Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Gray matters: Impacts of electrode placement on responses recorded during natural speech listening
Poster Session D, Saturday, October 26, 10:30 am - 12:00 pm, Great Hall 3 and 4
Anna Mai1, Andrea Martin1,2; 1MPI Psycholinguistics, 2Donders Centre for Cognitive Neuroimaging
Intracranial electroencephalography (iEEG) is increasingly recorded using intracerebral depth electrodes (i.e., stereo EEG) (Engel 2018, Abou-Al-Shaar et al. 2018, Miller et al. 2021) that record from a combination of gray matter (GM) and white matter (WM) sites. Previously, WM sites were assumed to be electrically neutral and used as reference channels (Mercier et al. 2022). However, activity recorded in WM contains signal from both nearby and distant GM (Mercier et al. 2017, Rizzi et al. 2021), increasing its relevance to models of cognition and language. This project enriches knowledge of WM’s role during language processing by showing how signal correlation varies not only as a function of distance between electrodes but also as a function of electrode placement in GM vs. WM, band frequency, and language familiarity. Intracranial EEG was recorded while nine English speakers with no Catalan familiarity listened to excerpts of conversational American English from the Buckeye Corpus (Pitt et al. 2007) and excerpts of conversational Catalan from the Corpus del Catalá Contemporani (Alturo et al. 2002). Stimulus-timelocked band-limited activity was subsequently extracted from the recordings. The similarity between activity recorded at pairs of electrodes was quantified using the Pearson correlation coefficient and was calculated for both phase and analytic amplitude of activity. Separate linear mixed-effects models were fit for phase and analytic amplitude. The correlation between sites (phase or amplitude) was modeled with participant identity as a random effect and combinations of electrode placement (GM vs. WM), language condition (English, Catalan, or silence), band (delta, theta, alpha, beta, gamma, or high gamma), and distance between electrodes as fixed effects. Models were compared within response variable type (phase or amplitude) using the Akaike Information Criterion (AIC). Best-fit models carried 100% of the cumulative model weight and had AIC scores >30 lower than other models. For both phase and amplitude, best fit models included all fixed effects, all two-way interactions, and the three-way interactions band*matter*distance and condition*matter*distance. Beta weights for GM>WM were negative and significant (p<0.0001), indicating that both phase and amplitude are less correlated across sites in GM than in WM. Similarly, weights for Catalan>Silence and English>Silence were negative (p<0.0015), indicating that responses during speech listening are less correlated than during silence. The interaction between GM>WM and English>Silence was positive (phase: p=0.031, amplitude: p=0.047), indicating that both phase and amplitude are more correlated across sites in GM when participants are listening to English. Results for other model features will be discussed further at the poster. Although correlations in WM and during silence are generally higher across sites, correlations in GM are higher when participants listen to a familiar language. These results most likely reflect a higher SNR in GM combined with better cortical tracking of a familiar language. Combined with effects of band and distance on phase and amplitude correlation across sites, these results paint a complex picture of the characteristics of language-related data recorded in WM and reiterate the impact of language familiarity on recorded response.
Topic Areas: Speech Perception,