Search

Semanticists have debated the extent to which modality and causation are related in natural language. This paper aims to promote a theory in which overt causatives share core components of meaning with deontic modals. We report a sentence recall experiment that suggests that priming can be used to target the high-level semantic representations shared between two syntactically distinct linguistic expressions. Our results show that it's possible to prime the production of the deontic modal 'had to' (e.g., ``George 'had to' go to the store"), with causative 'made' (e.g., ``Jane 'made' George go to the store''), suggesting that the two expressions share a component of their meaning. Our results contribute to the methodological development in experimental semantics by establishing the utility of the priming effect to target meaning.

As mental representations are standardly thought to underlie all cognitive processes, a major goal of cognitive science has been to uncover representations. Methods for representation learning from behavioral data often model choice or reaction time data alone, but not jointly, leaving out potentially useful information. Here we develop two models of choice and RT in the odd-one-out task, including one based on the Linear Ballistic Accumulator. Parameter recovery simulations show joint modeling of choice and RT with LBA recovers representations more accurately than modeling choice alone with softmax. However, on two empirical datasets of images and words, joint models performed no better than choice-only models, despite a significant correlation of reaction time with two measures of similarity and choice difficulty in both datasets. We speculate on reasons for the unrealized promise of joint modeling of RT and choice in representation learning.

Index systems are central to our everyday and intellectual lives. Their ubiquity and diversity make them an important class of cognitive artifacts, the study of which has implications for our understanding of representational systems in general. This paper builds schema-theoretic network models of the nature of the memory structures, that underpin the interpretation of indexing systems. We identify four common classes of index systems. Using Representation Interpretation Structure Theory, we ex-plain how the four basic classes can be responsible for the substantial diversity among index systems.

In similarity comparisons, people often show a preference for one direction over the other. Bowdle and Gentner (1997) proposed the base systematicity advantage account to explain this—namely, that people prefer similarity comparisons in which the more systematic item serves as the base. Results from a series of studies supported this account. However, the studies only covered literal similarity comparisons. The question of whether analogical comparison follows the base advantage pattern remained untested. Therefore, the present study investigated this question for analogical comparisons. We tested the prediction that a comparison will be preferred when the more systematic item serves as the base. This prediction was supported. We also found support for a further prediction: namely, that inferences were projected from the systematic to the less systematic passage. Further, these inferences spontaneously arose even when not requested. The overall results from these processes are consistent with the base systematicity advantage account.

Despite the exceptional performance of large language models (LLMs) on a wide range of tasks involving natural language processing and reasoning, there has been sharp disagreement as to whether their abilities extend to more creative human abilities. A core example is the interpretation of novel metaphors. Given the enormous and non-curated text corpora used to train LLMs, a serious obstacle to designing tests is the need to obtain novel yet high-quality metaphors that are unlikely to have been included in the training data. Here we assessed the ability of GPT-4, a state-of-the-art large language model, to provide natural-language interpretations of novel literary metaphors drawn from Serbian poetry and translated into English. Human judges—blind to the fact that an AI model was involved—rated metaphor interpretations generated by GPT-4 as superior to those provided by a group of college students. In interpreting reversed metaphors, GPT-4, as well as humans, exhibited signs of sensitivity to the Gricean cooperative principle. These results indicate that LLMs such as GPT-4 have acquired an emergent ability to interpret literary metaphors.

Understanding how people represent similarity relations between concepts is one of the most fundamental problems in cognitive science, with implications for many theories of learning and reasoning. Human judgments of similarity violate basic metric assumptions, leading to effects such as judgment asymmetry and the triangle inequality. These effects have been difficult to capture with modern geometric representations of conceptual structure such as vector embeddings. Here we introduce a similarity function related to a feature-based view of concepts. We show how this function can be applied to geometric representations and that the resulting algorithm can account for classic judgment effects. Using representations extracted from a Large Language Model, we computed the predictions of this approach to similarity relations among a set of everyday concepts (world countries), and evaluated these predictions against human judgments of similarity in a behavioral experiment. The model's predictions correlate with human judgments. These results offer insight into human judgments of similarity relations and the design of algorithms that align with human reasoning.

The precise mechanisms precipitating the process of representational change in problem solving have been investigated for nearly a century. One current hypothesis is that analyzing the unchanging elements of previous attempts may facilitate restructuring. We investigated this hypothesis by providing solvers with three common examples of unsuccessful problem attempts, their own problem attempts, or no previous attempts. The prior attempts conditions eliminated the need to rely on working memory to access previous unsuccessful attempts. While there was no evidence for an overall effect of the prior attempts conditions, cognitive reflection was identified as a reliable predictor of restructuring and solving. Eye-tracking data were collected to further investigate the contributions of these systems to fixations while solving. The current study is an exploratory analysis of this data, with analyses focusing on participants' fixations on problem-irrelevant space and unsuccessful attempts.

