Recent ucm items

Memory, Arousal, and Temporal Binding: A Computational Model

Thu, 6 Aug 2026 00:00:00 +0000

In the interval estimation task, participants are asked to estimate the time interval between two events. Interval estimates tend to be smaller if the first event is a voluntary action. The cognitive mechanisms that generate this temporal binding effect remain unclear. In this study, we propose a computational model of this phenomenon. We hypothesize that participants perform this task by comparing the availability of experimental events in memory. More specifically, we claim that temporal binding effects may be attributed to differences in emotional arousal at the time of encoding due to the type of event. For instance, voluntary actions may be accompanied by slightly higher arousal, leading to stronger traces of the first event relative to the involuntary case.

Four Stress Phenotypes Revealed Through Multimodal Thermal Imaging: The FenoStressNet Framework

Thu, 6 Aug 2026 00:00:00 +0000

Stress is typically conceptualized as varying along a single intensity dimension, yet individuals with identical "high stress" labels often exhibit divergent responses. This work challenges this unidimensional model through FenoStressNet, a multimodal framework integrating facial thermal imaging, physiological monitoring, psychological assessment, and explainable AI. Analyzing 120 participants across controlled stress-induction tasks (19,200 thermal images), this study identifies four reproducible stress phenotypes that cross-cut traditional intensity categories: Cognitive-Dominant (elevated cognitive appraisal, moderate physiology), Physiological-Reactive (rapid autonomic arousal), Integrated-Responder (synchronized multimodal activation), and Discordant-Denier (high physiological response with low subjective awareness). These phenotypes exhibit distinct facial thermal patterns and differential weighting of cognitive, affective, and physiological components. These findings suggest...

Stats Wars: Return of the Bayesians

Thu, 6 Aug 2026 00:00:00 +0000

In view of the ongoing replication crisis and various frustrations about classical null-hypothesis significance testing, calls for a statistical reform have increased. Often, it has been argued that the Bayesian approach offers a promising alternative to classical statistics, but at the same time, both approaches are sometimes hard to compare. This paper offers some basic steps for establishing a more decisive common ground and take steps towards resolving some fundamental conceptual issues, by introducing the frameworks of epistemic accuracy and efficient experimentation. Efficiency (i.e. the ratio between the accuracy of conclusions, and the experimental resources needed) is of central importance to good scientific inquiry. This paper shows how considerations of efficiency and accuracy motivate a broadly Bayesian approach, which however can also make room for frequentist tenets, and thereby push us towards a more unified and powerful framework for scientific inference.

Efficient compression in developmental trajectories of color naming

Thu, 6 Aug 2026 00:00:00 +0000

Convergent evidence suggests that the adult lexicon has evolved under pressure to efficiently satisfy a tradeoff between compressibility and informativeness, formally known as the Information Bottleneck (IB) principle. No work to date, however, has tested whether the same principle also impacts how children learn word meanings during development. Here, we address this open question in the domain of color by collecting color naming data from Hebrew speaking children across development (ages 3-12), and comparing it to color naming data from adults. Across ages, we find that children achieve near-optimal IB tradeoffs, with their color naming systems becoming more complex (less compressible) and more adult-like with age. Importantly, this pattern holds even though children's early color naming systems diverge from those of adults, implying that their efficiency is not merely a reflection of the efficiency of the adult system. These findings suggest that children's developmental trajectories...

Hierarchical Temporal Structure in Carnatic Improvisation: Interaction and Rhythmic Constraint

Thu, 6 Aug 2026 00:00:00 +0000

Hierarchical Temporal Structure in Carnatic Improvisation: Interaction and Rhythmic Constraint

Does Episodic Memory Help Close the Lexical Frequency Gap in Sensitivity to Syntactic Contrasts? A Test Using Retrieval-Augmented Language Models

Thu, 6 Aug 2026 00:00:00 +0000

Grammatical knowledge and how it is empirically tested are typically considered robust to the frequency of the lexical items used in the expressions. Complementary Learning Systems theory proposes that hippocampal episodic memory, which enables rapid encoding and retrieval of specific experiences, allows learners to leverage those experiences when processing rare patterns. We test the hypothesis that robustness to lexical frequency can arise via such an episodic memory mechanism by evaluating whether retrieval-augmented language models (specifically, k-nearest-neighbor language models that augment parametric neural networks with explicit instance storage), help close the lexical frequency gap in syntactic contrasts that vanilla language models exhibit. Using syntactic contrasts with frequency-stratified test items, we find that retrieval augmentation leads to improvements for test instances containing low-frequency lexical items, consistent with episodic memory compensating for...

Cinematic Techniques for Controlling Attentional Focus Enhance Emotional Engagement in Interactive Narratives

Thu, 6 Aug 2026 00:00:00 +0000

To enhance emotional engagement in interactive narratives, we propose an Arousal Induction Model that integrates cinematic techniques with the psychological theory of misattribution of arousal. The model combines attentional focus control (using over-the-shoulder shots and match-cuts) with arousal induction via auditory and visual effects.We evaluated this model through an RPG-based experiment ($N=49$). Results showed that the arousal-induced staging significantly improved participants' alignment with the protagonist's interpersonal preferences and increased subjective story evaluations. Physiological analysis (CSI) confirmed successful sympathetic activation during staged events.These findings suggest that simultaneously intervening in cognitive attention and physiological arousal can effectively deepen emotional engagement and improve experience quality. This research provides a framework for designing more moving and immersive interactive narrative experiences.

