UC Berkeley PhonLab Annual Report
Speech Production Patterns in Producing Linguistic Contrasts are Partly Determined by Individual Differences in Anatomy
- Author(s): Johnson, Keith
- et al.
This study explored correlations between (a) measures of vocal tract anatomy and (b) measures of articulatory/linguistic contrasts in vowels and coronal fricatives. The data for the study come from the Wisconsin X-Ray MicroBeam Database (Westbury, 1994). The anatomical measures included vocal tract length, oral cavity length, palate size and shape, as well as measures of maximal tongue protrusion and jaw wagging amplitude. Measures of the articulatory vowel space included the range of x and y location at vowel midpoints for four pellets on the tongue, the interpolated highest point of the tongue, and the locations of pellets on the upper and lower lips and to the lower incisor. For each of these clouds of vowel midpoint measurements, the orientation of variation was also measured. For fricatives, measures of tongue advancement and tongue tip lowering were taken. The results showed that the articulatory vowel space was related to both the length of the vocal tract, and to the shape of the palate, while fricative variation was related to palate parameters alone. In simple correlations, the percentage of articulatory variance between segments that could be predicted by anatomical characteristics was modest; never more than 36% for vowels and 25% for fricatives. Canonical correlation analysis found two anatomical factors that predict articulatory patterns jointly in vowels and coronal fricatives. The first canonical variable found a relationship between vocal tract length/palate depth and vowel tongue vertical range and jaw motion. Talkers with long vocal tracts and deep palates showed large tongue vertical range and small jaw range. The second canonical variable found a relationship between palate depth and tongue tip raising in coronal fricatives. Talkers with more shallow palate tended to have a tongue-tip up posture in fricatives. Phonetic tagging for the XRMBDB is made publicly available by this project.