Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The L2 Advantage in Neural Speech Tracking: Evidence from a Multilingual Cocktail-Party Paradigm
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Maros Filip1,2, Kateřina Chládková1,2; 1Charles University, 2Czech Academy of Sciences
Background: How does spoken communication succeed in a multilingual multitalker noise? As for speech comprehension in multitalker babble, listeners’ brains can accurately track the speech stream that they attend to (target) while attenuating the tracking of the unattended speech streams (maskers). The strength of neural speech tracking can be measured with EEG (electroencephalography) as the brain’s oscillatory activity that aligns with the attended speech stream (Kaufman & Golumbic, 2023). This study aims to test to what extent language similarity and language familiarity influence the accuracy of neural speech tracking. Objectives: The general premise is that the brain will track the attended speech stream more robustly than the unattended speech stream. We predict that greater familiarity with the target will be associated with more robust neural tracking, and that greater similarity between the masker and the target language will be associated with less robust tracking (linguistic similarity hypothesis; Brouwer, 2012). To test this, we implemented a multilingual multitalker cocktail-party design with various language combinations: Czech (CZ, native), English (EN, proficient L2), and Dutch (NL, unfamiliar to listeners). This means that each participant was exposed to four different language combinations (target–masker: CZ–CZ, CZ–EN, CZ–NL, EN–CZ). Method: The materials were 10-min podcast clips. Participants (N=64) were native speakers of Czech with intermediate proficiency in English, assessed using LEXtale (Lemhöfer & Broersma, 2012). During the experiment, participants listened dichotically to two simultaneous podcasts and were instructed which one to attend to. EEG was recorded throughout the experiment. MATLAB and the mTRF toolbox (Crosse et al., 2016) were used to fit backward (decoding) multidimensional temporal response functions (mTRFs). The mTRFs were used to reconstruct the speech envelopes for different conditions. Reconstruction accuracy was quantified as correlation coefficients between the reconstructed envelope and the target or masker envelope. Results: The data were analyzed using a linear mixed-effects model with correlation coefficient as the dependent variable, and language combination and stream (target vs. masker) as fixed effects, with random intercepts for participants. Across all conditions, participants tracked the target significantly more robustly than the masker (β = -0.029, SE = 0.006, p < .001). Language familiarity with the masker was not found to affect tracking of the native target: participants’ tracking of the Czech target was comparable across all conditions, i.e., when the masker was unknown (Dutch), foreign (English), or native language (Czech). Contrary to predictions, however, participants tracked English (a less familiar, non-native language) more accurately (β = 0.023, SE = 0.006, p < .001) than they tracked Czech (a more familiar, native language), with the masker being Czech in both conditions. The superior neural tracking of a second language could be explained by increased cognitive effort. In contrast, if the target is native and easy to process, the masker may be more distracting. The greater neural tracking in non-native English compared to native Czech could also be due to more salient language-specific prosodic cues in English. The discrepancy with previous studies supporting the linguistic similarity hypothesis might lie in methodological differences (dichotic vs diotic listening).
Topic Areas: Speech Perception, Multilingualism