Motion to Blame: Different Causal and Moral Attributions Evoked by Perceived Chasing and Following

Thu, 6 Aug 2026 00:00:00 +0000

Humans perceive rich social interactions from motion. We investigated whether humans can perceive subtle intention differences in motion displays even when kinematics are highly similar, and whether the recovered intention shapes downstream causal and moral judgments. By grounding different intentions in a planning algorithm with distinct reward functions, we generated "chasing" and "following" trajectories, each ending with the front agent falling into an accident. In an online responsibility task, participants perceived different intentions and assigned blame accordingly: in following, responsibility concentrated on the leading front agent, whereas in chasing, responsibility shifted toward the rear chaser. In a matched causal-attribution task, the front agent was judged to be the primary cause of the motion in both displays. These findings suggest that humans can perceive subtle differences in intention from motion, which scaffolds moral evaluation, but that this influence cannot...

The Representational Geometry of Number

Thu, 6 Aug 2026 00:00:00 +0000

A central question in cognitive science is whether conceptual representations converge onto a shared manifold to support generalization, or diverge into orthogonal subspaces to minimize task interference. While prior work has found evidence for both, a mechanistic account of how these properties coexist and transform across tasks remains elusive. We propose that representational sharing lies not in the concepts themselves, but in the \emph{geometric relations} between them. Using number concepts as a target domain and language models as high-dimensional computational testbeds, we show that number representations preserve a stable relational structure across tasks. Task-specific representations are embedded in distinct subspaces, with low-level features like magnitude and parity encoded along near-orthogonal axes. Crucially, we find that these subspaces are largely transformable into one another via linear mappings, indicating that task-specific representations, despite being located...

Decoding Neural Dissonance: From Sensory Mismatch to Model Neural Recalibration for Cybersickness Prediction

Thu, 6 Aug 2026 00:00:00 +0000

Cybersickness is a major barrier to the widespread adoption of virtual reality, arising from neural dissonance between visual and vestibular stimuli. Although kinematics-based deep learning enables non-intrusive detection, existing models often lack neurophysiological grounding and fail to capture dynamic sensory recalibration and continuous symptom accumulation. To address these limitations, we propose the Neural Sensory Conflict Network (NSCNet), a biologically inspired framework grounded in Sensory Conflict Theory. Specifically, NSCNet incorporates a Conflict Alignment Embedding to model the integration of discordant sensory inputs, a State-Space Re-entrant Experts module that combines selective state-space modeling with Re-entrant Experts to capture the temporal dynamics of neural dissonance and functional modularity, and a Dynamic Sensory Reweighting mechanism that approximates adaptive gain control in the central nervous system. Experiments on the MSCVR and VR Cybersickness...

Processing of, and Adaptation to, Nonbinary Pronouns

Thu, 6 Aug 2026 00:00:00 +0000

In recent years, the pronoun system of English has begun to accommodate gender diverse identities. The present study investigates how individuals process and adapt to two novel pronouns: nonbinary they and the neopronoun ze. Using a Web-based maze task, we compared processing of these pronouns relative to binary s/he pronouns. We also examined how reading behavior adapted to repeated exposure to each pronoun. Perhaps unsurprisingly, the rarer ze elicited greater processing difficulty than singular they overall. However, participants adapted more quickly to ze than to they. We propose that ze may be more easily learned than singular they because it does not compete with plural they.

Socially Motivated Observational Learning: Children Preferentially Learn from Overheard Speech Addressed to Their Own Mothers

Thu, 6 Aug 2026 00:00:00 +0000

When does overheard speech support early word-learning? The present study provides preliminary evidence that children's closest social partners (here, mothers) structure attention to overheard speech. In a Tseltal Mayan community where overhearing is central to socialization, 75 mother-child dyads participated in overhearing-sessions exposing them to two novel words. In each session, one child's mother was directly addressed ("Mother-Addressed" condition), while the second mother merely observed ("Mother-Unaddressed" condition). Immediately after, both children completed two gaze-based tests of word recognition and learning. Results show that, compared to Mother-Unaddressed children, Mother-Addressed children exhibited (i) greater one-shot recognition of the novel word whose referent was visible during the overhearing-session; and (ii) a word-learning advantage on cross-situational familiarization trials. Insofar as who is being spoken to matters, the present study finds that...