Objects in the world are represented at multiple hierarchical levels of abstraction. For example, you can identify a four-legged creature as an animal, or a dog, or specifically as a Cocker Spaniel. While there has been extensive work examining the relationships between hierarchical category levels, it is unclear how such representations interact during categorization. That is, do individuals process category levels serially or are category levels processed in parallel during categorization? Here, we had participants learn categorization rules for four categories of novel creatures. We examined patterns of errors that participants made in a forced response task, where we manipulated the amount of time participants had to make responses on a trial-by-trial basis. Our results indicate that participants process category levels in parallel, rather than serially resolving superordinate levels before subordinate levels. Parallel processing of category levels could underpin the remarkable flexibility with which we access and deploy category information.

Verbal descriptions of spatial configurations open a window to a specific aspect of visual cognition relevant to the interpretation of topological relations in the visual world. The present study reports an experimental investigation of the production of spatial prepositions by human participants while they verbally described visual stimuli within a stimuli battery commonly utilized in relevant research. The analysis of participants' eye movements revealed a relationship between the variety of spatial terms in the given language and native speakers' fixation patterns on the stimuli. A broader spectrum of spatial expressions, describing the same visual scene, is related to longer and more frequent fixations on the stimuli. The findings reflect cognitive processes, as indicated by oculomotor control variables, related to the verbal expression of spatial relationships.

Whether travelling, playing games, or debugging code, any situation where an agent desires change can be framed as a problem. Despite this ubiquity, there is no unifying framework describing how people reason backwards when solving problems. We introduce AND/OR trees, which chain together subgoals and actions to attain them, as a way to represent this process. To investigate whether actions from AND/OR trees were predictive of human behavior, we conducted a study in which participants solved deterministic, long-horizon puzzles. AND/OR trees were able to explain most of the actions the participants took. Next, we modeled search through these trees using a psychologically plausible, single-parameter search algorithm. We fit this model to the data of individual participants and found that it captures trends in summary statistics of human play. Our results show the promise of AND/OR trees as a representation for backward reasoning in problem solving.

The nature of consciousness has been a long-debated concept related to human cognition and self-understanding. As AI systems become more capable and autonomous, it is an increasingly pressing matter whether they can be called conscious. In line with narrative-based theories, here we present a simple but concrete computational criterion for consciousness grounded in the querying of a virtual self-representation. We adopt a reinforcement learning (RL) setting and implement these ideas in SubjectZero, a planning-based deep RL agent which has an explicit virtual self-model and whose architecture draws similarities to multiple prominent consciousness theories. Being able to self-localize, simulate the world, and model its own internal state, it can support a primitive virtual narrative, the quality of which depends on the number of abstractions that the underlying generative model sustains. Task performance still ultimately depends on the modeling capabilities of the agent where intelligence, understood simply as the ability to model complicated relationships, is what matters.

Emotion space models are frameworks that represent emotions in a multidimensional space, providing a structured way to understand and analyze the complex landscape of human emotions. However, the dimensional representation of emotions is still debatable. In this work, we are probing the higher dimensional space constituted by emotion labeling done by participants from India on multimedia stimuli.Our approach formalizes the study of emotion in the investigation of representational state spaces capturing semantic variation in emotion-related response.We have created a high-dimensional space of emotional ratings by participants to represent emotional stimuli. Using t-SNE, we have projected the higher dimensional space into two dimensions. We observed that the structure of emotional categories and clusters formed of these emotional categories is similar to Russell's circumplex model. The transition from the blended complex emotional states to the discrete emotional states is projected out from the centre, and discrete emotional states occur in the periphery.

Searching through memory is mediated by complex interactions between the underlying mental lexicon and the processes that operate on this lexicon. However, these interactions are difficult to study due to the effortless manner in which neurotypical individuals perform cognitive tasks. In this work, we examine these interactions within a sample of prelingually deaf individuals with cochlear implants and normal hearing individuals who were administered the verbal fluency task for the "animals" category. Specifically, we tested how different candidates for underlying mental lexicons and processes account for search behavior within the verbal fluency task across the two groups. The models learned semantic representations from different combinations of textual (word2vec) and speech-based (speech2vec) information. The representations were then combined with process models of memory search based on optimal foraging theory that incorporate different lexical sources for transitions within and between clusters of items produced in the fluency task. Our findings show that semantic, word frequency, and phonological information jointly influence search behavior and highlight the delicate balance of different lexical sources that produces successful search outcomes.

