-occur with, then we probably also do not want categorial features for functional categories such as [T], [C] or Chomsky’s (1995) [D]. Actually, I will be following Hegarty’s (2005) radical approach here (see also Harley & Ritter 2002 and Borer 2005) that functional heads are indeed just structured bundles of features, not primitive categories. On top of that, Ouhalla (1991), Grimshaw (1991) and van Riemsdijk’s (1998) Categorial Identity Thesis suggest that functional heads bear the categorial specification of the lexical material (i.e. the ‘noun’ and the ‘verb’) in their complement. Panagiotidis (2002: Ch. 5) builds on this thesis and offers a version thereof where functional heads bear the uninterpretable version of the categoriser’s [N] or [V] feature: a [uN] or a [uV]. This is the Categorial Deficiency hypothesis.

The existence of uninterpretable categorial features as ‘typing’ functional heads has a number of welcome results. On a conceptual plane, categorial features are now fully assimilated to other LF-interpretable features, with their own uninterpretable versions. Moreover, the contrast in the interpretability of categorial features provides an explanation for the lexical-functional distinction being real and sharp – a well-known empirical result. Even better, no matter how many functional categories are hypothesised, motivated and discovered, no proliferation of parts of speech is necessary. Categorial Deficiency may also entail that the repertory of functional heads can vary cross-linguistically (Thráinsson 1996, Bobaljik & Thráinsson 1998 – contra Cinque 1999; Borer 2005): maybe each grammar has its own functional categories, bundles of features typed by [uN] or a [uV]. Beyond the theory of grammar, if [uN] or a [uV] type particular elements, functional elements, we would expect these to be acquired later in L1 and harder in L2 – again, a well-known empirical result.

Turning to some empirical consequences of Categorial Deficiency, we can explain why there are no projections without a categoriser (although there are root-less projections: semi-lexical elements; see Emonds 1985, Schütze 2001, Harley 2006): these would contain no interpretable categorial feature to check off the uninterpretable versions of the functional elements in the tree. This leads us to a general conception of the functional-lexical relation within derivations as one of categorial Agree. This conception in turn enables us to

a. explain why categorisers always merge below functional material and never get sandwiched by it;
b. reduce most instances of head-movement to the overt reflex of a categorial Agree operation;
c. derive the empirically plausible thesis that uninterpretable features cannot act as Goals for Agree.