What can prosodic transfer tell us about L2 prosody acquisition? Data from Brazilian Portuguese Learners of English

Abstract

This dissertation investigates the acquisition of prosody in a second language (L2), focusing on prosodic transfer from Brazilian Portuguese (BP) to English, based on production and perception data from broad-focus declarative sentences and neutral yes/no questions. The study aims to analyze (i) the production of English yes/no questions by Brazilian learners of English, identifying the occurrence of L1 prosodic influence; (ii) the perception of these productions by native English listeners, in order to assess the extent to which prosodic influence affects sentence-modality interpretation; and (iii) the factors responsible for perceptual difficulties and for processing costs associated with mismatches between contour and function. In BP, neutral yes/no questions are typically realized with a rising–falling contour (L+H*L%) (Moraes, 2008; Castelo and Frota, 2016; Castelo et al., 2018), whereas in North America the unmarked pattern is rising (LH*-H%) (Hedberg, Sosa and Görgülü, 2014). These interrogatives were therefore selected because they present a clear intonational contrast between the two languages, making them particularly suitable for investigating prosodic transfer. The motivation for this study lies both in the scarcity of work addressing influence at the prosodic level and in the absence of L2 acquisition models that explicitly incorporate prosody as a structured component of grammar and communicative competence. The research comprises one pilot study and two production and perception experiments. Production data, obtained through self-paced contextualized reading tasks, show that Brazilian speakers (from MG, SP, and RJ) frequently transfer the BP rising–falling contour to English neutral yes/no questions. The perception experiments, conducted with native English listeners (USA, Canada, and England) using low-pass filtered stimuli from native English and Brazilian speakers, demonstrate that L+H*L% nuclear contours are systematically interpreted as declaratives when occurring in English yes/no questions, resulting in a significant decrease in question-identification accuracy. Binomial mixed-effects statistical models indicate that the low accuracy rate (≈ 40%) is not globally associated with the speaker’s native language, but rather with specific prosodic properties of the stimuli, especially the rising–falling nuclear contour (z = -3.19, β = -3.07, p = 0.0014). The results motivate the hypothesis that prosodic categorization depends on phonological and/or acoustic memory systems that mediate the link between the acoustic signal and phonological interpretation. Neurocognitive studies support this view by suggesting that prosodic processing relies on integrative mechanisms that bind auditory encoding to more abstract interpretive routines, reinforcing the claim that L2 prosody acquisition involves the gradual construction and reorganization of form– function mappings. Finally, the findings have applied implications for speech technologies: text-to-speech and automatic speech recognition systems (TTS/ASR) aimed at naturalness and communicative reliability should model prosodic inventories that are sensitive to language and variety. Taken together, the results show that prosodic transfer can compromise sentence-modality interpretation in L2 and point to the need to integrate prosody into theories of L2 acquisition and into language teaching.