Surprisal in non-native speech sound learning: A replication and extension of Nixon (2020) with Japanese vowels, Mandarin fricatives/affricates, and Southern-Min tones

Non-native speech acquisition involves more than statistical tracking of speech cues. This study explores how surprisal (i.e., prediction-error) drives speech sound acquisition, replicating and extending Nixon's (2020) error-driven learning framework. We examine surprisal in the learning of Southern-Min tone contrasts and extend the experiment to Japanese vowel length and Mandarin fricatives and affricates, representing both suprasegmental and segmental contrasts. A total of 146 native English-speaking participants (51 learning Japanese, 41 Mandarin, 54 Southern-Min) participated in one of two stimuli presentation orders: discriminative (speech-soundthen image) or non-discriminative (image then speech-sound). This study makes two key contributions: First, it replicates Nixon's (2020) framework by expanding the generalizability of error-driven learning across different speech contrasts. Second, it includes a methodological extension by incorporating eye-tracking to operationalize surprisal, providing real-time insights into cognitive processes during training. Results show that surprisal, particularly in low-frequency items, facilitates learning in the discriminative condition. These findings challenge the view that speech-sound learning is purely statistical and suggest that eliciting surprisal enhances the learning of challenging speech contrasts, such as Japanese vowel length, Mandarin fricatives, and Southern-Min tones, with broader implications for the acquisition of non-native speech sounds (Burnham, 2013; Kim, 2023).