Questionnaire study design
Psycholinguistic work has thus far been based on generalisations from corpus data, conclusions drawn from artificial language learning algorithms, and, finally, comprehension experiments involving human subjects. In the latter, the experimenter makes up nonce forms and in some way or other gets the participants to categorise them as nouns or verbs (see e.g. Don & Erkelens 2008). To date, there is however no study that explores what speakers do in a production task. Since production is of course an important part of language processing, this is what the questionnaire used by Hollmann (submitted) and the present study sets out to do.
The questionnaire used for this study was designed in a straightforward manner: participants were asked to devise three new English nouns and three new English verbs, and to use each novel form thus created in a sentence. The questionnaire was split, with half the participants starting with the nouns, and the other half with the verbs.
In terms of the phonological analysis, the assumption is that the forms produced somehow correspond to the prototypical sound structure of nouns and verbs, stored as part of the participants' representation of these categories (and described in especially Section 4, below).
For Hollmann (submitted) the sentences merely served two purposes: first, to provide a larger sample of each participant's handwriting, making it easier to read the novel nouns and verbs; second, to check whether the participant is aware of the distinction between these two lexical categories. In the present study the use of the novel forms is interpreted as an indication of their prototypical distribution.
The questionnaire was distributed among 80 first-year linguistics students at Lancaster University. Among them, there were eight non-native speakers, whose responses were excluded from the analysis. It is deemed unlikely that the students, in completing the questionnaire, tapped into any meta-knowledge of phonological and/or distributional properties of nouns and verbs, because at this early point in their studies they have not been exposed to any of the literature on these issues.
Phonological and distributional properties and scoring schemes
The phonological properties distinguishing English nouns and verbs that emerge from the literature (for references see Section 2.3 above and Hollmann submitted for more discussion) are word length (in syllables), mean syllable length (in phonemes), final obstruent voicing, nasal consonants, stressed vowel advancement, and stressed vowel height. The sound structure of nouns and verbs is such that nouns are generally longer compared to verbs, have longer syllables, have unvoiced final obstruents, have more nasal consonants, contain fewer front vowels, and more low vowels. In addition to these, Hollmann (submitted) finds evidence that the presence vs. absence of a final obstruent may be a further factor distinguishing nouns from verbs, as will be explained below.
The scoring scheme used in Hollmann (submitted) and here was inspired by the one devised by Monaghan et al. (2005). Each novel form was scored for each of the seven parameters. The scheme takes into account the different nature of the variables in question.
Final obstruent voicing and the presence vs. absence of final obstruents are of the ordinal kind, with only two values. Thus, the scale used here is [0, 1]. For the voicing variable, 0 is given for a voiced final obstruent and 1 for an unvoiced one. (Forms that do not end in obstruents simply receive no score here.) The presence
Figure 1. Vowel position, based on Roach's (2004: 242) description of RP
of a final obstruent is given a score of 1, absence is rated 0. (Needless to say all forms receive a score here.)
Word length (in syllables), mean syllable length (in phonemes), and nasal consonants all represent continuous scales, but the scale starts at 1 for word and syllable length. By contrast, it ranges from [0-1] for nasals, where 0 represents no nasals at all, and 1 stands for the (hypothetical) case of all sounds having a nasal quality.
Stressed vowel advancement and height, finally, are in principle also continuous, but these parameters were operationalised by distinguishing three values for each: front (0), mid (1), and back (2), and high (0), mid (1), and low (2), respectively.
Vowel position was determined with reference to Roach's (2004) phonological description of RP, as indicated in Figure 1, below. Diphthongs were analysed as mid both in terms of frontness and height, since none of them fall clearly into the front/back or high/low zones as indicated in Figure 1. Not all participants speak RP, but as an approximation this method is at least superior to the one used by Monaghan et al. (2005), who do not specify how exactly they score vowel position, and who also conflate British and American English data, despite well known differences in the realisation of certain vowels (cf. e.g. the difference in pronunciation of the stressed vowel in tomato).
In my calculations of the statistical significance of the differences observed, the nature of the variables was taken into consideration: The Mann-Whitney [/-test and t-test were used for most of the parameters, as they are ordinal, interval, or ratio variables. Final obstruent voicing and presence vs. absence of a final obstruent are nominal variables, and so the chi-square and Fisher's exact test were appropriate here. (For details concerning different kinds of linguistic variables and suitable statistical tests see e.g. Butler 1985 and Gries 2009.)
Although the questionnaire responses were written, the pronunciation of most of the forms was straightforward. For example, there is really only one way, in terms of stress and other aspects of pronunciation, to produce glop (noun 1, participant 10), dandrel (noun 2, participant 32), or fludder (verb 3, participant 52). However, there were 25 forms whose pronunciation was ambiguous, also in view of standard descriptions of English word accent (Chomsky & Halle 1968, Fudge 1984, Cruttenden 2008). For these forms I had three native speakers of English read out these words, supplying them with a nominal or verbal context as appropriate. In most cases they either all agreed or there was agreement between two of them. Yet there were also seven forms which yielded three different pronunciations. For example, daes (noun 1, participant 30) was pronounced as [dais], [dœz], and [deis]. These forms were omitted from the analysis of the parameters that were affected by the ambiguity (in this case vowel advancement and height as well as final obstruent voicing).