Francesca Capuano

While substantial research has addressed the challenges of negation processing in humans, the parallels between human cognitive models and large language models (LLMs) in this regard remain less explored. This paper investigates how GPT-4o processes negation, drawing comparisons to human negation processing, particularly the two-step model of negation. In two experiments using image-sentence pairs, we assess the model’s ability to handle affirmative and negated sentences. Our findings reveal that, like humans, GPT-4o struggles more with negation, exhibiting higher rates of incorrect inferences after negation, and a systematic bias toward completions associated with the negated state of affairs. This study highlights the qualitative similarities between AI and human processing of negation, offering insights into the limitations of current LLMs and suggesting future directions for improving their cognitive alignment with more complex linguistic constructs.

Negation in natural language is not interpreted as a simple logical complement but is resolved selectively, depending on how linguistic and conceptual structure constrain relevant alternatives. Psycholinguistic research suggests that negation facilitates consideration of an alternative only when a privileged counterstate is available, either through contextual restriction or graded prominence. We use text-to-image generation as a complementary method to probe this selectivity under forced representational commitment: when negated language must be rendered visually, a concrete state must be instantiated. Across two preregistered experiments, participants evaluated how well generated images matched affirmative and negated prompts. Experiment 1 shows that negation is resolved more successfully for predicates encoding conventionalized contrasts than for those with open alternative spaces. Experiment 2 shows that, within open spaces, negation is facilitated when a highly prominent alternative exists. Together, the results support an account in which binary advantages arise from how strongly a counterstate is privileged.

LLM bias research has largely focused on content-level associations, overlooking biases arising from linguistic form and pragmatic choice. One such phenomenon is the negation bias, where negated expressions are preferentially used for stereotype-inconsistent behavior and trigger systematic inferences. While LLMs’ handling of negation in semantic tasks is well studied, it is unclear whether they reproduce these pragmatic inferences in social contexts. We replicate Beukeboom et al. (2010) with Mistral-7B-Instruct-v0.2, varying the valence and polarity of trait descriptions. The model inferred more negative expectations from negated negative traits (e.g. not stupid) and more positive expectations from negated positive traits (e.g. not smart), and showed stronger situational but weaker dispositional attributions. These results mirror human negation-bias patterns, suggesting that instruction-tuned models can produce pragmatic-like bias in judgment tasks. The findings underscorethe need to extend LLM bias evaluation beyond content to linguistic form.

The study explores whether Large Language Models (LLMs) exhibit negation-induced forgetting (NIF), a cognitive phenomenon observed in humans where negating incorrect attributes of an object or event leads to diminished recall of this object or event compared to affirming correct attributes (Mayo et al., 2014; Zang et al., 2023). We adapted Zang et al. (2023) experimental framework to test this effect in ChatGPT-3.5, GPT-4o mini and Llama3-70b-instruct. Our results show that ChatGPT-3.5 exhibits NIF, with negated information being less likely to be recalled than affirmed information. GPT-4o-mini showed a marginally significant NIF effect, while LLaMA-3-70B did not exhibit NIF. The findings provide initial evidence of negation-induced forgetting in some LLMs, suggesting that similar cognitive biases may emerge in these models. This work is a preliminary step in understanding how memory-related phenomena manifest in LLMs.

Negation is often used to contradict or correct (e.g. There is no dog here.). While rejecting some state of affairs that is presumed to hold for the recipient (e.g. There is a dog here.), the speaker might implicitly suggest a set of plausible alternatives (e.g. There is a wolf instead.). Prior work indicates that alternatives are highly relevant to the comprehension of sentences involving focus: in priming studies, listeners infer plausible alternatives to focused items even when they are not contextually available. So far it is unclear whether negation similarly activates an automatic search for plausible alternatives. The current study was designed to investigate this question, by looking at the activation levels of nouns after negative and affirmative sentences. In a series of priming experiments, subjects were presented with negative and affirmative sentences (e.g. There is an/no apple.), followed by a lexical decision task with targets including plausible alternatives (e.g. pear), as well as semantically related but implausible alternatives (e.g. seed). An interaction of Sentence Polarity and Prime-Target Relation was expected, with negation facilitating responses to plausible alternatives. Results of the first experiment were numerically in line with the hypothesis but the interaction just missed significance level. A post hoc analysis revealed the expected significant interaction. Possible roles of sentential context and goodness of alternatives are discussed. A further experiment confirms that the goodness of alternatives is in fact critical in modulating the effect.

Stockholm University Press is an open access publisher of peer-reviewed academic journals and books. We aim to make journals and books affordable, and to enable the widest possible dissemination so that researchers around the world can find and access the information they need without barriers.In partnership with our authors and series editors, we publish in the humanities, social sciences and the natural sciences. Our main focal points are rigorous peer review, open access and global dissemination.

Studies on the spontaneous production of negation suggest that it can be modulated by pragmatic principles of successful communication such as informativity and relevance. The present study investigates whether negation production is additionally modulated by a more general principle of effort minimisation. In a series of circles games, subjects were presented with pairs of circles and asked to complete a sentence that would allow a listener to identify one of the two circles. Negation was only produced when an affirmative description for the circle at issue was harder, i.e. there was no simple intuitive way to describe the circle's pattern. The length of the concurrent descriptions did not strictly influence the production of negation. The results suggest that the use of negation becomes more frequent as the effort to produce it decreases with respect to a concurrent affirmation, even at the cost of greater informativity of affirmation.

The current study originates from inconsistent findings within the framework of embodied language processing, specifically in the reading-by-rotating literature: whereas some studies report a match advantage (e.g., Zwaan and Taylor (J Exp Psychol 135:1, 2006)), i.e., shorter reading times when the direction of a linguistically conveyed manual rotation matched rather than mismatched the direction of an actually to be performed manual rotation Claus (Acta Psychol 156:104–113, 2015) found a mismatch advantage. The current study addresses two explanations that were previously discussed as potentially responsible for this inconsistency: on the one hand, differences in the knob devices employed; on the other hand, differences in the perspectives adopted by the readers depending on the number of characters involved in the narratives. Concurrently, the study exploits individual differences in motoric experience to explore the experiential basis of action-sentence compatibility effects. The results are inconclusive with respect to the two explanations. However, in their overall picture, they contribute suggestive considerations for the ongoing debate on action-simulation effects by pointing to the potential role of interindividual variation in motoric experience.

Conversational negation often behaves differently from negation as a logical operator: when rejecting a state of affairs, it does not present all members of the complement set as equally plausible alternatives, but it rather suggests some of them as more plausible than others (e.g., “This is not a dog, it is a wolf/*screwdriver”). Entities that are semantically similar to a negated entity tend to be judged as better alternatives (Kruszewski et al., 2016). In fact, Kruszewski et al. (2016) show that the cosine similarity scores between the distributional semantics representations of a negated noun and its potential alternatives are highly correlated with the negated noun‐alternatives human plausibility ratings. In a series of cloze tasks, we show that negation likewise restricts the production of plausible alternatives to similar entities. Furthermore, completions to negative sentences appear to be even more restricted than completions to an affirmative conjunctive context, hinting at a peculiarity of negation.