Barbara Kaup

During the last decades there has been an ongoing debate about how people process language. Traditionally, language processing has been seen as being based on amodal and abstract symbols. More rece...

A lot of research suggests a close link between spatial attention and visuospatial working memory (Awh & Jonides, 2001; Cowan et al., 2005; Postle, 2006). It has been shown, for instance, that a vi...

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.

To be able to predict something means to generate reliable expectations about unknown cases based on data or models. To be able to explain a psychological phenomenon means to reconstruct the mental mechanisms and representations that produce it. In the context of AI and psychology, these two capacities are often conflated. We argue: prediction is not explanation—and uncovering the underlying cognitive processes and structures must remain the central goal of psychology, despite the successes of generative AI models in some behavioral prediction tasks.

The perception of temporal events on a unidirectional line represents a cognitive ability previously thought to develop between ages five and nine. However, prior studies employed tasks requiring memory and attention skills, which may confound task performance. We assessed linear time perception in 65 children aged 3 to 6 years using a game-like task with spatial representation and verbal measures. While we hypothesized that age, script exposure, and temporal-terms vocabulary would predict performance, only age emerged as a significant predictor. Contrary to previous findings, preschoolers already demonstrated unexpectedly high performance. Despite participants' exposure to right-to-left or two writing systems (right-to-left and left-to-right), the directionality of spatial representations showed no correlation with language directionality, with most children organizing time from left-to-right. This suggests a potentially innate predisposition toward left-to-right temporal directionality, challenging assumptions about the role of cultural influences in linear time perception.

Designing effective language learning settings requires an understanding of the processes taking place in language learning and the way they interact. One important issue concerns the interaction between meaning and grammar. A number of studies have shown a beneficial effect of semantics in grammar learning. What is unclear, however, is how far this effect may be influenced by the presentation formats of the semantic content. In two experiments, participants performed rule search tasks on Latin sentences. In Experiment 1, we presented semantic information in the form of naturalistic photographs, whereas in Experiment 2, the semantic information was implemented by quasi-translations. The control groups did not receive any semantic information. Learning performance was assessed by a grammaticality-judgment task combined with a source-attributions task. In both experiments, participants in the with-semantics group outperformed the respective control groups. Yet, only in Experiment 1, participants report having more explicit than implicit knowledge. We argue that semantic information boosts the acquisition of grammatical structures regardless of the presentation format. Furthermore, we suggest that, consistent with multimedia learning theories, the pictorial presentation format of Experiment 1 helped to use working memory capacity efficiently, which may have led to the generation of more explicit knowledge.

The current study aimed to evaluate the effect different modalities (pictures and words) of food stimuli have on inhibitory control under different homeostatic states. To this end, the homeostatic state was altered by asking participants to fast for 16 h (n = 67) or eat lunch as usual (n = 76) before completing an online stop-signal task with modal (pictures) and amodal (words) food and valenced-matched non-food stimuli. The inclusion of non-food stimuli allowed us to test the food specificity of the effect. We found a significant Group × Modality × Stimulus Type interaction (F(1,141) = 5.29, p = 0.023, η_p² = 0.036): fasted individuals had similar inhibitory capacity for modal and amodal food stimuli but better inhibitory capacity for non-food words compared to images, while there were no inhibitory differences in dependence on either modality or stimulus type in satiated individuals. Thus, we were able to show that inhibitory capacities to modal compared to amodal stimuli depend on participants’ current state of fasting. Future studies should focus on how this lowered inhibitory capacity influences food intake, as well as the role of stimulus valence in cognitive processing, to clarify potential implications for dieting and weight loss training.