Poster E54, Monday, March 26, 2:30-4:30 pm, Exhibit Hall C
The neural encoding of thematic roles
Jayden Ziegler1, Miriam Hauptman2, Jesse Snedeker1, Evelina Fedorenko3,4; 1Harvard University, 2Tufts University, 3Harvard Medical School, 4Massachusetts General Hospital
The same event can be described in different ways. For example, dog is the subject in the “The dog chased the man,” but it is the (oblique) object in “The man was chased by the dog.” Yet, in both cases, the dog is the biter, or doer of the action. To capture this regularity in meaning, linguists posit thematic roles (agent=doer, patient=doee). Previous work (Frankland & Greene, 2015; F&G hereafter) has shown that a classifier trained on fMRI data acquired while participants read sentences like these can reliably (and dissociably) decode the identity of the agent or patient (whether dog or man in either case), regardless of syntactic position (subject vs. object). This decoding was localized to two non-overlapping regions within the left temporal lobe. We ask: What information do these agent- and patient-selective regions encode? We replicated F&G’s classification analysis with the addition of a third condition: intransitive sentences (“The cup broke”). On the thematic roles hypothesis, cup is a patient and should therefore get decoded by the patient-selective region. Data collection and analysis is ongoing. In N=16, we observe reliable and dissociable classification of agents and transitive patients (both p<0.001), as in F&G. We also see reliable classification of intransitive patients (p<0.001), but not in the same region that decodes transitive patients (p=0.997). Our findings (tentatively) do not support the hypothesis that the regions discovered by F&G encode thematic roles. We suggest other possibilities for what these regions may encode (e.g., the causal structure of events).
Topic Area: LANGUAGE: Semantic