Analogical reasoning is one of the most common ways individuals bring previous experience to bear on unfamiliar situations. Most theories describe this process as a structured comparison that involves mapping the relational properties between a familiar source and unfamiliar target. This both allows the transfer of useful inferences from the source to the target and highlights the common structure shared by both analogs, represented by an abstract schema. This schema can help with identifying and reasoning about structurally similar situations in the future. While researchers have studied how representations of source and target analogs undergo alterations as a result of this mapping process, little attention has been paid to how the abstract schemas thought to guide future analogical reasoning might similarly change with use. We explore this question in two experiments and present evidence that suggests abstract schemas do indeed drift under certain conditions.

The present study examined the role of detecting the initial causal system model followed by engaging in active vs. passive structural alignment in recognizing the key causal principles in subsequent novel examples. The results echo prior research on the benefit of analogical comparison in learning relational categories: participants who were prompted to compare outperformed participants in the baseline condition. Moreover, while the accurate representation of the causal system predicted noticing the relational structure in novel examples, making more accurate relational mappings made participants more likely to notice the structure above and beyond having an accurate representation. These findings offer insight into the role of active vs. passive analogical comparison and have implications for conditions that might support learning of relational categories.

The process of adaptation to the partner in the course of an interaction is still not well understood. In the case of explanatory dialogues, to provide satisfying explanations, explainers have to consider the needs of the explainees. This requires mental representations of the explainees, i.e., “partner models”. Little is known about whether and how modifications of partner models during an explanation take place. We assumed that they get informed by the interactive behaviour of the explainee and investigated partner models in relation to explainees' verbal moves. A total of 59 dyadic explanations were investigated in an observation study. The comparison of the partner models before and after the explanation showed changes regarding, e.g., knowledge, interest in the explanation, cooperation, and mood. Moves such as questions as well as summarising and paraphrasing information given by the explainees were associated with the partner model dimensions interest in the explanation and co-construction.

Over the last years, advancements in deep learning models for computer vision have led to a dramatic improvement in their image classification accuracy. However, models with a higher accuracy in the task they were trained on do not necessarily develop better image representations that allow them to also perform better in other tasks they were not trained on. In order to investigate the representation learning capabilities of prominent high-performing computer vision models, we investigated how well they capture various indices of perceptual similarity from large-scale behavioral datasets. We find that higher image classification accuracy rates are not associated with a better performance on these datasets, and in fact we observe no improvement in performance since GoogLeNet (released 2015) and VGG-M (released 2014). We speculate that more accurate classification may result from hyper-engineering towards very fine-grained distinctions between highly similar classes, which does not incentivize the models to capture overall perceptual similarities.

It remains controversial whether the visual system encodes abstract relational roles such as Agent and Patient in visual events. The present experiment tested whether abstract role bindings induce priming effects across consecutive events. Each trial included a static target image preceded either by a brief silent video of a priming event or by an audio-visual presentation of an English sentence describing the same event. Example sentence: “The red goat on the left knocked down the blue goat on the right.” 64 videos counterbalanced 4 event types: launching, deforming, breaking, and a relationally ambiguous control. The set of static targets were the final frames of the same videos. The role bindings were either repeated, switched, or ambiguous across the target and prime. The dependent variable was the latency on a color-localization task (e.g., whether the red animal was on the left or on the right). Whereas the linguistic primes had no statistically significant effect on the latency of the visual task, the role bindings of the video primes did have an effect: The latency on unambiguous trials (which required role binding) was significantly greater than that on ambiguous trials (on which at least one component lacked clear relational roles). This suggests the visual system is sensitive to (the ambiguity of) the role bindings of abstract relations.

Executive functions guide episodic memory to retrieve information essential for adaptive behavior. The prefrontal cortex achieves this by influencing hippocampal processing through anatomical projections targeting the entorhinal cortex and area CA1. However, most computational models of the hippocampus overlook this cognitive control, either neglecting it or implementing implausible direct connections to the hippocampus. This paper explores the contextual control of associative memory implemented by modern Hopfield networks, within a hippocampus-inspired autoencoder. Our experiments underscore the importance of proximity between prefrontal afferences and the locus of memory storage for efficient contextual modulation of episodic memory, challenging the standard model of hippocampal processing. These findings not only advance our understanding of higher-level cognition but also provide design principles for more adaptive machine learning algorithms.

Traditional conceptions of inference emphasize explicit following of logical rules, often tied to the possession of natural language, thereby implying that non-human animals cannot make inferences. However, comparative research shows extensive evidence of the success of several species of non-human animals in nonverbal reasoning tasks, putting pressure on the traditional view. We deny two traditional assumptions about inference: the lingualism of thought, and the requirement of explicit rule following. We suggest instead a multidimensional model of inference illustrated through several case studies. Thereby, we categorize informational transfers across three dimensions by marking the degree of context-independence, the format of representation, and the type of perspectivity involved. By allowing for a more nuanced interpretation of empirical data than the traditional view, our framework is able to accommodate inferential behaviors of both linguistic and non-linguistic agents, and shed light on varied manifestations of inference across species and developmental stages.

