A produção do contraste de duração por aprendizes brasileiros de inglês como L2: um modelo de Máxima Entropia para ponderação de pistas

Abstract

This study investigates how Brazilian learners of English as a second language produce the length contrast between long-tense vowels ([i, u]) and short-lax vowels ([ɪ, ʊ]), focusing on cue-weighting strategies. Given that Brazilian Portuguese lacks a phonemic vowel length contrast, a question that arises is how learners process this feature in English. The study evaluates the influence of acoustic cues such as vowel duration, F1, and F2 in the learners’ production across different proficiency levels. Using Pillai scores to assess category separation and a Maximum Entropy (MaxEnt) model to estimate cue weights, our analysis reveals how these cues are integrated into learners’ phonological systems. The results show that lower proficiency learners heavily rely on vowel duration, while higher proficiency learners incorporate spectral cues (F1, F2) more consistently, especially for front vowel contrasts ([i, ɪ]). For back vowels ([u, ʊ]), however, even advanced learners show limited cue integration, as indicated by significant overlap in their acoustic space. Pillai scores demonstrate greater category separation for advanced learners, particularly in front vowels, but inconsistencies remain in back vowel distinctions. The MaxEnt analysis highlights that duration receives higher weights for back vowel contrasts, while F1 and F2 play more significant roles for front vowel contrasts at higher proficiency levels. These findings suggest that while learners progressively adjust their cue-weighting strategies as proficiency levels increase, L1 transfer effects remain prominent, particularly in the reliance on vowel duration, thus contributing to our understanding of how L2 phonological contrasts are developed.
Keywords: Cue Weighting. Length Contrast. Maximum Entropy Model. L2 Phonology Acquisition. Vowel.