The Shadow of the Past: Amortized Inference and Belief Revision using Chess as a Model System

Thu, 6 Aug 2026 00:00:00 +0000

How do humans allocate scarce cognitive resources across a stream of related decisions? Russek, Acosta-Kane, van Opheusden, Mattar, and Griffiths (2025) predict reaction time on each chess move from the within-move Benefit of Computation, treating moves as if they were drawn independently. We ask whether those costs also depend on computation carried forward from the previous move. In chess, a player who has calculated a likely opponent reply can often reuse that work; a surprising reply should force revision. On 135,608 quality-filtered Lichess puzzles attempted by 74,609 unique users, the within-puzzle log-RT spike is +0.309 (Cohen's __ = 0.235; paired __ = 86, N = 135,608; BF10 > 10100). Hierarchical regression climbs from __2 = 0.011 (Russek-static) to 0.282; the amortization interaction is large when predictability is operationalized by the human-style Maia-2 (__ = _0.054, __ < 10_27) and not significant under Stockfish best-move predictability. A threeparameter Bayesian...

Syntactic Prominence and Pragmatic Bias in Turkish Subject Anaphora: Humans and Large Language Models

Thu, 6 Aug 2026 00:00:00 +0000

We examined whether large language models (LLMs) responded to syntactic prominence of discourse entities and pragmatic biases in Turkish subject anaphora resolution similar to human judgments. Using an offline comprehension task with native speakers, we showed that null and overt pronouns responds to syntactic prominence and pragmatic bias. We then evaluated several autoregressive LLMs on the same materials. While GPT-4o correlated with human responses, LLaMA-4 more closely approximated the interaction between syntactic prominence and pragmatic biases observed in human data, raising questions about the relationship between model performance, scale, and human fit. We also found that all models mostly differed from humans in response variability, with model responses tending to be more deterministic. Finally, the results indicated partial but limited alignment between human and model anaphora resolution in Turkish.

MoESleepNet: A Multi-view Mixture-of-Experts Model for Single-Channel EEG Sleep Stage Classification

Thu, 6 Aug 2026 00:00:00 +0000

Accurate sleep staging plays a crucial role in diagnosing patients' sleep health. Numerous studies have confirmed that data from different views can highlight distinct characteristics. Based on the time-domain (TD) EEG, we constructed two additional views: the Power Spectral Density (PSD) and the time-frequency representation (TFR). Notably, there is no universally applicable model suitable for all types of data. That is to say, the characteristics of data should align with the model structure. Therefore, we designed specialized expert models for learning different views. However, directly combining features extracted from different views often results in excessive redundancy, especially for different views of the same data which the underlying data remains essentially identical. To address these issues, we proposed a multi-view mixture-of-experts (MoESleepNet) model for sleep staging, which achieves the best performance on SleepEDF20, SleepEDF78 and SHHS single-channel datasets....

Investigating Mechanisms of Social Offloading using a Joint Negative Priming Task

Thu, 6 Aug 2026 00:00:00 +0000

When sharing tasks, individuals may reduce internal demands by offloading task processing to others – social offloading. Studies have revealed facilitated performance when people believe a jointly acting partner is responsible for task distractors. A proposed mechanism underlying this effect is selective attention. We investigated this using a joint location negative priming (NP) paradigm. In NP tasks, participants initially respond to targets while ignoring distractors (prime phase); at a subsequent probe phase, responses are typically slower to targets in prime distractor locations, reflecting lingering inhibition – the NP effect. Participants (N = 80) completed the task either alone, or believing a partner was responding to distractors. Following a selective attention account, we predicted prime target facilitation followed by increased NP in the joint condition. While results did not reveal facilitation, NP was significantly increased. We propose that these findings demonstrate...

How do iconic co-speech gestures contribute to the truth-conditions of assertions: A surprisal-based ERP investigation targeting N400 and late positivity effects

Thu, 6 Aug 2026 00:00:00 +0000

To investigate how co-speech gestures modulate linguistic understanding in comparison to specific verbs with similar contents, we conducted an EEG experiment to measure semantic surprisal by exploring the amplitude changes in the N400 component for both information types. We used videos of a person uttering (i) specified sentences, whose verb semantically denoted a specific action, and (ii) underspecified sentences with an unspecific verb accompanied by an iconic co-speech gesture that indicated the specific action of the previous condition. The subsequent sentence contained an instrument noun as target, which either matched or mismatched the specific action. We measured ERPs on the target noun for both information types and found an N400 effect for mismatching target nouns as well as a late positivity effect for gesture conditions. Crucially, no interaction between information type and congruency was observed in the N400 time-window, indicating similar semantic surprisal induced...

Self-directed gameplay reveals common and divergent patterns in human problem-solving

Thu, 6 Aug 2026 00:00:00 +0000

A central goal of cognitive science has been to find universal principles that explain the remarkable flexibility of human problem-solving. However, people display a diversity of problem-solving strategies, prior experience, domain-specific knowledge, and preferences. Which aspects of problem solving are shared across individuals, which are individual-specific, and which are adapted to particular domains? Most studies of human problem-solving rely on brief laboratory tasks in one or a few domains and coarse behavioral measures, making it difficult to robustly explain individual cognitive processing. We propose that longitudinal, self-motivated gameplay on multiple tasks---combined with detailed process-tracing---can help fill these gaps. We introduce mitpuzzles.com, a public platform hosting a suite of constraint-based logic puzzles (e.g., Minesweeper, Sudoku, and Nonograms), instrumented to collect detailed behavioral data including mouse-tracking. We present preliminary analyses...

