Tracking the behavioral and neural dynamics of semantic representations through negation

Abstract

Combining words and composing meanings is at the basis of human language. However, how the brain constructs meaning online is not well understood. We develop an approach to address this puzzle that exploits the ubiquitous operation of negation. Although negation has been extensively debated among philosophers, psychologists, logicians, and linguists, the cognitive mechanisms supporting this elementary function remain largely unknown. Here, we track the compositional effects of negation ("not") on scalar adjectives (e.g., "good") in parametrically designed behavioral and magnetoencephalography (MEG) experiments. The behavioral data show that participants first interpret negated adjectives as affirmative and then modify their interpretation towards the opposite meaning. Decoding analyses of neural activity further reveal that negation does not invert the representation of adjectives but rather weakens their representation, as soon as lexical-semantic representations are encoded in the brain. This suppression mechanism is supported by desynchronization of beta-band activity elicited in sensory-motor areas during the processing of the adjective.

Multivariate pattern analysis of EEG reveals nuanced impact of negation on sentence processing in the N400 and later time windows

“…Thus, the fact that the N400 is enhanced by negation indicates that negative operators render the lexical access of single words more difficult. This interpretation is also well in line with the fMRI, behavioral, and recent MEG-MVPA literature that shows an inhibitive effect of negation (Beltrán et al, 2021;Papeo et al, 2016;Zuanazzi et al, 2022). However, additionally with MVPA, we observed that this inhibition effect on semantic retrieval depends on our semantic relatedness manipulation (i.e., semantic integration processes).…”

“…Therefore, it could be the case that the inherent limitation of the univariate ERP analysis restricts prior and current ERP analyses from properly detecting the modulative effect of negation on the semantic N400. Interestingly, our finding also echos a recent MVPA study that investigates the effect of negation on the processing of adjectives using MEG (Zuanazzi et al., 2022). By directly comparing low versus high adjectives as the baseline, with cross‐decoding and DAT together, Zuanazzi and colleagues also reported a modulative role of negation.…”

“…For ERPs, we hypothesized that (1) across sentence types, it is the semantic relatedness between the contextual word and the target word (e.g., robin-bird/vehicle and read-newspaper/knife), but not the truth value of a sentence that would lead to robust N400 effects for the unrelated words (Fischler et al, 1983;Palaz et al, 2020), whose amplitudes will not likely to be affected by sentence polarity (Dudschig et al, 2016); (2) regarding the effect of negation, a direct comparison between negative versus positive sentences might lead to an N400 effect (Lüdtke et al, 2008), but we may also observe no main effect of negation in the N400 window, as reported in most comparable studies (Dudschig et al, 2019;Haase et al, 2019;Palaz et al, 2020); (3) following most prior studies, in the later time windows, we may not observe reliable late effects for semantic congruency in the P600 time window; however, for the main effect of negation, we may also observe a P600 effect for negative sentences (Lüdtke et al, 2008). For MVPA, following Zuanazzi et al (2022), we hypothesized that (1) across sentence types, we could observe modulation effect of negation on the decoding of congruent versus incongruent target words; (2) for the main effect of negation, negative versus positive sentences should be decodable; however, here we refrain from hypothesizing how the decoding time series vary as a function of either sentence type of semantic congruency, due to the lack of directly relevant empirical literature and theoretical framework; (3) based on a previous MVPA study, for semantic congruency we expect significant decodability in the P600 time window; however how it will be affected by polarity remains to be explored. Also, given the lack of literature, whether the P600 window will be sensitive to the main effect of negation remains to be explored.…”

“…MVPA has also been employed to investigate phrasal negation. In an MEG study (Zuanazzi et al., 2022), the authors used the DAT analysis to investigate the impact of negation on scalar adjectives (e.g., decoding low vs high) and reported reduced decoding performance for the negated versus affirmative adjectives. When examining the main effect of negation, they also showed that affirmative versus negative phrases are decodable, suggesting that the presence of a negation marker directly impacts upon the processing of adjectives.…”

“…For MVPA, following Zuanazzi et al. (2022), we hypothesized that (1) across sentence types, we could observe modulation effect of negation on the decoding of congruent versus incongruent target words; (2) for the main effect of negation, negative versus positive sentences should be decodable; however, here we refrain from hypothesizing how the decoding time series vary as a function of either sentence type of semantic congruency, due to the lack of directly relevant empirical literature and theoretical framework; (3) based on a previous MVPA study, for semantic congruency we expect significant decodability in the P600 time window; however how it will be affected by polarity remains to be explored. Also, given the lack of literature, whether the P600 window will be sensitive to the main effect of negation remains to be explored.…”