Authors: Sophie Arana1,2,Jan-MathijsSchoffelen1,Tom Mitchell3, Peter Hagoort1,2
1Donders Institute for Brain, Cognition and Behaviour, Centre for Cognitive Neuroimaging, Radboud University Nijmegen, The Netherlands
2Max Planck Institute for Psycholinguistics, Nijmegen, The Netherlands;
3Machine Learning Department, Carnegie Mellon University, Pittsburgh, PA, USA;
Sentence level comprehension depends on both semantic and syntactic information. Various paradigms have focused on distinguishing between the two kinds of processing, characterizing them separately in terms of contributing brain areas or oscillatory dynamics.
In practice, both semantics and syntax are often closely entangled. A complete account of the neural mechanisms underlying sentence comprehension must therefore also address the interaction between the two. In this study, we specifically investigate how semantic information affects the processing of structurally ambiguous phrases.
While prepositional phrases have an ambiguous syntactic structure, preceding semantic information usually determines whether they are interpreted as verb or noun attachments. We aim at classifying from MEG recorded brain data whether participants were reading sentences with verb attached or noun attached phrases. While the prepositional phrase is held constant across two sentences, we manipulate attachment decision through semantics by either changing the verb (condition 1) or agents (condition 2) of the preceding sentence context.
We test the hypothesis that depending on the attachment decision, we will find a reactivation of either the preceding verb or noun following presentation of the prepositional phrase. To this end, we will investigate when in time during sentence reading we can successfully decode the verb or noun given the corresponding multivariate neural patterns.
This study provides insights into neural processes of sentence structure building and its interaction with semantic processing. In addition, successful classification of attachment structures based on MEG data may enable new syntax paradigms in the future.