Voice assimilation in French obstruents: categorical or gradient?

This work contributes to the issue of categoricity versus gradiency in natural assimilations. We focused on voice-assimilation in French and started from the assumption that the main cue to obstruent voicing is glottal pulsing. We quantified glottal pulsing continuously with a single acoustic measure — the proportion in duration of voiced portion(s) within a consonant — which we call v-ratio. We used a large corpus of French radio and television speech to compute v-ratios for all the obstruents appearing in word-final to word-initial obstruent contacts. The results were analyzed in terms of v-ratio distributions, which were compared with theoretical distributions predicted by two contrasted hypotheses on the mechanisms of assimilation: categorical switch versus v-ratio shift. The comparisons strongly suggested that, although voicing itself can be incomplete, voice assimilation is essentially categorical in terms of v-ratio. We discuss this result in the light of recent perceptual data showing sensitivity to extremely subtle acoustic differences: secondary cues to voicing do not seem to follow the same pattern of categoricity as glottal pulsing.