Semantic Coexistence of Heart and Brain in Chinese: Computational Evidence from 3.4 Billion Tokens

Thu, 6 Aug 2026 00:00:00 +0000

Cultural concepts of mind vary dramatically across societies. In Chinese thought, 心 (xīn, “heart-mind”) historically served as the seat of cognition and emotion, while 脑 (nǎo, “brain”) played minimal role. Modern neuroscience introduced brain-centered models, raising questions about whether scientific concepts replace or coexist with folk theories. How does scientific knowledge reshape traditional mental vocabulary in non-Western contexts? Here we show, through distributional semantic analysis of 3.4 billion Chinese tokens from classical texts to social media, that xīn and nǎo function as parallel rather than competing concepts. Contrary to eliminative materialism, xīn maintains its integrated multi-dimensional structure as a living folk concept, while nǎo enters as specialized medical terminology. This coexistence varies systematically: encyclopedic texts show convergence in biomedical contexts, while vernacular...

What reusable shortcuts do people propose when solving assembly problems?

Thu, 6 Aug 2026 00:00:00 +0000

Humans readily extract statistical regularities from perceptual experience (e.g. a cook noticing which ingredients often appear together). How does such learning guide peoples' performance on procedural tasks (e.g. preparing various dishes)? Here we examine what shortcuts people propose to help them to complete assembly problems more efficiently by eliminating repeated subroutines. Participants (N=301) repeatedly assembled tangram-like shapes, and could create composite tiles for future use. Some participants assembled a sequence of tangrams where certain pairs of tiles recurred consistently (Highly structured); the remaining assembled tangrams with less predictable tile arrangements (Less structured). Participants exposed to Highly structured sequences created tiles that tracked the frequency with which those tile pairs recurred across tangrams, and doing so was accompanied by greater efficiency in assembly. Taken together, these findings suggest that statistical learning guides...

Open dynamics of thought and memory with learning in a deformable landscape.

Thu, 6 Aug 2026 00:00:00 +0000

Classical attractor networks such as Hopfield models treat memories as static points in a fixed energy landscape. While foundational, this framework cannot capture how cognition unfolds through time as agents interact with their environment and recall structured experiences. In this work, I present an open dynamical systems model of memory implemented in an interactive physicsbased environment, where memories are not point attractors but attracting trajectories carved into a deformable potential landscape through experience dependent learning. A state variable evolves under the combined influence of learned landscape gradients and an external driving force, producing time-resolved recall paths rather than instantaneous convergence. I show that this system naturally produces graded recall, intrusion errors, and interference from basin geometry and drive dynamics alone. This framework reframes memory as a dynamical process embedded in perception/action loops and offers a bridge...

False Memory of Grammatical Constructions: Evidence for Structured Construction-Level Generalizations

Thu, 6 Aug 2026 00:00:00 +0000

Three experiments and a control task combine to indicate that grammatical constructions are implicitly represented in memory, with prototypicality emerging from distributional experience, alongside significant verbatim memory. Participants are exposed to instances of a construction and, following a delay filled by an unrelated language task, falsely endorse novel instances of the same construction more often than paraphrases in a recognition task. Critically, false memories are particularly common for lures containing an unwitnessed word that is distributionally prototypical of the construction, compared to frequency-matched controls. A control study excludes the possibility that the latter effect is simply a word-level effect. Finally, we simultaneously find memory for specific exemplars. Overall performance is above chance yet implicit: participants underestimated their accuracy, which was unrelated to age or education.

Who Did What to Whom? Visually Grounded Role Assignment in Humans but Not in Vision--Language Models

Thu, 6 Aug 2026 00:00:00 +0000

Event role assignment is central to event understanding in both language and vision. Crucially, the foundation of this semantic structure is likely rooted in visual experience. However, it remains unclear whether recent vision–language models (VLMs) can attain this fundamental human cognitive capability. To compare humans and VLMs, we conducted two studies using Heider–Simmel–style animations that minimize object and scene semantics. In Study 1, humans identified roles near ceiling (~97%), whereas VLMs were less accurate and less stable across actions (GPT-5: 84%; GPT-4o: 47%). In Study 2, we introduced a Stroop-inspired visual–linguistic mismatch by pairing animations with occasionally incongruent role statements. Humans' role judgments remained highly vision-consistent (92.7%), but VLMs shifted away from the visual event under conflict (GPT-5: 46.9%; GPT-4o: 29.7%), indicating heavier reliance on linguistic cues. In Study 3, a mismatch-shape baseline left both models perfectly...

A Mechanistic Explanation for the Inverted Face Effect

Thu, 6 Aug 2026 00:00:00 +0000

The Inverted Face Effect has been the subject of a great deal of research and controversy in vision science. The nature of the effect, whether it involves holistic processing or not, where in the visual system it resides in the cortex, and whether it is a result of expertise and can apply to other phenomena, has been debated for decades. However, no mechanistic explanation for the effect has been put forward. Here we show that a model that includes multiple fixations on the face driven by a simple salience operator, a foveated retina, and the log-polar mapping from the visual field to V1 can explain this effect. We also simulate Yin's 1969 recognition experiment and show good agreement with our model.

