Representation of Phonological Alternations in a First and a Second Language: A Preliminary Report

The first language word recognition system copes easily with the great variability of spoken language. A particular kind of context-dependent variation is phonological assimilation. In French, for example, there is a process of voicing and devoicing assimilation. For example, jupe [yp] ‘skirt’ can be pronounced [yb] when followed by an obstruent of opposite voicing, like /g/ in grise [iz] ‘grey’ (voicing assimilation). Conversely, robe [ob] ‘dress’ is pronounced [op] when followed by a voiceless obstruent, such as /s/ in sombre [sb] ‘dark’ (devoicing assimilation). When the context is a sonorant (/, l, m, n/), neither voicing nor devoicing occur: robe rouge [ob]+[u] ‘red dress’, jupe noire [yp]+[nwa] ‘black skirt’. This process is regressive and systematically occurs across word boundaries, when both words are in the same prosodic domain of the phonological phrase. A possible representation of regressive voicing assimilation in French is an autosegmental spreading and delinking operation (Goldsmith, 1990; Mester & Itô, 1989). Because French has no final devoicing otherwise, it seems that the feature [–voice] can spread from one obstruent to the preceding one, such that French needs to assume a contrastive feature specification of [+voice] and [–voice] for obstruents, and symmetrical spreading of both specifications (Cohn, 1995; Féry, 2003; Wetzels & Mascaró, 2001). Sonorants in French can remain underspecified for the voicing feature (Cho, 1999; Lombardi, 1991; see Darcy, 2006 for a review and typological overview). Because such modifications alter a word’s form across word boundaries and can potentially create ambiguity, they need to be compensated for during the process of word recognition. Previous research established that listeners are able to use abstract phonological knowledge of such alternations in their first language (L1) to compensate for assimilations (Gaskell & Marslen-Wilson, 1996; Darcy et al., in press; Darcy & Kügler, 2007). Other interpretations tend to support a universal and early processing capacity involved in compensation for alternations instead of abstract mental representations (Gow, 2002; Mitterer & Blomert, 2003). However, these mechanisms work best if the stimuli presented are gradient rather than categorical (see Darcy & Kügler, 2007). Darcy and Kügler presented results from French listeners hearing categorical occurrences of French voicing and devoicing assimilations. If fully neutralizing, such assimilations are predicted to be challenging for lexical access: reinterpretation of fully, categorically assimilated segments is necessary in order to have successful access to the stored lexical representation. In such a case (for example in the case of [op] where [p] resulting from underlying /b/ is completely voiceless), reinterpretation of a surface [p] into /b/ could depend on prior linguistic knowledge of the alternation in French (voicing/devoicing assimilation), because there are no available auditory or phonetic cues in the stimuli. If on the contrary assimilated segments still carry traces of the underlying sound, like in the case of incomplete or partial assimilation, such reinterpretation can be achieved without specific linguistic knowledge of alternations, using phonetic cues present in the signal (Gow, 2002; Mitterer & Blomert, 2003). In Darcy and Kügler’s study, such phonetic cues have been shown to be absent in acoustic analysis of the stimuli, which were produced naturally. The speaker was instructed and trained to pronounce sentences as naturally as possible, where assimilatory changes were orthographically indicated (e.g. for robe ‘dress’ /ob/ in the sentence ...[op]+[sal]... ‘dirty dress’, the sequence was written). The changes made were comparable across conditions. Special attention has been given to not overemphasize the key word pairs. Details of the acoustic analysis are presented in Table