An instrumental study of alveolar to velar assimilation in slow and fast speech using EPG and EMA techniques
This thesis evaluates the widely-held notion that place assimilation is (i) more frequent at faster rates of speech and (ii) a gradual phonetic process. The latter view is based on previous small-scale EPG studies which showed evidence of partial alveolar assimilations lacking complete stop closure on the alveolar ridge but with a residual tongue body gesture. For the present study, EPG data from 10 speakers were collected. Two experimental sequences, /n#k/ and /ŋ#k/, embedded in meaningful sentences, were produced by subjects 10 times each in a slow/careful style and 10 times each in a fast/casual style. The first sequence captures the potential site of assimilation and the second is a lexical velar-velar sequence with which cases of complete assimilation can be compared. The results showed that, overall, assimilation was more frequent in fast speech than in careful speech, although timing analysis revealed that assimilation is not the automatic consequence of rate-induced changes in intergestural timing of /n#k/. In fast speech, six of the ten speakers showed relatively consistent assimilatory preferences: they either produced only complete assimilations or they never assimilated. However, four speakers showed considerable intra-speaker variability. Two of the four produced either full alveolars or complete assimilations in the manner of a categorical opposition (complete assimilations were indistinguishable from control /ŋ#k/ sequences). The other two speakers produced a continuum of forms that could be ranked from full alveolars to complete assimilations via partial assimilations. Using the same stimuli, a follow-up combined EPG/EMA study was carried out, the purpose of which was to look for reduced coronal gestures undetectable in tongue-palate contact-only data. Two 'categorical' assimilators were re-recorded and these gestures were not found. This supports the interpretation that for some speakers assimilation is determined at a higher level through the application of a cognitive rule, while for others variation is 'computed on-line' during speech production itself. Current phonological models of assimilation are found to be unable to capture both gradient effects and more radical feature-sized substitutions under a single framework.