Languages systematically differ in both coarticulatory patterns and perceptual organization. Supported in part by NSF grant SBR 9319597, "Cross-language study of perception of coarticulated speech," we have explored the possibility that some language differences in speech perception are linked to language differences in coarticulatory structure. The specific hypothesis tested was that language-specific patterns of coarticulation give rise to listener expectations concerning the acoustic consequences of coarticulation, and that such expectations in turn lead to language-specific patterns of perceptual accommodation to coarticulated speech.
Acoustic and perceptual experiments conducted in our lab support this hypothesis for coarticulatory vowel nasalization and vowel-to-vowel (V-to-V) coarticulation. V-to-V coarticulation in Shona and English illustrates our approach and findings. Acoustic analyses showed different coarticulatory patterns for the two languages: English exhibited more extensive carryover V-to-V coarticulation (in F1 and F2) than did Shona; and unstressed vowels in English, but not Shona, coarticulated more than did stressed vowels. Subsequent perceptual tests conducted in Zimbabwe and the U.S. manipulated these coarticulatory effects, testing Shona and English listeners' discrimination of both Shona and English vowels spliced into coarticulatorily appropriate as compared to inappropriate contexts. (For example, /a/ excised from /abi/ - i.e., [ai] - was spliced into different /_bi/ tokens, creating appropriate [aibi] sequences, but was also spliced into /_ba/ contexts, creating inappropriate [aiba] sequences. The design is similar to one previously used by Carol Fowler.) If listeners use coarticulatory context to achieve perceptual constancy, then acoustically distinct but phonologically identical vowels in appropriate contexts (e.g., the initial /a/s in [aibi] and [aaba]) should sound similar, because the acoustic differences will be (correctly) attributed to context. This should not happen for the vowels in inappropriate contexts (e.g., [aiba]) because, lacking a coarticulatory source, their acoustic structure should not be heard as due to context. This is precisely what we found, but primarily for native(-like) coarticulatory effects: the more the coarticulatory effects resembled native-language patterns of V-to-V coarticulation, the more likely Shona and English listeners were to ascribe these effects to their coarticulatory source. We tested the generality of these findings with synthetic vowel continua embedded in different vowel contexts and again found evidence of parallels between language-specific patterns of coarticulation and language-specific patterns of perceptual responses to coarticulated speech.
That perception of coarticulated speech is attuned to native-language coarticulatory patterns was also demonstrated for vowel nasalization. Acoustic analyses of anticipatory nasalization in English and Thai showed more nasalization in English (i.e., temporally and spectrally more extensive effects of a nasal consonant on the preceding vowel). A perceptual rating test in which Thai and English listeners judged the relative nasality of oral and nasalized vowels spliced into nasal (N_N) and oral (C_C) contexts showed that English listeners were more likely than Thai listeners to attribute the acoustic differences between the vowels to their coarticulatory context. Discrimination testing (using the paradigm described for V-to-V coarticulation) yielded less robust cross-language perceptual differences, but those that were found were consistent with native-language coarticulatory structures.
Font troubles? Download the pdf.