You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster D68, Thursday, November 9, 6:15 – 7:30 pm, Harborview and Loch Raven Ballrooms

Cortical responses to linguistic features in natural story comprehension

Katerina Danae Kandylaki1, Hugo Weissbart1, Tobias Reichenbach1;1Imperial College London

Previous research has reported neural tracking of linguistic features such as phrase structure in spoken language (Ding et al., 2016). However, the employed stimuli had an artificially-imposed periodic timing of words. It thus remained unclear how the obtained results generalise to natural language, in which the timing of syllables and words is irregular. Here we aimed to quantify neural response to linguistic features in natural spoken language comprehension. Previous research on neurological diseases and experiments on aphasia patients points to separate neurocognitive mechanisms for agrammatism and anomia. We therefore tested the hypothesis that neural responses to function words differ from those to content words. Moreover, previous findings on evoked response potentials (ERPs) highlight a different neural response to the word categories of noun/verb (Federmeier et al., 2000), so we tested the hypothesis that cortical responses are sensitive to this grammatical and semantic feature. We used short stories as stimuli; the stories were written in English and spoken out by volunteer speakers (see librivox, public domain). We then recorded cortical responses from 11 participants (5 female) while participants listened to the stories using electroengephalography (EEG). The EEG recordings were then correlated to linguistic features of the spoken language stimuli. First, we applied tools from computational linguistics to tag the texts according to the word category: a. function/content words, b. noun/verb. We then used forced alignment to align the features to the acoustic signal. The EEG responses were filtered between 0.1 and 4 Hz since these low frequencies, mostly in the delta band, match the rhythm of word-based linguistic features. Finally, we used the MNE python toolbox to model the relation between the EEG responses and the linguistic features. Using linear regression, we modelled the word categories as continuous factors and included word onset as a control factor. We also included the word frequency as additional control factor since word frequency can affect ERPs. Word frequency metrics were extracted from Google Ngrams. A cortical response to word onset was found at frontal electrodes bilaterally at a latency of around 400 ms after stimulus onset. A neural response to word frequency emerged around 100 ms after word onset; the response was left lateralised on frontal electrodes. The grammatical distinction of function/content words elicited a sustained cortical response that was mainly frontally and bilaterally distributed. The noun/verb distinction elicited a bilateral response on mainly temporal and occipital channels about 300 ms after word onset; this response was left lateralised at 400ms post stimulus onset. In summary, we found significant cortical responses in the delta frequency band for the linguistic features word frequency and grammatical category. Neural responses to linguistic features of natural spoken language, with an irregular timing of words, can hence be successfully recorded with EEG. The findings point towards a role of the delta frequency band in the predictive coding of spoken language.

Topic Area: Perception: Auditory

Back to Poster Schedule