Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Optimizing the multivariate stimulus-response framework for identifying EEG responses to statistically-dependent stimulus representations

Poster Session B - Sunday, March 8, 2026, 8:00 – 10:00 am PDT, Fairview/Kitsilano Ballroom

Konrad Dapper¹ (), Sarah Hollywood¹, Taylor Dool¹, Blake Butler^1,2, Marc Joanisse^1,3; ¹Department of Psychology, University of Western Ontario, London, Canada, ²National Centre for Audiology, University of Western Ontario, London, ON, Canada, ³Haskins Laboratories, New Haven CT, USA

An emerging approach to studying speech-related EEG processing involves fitting a multivariate forward model (mTRF) in which stimulus properties are used to predict neural electrical responses. A key question in that regard is how to represent the stimulus as input. Two attributes of speech stimuli commonly used as inputs to forward models are the spectrogram of speech sounds and binary feature vectors representing its phonetic properties. Both of these are known to contribute to the neuroelectric response evoked by speech, individually and together. However, a significant challenge is how to best isolate their respective effects, given that they share some degree of mutual information. To address this, we propose enhancements to the mTRF framework using a new statistical method. Forward model specificity is increased through cyclic permutation. Implementation requires tackling three technical challenges: effectively managing correlations among channels in multi-sensor EEG data; reducing effects of endogenous drift and recording artifacts; and reducing the noise sensitivity of the riddge regression hyperparameter via numerical simulation. The effectiveness of this approach is tested on EEG data of natural speech listening in 24 adults with normal hearing. Our method identifies neural responses specific to phonetic and acoustic input variables, even when data were insufficient for the conventional mTRF model to achieve the same. We discuss the benefits of this approach to developing multivariate EEG models with greater sensitivity and stimulus-specificity.

Topic Area: LANGUAGE: Other