Poster B69, Sunday, March 26, 8:00 – 10:00 am, Pacific Concourse
The brain dissociates between different levels of prediction during language comprehension
Gina R. Kuperberg1,2,3,4, Edward W. Wlotko1,5, Simone J. Riley1, Margarita Zeitlin1, Maria Luiza Cuhna-Lima1; 1Tufts University, 2Massachusetts General Hospital, 3Athinoula A. Martinos Center for Biomedical Imaging, 4Harvard Medical School, 5Moss Rehabilitation Research Institute
Comprehenders continually use context to generate probabilistic predictions. We asked whether and how these predictions influence neural processing of incoming words. Mini-discourse contexts varied in their constraint for a specific lexical item or event structure. Event-related potentials were recorded to nouns that fulfilled lexical predictions, that violated lexical predictions but were plausible, or that violated the preceding verb’s selectional restrictions ('They cautioned the SWIMMERS / TRAINEES/ DRAWER' following a context about lifeguards and sharks). Semantic facilitation was reflected by a selective reduction of the N400 on predictable nouns. Plausible nouns that violated lexical predictions selectively elicited a late anteriorly-distributed positivity, whereas nouns that violated selectional restrictions elicited a late posteriorly-distributed positivity. These dissociable neural signatures of prediction violations at different levels of representation provide support for a hierarchical generative architecture in which bottom-up information is continually evaluated against top-down predictions at multiple levels of representation to support ongoing comprehension.
Topic Area: LANGUAGE: Semantic