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

Poster B41, Wednesday, November 8, 3:00 – 4:15 pm, Harborview and Loch Raven Ballrooms

Multivariate pattern analysis reveals semantic information in brain areas activated for nonwords

Hillary Levinson1, Samantha Mattheiss1, William W. Graves1;1Rutgers University

The neural basis of semantic cognition has been investigated using univariate analysis of functional magnetic resonance imaging (fMRI) data for at least the past 20 years. These analyses have proven to be very effective in revealing neural regions involved in the putative neural semantic network, which significantly overlaps with the default mode network (DMN). However, there have been some inconsistencies across fMRI studies in terms of the primary regions involved in semantic processing. These discrepancies have most often been found in studies that manipulate the level of task difficulty, where increasing levels of difficulty activate regions outside of the putative semantic network/DMN, such as the task positive network (TPN). We recently observed this pattern in a lexical decision task with high and low imageability words, where the word-nonword contrast revealed nonword activation primarily in the DMN (including angular gyrus and posterior cingulate), and word activation in the TPN (including inferior frontal junction, intraparietal sulcus, and ventral occipitotemporal sulcus). Here we investigated whether the putative semantic areas activated for nonwords also encoded semantic information. This was determined by classifying high and low imageability words using multivariate pattern analysis (MVPA), implemented in the PyMVPA suite. We trained a Sparse Multinomial Logistic Regression classifier on fMRI data restricted to the nonword contrast to determine whether participants were reading high or low imageability words. It reliably classified imageability category at 83.3% accuracy (p < .05 by Monte Carlo simulation). This suggests that semantic information is present even in areas activated by meaningless nonwords. Although previous activation of putative semantic areas by nonwords was presumably due to difficulty effects, this analysis shows difficulty effects and semantic information can co-localize in the same neural network.

Topic Area: Meaning: Lexical Semantics

Back to Poster Schedule