Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Lightning Talks
Comparing neural measures of prediction between native and second language listeners in continuous speech
There is a Poster PDF for this presentation, but you must be a current member or registered to attend SNL 2023 to view it. Please go to your Account Home page to register.
Poster C32 in Poster Session C, Wednesday, October 25, 10:15 am - 12:00 pm CEST, Espace Vieux-Port
Craig Thorburn1, I.M Dushyanthi Karunathilake1, London Dixon1, Ellen Lau1, Jonathan Simon1; 1University Of Maryland
When listening to speech, speakers continuously anticipate upcoming phonemes, words and concepts in their native language using prior context (Ferreira and Chantavarin, 2018), integrating information at the sublexical, lexical and sentence level. However, second language learners have been shown to rely less on syntactic information and previous input to predict upcoming material, including not using grammatical gender to restrict prediction of an upcoming noun (Lew-Williams & Fernald, 2010), waiting longer to predict in filler-gap contexts (Omaki & Schulz, 2011) and not exhibiting the same evoked responses as native speakers (Hahne & Friederici, 2001). These differences are generally taken as evidence that listeners do not actively predict upcoming material in a second language. However, more recently it has been suggested that task-related effects which alter online processing, as well as differences in the distributional information that listeners have of their second language could give rise to results which might falsely indicate that these listeners do not form predictions (Kaan, 2014). In order to investigate these mechanisms during online processing in a second language, we analyze continuous naturalistic data where native Sinhala and Mandarin speakers listen to an English audiobook while magnetoencephalography (MEG) responses are recorded. Using multivariate Temporal Response Function (mTRF) analysis (Ding and Simon, 2012) on the continuous speech and neural responses, and comparing to a corpus of native English speakers listening to the same audiobook (Brodbeck et al. 2022), we demonstrate that second language listeners exhibit very similar responses to native speakers and do predict upcoming input integrating phoneme, lexical and sublexical level contextual information. Responses in Sinhala and Mandarin speakers are similar to each other and those of native English listeners. We see few differences in how the neural responses can be accounted for by prediction features and the time lag of the mTRF response. Small differences in lateralization can be seen with second language learners showing an increased bias towards prediction-modulated speech responses in the left hemisphere over native speakers. These results provide evidence that second language listeners may indeed leverage prediction in similar ways as native listeners during online continuous listening tasks. We posit that these results differ from previous results for two reasons. First, task effects might play a role in previous results and second language listeners may show more markers of prediction during a continuous listening paradigm than constrained experimental contexts. Secondly, knowledge of statistical information of a language — ie. phoneme and lexical transition probabilities — likely plays an important role in whether prediction is observed in second language learners. The participants in this experiment were advanced learners of English who currently study at an English-speaking university, meaning they likely have a robust representation of the distributional information of English, resulting in similar neural responses to native speakers. We conclude that naturalistic listening studies are particularly useful in investigating mechanisms in second language listeners and that this work should go hand-in-hand with controlled experiments in order to uncover how these processes differ between language backgrounds.
Topic Areas: Multilingualism, Speech Perception