Marco Aurelio Silva Fonseca
East Asian Studies, UCI
“L2 speech: production, perception and computational modeling”
Abstract: The acquisition of second language (L2) speech sounds can be challenging to L2 learners. It is well established that phonetic similarities and differences in phonemic status between a learner’s first language (L1) and L2 influence the acquisition process (Best, 1995; Flege, 1995). In this talk I present how research in L2 phonology can be approached through production and perception experiments, alongside computational modeling. Specifically, I employed long short-term memory recurrent neural networks (LSTM RNNs, Hochreiter and Schmidhuber, 1995) and examined the role of distinctive features within this framework. While prior work has argued that distinctive features are not essential for modeling phonotactic patterns in a language (Mirea and Bicknell, 2019), I provide additional evidence derived from bilingual models and principal component analysis (PCAs).
I will also present the findings of an experiment that I conducted with 22 L2 learners of Japanese who are native speakers of English, focusing on the acquisition of segmental (vowels and consonants) and suprasegmental (pitch accent) contrasts. Results indicate that although L2 learners are able to perceive differences in pitch accent, they encounter difficulties in assigning meaning to pitch-accent minimal pairs. Finally, I will discuss a planned experiment on L2 vowel production, specifically examining the production of English vowels by native Japanese speakers and Japanese vowels by native English speakers.
May 12, 2026
11:00AM-12:20PM
Location: SSPB 1222
Zoom link: https://uci.zoom.us/j/91644840724