Cognitive linguistic models of word classes are normally thought of as semantic (Cognitive Grammar) or semantic-pragmatic (Radical Construction Grammar) in nature. However, I showed in Section 2.2.1 that Langacker may actually be interpreted as leaving open the possibility that distributional properties are part of his noun and verb schemas. I emphasised firstly that his commitment to purely semantic characterisations is less than full, and secondly drew attention to his
suggestion that schematic units emerge out of more specific constructions ("de-contextualization") without the meronymic links being completely broken. This means that the noun and verb schemas may be connected with constructions that they frequently occur in, and thus that these distributional facts are part of our knowledge of word classes. With regard to Croft, I noted in Section 2.2.2 that the discourse propositional act functions are ultimately represented as syntactic constructions. From a usage-based view it is reasonable to assume that these constructions, although they may be language-particular, are stored as part of speakers' knowledge of the word classes that they help to define.
The results from this production study point us in the direction of what are likely to be the main constructions that are stored as part of (English) speakers' knowledge of nouns and verbs. For nouns, a high number of examples of subject and object status were found. Considering these findings in detail, I observe that the nature of the task may have influenced the frequency of objects as against subjects. Participants were asked to create new nouns, and thus presumably often thought of new 'objects' (in Croft's terms) or 'things' (in Langacker's sense). Now because of the lack of familiarity with these new entities, they may have been inclined to position the nouns in object position, which is after all often associated with new information. This suspicion is supported by the observation that 10 instances of novel nouns in object position but only three novel noun subjects contained the modifier new; compare (13-14), below:
(13) Do you like my new proogle? It's very fluffy so I like it. (noun 2, participant 7)
(14) My new zog followed me round all day. (noun 3, participant 37)
These numbers are too low for meaningful statistical analysis, but they are suggestive. Therefore, the subject slot may be a more important aspect of the noun schema than the object position.
Either way, statistical analysis suggested that participants produce fully referential nouns significantly more often than partly or non-referential ones. This is in line with Croft (1990, 1991, 2001), although as we have seen he does not explicitly associate specific syntactic constructions with the pragmatic dimension of his definition of nouns.
However, interestingly, an even stronger association was found between novel nouns and determiners, which I argued are not necessarily enough for full referential status of the NP. This could be seen as a partial vindication of the structuralist and generative approach, which often treats the Det _frame as a distributional criterion for nounhood.
The conclusion regarding nouns is that the schema (to the extent that it exists independently; see fn. 10, above) emerges out of, and is connected to, the subject and object slots in the [X Verb (Y)] construction, and has a perhaps even stronger link to the [Det X] construction. The links to these constructions are part of (English) speakers' knowledge of nouns, in addition to their semantic and phonological properties.
As regards verbs, the way in which they were used by my participants suggests that a connection exists with the construction(s) in which they are, or are part of, finite predications. In usage-based terms, the verb schema arises out of, and is stored in connection with, the construction [X V-TNS (Y)], where TNS stands for 'tense' and V for either a single verb or an auxiliary and main verb string. A full network representation of the verb (and noun) category would thus go beyond Figure 2, above, and incorporate links with the relevant constructions outlined in this section.
The proposals made here agree to some extent with the position taken by Croft (1990, 1991, 2001), although there the propositional act functions are seen as pragmatic ("definitely not formal"; 1990: 248), with no explicit recognition that certain syntactic constructions performing these functions may be stored as part of our knowledge of nouns and verbs. My suggestion is in fact a more fleshed-out version of Taylor's (2002) view. Just as we saw was the case in relation to phonological properties of nouns and verbs (see Section 4.2, above) Taylor is also more explicit about the status of distributional facts than other cognitive linguists. He refers to Croft's work but seems to take a (partly) syntactic view of constructions, in contrast to Croft's pragmatic conception:
[T]here is an important sense in which the categories of adjective and noun (and indeed the other word classes) must be understood with respect to the constructional schemas in which they occur (Croft 1999). This is not to deny the possibility of entertaining construction-independent characterizations of the word-classes, in terms of the nature of the concepts that the words designate, for example (...). Ultimately, however, a word class emerges as a function of its role within a constructional schema. (Taylor 2002: 563)
Interestingly, another conclusion one might draw from this study is that there is evidence for the cognitive realism of some aspects of the structuralist-generativist characterisation of nouns and verbs.
-  One might wish to suggest that the relevant construction is simply [V-TNS]. That is less likely, however, because (at least in English) tensed verb or verb phrase configurations hardly ever occur in isolation, but normally with a subject and possibly one or more other arguments.
-  See Section 2.2.2, above.