Are More Tokens Rational? Inference-Time Scaling in Language Models as Adaptive Resource Rationality

Thu, 6 Aug 2026 00:00:00 +0000

Human reasoning is characterized by the rational use of resources to optimize performance under constraints. Recently, inference-time scaling has improved the reasoning performance of Large Language Models by increasing test-time computation. Instruction-tuned (IT) models explicitly generate long reasoning traces, whereas Large Reasoning Models (LRMs) are trained via reinforcement learning to discover reasoning paths that maximize accuracy. However, it remains unclear whether resource-rationality can emerge from such scaling without explicit rewards related to computational costs. We introduce a Variable Attribution Task (VAT) in which models infer which variables determine outcomes given candidate variables, input–output trials, and predefined logical functions. By varying the number of candidate variables and trials, we systematically manipulate task complexity. Both models exhibit a transition from brute-force to analytic strategies as complexity increases. IT models degrade...

Parameter-Driven Consensus-based Filtering Improves Collective Judgment Reliability in Crowdsourced Annotation

Thu, 6 Aug 2026 00:00:00 +0000

Crowdsourced annotation can be viewed as a form of distributed human judgment in which individual reliability and task difficulty jointly shape collective decisions. Probabilistic aggregation models estimate latent annotator reliability and task difficulty, but are typically used only to weight judgments or infer labels, not to guide selective dataset refinement. We propose a replicated-dataset consensus filtering framework that improves collective judgment reliability by removing unreliable annotators and difficult tasks based on latent reliability-difficulty parameters. Instead of relying on a single dataset estimate, the method constructs multiple replicated datasets by small random label removal, re-estimates latent parameters on each replicated dataset, and removes components that are consistently selected across replicated datasets. Filtering is applied iteratively with entropy-based stopping conditions. The framework is model-agnostic and can be combined with any probabilistic...

Falsificationism – the incomplete story about choosing good experiments

Thu, 6 Aug 2026 00:00:00 +0000

Falsificationism remains a popular framework considered by many scientists as 'the normatively correct model of science', and of human cognition as a kind of 'lay science'. In this paper, we demonstrate some fundamental conceptual shortcomings and gaps in this normative prescription. While some of the issues presented here connect to well-known arguments from the philosophy of science, we connect these issues to the topic of efficient information search. Specifically, we show that a strict falsificationist rejection of the idea of 'confirmation' of scientific hypothesis (more specifically, entertaining and updating beliefs in those hypotheses, based on evidence) makes scientific research and information search overall less efficient, compared to an accuracy-based Bayesian approach.

The Role of Comparison and Category Confusability on Concept Acquisition

Thu, 6 Aug 2026 00:00:00 +0000

Comparison can support category learning by highlighting shared similarities or diagnostic differences between categories. We examined how comparison type (match vs. contrast vs. single-item control) influences featural category learning and whether these effects depend on category confusability (low vs. high). On training trials, the match condition was shown co-presented exemplars from the same category, whereas those in the contrast condition were shown co-presented exemplars from different categories; the control condition was shown individual exemplars. Participants learned artificial categories in a mixed design with comparison type manipulated between subjects and confusability within subjects. Learning was assessed via training performance, embedded endorsement, and posttest endorsement measures. Across all measures, contrast outperformed match, whereas control performed comparably to contrast. Notably, even with fewer exposures, single-item presentation still outperformed...

What comes to mind? Ad hoc categories as contextual reweighting of a stable high-dimensional semantic space

Thu, 6 Aug 2026 00:00:00 +0000

Humans rapidly construct ad hoc categories, e.g., generating "beet" for "vegetables to paint with". Building on feature-based accounts of category representation, we propose an analytic framework in which ad hoc categories are constructed by shifting the base category's representation (e.g., VEGETABLE) along modifier-relevant dimensions (e.g. suitable for painting) in a high-dimensional semantic space, without an explicit hand enumerated feature search. We operationalize this with off-the shelf semantic embeddings, fitting per-category sparse linear models that predict item generation across 20 categories. Across 9 categories tested for compositional transfer, learned modifier directions improved predicted retrieval when physical constraints generalize across domains (e.g., portability for "that could fit in your pocket"), while transfer was weaker for context-dependent modifiers. The learned axes align with human-rated features, with top-weighted features matching modifier semantics....

Valence is more prominent than truth conditional meaning

Thu, 6 Aug 2026 00:00:00 +0000

Valence, the positive/pleasant or negative/unpleasant value of information, grounds our experience with the world. What specific role does valence play in constructing meaning? Some have downplayed it, centering the referential relationship between words and the world, "truth-conditions". Others have taken valence to be central to meaning. Empirically, the jury is still out. Yet, few have examined how these two types of information compete for prominence within the same word (the Valence Prominence Hypothesis, that valence usually wins). In a first set of studies, we find that valence is significantly faster than truth-conditional categorization. In the second set of studies, we find that valence and not truth-conditional similarity drive judgments. Further, we tested valence against 4 different informational domains fundamental to development and cognition. Animacy stood out as the domain which most closely parallels Valence in its prominence, suggesting a link between meaning...

