
Abstract This study examines whether audio-capable large language models (LLMs) align with human patterns of L2 English accentedness judgment. Google Gemini Flash and Gemini Pro rated the L2 English speech of Brazilian Portuguese speakers from the CSLU Foreign Accented English Corpus ( Lander, 2007 ) and were compared against native American English listeners. Both models assigned systematically more severe ratings, using the highest category of the rating scale at frequencies not observed in the human data. The shift was most pronounced in Gemini Pro ( OR = 8.25), suggesting that greater model capacity does not yield closer alignment with human judgment. The two models also failed to agree with each other. These divergences raise concerns about the validity of LLM accentedness ratings as proxies for human judgment.
This study examines whether audio-capable large language models (LLMs) align with human patterns of L2 English accentedness judgment. Google Gemini Flash and Gemini Pro rated the L2 English speech of Brazilian Portuguese speakers from the CSLU Foreign Accented English Corpus ( Lander, 2007 ) and were compared against native American English listeners. Both models assigned systematically more severe ratings, using the highest category of the rating scale at frequencies not observed in the human data. The shift was most pronounced in Gemini Pro ( OR = 8.25), suggesting that greater model capacity does not yield closer alignment with human judgment. The two models also failed to agree with each other. These divergences raise concerns about the validity of LLM accentedness ratings as proxies for human judgment.
Abstract Phonetic and prosodic variability is a defining characteristic of Spanish heritage speakers. This paper presents a synthesis of research on segmental, vocalic, and suprasegmental features in heritage Spanish and argues that such variability reflects systematic bilingual adaptation shaped by task demands, literacy experience, and patterns of language use. Heritage speakers typically preserve core phonological contrasts and demonstrate high perceptual sensitivity, even when phonetic realization varies across contexts. Particular attention is paid to prosodic phenomena, especially stress and intonation, given their role in intelligibility, pragmatic meaning, and pronunciation assessment. On the basis of this evidence, the paper examines implications for pronunciation assessment and placement practices, highlighting the risk of misclassification when evaluation relies primarily on literacy-based tasks. The final section translates empirical findings into pedagogical implications, advocating intelligibility-oriented pronunciation targets, contextualized instructional approaches, and greater integration of phonetic research into teacher education.
Abstract The present study investigated the effectiveness of classroom-based perception training and different types of corrective feedback on learners’ perception of the German vowel pairs /ʊ/–/ʏ/ (as in Mutter – Mütter ) and /ɔ/–/œ/ (as in Tochter – Töchter ). Participants included 68 L2 learners of German in intact university courses. They completed an in-class perception training and were assigned to one of three feedback conditions: target feedback, nontarget feedback, or a combination of both. Results from an AX task revealed that learners with low prior perceptual accuracy benefited from the training, whereas learners with higher prior perceptual accuracy did not. No differential effects emerged among the three feedback types, suggesting that in ecologically valid classroom contexts, the mere presence of feedback may be more influential than its specific form. These findings highlight the pedagogical value of early perception-focused pronunciation instruction in classroom contexts.