Category Similarity in Multilingual Pronunciation Training

Jacques Koreman


Learners with different native languages (L1) meet different challenges when they learn a foreign language (L2). The Speech Learning Model and the Perceptual Assimilation Model PAM-L2 have led to important insights about these challenges. Among other things, they have shown that the learnability of L2 sounds depends on their similarity to sounds in the L1: L2 sounds are more likely to lead to the formation of new phonetic categories if they differ strongly from L1 categories than if they are similar. The similarity of sounds is hard to quantify objectively, especially if the aim is to do this for many L1-L2 pairs. This limits the models’ practical applicability. The multilingual pronunciation training platform CALST offers exercises for all new L2 sounds. Two implementations of category (dis)similarity are proposed to identify new sounds, one at the level of functional similarity maintaining all L2 phonemic contrasts, the other based on a more fine-grained, multilingual similarity measure, where L2 sounds are considered new if they can contrast phonemically with the most similar L1 sound in any one language. This level of granularity reflects phonetically salient differences between sounds which, when perceived and produced adequately, suffice for high intelligibility and comprehensibility in L2.


 DOI: 10.21437/Interspeech.2018-1938

Cite as: Koreman, J. (2018) Category Similarity in Multilingual Pronunciation Training. Proc. Interspeech 2018, 2578-2582, DOI: 10.21437/Interspeech.2018-1938.


@inproceedings{Koreman2018,
  author={Jacques Koreman},
  title={Category Similarity in Multilingual Pronunciation Training},
  year=2018,
  booktitle={Proc. Interspeech 2018},
  pages={2578--2582},
  doi={10.21437/Interspeech.2018-1938},
  url={http://dx.doi.org/10.21437/Interspeech.2018-1938}
}