Previous scholarship on word classes
This section discusses previous scholarship on word classes in roughly chronological order, starting with the structuralist-generative approach (itself seen to be a response to earlier work), then moving on to the cognitive linguistic approach, and finally discussing relevant work in psycholinguistics.
The structuralist-generative approach
Structuralist and generative grammarians often preface their treatments of word classes with critiques of overly simplistic descriptive work especially related to word classes in English. Palmer disproves Nesfield's (1908: 5) suggestion that a noun "gives a name to some individual person or thing or to some kind of person or thing" with reference to abstract nouns and nominalisations such as His suffering was terrible (1971: 39). The usual position in the structuralist and generativist approach has been that word classes should be defined distributionally, i.e. in terms of morphological and syntactic criteria. With respect to morphology, Chomsky (2002) notes that "in the sequence 'Pirots karulize elastically' we know that the three words are noun, verb, and adverb by virtue of the s, ize, and y, respectively" (104). In relation to syntax, for nouns, one criterion that is normally given is that they may follow determiners (e.g. Aarts 2007); for verbs, co-occurrence with a subject is often mentioned (e.g. Baker 2003).
Chomsky (1981: 48) defines nouns and verbs in terms of the features [+N] and [+V], with nouns being [+N, -V] and verbs being [-N, +V]. As noted by Francis & Matthews (2005: 283), this implies that "[c]ategories are not defined by distributional patterns in particular languages; rather, distributional patterns are derivative of the feature specifications". Yet more recent work in the generative tradition (e.g. Baker 2003, Aarts 2007) by and large places distributional facts centre stage. Aarts states that
in the case of grammatical form classes, we might say that a particular item belongs to some class or other by virtue of the abstract syntactic framework to which it belongs, and the attributes that make us classify a particular element as, say, an adjective (attributive/predicative position, gradability, etc.) are themselves part of that syntactic framework. (2007: 88)
Baker admittedly suggests that semantics play a role in the acquisition of the noun, verb, and adjective categories (2003: 300), but claims that once the word classes have been set up with the help of their distinctive meanings, the connection with their semantic basis is severed and they remain as purely syntactic entities. In so doing, he invokes Macnamara's old ladder metaphor: "The child climbs to grammar on a semantic ladder and then kicks the ladder away" (1982: 134, cited by Baker 2003: 297).
In contrast to Aarts and Baker, both of whom explicitly formulate word classes as categories in speakers' mental grammars, another recent approach on the generative end of the theoretical spectrum, Distributed Morphology (e.g. Halle & Marantz 1993, 1994, Marantz 1997, Harley & Noyer 1999), denies the psychological realism of lexical categories.
Especially revealing in relation to word classes is the exchange between Barner & Bale (2002, 2005), both proponents of this theory, and Panagiotidis (2005), who criticises their suggestions within a generative theoretical context. In discussing nouns and verbs, Barner & Bale suggest that what is stored are roots, which are underspecified as to their word class. It is only by virtue of being placed in certain syntactic slots (called "Extended Projections") that they acquire the status of nouns or verbs. One of Panagiotidis's objections is that in conversion processes, the semantics very often change in unpredictable ways. For example, when spider is used not as a noun but as a verb, it means to move in the characteristic manner of a spider. But, Panagiotidis goes on, the OED gives a second meaning, i.e. "to catch or entrap after the manner of a spider" (2005: 1187). The author adds that spider as a verb does, however, not mean other things that are characteristic of arachnids, such as "having eight legs, devouring one's mate or spinning webs" (Panagiotidis 2005: 1187), and concludes that the relation between the noun and verb meanings is far from straightforward.
The Distributed Morphology view that word classes are mere epiphenomena, linguists' generalisations over patterns observed in language, is rejected in the present chapter. The first reason lies within the framework of Distributed Morphology itself. Hailed as a very parsimonious model of grammatical representation and acquisition (see e.g. Barner & Bale 2002: 771), Distributed Morphology assigns all idiosyncrasies, such as the unpredictable meanings of a verb like spider, to a component called the Encyclopedia (Barner & Bale 2005: 1177-1178). Harley defines the Encyclopedia as a list of idioms, where an idiom may be "any expression (even a single word or subpart of a word) whose meaning is not wholly predictable from its morphosyntactic structural description" (1999: 8). The sheer wealth and variety of idioms discussed by construction grammarians such as Fillmore et al. (1988) and Nunberg et al. (1994) suggests that idioms do not merely constitute a marginal part of our grammatical knowledge (see also Croft & Cruse 2004). Therefore, the Encyclopedia must be of a size that subverts Distributed Morphologists' fundamental claim of parsimony in representation and acquisition. Furthermore, whilst philosophers of science suggest that theories should be maximally simple (see e.g. Kuhn 1977: 322), this should not go at the expense of consistency with theories from related disciplines (ibid.). In this connection, it should be noted that even if Distributed Morphology were parsimonious in terms of storage (which it is not), this is not necessarily consistent with cognitive psychologists' finding that storage actually involves considerable degrees of redundancy (see e.g. Barsalou 1992: 121, Solomon & Barsalou 2001).
The second reason why nouns and verbs are taken here to be somehow psychologically real is the significant amount of evidence from psycho- and neuro-linguistics pointing in this direction. Some of the psycholinguistic evidence is discussed in Section 2.3 below. Here, let me briefly mention two kinds of neuro-linguistic work suggesting that (certain) categories are represented differently in the brain. First, studies on selective impairments show that one category (more often verbs than nouns) can be affected, with other knowledge remaining more or less intact (see e.g. Myerson & Goodglass 1972, Miceli et al. 1984, Daniele et al. 1994). Second, electrophysiological measurements suggest that verbs are associated with activity in the frontal region, nouns with the temporal/occipital regions (e.g. Brown & Lehmann 1979, Dehaene 1995, Preissl et al. 1995, Pulvermuller et al. 1996).
The psycholinguist Ferreira (2005) sketches the general disenchantment in her field with formal syntactic theory. Although she emphasises the empirical weaknesses in the Minimalist Program, the neglect of evidence concerning the psychological realism of word classes means that her criticism could be extended to Distributed Morphology.
-  This criticism can be extended to other generative accounts of word classes. Baker, for example, reviews Dixon (1982), Hopper & Thompson (1984), Givön (1984), Langacker (1987a), and Croft (1991). He suggests that although "[t]hese functionalist approaches undoubtedly contain important grains of truth" (2003: 14), a "crisper ( ) theory of the lexical categories would be inherently desirable" (ibid.:24). He fails to define "crispness" in relation to scientific theory, but elsewhere calls his own characterisations of the word classes "crisp and simple" (ibid.:301). To the extent, then, that he favours his account on grounds of Kuhnian simplicity, he would have to explain why this outweighs the lack of consistency with cognitive psychological findings concerning redundancy in storage.
-  The latter associations are perhaps less straightforward than they might seem, however: there is some evidence that these differences in brain activity depend on the nouns and verbs occurring with morphological inflections (e.g. Tyler et al. 2004).