Search Abstracts | Symposia | Slide Sessions | Poster Sessions
When no doesn't mean no: negation as a challenge for predictive-coding approaches to language processing
Poster Session B, Friday, October 25, 10:00 - 11:30 am, Great Hall 3 and 4
Matthias Schlesewsky1, Sophie Jano, Ina Bornkessel-Schlesewsky, Barbara Kaup; 1University of South Australia, 2University of Tuebingen
Incremental processing is a basic tenet of psycholinguistic research, positing that linguistic input is interpreted in real time. Neurobiologically, incrementality has been explained via predictive coding: if only the unpredicted portions of the linguistic input are encoded, this could explain the rapidity of language processing (Pickering & Garrod, 2007). Negation has long been a thorn in the side of incrementality. Fischler et al. (1983) first reported that N400 effects in simple sentences ("A robin is (not) a bird/vehicle") were unaffected by negation. This pattern has proven to be highly consistent across a range of manipulations involving negation (e.g. Wiswede et al., 2013; Dudschig et al., 2018), thus leading to the suggestion that negation may not be processed in real time (e.g. Herbert & Kissler, 2014). These findings are surprising since negation can have a profound effect on our interaction with the environment (e.g. "This mushroom is (not) poisonous."). Moreover, they appear to challenge a view of the brain as a "predictive machine" which constantly seeks to verify or falsify the predictions about the environment generated by its current internal model. Here, we examined the effect of negation on incremental predictive processing by focusing on prediction matches in addition to prediction mismatches. To this end, we used antonym relations (e.g. "The opposite of black is white/yellow/nice"). Expected antonyms, as the sole appropriate continuation, elicit a target-related P3, while the two unexpected continuations engender a graded N400 depending on their relation to the antonym (unrelated, "nice" > related, "yellow") (Roehm et al., 2007). Forty speakers of Australian English (33 female; mean age: 24.1, range: 18-36) were presented with negated and non-negated sentences as in the above example using rapid serial visual presentation while their EEG was recorded. Antonym sentences were interspersed with an equal number of non-antonym sentences (50% of which were negated). From a predictive-coding perspective, one would hypothesise no observable antonym-P3 effect for the negated antonym conditions: even if a negation cannot be fully integrated into the current sentence meaning incrementally, it should, from a predictive coding perspective, at the very least eliminate the status of the antonym as a predicted target word. In other words, while the negation makes it difficult to formulate predictions for the continuation of "The opposite of black is not ...", it should rule out "white" as the predicted next word. Contrary to the hypotheses, we observed indistinguishable EEG responses to negated and affirmative antonym sentences: both showed the previously observed graded N400 effect (unrelated > related) accompanied by a P3 for the antonym condition. Neurobiologically, the P3 has been linked to the release of noradrenaline from the brainstem locus coeruleus in response to motivationally significant stimuli (Nieuwenhuis et al., 2005; for language, see Sassenhagen et al., 2014). It thus occurs in response to stimuli that are highly relevant for our interaction with the environment. Strikingly, this response is not modulated by negation in real time. This finding poses a challenge for predictive coding as a fundamental mechanism of human information processing.
Topic Areas: Syntax and Combinatorial Semantics, Meaning: Lexical Semantics