Thomas R. Sawallis
IASCP, 63 Dauer, Univ. of Florida, Gainesville, FL 32611
An experiment is presented concerning perception of edited consonant length, voicing amplitude, and burst+aspiration amplitude in a corpus of natural tokens of French intervocalic [t, d]. The editing is designed using the statistical distributions of those cues as previously measured in the selfsame corpus. This design largely preserves the natural variation of the nontarget cues, and is an advantage over the traditional use of synthesis or prototype based series. Then, mathematical tools borrowed from signal detection theory are used on the perceptual test results to derive sensitivity measurements for the cues. Measurements of this type can be compared autonomously, without trading-relationship-style equivalences, across cues, environments, languages, etc., and this property makes them amenable to use not only in speech perception, but also in formal phonology and general linguistics. In this paper, they are used to explore differential use of cues of listener groups based on native language (French versus other native languages) and on linguistic training (``naives'' versus phoneticians, linguists, language teachers, etc.).