Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Evaluation of feature-based and gestural theories of Phonology on MEG signal through TRFs
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
José Antonio Gonzalo Gimeno12, Jose J. Pérez-Navarro13, Ihintza Malharin12, Phoebe Gaston4, Christian Brodbeck4, Emily Myers5, James Magnuson517, José María Lahoz-Bengoechea6, Nicola Molinaro7; 1Basque Center on Cognition, Brain and Language, Spain, 2University of the Basque Country, Spain, 3University of Geneva, Switzerland, 4McMaster University, Canada, 5University of Connecticut, United States, 6Complutense University of Madrid, Spain, 7Ikerbasque, Spain
During speech perception, listeners continuously integrate acoustic information with higher levels of linguistic representation. This integration unfolds dynamically over time and shapes neural activity measurable with magnetoencephalography (MEG). At the phonological level, competing theories propose different representational units for speech sounds, including phonetic features (Mesgarani et al., 2014) and articulatory gestures (Chartier et al., 2018). Rather than treating these accounts as mutually exclusive, the present study investigates whether they contribute at different stages of the speech processing hierarchy. Temporal Response Functions relate stimuli descriptors to neural activity over time, making it possible both to evaluate the explanatory power of different models and to identify the temporal windows in which their effects emerge in the MEG signal (Brodbeck et al., 2023). In the present study, phonological descriptors derived from feature-based and articulatory gestural theories are regressed against MEG recordings obtained during speech perception. Model performance is assessed through prediction correlations reflecting the proportion of neural signal variance explained by each representational framework, while TRF weights are used to compare the temporal dynamics of their contributions to neural activity. To evaluate the generalizability of these effects, we have analyzed several MEG datasets from active listening experiments conducted across different languages. This cross-linguistic approach allows us to examine whether the relative contribution of phonetic features and articulatory gestures varies as a function of perceptual and contextual demands. Both phonetic features (df=35, tmax=9.36, p<0.001) and articulatory gestures (df=35, tmax=10.08, p<0.001) contribute to explain MEG activity in a speech in noise listening experiment in Spanish, as compared to models composed only of an acoustic baseline of the stimuli. When validating these models in a clean speech listening experiment, only feature-based regressors (df=39, tmax=4.46, p<0.001) showed any significant difference (df=39, tmax=4.46, p=0.684, for the articulatory gestures). It may be the case that, under favourable listening conditions, semantic top-down information may reduce the need for detailed phonological processing, whereas acoustically challenging conditions may increase reliance on deeper phonological analysis, reflected in stronger or temporally extended neural effects. Future work will further investigate under which conditions phonetic features and articulatory gestures account for greater proportions of neural signal variance and to what extent the relative contribution of each representational level is modulated by listening difficulty and semantic context. In addition, we will examine the temporal dynamics of the different phonological levels, and hypothesize that phonetic feature representations may exert earlier effects on the neural signal, whereas articulatory gestural representations may emerge at later stages of phonological integration. Overall, these findings suggest that experimental approaches to phonological representation can not only inform debates concerning the nature of phonological units, but also provide insight into the temporal organisation of early speech and phonological processing.
Topic Areas: Speech Perception, Phonology