Matar, S. 1, 2 , Dirani, J. 2 , Pylkkänen, L. 1, 2 & Marantz, A. 1, 2
1 New York University
2 NYU Abu Dhabi Research Institute
Predictive processing during language comprehension was previously shown to occur on a lexical level, including for specific lexical items. Correlations with lexical entropy and surprisal have been shown in various regions of the frontal and temporal lobes across both hemispheres. However, evidence for syntactic predictive processing (e.g., prediction of syntactic categories of upcoming words) has mostly been restricted either to violation paradigms, which limit the interpretability of results due to stimuli ungrammaticality, or to fMRI studies that use uncontrolled auditory texts, for which different parsers quantify surprisal for each word.
In this study, our aim was to look for evidence for syntactic predictive processing outside of the realm of violation paradigms. We chose a well-controlled paradigm in Standard Arabic, where sentences contained a subject modified by a gender-congruent or gender-incongruent adjective. Changing the congruence of the adjective varied: i) how surprising the adjective is, and ii) the local syntactic entropy of the sentence, because the gender-incongruent adjective was much more predictive of an upcoming second noun. All the sentences were grammatical. If comprehension is syntactically predictive, we expected the minimal differences between conditions to correlate with differences in neural activity, in regions sensitive to informational measures, such as surprisal and entropy.
In a magnetoencephalography (MEG) experiment, we had 27 participants read sentences in a rapid serial visual presentation. We found a higher level of activation for the gender-congruent cases in a right-hemispheric frontal spatiotemporal cluster (175-270ms after adjective onset). Additionally, on the word that follows the adjective, we found a modulation of the visual M100 in the left occipital lobe; this component was smaller for the incongruent cases, in which the appearance of a second noun was highly predictable. These findings suggest that the brain does indeed use syntactically predictive cues to facilitate fast on-line language processing.