Search

Compositionality has been argued to a necessary component of interpreting language, yet there appear to be many linguistic phenomena that do not overtly exhibit semantic compositional behavior. One of the challenges involves the phenomena of contextual modulations referred to collectively as semantic coercion or logical metonymy. In this paper, we present a computational model that provides the “compositional flexibility” in the interpretation of a verb with its arguments, for such coercive contexts in English. Specifically, we argue that such constructions typically have surface structural correlates in the form of dense paraphrases, and that these forms can be used to model the masked content in the coerced compositional context. We present preliminary results using a transformer architecture on a masked completion task. Our results show that modeling logical metonymy is a challenging task but can be substantially improved by fine-tuning through dense paraphrasing.

Human communication usually exhibits two fundamental and essential characteristics under environmental pressure, efficiency, i.e., using less communication frequency to achieve comparable performances in cooperation; and robustness, i.e., maintaining a relative performance when communicating in a complicated environment. Since a critical goal of designing artificial agents is making them human-like in many scenarios. How artificial agents could learn a human-like communication mechanisms in terms of efficiency and robustness is a long-existing problem which has not yet been solved. Reinforcement Learning, due to its trail-and-error paradigm, provides a promising framework in solving the above research problem. With reinforcement learning, this paper develops architectures that help agents learn an efficient and robust communication, and carries out extensive experiments which uncover that artificial agents are cognitively capable to learn such human-like communication protocol in various environments (tasks).

The opportunistic assimilation hypothesis posits that struggling and failing to solve a problem creates failure indexes, or long-term memory traces of the problem, that creates sensitivity to environment hints that trigger insight experiences. Past laboratory research has cast doubt on the usefulness of such hints during incubation breaks, but laboratory work is limited in its ability to recreate the diversity of stimuli in everyday life the opportunistic assimilation hypothesis requires. The current paper evaluates the insight experiences of over 150 participants who solved an insight math puzzle outside the lab for the presence of incidental hints that aided with problem solving. Across two studies, participants reported that chance hints in the wild had helped them to solve the puzzle and triggered insight moments. This suggests that opportunistic assimilation may play a role in everyday insight experiences and should not be discounted in future research.

It is intuitive to believe that humans take considerations of mental effort into account when making decisions. However, it has proved difficult to differentiate theories of mental effort in the absence of direct measurements of this psychological construct. Existing measurements of mental effort using response times and revealed preferences have low reliability. In this paper, we present a new experimental task - selecting between visuomotor lotteries using eye-tracking for sampling lotteries - that enables direct measurement of mental effort. Unlike response time-based measures, effort measurements in this task are not confounded by actual effort allocation. Unlike revealed preference-based measures, effort measurements in this task are acquired on a natural scale unitized by automatic visual selection processes. We also report results from a simple experiment conducted using this task, which reproduce existing findings of costly effort-aversion, and also demonstrate adaptive adjustment of mental effort.

Recent research on thick terms like ‘rude’ and ‘friendly’ has revealed a polarity effect, according to which the evaluative content of positive thick terms like ‘friendly’ and ‘courageous’ can be more easily cancelled than the evaluative content of negative terms like ‘rude’ and ‘selfish’. In this paper, we study the polarity effect in greater detail. We first demonstrate that the polarity effect is insensitive to manipulations of embeddings (Study 1). Second, we show that the effect occurs not only for thick terms but also for thin terms such as ‘good’ or ‘bad’ (Study 2). We conclude that the polarity effect is indicative of a pervasive asymmetry that holds between positive and negative evaluative terms.

Fast and efficient correction of speech errors is essential to effective communication. Yet, despite several accounts of error detection, no computational account exists to explain how humans repair their speech errors. This paper proposes the first such model. We demonstrate that a simple automatic mechanism can form the basis of most repairs. We then demonstrate that augmenting the model with a conflict-based monitoring-control loop allows it to capture more nuanced findings in human speech error repair data.

