Presentation
Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams
Retrieval precedes evaluation: An MEG study on implausible gaps in English
Poster A7 in Poster Session A, Thursday, October 6, 10:15 am - 12:00 pm EDT, Millennium Hall
Dustin A. Chacón1,2, Liina Pylkkänen1,3; 1New York University Abu Dhabi, 2University of Georgia, 3New York University
[INTRODUCTION] Language comprehension is predictive. Brain recordings differentiate predicted words vs. unpredicted words resulting in semantic anomalies <400ms, e.g., I like my coffee with sugar and *socks*. However, results are mixed when the semantic anomaly arises from an implausible thematic relation, e.g., ‘meal’ as agent of ‘devoured’ in ‘the hearty meal *devoured*...’, suggesting this information is not integrated into verb predictions. Some results suggest detection of thematic relation anomalies are delayed (>600ms), but others show early responses in sentences with filler-gap dependencies (e.g., ‘city/book’ in ‘which city/book did you *write*’). We conducted a magnetoencephalography (MEG) experiment to compare neural responses to thematic relation anomalies in sentences with filler-gap dependencies and sentences without. We found that left anterior temporal lobe (LATL) activity responds to implausible thematic relations ~700ms post verb-onset for sentences without filler-gap dependencies, and ~1000ms for sentences with. Left inferior frontal gyrus (LIFG) activity showed sensitivity to filler-gap dependencies ~700ms post verb-onset. This pattern fits the hypothesis that detecting implausible filler-gap dependencies is delayed because the filler must be retrieved from short-term memory before its meaning can be evaluated. [METHODS] MEG activity from 30 English-speaking adults was recorded in a 2×3 within-subjects design of Question Type (Wh-Subject, Wh-Object, Yes/No) and Plausibility (Plausible, Implausible). In Wh-Object and Wh-Subject sentences, there was a filler-gap dependency. In Plausible sentences, the verb’s thematic roles assigned to the filler or subject were plausible, and implausible for Implausible sentences. Sentences (49 sets) were displayed in randomized order, word-by-word, with each word displayed 300ms on/300ms off. After each sentence, participants registered one of two responses or “N/A” for uninterpretable stimuli. Wh-Object, {Plausible/Implausible} Which {sandwiches/forks} do hungry kids *eat voraciously*? Wh-Subject, {Plausible/Implausible} Which {kids/forks} in the kitchen *eat voraciously*? Yes/No, {Plausible/Implausible} Do you think that {kid/forks} *eat voraciously*? [RESULTS] We conducted a two-stage regression analysis, regressing on experimental factors and nuisance variables, then conducting cluster-based permutation tests on the coefficients. Analyses were conducted in 4 time windows (300-500ms, 600-800ms, 800-1000ms, and 1000-1200ms post verb-onset, ‘eat’) in LIFG and bilateral frontotemporal regions. [QUESTION TYPE] There was a cluster showing an effect of Question Type in LIFG ~700ms, in which Wh-Object and Wh-Subject diverged from Yes/No (p = 0.047). We propose this LIFG activity reflects retrieval of the filler phrase from short-term memory. [PLAUSIBILITY] There was a cluster showing an effect of Plausibility in LATL ~700ms (p < 0.01), although pairwise comparisons show that this effect is driven by greater activation in Yes/No, Implausible sentences compared to Yes/No, Plausible sentences (p < 0.0001). A later cluster showed an interaction of Question Type x Plausibility in LATL ~1000ms (p < 0.01). This cluster shows a trend of greater activation for Implausible sentences over Plausible, with a polarity reversal for Wh-Subject conditions (all ps < 0.1). [CONCLUSION] Language processing is predictive, but thematic role information may not be integrated into verb predictions. We show that detecting thematic role anomalies is delayed generally (>600ms), but specifically for filler-gap dependencies (>1000ms), likely due to the need to retrieve the filler from short-term memory.
Topic Areas: , Methods