The relation between perceptual and production categories in acquisition

Results from two experiments on acquisition of the word initial voicing contrast are compared, one on perception, one on production. Both involved monolingual and bilingual (French/English) subjects. Monolinguals are shown to have frequent disparities between their use of parameters in production and perception. Furthermore, bilingualism is found to affect subjects' production and their perception differentially, in a way which suggests that there cannot be isomorphism between their production and perceptual representations. However, the ongoing interference between the phonetic categories in bilinguals' two languages also argues that at the phonological level there is identity between cognates in their two languages. This pattern of phonological identity but phonetic difference is taken as support for models of speech processing in which distinct representations are attributed to phonology and phonetics, and to production and perception. 1. REPRESENTATIONS IN SPEECH Accounts of the human speech faculty differ fundamentally in the way they model the relationship between speech production and perception; these differences furthermore often reflect equally deep-seated disagreements concerning the relation between phonetics and phonology. In both cases, a crucial question concerns the number of different representations postulated. For the production/perception dichotomy, the division is between approaches which attribute independent representations to production and perception (e.g. Keating's model [7]) and those which deny the existence of a dichotomy, postulating a single (usually articulatory) representation for both [2,3,4,9]. The latter approach favours perceptual models such as Direct perception and the revised Motor Theory [3,9]. For the phonetics/phonology relation, the essential contrast is between models which postulate independent phonetic and phonological representations (termed "translation models" by Nolan, [10]) and those in which there are no independent phonetic representations, the phonetic realisation of phonological units being seen rather as an automatic consequence of their very specification (hereinafter "unity models"). A recent example of a translation model would be that presented by Keating [7]; unity models are those proposed or implied by approaches such as action theory and articulatory phonology [4,2]. Unity models are often associated with views of phonetics involving a single representation, that is, with perception based on articulation. In this paper, evidence from two experiments on phonetic acquisition is used to argue, for both dichotomies, in favour of models with more rather than fewer distinct representations. The arguments come from three related sources (i) Disparities in the patterns of acquisition by monolinguals in perception and production; (ii) differences between monolinguals and bilinguals attributable to bilinguals' identifying two different phonetic categories (e.g. French [k] and English [kh]) with each other at the phonological level; (iii) disparities in the effect of bilingualism on perception and production. Single and dual-representation models of phonetic processing may be distinguished according to the predictions they make concerning acquisition in general, and in particular, concerning the effect of bilingualism on processing. Single representation models would predict parallelism in the development of categories in both production and perception, and thus that both should be affected equally by bilingualism; dual representation models would allow for disparities between the two. Data from bilinguals can also throw light on the phonetics-phonology relation. Logically, bilinguals could show one of three conditions: (i) completely merged phonetics and phonology; (ii) completely distinct phonetics and phonology; (iii) divergent phonetic realisation in their languages but evidence that cognate phonological categories are identified with each other cross-linguistically. The first two conditions would not allow us to adjudicate between translation and unity theories, but the latter would be embarrassed by the third type, which implies a distinction between phonological and phonetic representations: if phonetic objects differ only because their phonological specifications differ, how can different phonetic objects be identified with the same underlying phonological representation?