Robots are extraordinary, category-defying entities. Machines that move autonomously, store and communicate information, display emotions, and cultivate social relationships pose a challenge to our most basic assumptions about what kinds of things exists in the world and how we should reason about them. As such, studies of lay people’s beliefs about robots offer new insights into the ordinary functioning of folk ontologies. In this paper, I propose that there are two ontological questions that human reasoners must grapple with in making sense of robots, or any other entity: Which kind of thing is it? and Which causal forces act on it? Each question highlights a distinct way in which robots are extraordinary—albeit, not exceptional—entities for the human cognitive system. A meditation on the dynamic interplay between these two ontological questions provides a new theoretical framework for understanding conceptual change at both the individual and the cultural-historical level.

Children may participate in the design of digital technologies as users, testers, informants, and design partners (Druin, 2002). Yet, more often than not, teachers, researchers, and engineers are design partners, while children remain as mere users and testers (Krumm, Means, & Bienkowski, 2018). This paper explores how children are designers of digital artifacts and learning analytics on an online platform called Knowledge Forum (Scardamalia & Bereiter, 2021). Eight children participated in a group interview to reflect on their online notes and drawings from grade 1 to grade 6, using the analytic tools to mine developmental patterns in their online discourse (e.g., vocabulary growth, diversity and complexity of ideas). Implications of this work are discussed by extending Druin’s framework to consider children’s roles in simultaneously improving usability, augmenting the impact of technology on their learning, and the advancement of learning theories within the context of hybrid learning in a post-pandemic world.

The question of why human language relies so heavily on vague terms has received a great deal of attention from philosophers, linguists, and more recently cognitive scientists, yet much less is known about their effect on other aspects of language use. In this paper, we propose a model for the interaction between vagueness and implicatures, an important pragmatic phenomenon, incorporating recent work in the RSA framework and insights from the philosophical literature on vagueness. We show that the model offers a good fit of data from earlier studies, and discuss the scope of the model more broadly.

Virtual Reality (VR) technology has been widely used in researching situations that require high ecological validity but are difficult to copy. In this paper, we compare the effect of 360-degree videos in VR head-mounted display and 2D computer screen on evoking four types of emotional states, and innovatively assess the effect of emotion evoking using the Go/No-go attention paradigm. Based on the eye movement data collected from a considerable number of participants (N=48), the result reaches marginally significance in the effect of immersion level of evoking (p=0.075), emotion’s level of arousal (p=0.019), and the interaction between them (p=0.037) under the condition of positive evoking. We find that immersive device can better evoke positive emotions with high arousal, while non-immersive device cannot manipulate the intensity of evoked emotions. Our study empirically demonstrate that high-arousal positive emotion evoking is more effective in VR than on a 2D monitor based on computational affection.

There has been a small but now growing interest in studying decision making in real-world contexts where part of the problem faced by decision makers is to generate candidate options they will actually decide between. While some of this work has employed large decision spaces where options are discrete and valuation is computationally tractable (e.g., chess), very little work has focused on genuinely open-ended decision contexts that more closely mirror mundane real-world decisions. This paper leverages large language models to investigate how people generate options when facing genuinely open-ended problems. Across three experiments, we apply semantic similarity and sentiment analyses to the options that participants sequentially generate for real-world decision problems. We find that the first options generated tend to be sampled from a relatively local region of semantic space and are typically of high value. As additional options are generated, they become increasingly dissimilar and are of lower value. These patterns held both at the level of individual option generation trajectories within a given participant and at the level of individual differences across participants.

A central component of the predictive coding theoretical framework concerns the comparison between predictions and sensory decoding. In the probabilistic setting, this takes the form of assessing the similarity or distance between probability distributions. However, such similarity or distance measures are not associated with explicit probabilistic models, making their assumptions implicit. In this paper, we explore an original variation on probabilistic coherence variables; we define a probabilistic component, that we call a "Bayesian comparator", that mathematically yields a particular similarity measure. A geometrical analogy suggests two variants of this measure. We apply these similarity measures to simulate the comparison of known, predicted patterns to patterns from sensory decoding, first in a simple, illustrative model, and second, in a previous model of visual word recognition. Experimental results suggest that the variant that is scaled by the norms of both predicted and perceived probability distributions yields better robustness and more desirable dynamics.

