Phoneme deletion and fusion in conversational speech

Major phonetic variations and reductions are observed in casual speech [1] with no effect on listener's comprehension. Nevertheless, if listeners are able to identify some kind of reduction (vowel deletion [2] [3]), they are deaf to others. Reduction may be attributed to a physiological phenomenon (undershoot) but is also related to language structure and communication constraints [4]. In this way, we hypothesize that reduction different forms are related to different phonological or linguistic processes. In this paper we observed two types of reduction which are deletion (one or more phonemes are not realized) and fusion (two or more phonemes are merged). Fusion is less perceptively salient. Deletion and fusion have been observed in a selection of familiar conversations [5] and have been annotated according two principles: for deletion, the unrealized segment is clearly identified; for fusion, several phonemes (at least two) cannot be distinguished individually and it is not possible to determine which is realized or not. First results show that fusion and deletion are quite frequent (around 12% of words and 7% of phonemes are affected) but depend on speaker's specificities. For fusion, all phonemes are merged with no preference for some of them. On the other hand, vowels are more often deleted (74%) than consonants. Reductions (deletion and fusion) most often appear in short function words (pronouns, determinants, prepositions, etc.). But more surprisingly, fusion affects several phonemes (two phonemes=54%, and 46% for more than two) overlapping two or more words, while deletion essentially affects a single phoneme (90%) within a single word. These results show that deletion seems to be precisely located and delimited, while fusion of phonemes is more disseminated and may overlap several words. Our hypothesis is that deletion is an internal word process while fusion would be determined by larger linguistic constraints (prosody, syntax, discourse).