Salience of Surface versus Higher-level Properties in Spatial Comparison at Different Levels of Scene Similarity

Thu, 6 Aug 2026 00:00:00 +0000

Identifying similarity plays a major role in spatial reasoning. Prior work has shown that identification of similarity and difference between spatial scenes involves reasoning about their properties at two different levels--surface-level properties, such as color and size, and higher-level compositional properties. The interaction between the use of these property types during reasoning and the degree of similarity or difference between spatial scenes has, however, been largely unstudied. We presented participants with in-progress Tangram puzzles at various stages of similarity to their target image. Participants were asked to identify the puzzle being built and provide an explanation of how they came to that conclusion. Analysis of explanations found an interaction between degree of similarity and preference for reasoning over surface-level or higher-level properties.

The Bilingual Advantage in Preschoolers: Does It Hold for Typologically Distant Languages in a Hybrid Cultural Context?

Thu, 6 Aug 2026 00:00:00 +0000

This study focused on the impact of bilingualism on executive function (EF) and theory of mind (ToM) in preschool children. Seventy-two children (age, 4-6 years) were involved: 36 Turkish monolinguals and 36 bilingual children whose first language was typologically distant and second language was Turkish. EF was measured using a Flanker task and ToM was measured using a False Belief task. Contrary to an expected bilingual advantage, monolinguals demonstrated better performance in EF task and were more likely to pass the ToM task. Among the bilinguals, the Arabic-Turkish speakers scored lower on ToM, which might be attributed to socioeconomic differences. The results raise the possibility that distant languages may lead to higher cognitive load and to fewer bilingual advantages. Moreover, the sociocultural context of Turkiye characterized by diverse cultural values and linguistic environments may moderate effects of bilingualism, revealing the need for models of bilingual cognition...

DEGSleepNet: A Dual Evolving Graph Network for EEG-Based Single-Channel Automatic Sleep Staging

Thu, 6 Aug 2026 00:00:00 +0000

The development of brain–computer interfaces (BCI) has provided a solid data foundation for sleep staging. Transformer-based methods have achieved moderate sleep staging by capturing temporal dependencies within time-frequency representations. However, Transformers with substantial computational overhead are limited to capturing pairwise temporal dependencies rather than group-wise temporal dependencies. To address these issues, we propose a Dual Evolving Graph Network (DEGSleepNet) for sleep staging. DEGSleepNet consists of multiple Mamba with one-dimension inverse discrete cosine transform (MambaIDCT) blocks and dual evolving graph (DEvoGraph) blocks. DEvoGraph consists of a time EvoGraph and a frequency EvoGraph. Time EvoGraph sequentially captures both group-wise and pairwise temporal dependencies, while frequency EvoGraph does the same for frequency dependencies. Extensive experiments demonstrate DEGSleepNet achieves the best performance on public sleep datasets. Notably,...

Adults equate stillness with learning

Thu, 6 Aug 2026 00:00:00 +0000

Learning is difficult to measure because it is not directly observable. Even in formal educational settings it's hard to tell whether a child might be thinking through a difficult problem or daydreaming. The same is especially true in informal settings like while watching TV at home—it's difficult to know whether a child is meaningfully engaged with what they are watching or whether they are overstimulated and unable to disengage. Caregivers and educators regularly rely on behavioral cues to infer kids' learning. Physical engagement, or sitting still, is often treated as a reliable signal of learning. However, stillness during screen-based media is not synonymous with learning in young children and may in fact indicate overstimulation and disrupted learning (Shepherd & Kidd, 2024). We asked 200 adults whether they thought the same child was more attentive and learning more from a video when sitting still compared to when making small fidgets, like twiddling their thumbs or...

Intuitive Judgement with Analytical Oversight: A Dual-Process Architecture for Complex Relational Understanding

Thu, 6 Aug 2026 00:00:00 +0000

Complex relational understanding tasks such as Document-level Relation Extraction require resolving semantic ambiguity amid a quadratic explosion of entity pairs, causing severe class imbalance and false negatives. Although large language models show strong reasoning ability, directly applying them to DocRE is inefficient and prone to hallucination. Inspired by dual-process theories of human cognition, we propose DocRE-Thinker, a hybrid framework integrating fast intuition with controlled deliberation. A fine-tuned backbone serves as System 1, efficiently generating candidate relations while estimating uncertainty. A frozen large language model acts as System 2 and is invoked only for high-uncertainty cases, performing uncertainty arbitration and attribute-based rule induction to justify relations. These symbolic rules are converted into supervision through a text-gradient feedback mechanism, enabling System 1 to internalize analytical reasoning in a rule-aware manner. Experiments...

