Language-specific aspects of the perception of illegal clusters

International audienceThis talk focuses on the perception of some illegal consonant clusters, as constrained by language-specific phonological constraints in the listeners' native language, that is, loosely speaking, by phonotactic constraints. Such constraints not only specify the permissible sequences of consonants, regardless of syllabic structure, but also the consonants that are allowed to occupy a given syllable slot (e.g., onset or coda). (Here, we restrict ourselves to the discussion to utterance-initial consonant clusters.) An early mention of how nonnative consonant clusters are perceived appeared in Polivanov's (1931) "La perception des sons d'une langue étrangère." Polivanov reported, for example, that Japanese listeners hear an epenthetic vowel in the initial cluster of drama and produce either dorama or dzurama when they attempt to pronounce this word. It's important to note that, in Polivanov' account, such epenthesis occurs at the perceptual level. That is, Japanese speakers/listeners pronounce drama as dzurama because they hear an /u/ in /dr/, not just because their output grammar disallows /dr/. Of course both aspects may be viewed as tightly linked: speech perception is largely driven by the listeners' native phonology. But what is the processing locus at which the native phonology operates? Recent studies in psycholinguistics or at the crossroad of psycholinguistics and loanword phonology have supported the view that phonological repairs are the reflexion of early, automatic phonetic perception (Berent et al., 2007; Dupoux et al., 1999, 2001; Hallé et al., 1998; Peperkamp & Dupoux, 2003). Certain aspects of this early phonetic perception may reflect universal trends (Berent et al., 2007) but we focus here on those perceptually motivated phonological repairs that are clearly language-specific. The probably most common type of repair is vowel epenthesis, as in Polivanov's example. Various other cases of within-cluster epenthesis are illustrated in Berent et al. (2007). Vowel prosthesis in sibilant + obstruent onset clusters is well documented in the loanword phonology (Fleischhacker, 2001) but not in the speech perception literature. We contributed to fill the gap in examining Spanish listeners' perception of /s/ + stop clusters in utterance-initial position. In Spanish, these sC clusters are illegal. Loans from English suggest that sC clusters are repaired with /e/ prosthesis (e.g., steak > esteak; smoking > esmoquin). But do Spanish listeners hear an initial /e/ in sC clusters? We used spoken words such as especial compared to words such as astuto and digitally removed the initial vowel, leaving pseudowords such as special vs. stuto. This material was presented to Spanish subjects for lexical decision (Hallé et al., 2008a). By and large, subjects were "fooled" by special but not by stuto pseudowords, suggesting they (perceptually) repaired sC with a prothetic /e/ (special > especial; stuto > *estuto), whether this produced a word or not (Fig. 1). Hence, this case of prosthesis is largely prelexical. Now, how automatic is it? To answer this question, we ran a visual masked priming experiment with the same materials presented as printed primes (Hallé et al., 2008b). Although the primes were not consciously visible, we found facilitation priming for special-ESPECIAL but not for stuto-ASTUTO, at prime-target SOAs larger than about 100 ms. These data suggest, among other things, that this phonological repair occurs in print as well as in speech, that it is out of conscious control, and that it requires some additional processing time (Fig. 2). Substitution repairs are much less documented in both loanword phonology and speech perception literature. Massaro and Cohen (1983) showed that the identification of an /r/-/l/ continuum was biased by "phonological context": more /r/ responses after /t, d/ than after /b, g/. But */tl/ is not perceived as /tr/: We have reported evidence for dental-to-velar perceptual shifts in word-initial /dl, tl/ clusters for both French and American but not Hebrew listeners, in line with the ban of /dl, tl/ in English and French but not Hebrew (Hallé & Best, 2007). These data confirmed earlier results obtained with /tl, dl/ nonwords produced by a native speaker of French (Hallé et al., 1998). Using an experimental design similar to that described above for Spanish, we also found that the dental-to-velar substitution repair at work during the overt presentation of /dl, tl/ pseudowords is largely prelexical: tlavier > clavier ('keyboard) vs. tlacard > *clacard, not placard ('closet') (Fig. 3). These data converge to suggest that phonological repairs occur at an early, automatic processing stage. Is this stage preceded by an even earlier stage at which phonetic perception is "faithful"? If such is the case, this stage would be even more deeply buried an more unavailable to consciousness than the phonological repair stage