14th International Congress of Phonetic Sciences (ICPhS-14)
San Francisco, CA, USA
In this study we investigate three properties of the left vowel context speech recognizer (CSR) in forced recognition mode was used. The which we hypothesize to underlie deletion of /r/ in postvocalic, CSR had the task to decide whether /r/-deletion was applied or not in preconsonantal position in Dutch spontaneous speech: vowel type a word. (schwa, full vowel), vowel length (long, short) and lexical stress (+,-). The accuracy of forced recognition in selecting the correct For each of five categories 90 instances with possible /r/-realizations variant was checked in an experiment  in which the CSR's were extracted from a large speech database containing man-machine responses were compared with those of nine expert listeners who dialogues in an automatic train-table inquiry system. The frequency of carried out the same task. For /r/-deletion the CSR turned out to select /r/-deletions in these 450 cases was investigated on the basis of variant the correct pronunciation variant in 75.6% of the cases, while for the selection by a CSR and human transcriptions of the same material. listeners this percentage varied between 74% and 93%. This indicated Loglinear analyses revealed that /r/-deletion was significantly more that there was a good correspondence between recognizer response frequent after schwa than after full vowels, and that vowel length and and human ratings for /r/-deletion. lexical stress were not significant. This appeared from both CSR data The results of forced recognition showed that in a corpus and human transcriptions. The only discrepancy between the two sets containing 214,102 words in which /r/-deletion could be applied of results concerns the absolute frequencies of /r/-deletion. Possible 16,865 times, it was actually applied in 47.6% of the cases. Moreover, explanations are discussed.
Bibliographic reference. Cucchiarini, Catia / Heuvel, Henk van den (1999): "Postvocalic /r/-deletion in Dutch: more experimental evidence", In ICPhS-14, 1673-1676.