ISCA Archive Interspeech 2018
ISCA Archive Interspeech 2018

The Retroflex-dental Contrast in Punjabi Stops and Nasals: A Principal Component Analysis of Ultrasound Images

Alexei Kochetov, Matthew Faytak, Kiranpreet Nara

Many languages of South Asia show a phonemic contrast between retroflexes and dentals across different manners of articulation. This contrast, however, tends to be less phonetically distinct and more variable in nasals. The goal of this paper is to examine the overall similarity of the retroflex-dental contrasts in Punjabi stops and nasals. Ultrasound tongue imaging recordings were obtained from 14 Punjabi speakers producing /ʈ,ɳ,t,n/ in the /ba_ab/ nonsense word context. Selected video frames were fed to a principal component analysis (PCA); the output was used for (1) training a linear discriminant model on one manner that discriminates place and (2) testing it on the other manner. The results showed 100% correct classification of the contrast (retroflex or dental) in stops and 92% correct classification in nasals in the training data. The classification was much poorer across different manners: on average 67% of stops and 57% of nasals were classified correctly based on training sets with nasals or stops, respectively. In both cases, retroflex responses were more common. These results suggest that the tongue configurations for Punjabi retroflex and dental consonants differ by manner of articulation. The contrast is also overall less robust in nasals than in stops, confirming previous reports.


doi: 10.21437/Interspeech.2018-1457

Cite as: Kochetov, A., Faytak, M., Nara, K. (2018) The Retroflex-dental Contrast in Punjabi Stops and Nasals: A Principal Component Analysis of Ultrasound Images. Proc. Interspeech 2018, 202-206, doi: 10.21437/Interspeech.2018-1457

@inproceedings{kochetov18_interspeech,
  author={Alexei Kochetov and Matthew Faytak and Kiranpreet Nara},
  title={{The Retroflex-dental Contrast in Punjabi Stops and Nasals: A Principal Component Analysis of Ultrasound Images}},
  year=2018,
  booktitle={Proc. Interspeech 2018},
  pages={202--206},
  doi={10.21437/Interspeech.2018-1457}
}