We explore how viewpoint phenomena interact with metacognition during dynamic, intertwined processes of thinking, speaking, gesturing, and improvising music. Taking a perspective on experienced or solely imagined situations involves physical and/or conceptual positioning within or outside a spatial, narrative or mental context, whereby speakers typically employ various bodily articulators to signal simultaneous or shifting viewpoints. Tapping into how viewpoint frames thought processes, we propose that shifting viewpoints are a metacognitive strategy to explore contextual possibilities through semiosis, embodied in gestures and other body movements. Changing viewpoints on an unfolding situation, including one's own mental activities, allows for both re-experiencing scenarios and exploring new ideas and perspectives. This theoretical groundwork prepares our empirical research into how metacognition and viewpoint jointly drive musical improvisation. Applying cognitive semantics and Peirce's semiotics, we present preliminary analyses of musicians' improvisation and their multimodal meta-discourses (including motion-capture data), thus exploring cognitive-semiotic processes in musical creativity.

We present a computational explainability approach for human comparison tasks, using Alignment Importance Score (AIS) heatmaps derived from deep-vision models. The AIS reflects a feature-map's unique contribution to the alignment between Deep Neural Network's (DNN) representational geometry and that of humans. We first validate the AIS by showing that prediction of out-of-sample human similarity judgments is improved when constructing representations using only higher-AIS feature maps identified from a training set. We then compute image-specific heatmaps that visually indicate the areas corresponding to feature-maps with higher AIS scores. These maps provide an intuitive explanation of which image areas are more important when it is compared to other images in a cohort. We observe a strong correspondence between these heatmaps and saliency maps produced by a gaze-prediction model. However, in some cases, meaningful differences emerge, as the dimensions relevant for comparison are not necessarily the most visually salient. In sum, Alignment Importance improves prediction of human similarity judgments from DNN embeddings, and provides interpretable insights into the relevant information in image space.

We identify three shortcomings of word vectors as representations of the full meaning of words: 1) the dimensions of the vectors are implicit and difficult to interpret, 2) the vectors entangle all the meanings and uses of words, and 3) the vectors are unstructured. We propose solutions to each of these shortcomings and explore the implications. Our goal is to integrate word, phrase, and clause level vectors representing fine-grained, associative aspects of meaning into grammatical analysis, to support the resolution of structural ambiguities that cannot be grammatically resolved.

Emotions can be insensitive to certain attributes of a situation. A large body of evidence shows that information on probabilities, large numerical counts, or intentions is frequently disregarded in the elicitation and regulation of emotions. To date, no existing theory comprehensively accounts for the features that tend to be overlooked by emotion. In this paper I call attention to the common denominator of such features: they cannot be perceived nor contribute to the iconic representation of events. For instance, the exceedingly low probability of a plane crash does not affect its imagistic representation (i.e., the iconic representation of the event is silent about the event's probability). I introduce the Iconic Prioritization Hypothesis, positing that the prioritization of the iconic format in emotion can explain the neglect of information that is representationally silent in this format. Emotion may favour iconicity as it is the format of immediate, first-hand evidence about our surroundings.

In visual cognitive neuroscience, there are two main theories about the function of the ventral visual system. One suggests that it serves to classify objects (H1); the other suggests that it generates intermediate representations from which people can generate verbal descriptions, actions, and other kinds of information (H2). To adjudicate these, we trained two deep convolutional AlexNet models on 330,000 images belonging to 86 classes, representing the intersection of Ecoset images and the semantic norms collected by the Leuven group. One model was trained to produce category labels (H1) , the other to generate all of an item's semantic features (H2). The two models learned very different representational geometries throughout the network. The representations acquired by the feature-generating model aligned better with human-perceived similarities amongst images, and better predicted human judgments in a triadic comparison task. The results thus support H2.

In the field of medical decision-making, understanding the treatment intentions of clinicians is crucial for effective treatment strategies. However, these intentions are often implicit and challenging to quantify. In this paper, we propose a novel two-module model to infer and implement clinicians' treatment intentions through treatment records. We construct the InfCTI module, which infers intentions and quantifies them numerically, and the ImpCTI module, which generates treatment strategies based on inferred intentions. Our experiments demonstrate that the treatment strategies obtained by ImpCTI reflect clinicians' intentions and the intention values obtained by InfCTI are reasonable. This model has the potential to improve the quality of care provided to patients.

