Phoneme, Phone Boundary, and Tone in Automatic Scoring of Mandarin Proficiency

Jiahong Yuan, Mark Liberman


Not every phone, word, or sentence is equally good for assessing language proficiency. We investigated three phonetic factors that may affect automatic scoring of Mandarin proficiency — phoneme, phone boundary, and tone. Results showed that phone boundaries performed the best, and within-syllable boundaries were better than cross-syllable boundaries. The retroflex consonants as well as the vowel following these consonants outperformed the other phonemes. Tone0 and Tone3 outperformed the other tones, and ditone models significantly improved the performance of Tone0. These results suggest that phone boundary models and phoneme- and tone- dependent scoring algorithms should be employed in automatic assessment of Mandarin proficiency. It may also be helpful to separate phoneme and tone scoring prior to the combination of individual scores, as we found that the worst phoneme and the best tone, with respect to automatic scoring of Mandarin proficiency, appeared in the same word.


DOI: 10.21437/Interspeech.2016-510

Cite as

Yuan, J., Liberman, M. (2016) Phoneme, Phone Boundary, and Tone in Automatic Scoring of Mandarin Proficiency. Proc. Interspeech 2016, 2145-2149.

Bibtex
@inproceedings{Yuan+2016,
author={Jiahong Yuan and Mark Liberman},
title={Phoneme, Phone Boundary, and Tone in Automatic Scoring of Mandarin Proficiency},
year=2016,
booktitle={Interspeech 2016},
doi={10.21437/Interspeech.2016-510},
url={http://dx.doi.org/10.21437/Interspeech.2016-510},
pages={2145--2149}
}