Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural tracking of phonetic predictability under varying attentional load
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Dominic West1, Sophie Forster1, Jeff Mitchell1, Ediz Sohoglu1; 1University of Sussex
Predictive processing is a theoretical framework that proposes the brain continuously generates predictions about incoming sensory input. It suggests that perception is not a passive process of receiving stimuli but an active process of minimising prediction error – the difference between expected and actual sensory input. Previous neuroimaging studies have reached opposing conclusions as to whether attentional focus is required for the predictive processing of speech. One explanation for these conflicting results is that the tasks used in experiments have varied in their level of perceptual difficulty or ‘load’. According to the Perceptual Load Theory of attention, in a task with low perceptual load, spare brain resources can ‘spill over’ to process task-irrelevant stimuli. In a task with high perceptual load, no spare capacity is available and task-irrelevant stimuli remain unprocessed. In this study we manipulate perceptual load using a visual task whilst participants listen to an audiobook. The task requires participants to track 1, 3, 4, or 6 moving dots within a larger set of identical dots. Trials last 5 seconds and the audiobook plays continuously throughout each 8 minute block. We also include a control condition whereby participants are instructed to attend to the audiobook and ignore the visual task. In this condition, participants are instructed to indicate with a keypress if they detect an occasional repeated sentence in the audiobook. We use EEG to measure changes in the neural tracking of speech predictability between conditions. In this way, we can directly assess the relationship between attention and predictive processing in the context of speech comprehension. To assess neural tracking of speech features we regress changes in phoneme surprisal against the EEG response to the audiobook and use ‘leave-one-trial-out’ cross-validation to compute the accuracy of each model. Phoneme surprisal represents how surprising a phoneme is, given the words that are consistent with the unfolding speech signal and their relative frequency in the language. We also compute acoustic models derived from the broadband envelope and the rectified derivative of the broadband envelope. Individual models demonstrate reliable neural tracking, as inferred by comparing observed model accuracies with those obtained after random shuffling of the EEG data. Unexpectedly, in preliminary analyses (N=45 participants), we find no reduction in the neural tracking of phoneme surprisal when participants attended to visual compared with auditory input, nor when comparing perceptual load levels during the visual task. In ongoing analyses, we will investigate whether speech predictive processing is modulated by attention in higher linguistic levels (e.g. word surprisal) and whether this lack of attention effect is consistent across other metrics (such as semantic dissimilarity).
Topic Areas: Speech Perception, Control, Selection, and Executive Processes