In the results reported here and in Section 5 the totals do not add up to 216 data points for both nouns and verbs (3 forms per category; 72 native speaker informants). This is because certain responses were excluded, e.g. because of pronunciation ambiguities, miscategorisation (as revealed by example sentences), or because they exemplified zero-derivation rather than the actual formation of new words.
The mean word length in syllables for nouns here is 1.80 (SD = 0.93); for verbs 1.42 (SD = 0.73). The t-test shows that the difference observed is very highly significant (df = 403; t = 4.51, p = 0.000004. The Mann-Whitney [-test suggests the same (U = 15466; group N had 202 points, V 203 points, p = 0.0000007).
Mean syllable length
Contrary to what one expects on the basis of the literature, syllables in nouns (3.20; SD = 1.02) turned out to be shorter than in verbs (3.50; SD = 0.87). The t-test (df = 403; t = -3.1, p = 0.0021) and the U-test (U = 16125.5; group N had 202 points, V 203 points, p = 0.0001) both show that the difference is very highly significant.
The significance tests were carried out in a two-tailed fashion: one-tailed tests to see whether verbs consist of longer syllables than nouns would be unsuitable, because that hypothesis would go against conclusions reached in the literature.
Final obstruent voicing
The mean score for nouns is 0.54 (SD = 0.50), while for verbs it is 0.65 (SD = 0.48). The chi-square test shows that p = 0.0639 (df = 1, ^2 = 2.318), i.e. the difference is almost significant. Fischer's exact test confirms this (p = 0.0849). Given how close these values are to p < 0.05 one suspects that with a larger sample size, the effect would probably be significant.
The mean value for nouns is 0.17 (SD = 0.22); for verbs 0.14 (SD = 0.19). The t-test here yields a p-value of 0.0489 (df= 403, t = 1.66). The Mann-Whitney U-test sheds a slight degree of doubt on the significance, as it yields a p value of 0.0731 (U = 18963; group N had 202 points, V 203 points). Still, with a larger sample size there is every reason to suspect that the difference would also show up as significant under that test.
Stressed vowel advancement
The mean for nouns here is 1.08 (SD = 0.81), as against 0.96 (SD = 0.79) for verbs. This difference is very nearly significant: the t-test gives a p-value of0.754 (df = 400; t = 1.44). The Mann-Whitney 17-test yields a [/-value of 18614.5 (group N had 199 points, V 203 points), which means that p = 0.0745.
Stressed vowel height
Here we fail to find a significant difference. The score for nouns is 1.17 (SD = 0.71). The one for verbs is very close, at 1.19 (SD = 0.65). It is thus not surprising to find a p-value of 0.377 with the t-test (df= 400; t = -0.314), and even 0.435 with the [-test (U = 20024.5; group N had 199 points, V had 203 points).
Presence vs. absence of a final obstruent
Nouns score a mean of 0.41 (SD = 0.49), whereas for verbs it is 0.54 (SD = 0.50). The chi-square test (two-tailed, since there was no hypothesis in the literature to start from; see Section 3.3.1, above) yields a ^2 value of 7.503 (df= 1), which corresponds to a p-value of 0.0062. Fisher's exact test confirms that the difference is significant (p = 0.0072).
Many of the observations made in previous scholarship are seen to be confirmed here for the first time in a study of production. Specifically, for all variables found to be relevant in the literature on lexical categorisation in English, the novel nouns and verbs displayed significant or nearly significant differences, with only vowel height and mean syllable length being exceptions. With vowel height, no difference was found, and regarding syllable length, the verbs produced by my participants actually contained longer syllables on average than the novel nouns.
It is not clear what the explanation for the exceptional behaviour of these two parameters may be, but it is more important to focus on the considerable degree of overlap between the findings from this study and previous work. The study also found evidence to suggest that verbs may end in obstruents more often than nouns
Figure 2. The verb network including specific verbs, phonologically partly specific subschemas, and the phonologically maximally abstract super-schema
do, which awaits confirmation in future research not only research on production but also on corpora and comprehension data.
In terms of a usage-based model of the noun and verb categories, then, there is clear converging evidence that phonological properties should be incorporated. Based on the phonological analysis of nouns and verbs in the CELEX corpus Taylor (2002) proposes that this be done in terms of phonologically partly specific "sub-schemas", i.e. representations that are more schematic than specific nouns and verbs yet less abstract than the "super-schemas" categorising all nouns and all verbs. Inspired by his suggestion, I suggest that my production data yield the subschemas for the category of verbs that are given in Figure 2, below. Of these subschemas, I hypothesise that the leftmost one, representing the prototypical verb as monosyllabic, is especially salient. This is because previous scholarship (e.g. Kelly 1992, 1996; Durieux & Gillis 2001) presents this parameter as particularly robust, which was supported by the statistical evidence found in the present study (see Section 4.1.1, above). The specific verbs in this network are the same ones Taylor uses in his figure (2002: 184), and I also follow his representation of the semantic pole of verbs as '...'. This should not be taken to mean that the sub-schemas and the verb super-schema are semantically empty: the meaning is PROCESS (in Langacker's sense of the term; see Section 2.2.1). More details about the various subschemas are provided by Hollmann (submitted).
A similar network could be given for nouns, but for reasons of space that is not attempted here.
There has been some discussion in the cognitive linguistic literature on schema-based as against exemplar-based approaches. In this connection, one may wonder whether the sub-schemas proposed by Taylor (2002) and adopted here might not be better thought of in terms of exemplars. However, I side with Langacker in suggesting that despite "differences in emphasis, detail, and methodology" schema and exemplar-based theories "are essentially equivalent" (2010: 138). Another way of putting this is to say that the schemas proposed by Taylor and in this study could be seen as "the relationships among a set of specific exemplars" (Croft 2012, Ch.12).
-  One instance of a sentence that reveals erroneous categorisation is That test was unbelievably ringical (noun 2, participant 66). There were 8 such cases.
-  One of the 3 instances of this is table: I am going to table you if you don't be quiet (verb 1, participant 47).
-  Lack of space prevents fuller discussion of the relative salience of the various sub-schemas and the super-schema, but Hollmann (submitted) argues that the noun and verb super-schemas may be extremely low in salience, and perhaps not be extracted at all. Their existence is exceedingly hard to disprove, but therein also lies their weakness: the sub-schemas allow one to make far more explicit predictions with regard to psycholinguistic experiments and in various other domains as well; see further Hollmann (submitted). For the relative importance of low-level schemas more generally see also Croft (2002), Langacker (2002: 118), and especially Dabrowska (2010).
-  This quote appeared in an earlier draft of Croft's monograph, but was not included in the published version. The point, however, still stands.