| Speech sound assimilation is a phenomenon widely attested across languages where speech units occurring in proximity in continuous speech become more like each other. From previous literature, there are opposite asymmetries in its direction of operation for segmental feature (e.g. consonant voicing, nasality, place of articulation, etc.) and tonal feature (i.e. fundamental frequency, F0) assimilations: the former one is dominantly leftward / regressive while the latter one mainly rightward / progressive. This dissertation presents a series of six experiments conducted to investigate the effect of speaking rate on directionality of speech sound assimilation, with Cantonese as the target language. Segmental feature assimilation was studied in the first four experiments: Experiment 1 elicited native speakers' production of target syllables containing two consecutive consonants embedded in a carrier sentence, under different speaking rates. We noticed assimilative contextual effects from onset consonants acting regressively on F2 (second formant) transition of the previous coda consonant, and that as speaking rate increases, this influence becomes more prominent. Interestingly, this regressive direction of influence agrees with the cross-linguistically dominant (leftward) direction of stop place assimilation; To study the perceptual consequences of the observed rate-induced formant variations, in Experiment 2, we presented to another pool of subjects a subset of samples from Experiment 1 for a consonant identification task. Identification error analysis showed that those acoustic variations indeed caused a corresponding confusion pattern at the listener side, rather than being perceived merely as allophonic variations; Experiment 3 and 4 presented our effort to draw additional empirical support from real-world language uses apart from behavioral data obtained under tightly controlled laboratory settings. Experiment 3 was a frequency analysis into a contemporary spoken corpus on consonant variations. Agreeing with our previous acoustic experiments, it was found that coda consonants were often influenced in an assimilative manner by its following onset consonant rather than the other way round; Experiment 4 compared two Cantonese spoken corpora constructed in different times for revealing the diachronic trajectory of consonants. Several well-attested changes surfaced in the comparison. Besides, frequency results were matched against results from simulation modeling to provide additional pieces of supporting evidence to our conclusion of regressive segmental feature assimilation; We tried to extend the methodology in Experiment 1 and 2 to study tonal feature assimilation in Experiment 5 and 6. Data from Experiment 5 revealed, under high speaking rate, a rightward shift of certain tonal features, in particular, F0 peak, relative to segmental boundaries. The corresponding perception Experiment 6 made use of samples from Experiment 5 to elicit subjects' perceptual responses when presented those rightward shifted F0 contours. Perception results indicated that listeners were able to recover the intended tones even given the distorted F0 contours, probably with the help of other co-varying acoustic cues present; On a slightly different topic, Experiment 7 presents an attempt to quantitatively document ongoing merger of two tone pairs in Hong Kong Cantonese, namely, T2-T 5 and T3-T6. Results showed statistically significant tone confusion rates across subject ages. Among the two tone pairs studied, the first one was more severely confused while the second one was found to merge at a slightly faster pace. To sum up, since speech rate fluctuation is inevitable in everyday oral communication, if the speech rate-induced acoustic variations obtained in production experiments are found to cause perceptual confusion at the listener side, as exemplified in Experiment 2, in the same direction as reported in the literature on historical sound assimilation processes, speech rate is highly probably one of the candidates driving diachronic speech sound assimilation processes, as found in Experiment 3 and 4. The whole dissertation attempted to re-construct the whole speech sound evolution process with experimental results from multiple perspectives. |
