One of the main ultimate objectives of speech perception research is the description, on a neurophysiological basis, of the mechanisms involved in human phoneme recognition. Direct probing is obviously out of the question, but some novel inspiration for hypotheses may be gleaned from human phonemic productions that elicit correct and robust identification in spite of acoustical patterns that blatantly infringe known (production and/or perception) "rules". To find a number of productions as such, the detailed analysis of a multinational database of 64 non-professional speakers was undertaken. Some productions showed to grossly violate the previously known "approximate rule" that states that in a stop-high vowel CV, a "slightly ascending F2 transition cues dental place, whereas a markedly ascending transition cues labial place". A biologically-plausible model of a processing structure ancillary to formant-transition perception is proposed that is compatible not only with these productions, but also with the more common productions that conform to the "approximate rule" mentioned. Furthermore, the model appears to explain why a slightly ascending transition (prevalent in dental-stop, high-F2 vowel CV's) may be perceptually similar to a markedly descending transition (prevalent in dental-stop, low-F2 vowel CV's).
Bibliographic reference. Sa Marta, Eduardo / Perdigao, Fernando / Vieira de Sa, Luis (1995): "Researching the processing structures of human phoneme recognition by analysis of natural stop-consonant-vowel utterances that elicit correct recognition through unusual acoustic patterns", In EUROSPEECH-1995, 2285-2288.