Representational overlap is debated as the neural basis of multitasking costs. Cognitive theories propose that overlapping task representations lead to an unintended exchange of information between tasks (e.g., crosstalk). Recently, modality-based crosstalk was suggested as a source for multitasking costs in multisensory settings. Robust findings of increased costs for certain modality mappings, even when both tasks use non-overlapping stimulus and response modalities, may be explained by crosstalk between the stimulus modality in one task and sensory action consequences in the concurrently performed task. This study (N = 54) employs functional neuroimaging, multivariate pattern analysis, and modality-specific interventions to investigate neural overlap in multitasking, emphasizing modality compatibility. Noteworthy, differences in single-task representations were found in the auditory cortex but not in fronto-parietal regions. Improved auditory decoding accuracy in modality-incompatible tasks predicted dual-task performance gains, eliminating modality-specific costs, exclusively for the modality-incompatible intervention group. This study provides neural evidence for modality-based crosstalk in sensory regions.

Implicit attitudes are often defined as residing beyond conscious awareness. This definition has been challenged by robust evidence demonstrating highly accurate predictions of implicit attitudes. However, relevant tests have all been conducted using well-known targets (e.g., racial groups), about which participants possess ample relevant knowledge. Therefore, accurate predictions may have emerged from inferential mechanisms rather than privileged first-person awareness. Here we probe participants' (N = 4,448) ability to report their own experimentally created implicit attitudes across four studies where implicit attitudes and their explicit counterparts (representing an obvious source of inference) were manipulated to shift in opposite directions. Predicted and actual implicit attitudes were either unrelated to each other, or predictive accuracy was limited to participants whose implicit and explicit attitudes were aligned. Echoing classic and contemporary accounts, these data suggest that implicit attitudes are (largely) unconscious, and successful implicit attitude predictions are likely subserved by inference rather than introspection.

Does grammatical gender influence the meaning of inanimate nouns? We examined word embeddings from distributional semantics models, representing meanings in a vector space. In 26 gendered languages and non-gendered English, we measured the meaning similarity of inanimate nouns to gendered anchor nouns like 'male' and 'female.' In gendered languages, noun meanings aligned more with the anchor noun congruent with grammatical gender. This effect persisted when comparing the same nouns across languages (e.g., 'cucchiaio' vs 'cuchara' vs 'spoon'). We propose that grammatical gender introduces a gender bias into lexical semantics through distributional similarities with anchor words, revealing masculine/feminine features even without direct sensorimotor experience. This suggests that embodiment in language processing may become statistically embedded in word usage patterns.

Real-world decision-making requires the generation of possible options. Humans are exceptionally good at navigating such potentially unbounded spaces: they typically generate their best options first and most idiosyncratic last (Srinivasan, Acierno, & Phillips, 2022). Recently, large language models (LLMs) have shown impressive communication and reasoning abilities, suggesting that they may now be mirroring some of the conceptual structures used by humans. Here, we explore if LLMs navigate option spaces similarly to humans. We compared series of human-generated options to those from an LLM using the semantic similarity of generated options across various open-ended contexts. While LLMs display some global patterns similar to humans, their option sequences follow different trajectories within the semantic space. Specifically, GPT-3 frequently revisits previous semantic clusters, whereas humans progress more linearly. Additionally, compared to humans, GPT-4 typically shows fewer revisits and shorter stays in a given semantic cluster, suggesting a more transient trajectory across the semantic landscape.

Human perceivers are very sensitive to which way others are facing, with head and eye gaze cues capturing attention (when directed at us), orienting attention (when directed elsewhere), and even influencing downstream judgments about others' social traits. But what causes us to see a shape as directed in the first place? Does the perception of a shape's facing direction depend mainly on its internal structure — or might it also be influenced by spatial context? In Experiment 1, observers briefly viewed a randomly oriented oval, and afterward used a circular slider to report which way they saw it as facing. A dot was always drawn near the oval — aligned with either its long or short symmetry axis. Observers were biased to see the oval as facing toward the dot, but this effect was much stronger when the dot was aligned with the oval's long (vs. short) symmetry axis, indicating that external elements interact with a shape's internal structure to determine its perceived facing direction. How automatic is this association between long-axis alignment and ‘towardness'? In Experiment 2, participants saw the same displays, and now made speeded keypresses to indicate whether the oval's long or short axis was aligned with the dot. In one block of trials, they pressed an anterior (further forward) key to report long-axis alignment, and a posterior (further back) key to report short-axis alignment. In another block, they responded with opposite key-mappings. Participants responded faster in the block where an anterior key was paired with long-axis alignment and a posterior key with short-axis alignment, suggesting an automatic bias to see long-axis alignment as facing towards. We conclude that the perception of facing direction is driven by the interaction of internal structure and external context, in a way which indicates the particular salience of long symmetry axes.

