Indranil Dutta, Charles Redmon, Meghavarshini Krishnaswamy, Sarath Chandran, and Nayana Raj (in prep.) Paper from 2019 ICPhS Proceedings: pdf
Measures of coarticulation from two models of consonant-vowel formant transitions – locus equations (LEs) and target-locus scaling (TLS) – are presented for Malayalam coronals. For both VC and CV transition models, LE slopes exhibited an ordering inconsistent with expectations based purely on estimates of articulatory complexity (i.e., retroflex < alveolar ≤ dental). In the CV context, LE slopes were found to be: tt < ʈʈ < t̪t̪, whereas in the VC context the ordering of dental and retroflex reversed to yield tt < t̪t̪ < ʈʈ. Alveolars consistently had the flatter slopes, a result predictable from their sparse representation in the lexicon. The retroflex < dental ordering in CV context is consistent with predictions based on articulatory complexity. The VC results suggest an asymmetry in coarticulation of retroflex stops with preceding and following vowels. These results suggest that coarticulatory resistance is mitigated by lexical imbalance. We discuss the production dynamics underlying these patterns through TLS models of the VC and CV formant trajectories.
In addition to the above acoustic work, we have now collected ultrasound data on six speakers (also a part of the acoustic study) and are examining the alignment between their tongue trajectories and acoustic formant-based models of coarticulation.