The Role of Holes in Whack-a-Mole: Investigating Procedural Learning with Graded Statistical Structures Across State Space Sizes

Thu, 6 Aug 2026 00:00:00 +0000

Procedural learning---the implicit acquisition of motor and cognitive skills through repeated practice---has been proposed to support abilities from typing to language acquisition. Traditional serial reaction time paradigms study procedural learning using stimulus sequences with simple underlying statistical structures: deterministic or with binary probability levels, involving only 4 possible states. We introduce a novel paradigm where participants learn sequences with graded probability levels, generated from transition matrices with a continuous gradient of transition probabilities. We found robust effects of both surprisal (transition probability from previous state) and entropy (uncertainty of next state) on reaction times across 5 state space sizes (4-8), with no interaction between surprisal and entropy, indicating learners tracked specific transition probabilities even at high-uncertainty states. These findings demonstrate that procedural learning is sensitive to fine-grained...

Primates solve risky decision-making problems via spatial computations

Thu, 6 Aug 2026 00:00:00 +0000

What are the neural mechanisms underlying risky decision making? How do the implementation details of these mechanisms impact observable behaviour? Here we hypothesise that non-human primates (NHPs) are re-using circuits for spatial navigation to solve decision making problems. This is motivated by recent findings of grid coding of probability and magnitude in NHP frontal cortex (Bongioanni et al., 2021; Veselic et al., 2025). Grid cells are an optimal code for space (Dorrell et al., 2023; Fiete et al., 2008; Sorscher et al., 2023) but this form of representation is not normatively optimal for risky-decision making. However, this might be a small price for efficiency: by embedding problems in space, NHPs can recycle spatial solutions to quickly solve new problems. First, we show that this spatial framework can replicate classical prospect theoretic findings (Kahneman & Tversky, 1979). Then, we show that the framework also predicts unique biases in the behaviour of NHPs, which...

Mental Rotation or Pattern Matching? Representational Structure and Angle-Dependent Behavior in Vision–Language Models

Thu, 6 Aug 2026 00:00:00 +0000

Vision-language models (VLMs) often perform well on spatial reasoning tasks, but it remains unclear whether their performance is supported by internal representations that track rotation angle. We study this question in a controlled same or mirror 3D object task using LLaVA-1.5-7B as an intervenable computational system. First, the model's decision margin varies systematically with rotation angle, mainly for rotated rather than mirrored stimuli. Second, we identify an angle-related direction in intermediate hidden representations whose activity covaries with behavior. Third, targeted projection ablations produce progressive flattening of the angle-margin relationship as intervention strength increases. Together, these findings provide mechanistic evidence that angle-sensitive internal geometry contributes to the model's behavior on this task. The results support a cautious interpretation of structured, intervenable spatial processing rather than a strong claim of human-like mental...

Use of symbolic inductive biases for few-shot learning in humans and machines

Thu, 6 Aug 2026 00:00:00 +0000

People are capable of learning with very little data. We argue that this is because people possess an symbolic inductive bias, consistent with the language of thought (LoT) hypothesis (Fodor & Pylyshn, 1988). To evaluate this claim, participants were asked to learn list functions by predicting how to transform an input list of numbers to an output list. We used participants' predictions on trials with no feedback to search for a congruent representation using a LoT model. The model was able to predict participants' heldout responses at above chance levels, even for participants who were unable to learn the function. Furthermore, LLMs and a neural network trained with a LoT inductive bias displayed similar patterns. These findings suggest that people and ML models display signatures of symbolic representation use to accomplish few-shot learning.

Your Brain Knows It's a Lie: Event Related Potentials Reveal the Role of Evidentiality in Deception Detection

Thu, 6 Aug 2026 00:00:00 +0000

Evaluating the credibility of a speaker's statement requires more than assessing factual accuracy; it also involves tracking cues to the speaker's epistemic commitment to the claim. Grammatical evidentiality, which encodes information sources, may therefore shape how statements are evaluated during language comprehension. Turkish marks the information source through obligatory evidential morphology and offers a unique window into how source information modulates real-time truth judgements. The present study examined the neural dynamics of processing of accurate and inaccurate statements in Turkish by investigating how direct (-DI) and indirect (-mI_) markers modulate language comprehension as measured by ERPs. Participants viewed short event videos and then judged the veracity of written statements describing the events while an EEG was recorded. ERPs time-locked to the evidentially-marked verbs showed differential electrophysiological responses associated with semantic integration...

A computational model of strategic punishment in divided societies

Thu, 6 Aug 2026 00:00:00 +0000

In divided societies, authorities often use punishment to establish shared norms while trying to maintain or signal their legitimacy. When facing a polarized audience, these goals may often be at odds. We extend the Rational Communicative Social Action (RCSA) framework (Radkani et al., 2022) to formally model an authority's strategic decision-making when using public punitive actions to pursue these goals. We distinguish between a communicative authority who aims to shape the audience's moral beliefs, and a reputation-aware authority who optimizes the audience's assessment of their own character. By simulating these agents against polarized audiences, we characterized the tradeoffs and dynamics of strategic punishment when facing diverse audiences with various forms and levels of polarization, which serves to generate systematic predictions for future experiments.