The ability to predict and understand other people's actions is critical for real-world social behavior. Here we hypothesized that representations of social roles (e.g., cashier, mechanic, doctor) enable people to build rapid expectations about what others know and how they might act. Using a self-paced reading paradigm and a variety of everyday roles, we show that the mere mention of a role (e.g., “mechanic”) supports real time expectations about what the person will do (e.g., in the mechanic case, take your car keys but not your cellphone) and the knowledge they might possess (e.g., in the mechanic case, having private information about your car). Moreover, people reported more surprisal when the events deviated from role expectations, and they were more likely to misremember what happened. Our results suggest that roles are a powerful route for social understanding that has been previously understudied in social cognition.

Non-invasive vagus nerve stimulation has proven effective in modulating parasympathetic autonomic nervous system activity and various cognitive functions.This study investigated the effects of transauricular vagus nerve stimulation (taVNS) on body ownership and interoception in healthy subjects using a within-subjects experimental design (active taVNS/sham).The rubber hand illusion (RHI) and the Heartbeat Counting Task (HCT) were employed.Cardiac activity was recorded throughout the procedure to measure physiological indices of heart rate (HR) and heart rate variability (HRV).Ownership for the fake hand was observed in both active and sham stimulation, as indicated by drift and scores on illusion-relevant items (Q1-Q3).HR and HRV showed no variations between synchronous/asynchronous RHI or between stimulation conditions. Active taVNS resulted in decreased interoceptive meta-awareness.Individuals with lower interoceptive abilities exhibited heightened susceptibility to RHI during active taVNS, possibly due to perturbation of interoceptive signals and increased reliance on exteroceptive signals in constructing body representation.

Children learn nouns more readily than verbs in early development. Research on candidate explanations for this noun advantage has suggested that, while noun meanings can be easily gleaned from their observational contexts, verb meanings require access to their syntactic constructions, which remain inaccessible until later in development (Gentner, 2006; McDonough et al., 2011; Piccin & Waxman, 2007). This study asks whether previous demonstrations of tenuous verb learning from their observational contexts are partly due to the assessment method. In an adapted version of the Human Simulation Paradigm (HSP; Gillette et al., 1999), we assessed verb learning using multiple tasks. When verb learning was assessed via a free-response task, learning was minimal, replicating the challenge of learning precise verb meanings via observational contexts. The findings from both a categorization and semantic similarity task, however, suggest that learners do acquire partial knowledge of both action and mental verbs via their observational contexts.

The theory of language prediction posits a competitive preactivation of semantic (meaning) and form (sound/grapheme) information, aiding in the selection of the most likely lexical candidate. Hypothetically, multiple semantic and form cohorts are preactivated before the actual lexical candidate is activated. This study explores this by examining young adults reading constrained sentences (discretely), with simultaneous electroencephalographic recording. Representational similarity analysis was conducted to assess word-specific, semantic-related, and form-related pair of sentences (focusing on the word preceding the expected word). To examine the temporality, cluster permutation and divergence point analyses were performed. The results indicated a semantic coactivation effect occurring before the phonological one and the recovery of the specific words. However, despite the phonological information being recovered before the word specific information, there were no significant differences in temporality. These findings indicate a semantic coactivation process for meaning selection during prediction, with form coactivation dependent on the expected word's selection.

The CONTROL IS UP metaphor is an embodied cognitive mechanism that helps western speakers reason about power relations in terms of vertical spatial organization, This paper explores its multimodal elaboration in visual manifestations of pyramidal structured arrangements and kinetic practices such as genuflexion in medieval miniatures in order to (i) demonstrate the role of multimodal representations of metaphors in reasoning, (ii) unveil power relationships between different societal strata.34 miniatures (12th ct. Liber feudorum maior) were analised with the “multimodal genuflexion test” to describe and measure (CAD-software) distances between kinetic elements (facial/manual gestures, postures).Results indicate that (i) power relations are not just vertically represented (kneeling & bowing), physical distance between characters and position of hands are crucial, (ii) there is a significant positive correlation [U = 61,119; p = 0.003] between the power figure and the degree of “body bending” (bowing, kneeling) and hand distance.

Cognitive scientists attribute representations to complex systems in order to explain their behavior. The shocking facility with which Large Language Models (LLMs) perform difficult linguistic and non-linguistic tasks has generated an increasing amount of speculation concerning what sorts of internal representations might underlie this behavior (whether personal, sub-personal, and of which kinds) and what properties such representations might have (for instance, whether they are grounded). This paper aims to elaborate and defend a conservative explanatory methodology, based on analyses of particular LLM behaviors, according to which attribution of sub-personal representations is key to explaining model performance which is robust, systematic, and flexible, especially in zero-shot settings, and that behavioral benchmarking alone is insufficient to resolve questions about representation due to the mutual underdetermination of performance and competence. The resulting view should help frame future explanations of LLM behavior, and provide an empirically grounded alternative to mere a priori speculation.

