Search Abstracts | Symposia | Slide Sessions | Poster Sessions
What is the role of pragmatic support in negation processing? Evidence from time-resolved EEG MVPA
Poster Session F, Friday, October 2, 2:45 - 4:45 pm, Wangari Maathai
Neemias Souza Filho1, Jacob Burger1, Azia Knox1, Abdulaziz Alromihi1, Gonzalo Resa Heras1, Tarleton Hill1, Arild Hestvik1; 1University of Delaware
Commonly found in the negation processing literature is that an appropriate pragmatic context mitigates the extra comprehension cost imposed by negation. Different mechanistic accounts are proposed for this pattern. In two-step negation processing models, contextual information would simply reduce the cost of two mandatory stages: representing the to-be-negated information and integrating the negative operator. Alternatively, contextual-pragmatic accounts posit that the first stage would not occur in supportive contexts. This predicts an algorithmic difference between how a sentence such as “Maria's car is not red” is processed with pragmatic support (e.g., correcting a friend's incorrect belief that an approaching red vehicle is Maria’s) and without (e.g., as a standalone statement about the car). We test this prediction using time-resolved multivariate pattern analysis (MVPA) of EEG data, training a naive classifier to discriminate between affirmative and negative trials within and across levels of pragmatic support. If negation is processed in one or two steps depending on pragmatic context, sentence polarity representations should not generalize across levels of pragmatic support. English speakers (N = 48, 40 female, 20.6 ± 2.2 years old) performed a sentence verification task based on 144 visual displays, each containing two different geometric shapes of different colors. Participants were told that a robot was being trained to identify shapes and colors. The robot reproduced one of the shapes on the display, doing so correctly on half the trials. Participants then saw a supervisor provide feedback to the robot in the form of affirmative or negative sentences. These were presented in chunks (e.g., |The square| |is/is not| |red| |on the screen|) and judged as true or false by participants. Three variables were manipulated, therefore. Most importantly, pragmatic support: the trials in which the robot produced an incorrect shape created the opportunity for corrections, a supportive context for negation use. Polarity (affirmative vs. negative) and truth value (true vs. false) of the supervisor’s sentence were the other manipulated variables. Continuous EEG was recorded with 65-channel nets, and MVPA procedures were time-locked to the onset of the color word in the supervisor’s feedback (i.e., “red” in “The square is red on the screen"). Polarity decoding revealed no above-chance clusters for supportive trials, indicating that there were no reliable differences in the signals elicited by affirmative and negative sentences. For non-supportive trials, polarity decoding was reliably above chance between 320-750 ms, indicating a significant effect of polarity (negative sentences elicited larger negativities, as corroborated by an ERP analysis). Despite these within-decoding differences, cross-decoding results (classifier trained on data from one pragmatic condition and tested on the other) revealed above-chance polarity decoding between 316-564 ms, suggesting that polarity representations generalize across pragmatic support levels. These results are consistent with the wide body of literature reporting that supportive pragmatic contexts mitigate the extra cost of negative sentences relative to affirmatives. Critically, the cross-decoding results reported here are evidence that polarity representations are similar across levels of pragmatic support, corroborating the fixed algorithm posited by two-step accounts of negation processing.
Topic Areas: Meaning: Discourse and Pragmatics,