Similarity and generalization as discounted integration over higher-order paths in semantic networks

Thu, 6 Aug 2026 00:00:00 +0000

Similarity and generalization are often assumed to reflect distance in an underlying representational space. In semantic networks, however, distance can be defined by both shortest paths or by multiple indirect pathways. Previous research has shown that non-shortest, higher-order paths influence similarity judgements. Here, we extend these findings by re-analyzing a publicly available dataset to compare shortest-path models with models integrating over discounted higher-order paths. The latter better accounted for similarity judgements; error analyses indicate that this advantage arose from integrating multiple indirect paths rather than relying on shortest connections. These random-walk models implement the same core computation as the successor representation (SR), which has been proposed to support human generalization. Consistently, an SR-based model outperformed alternatives in accounting for inductive generalization judgements from the same dataset. These findings suggest...

From Finding to Feed: Translating Empirical Research for Social Media

Thu, 6 Aug 2026 00:00:00 +0000

Cognitive scientists spend much of their professional training learning how to explain empirical research to one another: in talks, papers, posters, grant proposals, and conference discussions. An integral part of our training is the communication and dissemination of scientific ideas at an expert level. Far less attention is given to translating that same work for communities beyond academia. This gap matters because cognitive science produces findings about learning, reasoning, attention, memory, communication, development, decision making, and human-technology interaction—topics that many members of the public already encounter in everyday life. Cognitive research is fascinating, relevant, and often meets urgent needs. Yet when research findings circulate without context, they can be misunderstood, oversimplified, or mistrusted. People may not trust what they cannot understand, and many misunderstandings begin at a basic level: what empirical research is or looks like, what...

Reflective Satisficing: Beyond the Classical Speed–Accuracy Tradeoff

Thu, 6 Aug 2026 00:00:00 +0000

Classical satisficing was grounded in single computation. However, when cognitive processing is delegated to external agents, cognitive modes (System 1/2) become dual and isomorphic, creating a 2×2 intersection. This study identifies four cognitive resource allocation patterns arising from this intersection as Reflective Satisficing (RS): co-constrained compromise that bypasses verification (Mode 1: S1×S1), control recovery through manual verification (Mode 2a: S2×S1), structural futility where effort fails to convert to accuracy (Mode 2b: S1×S2), and optimal allocation leveraging external resources as epistemic scaffolding (Mode 3: S2×S2). The Speed-Accuracy Tradeoff (SAT) predicts that deliberation improves accuracy. Extended cognition expands this landscape in both directions. Mode 3 (77% prevalence) achieves gains beyond the classical curve—deliberation augmented by epistemic scaffolding. Yet we also identify a region where SAT breaks:...

Can we assess consciousness in AI?

Thu, 6 Aug 2026 00:00:00 +0000

Progress in AI has led to a lively debate about the possibilities of creating and assessing consciousness in AI, building on Putnam's conjecture that computational functionalism is more plausible than a biological view of consciousness. I discuss Butlin et al.'s (2025) proposal for extracting computational indicators of consciousness from neuroscientific theories and critically evaluate computational functionalism in light of the alternative, biological naturalism. While computational functionalism allows for AI consciousness in principle, it is questionable whether conscious AI systems are practically possible, and whether we are in an epistemic position to judge that a specific AI is conscious. Putative computational indicators extracted from neuroscientific theories of consciousness can never be as well (or even better) supported by empirical evidence than their neurobiological realizers measured in human brains. All potential markers of consciousness that are typically...

Psychological Heterogeneity in User Preferences of Real-Time AI Mediation As Cognitive Scaffolds: A Latent Class Analysis

Thu, 6 Aug 2026 00:00:00 +0000

User studies of AI-mediated communication typically report average effects, while interview-based work remains dominated by thematic narrative or Likert clustering, leaving the structured heterogeneity expressed in post-task interviews unquantified. We applied latent class analysis as a methodological bridge that converts qualitative interview codes into discrete, domain-specific user profiles, enabling design reasoning about for whom, on which dimension, and under what conditions real-time AI scaffolds help or burden. As a validating scenario, 29 non-native English speakers interacted with XPLAIN, a Wizard-of-Oz proactive scaffold in Zoom's sidebar, to bridge gaps in linguistic and cultural knowledge during a collaborative task. Across eight thematic domains, two-class solutions were consistently best-fitting (BIC); class membership was largely independent across domains, indicating multi-dimensional rather than global user types (e.g., longer English immersion was associated...

The Cost of Inline Definitions: Vocabulary Support in English-Language Math Problem Solving

Thu, 6 Aug 2026 00:00:00 +0000

Many non-native English speakers study mathematics in English, so they must learn both the math and the academic language used to express it. Large language models (LLMs) might help by producing worked solutions while also explaining difficult words and phrases, but it is unclear whether doing both harms mathematical accuracy. We test whether adding inline vocabulary explanations changes solution correctness. Using the CEFR-J vocabulary profile, we identify terms likely to need clarification and evaluate four open LLMs (2.7B–20B parameters) on English math problems in two conditions: standard solving vs. solving with embedded vocabulary support. On MMLU Elementary Mathematics and GSM8K, vocabulary scaffolding consistently reduces accuracy, with drops up to 16.2 percentage points. These findings reveal a trade-off between language assistance and reasoning performance. Educational interfaces may need to balance these goals carefully or separate language support from problem solving...

LLM-generated possibilities increase blame attribution