Sequential information is encoded through various systems, among which, chunking, rule recognition and nested tree structures. However, the computational and neural mechanisms connecting these systems remain largely unknown. Dehaene et al. (2022) propose that humans possess internal languages governed by symbolic rules, coined Language of Thought (LoT). Based on this assumption we developed the Language of Thought (LoT) algorithm, which processes sequences and produces descriptions as minimal programs. In an online experiment, participants reproduced spatial sequences. Structured sequences, defined by temporal regularities, were notably better reproduced than controls for temporal structure. Participants demonstrated the ability to compress structured sequences in working memory. Response times and performance suggested chunking around a repetition rule. Further analysis, suggested hierarchical organization of those chunks, following a syntactic rule - recursive repetition. LoT-complexity, equal to minimal description length (MDL) of the sequence in our LoT, outperformed other information theory models, aligning best with the data.

Learning a function from input-output pairs often follows exemplar or rule learning. Speed of exemplar learning and generality of rule learning were suggested to be promoted by visually aligning familiar and unfamiliar input-output exemplars. To test this, undergraduates (n=47) were randomly assigned to Full-, Partial- and No-Alignment groups. On each trial, students estimated fractions on number lines, and functions used to generate estimates were examined on 9 trial blocks. On pretest, estimates were (incorrectly) a linear function of denominators alone (Full 50%; Partial 50%; No 40%); on post-test, estimates were (correctly) a linear function of the whole fraction (88%; 44%; 47%). Virtually all change in the Full group occurred after training just two exemplars (75%). Also, regression to the denominator-only function differed across groups (0%; 38%; 33%). Finally, Full-Alignment group generalized to untrained problems more broadly than other groups. Findings demonstrated efficiency of visual alignment in function learning.

Humans are born with an intuitive representation of the physics world. How accurate is intuitive physics? Researchers from education focus on the failures, students' errors and misconceptions while cognitive psychologists argue humans anticipate and manipulate physical environments in ways betraying veridical knowledge of classical mechanics. One solution is to hypothesize there are distinct systems of “cognitive physics” with different limitations and deployment in the tasks favored by the two literatures. The goal of current study is to gather evidence from psychometric studies by estimating how many distinct factors explain performance on intuitive physics assessments. We build an augmented concept inventory including several previously-validated concept inventories, around 120 items. The pilot study indicated that participants recruited online from Prolific displayed expected accuracy on the tasks. We are now collecting around 1,000 participants and applying multidimensional item response theory (MIRT) analyzes to identify the latent structure.

Humans are often involved in complex social relationships, where they exhibit biased behavior when they process information from neighbors (e.g., irrational DeGroot learning) and cognitive biases on perceiving social network structures (e.g., egocentric biases, network centrality, etc.). But little is known about the cognitive reason behind. Here we purpose a unified computational framework (reduced representation model, RRM) to deal with the problems, which assumes people represent an optimal reduced network based on the trade-off of utility and cognitive cost for the representation, and make rational inference on it, where DeGroot-like behavior emerges. We did simulations to show RRM can provide an underlying explanation for DeGroot model and human perceptual biases, and tested model predictions in previous dataset (n=209), lab experiment (n=248) and field data. Our work provides an optimal way to depict social network representation when considering human cognitive limitations, and may help understand widespread human biases in social environments.

Previous work has found that when adults are asked for "the first thing that comes to mind", they will provide something that falls between the descriptive average and the prescriptive ideal. In two experiments, we tested whether children would also be influenced by prescriptive information in their first-to-mind judgments, but also when they were asked to predict a randomly sampled item. In Experiment 1, providing information about whether being longer or shorter made a fictional tool better or worse led children to provide judgments that were biased toward the prescriptively 'best' tool, regardless of what they were asked for, while adults ignored prescriptive information when asked for a random sample. Experiment 2 replicated this result but further showed that the effect was specifically driven by information about which objects were prescriptively good, and did not also arise when the only salient information was about which objects were prescriptively bad.

Random generation studies have shown that people struggle to be unpredictable – they slowly and effortfully produce autocorrelated sequences instead. However, true random processes (such as radioactive decay) are also not instantaneous. In this project we explore how long it takes people in a random generation task to be random. We do so in two experiments asking people to draw samples from naturalistic domains (lifespans and heights), manipulating either the rate of generation or the requirement to be random (within participants). Irrespective of pace or instructions, we find that people can produce a random sample every four to five seconds. Additionally, the time a person needs to produce random samples is consistent across conditions, but varies widely between people. Following previous literature, we model random generation performance as an autocorrelated sampling algorithm, giving a process level account of how people do these tasks and why they need time to be random.

