In L2 acquisition, learners face the challenge of detecting articulatory sources from ambiguous auditory signals of non-native phonemes. The presence of many-to-one articulatory-acoustic mappings leads to multiple solutions in phonemic formation. This study investigated the acquisition of non-native Cantonese vowels /ɛ, œ, ɔ/ by Mandarin learners. Nineteen native Mandarin speakers (13F, 6M, mean age: 23), who had received formal Cantonese training, participated in perception, elicitation, and imitation tasks of Cantonese and Mandarin. Synchronized ultrasonic tongue images, lip videos, and speech audio were recorded and analyzed. Results showed that while learners achieved target-like production for all three vowels in imitation, accuracy and consistency varied significantly between the more peripheral vowels /ɛ, ɔ/ and the less peripheral vowel /œ/ in elicitation. /ɛ, ɔ/ exhibited less cross-speaker variance and more target-like realizations, despite occasional diphthongization caused by L1 transfer (e.g., /ɛ/ as /ɛj/ and /ɔ/ as /wɔ, ow/). In contrast, vowel /œ/ demonstrated considerable inter-speaker and intra-speaker variability, with variants including œ, ɤ, ɔ and most frequently /ɔy/, where the rounding and fronting gestural timing varied among speakers. These findings suggest that greater ambiguity in articulatory-acoustic mapping contributes to variability in forming novel vowel categories.
Liu et al. (Wed,) studied this question.