Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Incremental versus global visual word recognition in the M170
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
This poster is part of the Sandbox Series.
Stefan Pophristic1, Alec marantz; 1New York University
This study compares the neural correlates of two models of word recognition to assess whether the initial stages of visual lexical processing incrementally or globally compute the morphological relationships within complex words. Previous neurolinguistic work (e.g. Wray et al. 2022) has shown that morpheme surprisal is correlated with decreased M170 amplitude, a Magnetoencephalographic (MEG) response component associated with the earliest stages of visual word processing. This surprisal measure has been taken as an index of obligatory morphological decomposition and tracking of incremental expectations of linearly adjacent morphemes during early stages of lexical processing in the Visual Word Form Area (VWFA) (e.g. tracking the relationship of “work” and “-ed” in the word “worked”). A set of psycholinguistic work (e.g. Milin et al. 2009) has explicitly disputed the relevance of this incremental surprisal measure, suggesting that a global relative entropy (RE) between the probability distribution of a word’s inflected forms (e.g. “work”: “works”, “worked,” etc.) and the class level distribution of identically inflected words (e.g. “stem”: “-s”, “-ed”, etc.) can better account for the data. This model predicts that for a word like “worked”, rather than computing the local relationship between “work” and “-ed”, lexical processing tracks information associated with the stem “work” regardless of whether the word appears with the ending “-ed” or “-s”. While proponents of both models have argued against the other model, this is the first study to directly test both. Participants completed a visual lexical decision task online (n=54) or in-person with simultaneous MEG recording (n=25) in Spanish. Data collection for both experiments is complete. The stimuli consisted of 84 verbal and pseudo-stems (distributed across 3 inflection classes), conjugated with 13 different inflectional endings. We created statistical models with either global RE or incremental surprisal values. For the behavioral response times, AIC model comparison was computed on lmer models fit with surprisal and RE values. For the MEG data, we will conduct the following analyses: 1) we will average activity over the space defined by an fROI (see Gwilliams et al. 2016) and the time-frame around the M170 and fit RE and surprisal GLM models to the averaged activity; 2) fit RE and surprisal GLMs to each spatial vertex and run a spatio-temporal permutation test over the GLM beta values. Comparing the models fit to behavioral response times, we find substantial improvement for surprisal compared RE models (AIC-Difference: 34.7). Our MEG data is still being analyzed. As an initial analysis to show the study is sufficiently powered, we averaged brain activity per participant over the entire BA37 and found significant effects of surprisal from 223-243ms and RE from 207-209ms. Our behavioral model comparison strongly suggests that lexical processing tracks incremental morpheme-by-morpheme rather than global stem information. However, from behavioral results alone, it is unclear whether our models are indexing early stages of lexical processing, or down-stream bottlenecks of syntactic and semantic processing. The analyses of the M170 response should reveal whether the outperformance of the surprisal models compared to RE models truly reflects these early processing stages.
Topic Areas: Morphology,