Human cognition is highly flexible– even when posed with novel questions and situations, we are able to manipulate our existing knowledge to draw reasonable conclusions. All human cognition requires flexibility, yet we lack a well-justified computational explanation for how humans might learn and manipulate conceptual knowledge in a way that allows for cognitive flexibility. Here, we develop and test a neural network model of how humans learn and use concept representations. The core of this model frames concepts as latent vector representations that are learned through observations across multiple context domains. The architecture we propose gives rise to a natural mechanism for generalization of conceptual knowledge between familiar domains. This work integrates findings and methods across cognitive science, neuroscience, and machine learning, and holds promise to advance the understanding of conceptual representations within each of these fields.

Round numbers are often interpreted approximately (Krifka, 2002), with "pragmatic halos" (Lasersohn, 1999) that encompass multiple permissible values. For example, stating "there were 200 people at the meeting" would be acceptable even if the exact count were 197 or 204. In line with the idea that larger numbers have more approximate representations (e.g., Cheyette & Piantadosi, 2020), we demonstrate that rounding and pragmatic halos are magnitude-dependent. First, an analysis of every single number in two large corpora (COCA, BNC) shows that indicators of rounding predict frequency (cf. Woodin et al., 2023), but crucially in interaction with magnitude, with round numbers over-represented for larger magnitudes. Second, we show that jigsaw puzzles often systematically deviate from what is advertised on the box in a way that depends on magnitude, e.g., a 1,000-piece puzzle may contain 1,024 pieces, whereas a 50-piece puzzle is more likely to contain the stated value exactly.

The Society for Social Studies of Science (4S) annually awards the John Desmond Bernal Prize to one or more individuals who have made distinguished contributions to the field of Science and Technology Studies. Past winners have included founders of the field, along with outstanding scholars who have devoted their careers to the understanding of the social dimensions of science and technology. This article is the revised text of Joan H. Fujimura's 2023 Bernal Lecture. [ABSTRACT FROM AUTHOR]

Candidate selection within parties is a key stage in the political process and provides an important frame for the degree of representation of social groups in parliaments. We seek to develop a better understanding of the effect of intersectionality on candidate selection processes. We do so by examining the effect of candidates' key socio-demographic characteristics, like their gender and ethnic background, on their chances of getting nominated by their party. We argue that features of the ideological background of the respective nominating party matter for the chances that women and aspirants with an ethnic background win the nomination as their party's district candidate. We make use of novel data from the 2021 German federal election that provides detailed information on the candidate selection processes of all major parties in the 299 election districts. By doing so, we apply existing theoretical expectations to an untested case and find that female competitors and aspirants with an ethnic background face difficulties being nominated, in particular in the case of parties with rather traditionalist societal policy positions. We also find that intersectionality matters: female aspirants of ethnic minorities are even less likely to be nominated by ideologically traditional parties. [ABSTRACT FROM AUTHOR]

Research has shown that deliberative mini-publics have the potential to mitigate some of the deficiencies of contemporary representative democracy. To this end, mini-publics are becoming increasingly connected to other institutions of the democratic system so that their outputs can be transmitted to policy-makers. This article argues that the literature has underappreciated that mini-publics' outputs are not simply transmitted from one institution to another. Instead, the transmission of outputs from mini-publics to other sites in the democratic system is always mediated by representation, which necessarily transforms outputs to some extent. The article identifies the following three main reasons for this: mini-publics' representatives must interpret, perform, and negotiate outputs. Drawing on a least-likely case-study of Agora's representation of the Brussels Citizens' Assembly, it illustrates how these issues are pervasive even when representatives are primarily interested in the literal and faithful transmission of mini-publics' outputs. The findings highlight how representatives struggle with representing mini-publics' outputs. [ABSTRACT FROM AUTHOR]

In this article, we explore the intricacies of veteran care and show how care practices come to incorporate veterans' 'self‐performances' to raise political attention and funding for future rehabilitation activities. By bringing into dialogue theories of care and theories of performance and representation, we illustrate how a seemingly classic form of care—veteran rehabilitation—takes the form of representative performance. Drawing on ethnographic fieldwork and interviews with the Danish Invictus Games team, we demonstrate how politics, research and TV documentaries are integrated into veteran care practices. Through this integration, mentally wounded veterans, while performing 'themselves' for shifting audiences with shifting agendas, come to assume the roles of both caregivers and care receivers. Crucially, we highlight that wounded veterans, while undertaking their personal rehabilitation journey, are curated into and (un)willingly positioned as representatives of others. By showing how caring for wounded veterans goes hand in hand with caring for fictive, future wounded veterans and for political, research and media agendas, this article offers new ways of thinking of and with care. [ABSTRACT FROM AUTHOR]