When people think about the features of common objects, like scissors, they often spontaneously recall a central feature: scissors cut things. They tend not to recall other features of scissors, e.g., that they have handles. The present paper posits a novel explanation for the behavior: the features people recall first and most often reflect semantic generalizations of kinds. A recent taxonomy of such generalizations suggests that people represent privileged links between kinds and their features known as principled connections (Prasada et al., 2013). Several tests diagnose principled connections: for instance, principled connections reflect norms, so one way to diagnose the presence of a principled connection is to test the acceptability of sentences of the form all normal Xs have feature Y, as in all normal cars have four wheels. We tested whether participants accept generalizations about the normality of features produced in a semantic feature production task (Experiments 1 and 2) as well as self-referential generalizations (Experiment 3). The experiments provided participants with generalizations about features listed first and most often as well as features that people list less frequently. They found that people readily accepted generalizations that diagnose the presence of principled connections. The results corroborate the view that principled connections help people recall the features of conceptual categories.

This paper proposes new empirical classifiers for evaluative concepts, including thin concepts like 'good' or 'bad' and thick concepts such as 'honest' or 'disgusting', based on quantitative corpus linguistics. Prior work in experimental philosophy has shown that sentiment analysis can be used to track differences between concept classes. Building on this, Task 1 investigates whether the relationship between sentiment and evaluativeness is parabolic rather than linear. Task 2 extends this question to the differences between evaluative and non-evaluative concept classes. The results of both Tasks show that the linear and the parabolic logistic regression classifiers perform equally well. Interestingly, this study also finds that adjectives attributed to animate entities (e.g. "generous customer") generally have a higher probability to be evaluative concepts than those attributed to inanimate entities (e.g."dry soil").

The present paper reports an experiment (N=254) testing two views of how reasoners learn and generalize potentially complex causal knowledge. Previous work has focused on reasoners’ ability to learn rules describing how pre-defined candidate causes combine, potentially interactively, to produce an outcome in a domain. This empirical-function learning view predicts that reasoners would generalize an acquired combination rule based on similarity to stimuli they experienced in the domain. An alternative causal-invariance view goes beyond empirical learning: it allows for the possibility that one’s current representation may not yield useable (i.e., invariant) causal knowledge –– knowledge that holds true when applied. Accordingly, because useable causal knowledge is the evident aspiration of causal induction, this view posits that deviation from causal invariance is a criterion for knowledge revision. The criterion shapes the empirical functions learned and retained. A discriminating test is whether reasoners would re-represent interacting causes as a whole cause that does not interact with other causes, even when in their relevant experience all (pre-defined) causes in the domain interact. Our results favor the causal-invariance view.

The purpose of this paper is to present a novel conceptualisation of an intuitive, primitive form of trust termed proto-trust. This concept is proposed in order to account for the many different senses, types and domains in which trust has traditionally been defined and theorised. A brief review of the literature on affective and intuitive trust is presented, informing the definition and formalisation of proto-trust. Following this, a preliminary empirical investigation of proto-trust is described, where intuitive trust assessments are compared to analytical trust decisions, under various attribution prompts. Results showed effects of attribution prompts on changes to trust assessments from intuitive to deliberative decisions. In addition, qualitative data are presented for the various reasons participants gave for their trust decisions. One of these reasons (emotional reaction) was found to affect the degree of difference between intuitive and deliberative trust assessments.

When socially evaluating a speaker, listeners partially rely on context-dependent expectations, giving greater social penalties for using a marked form in a less expected context. The nature of listeners’ expectations can be based on the context in which a form is produced in the current utterance (local use), as well as cumulative information about the context in which a form tends to be produced (typical use). This paper asks which of these kinds of expectations about the English sociolinguistic variable (ING), as in talking vs. talkin, is most relevant to listeners in making social judgments. Results indicate that (ING) words’ typical grammatical functions (as a noun vs. a verb) contributed to social judgments, while the local grammatical use of (ING) words did not, supporting usage-based theories and raising new questions about the cognitive mechanisms that underlie social evaluations based on speech.

In many real-world settings, it is crucially vital for agents to learn to communicate and cooperate. Different cooperation models have been proposed to represent cooperative relations among agents. However, the intensity of the cooperative relation has not received much attention. In particular, how it varied with spatial-temporal information has not been studied deeply. In this paper, we propose a temporal dynamic weighted graph convolution based multi-agent reinforcement learning framework (TWG-Q). We design a weighted graph convolutional network to capture cooperative information among agents. On top of that, a temporal weight learning mechanism is introduced to characterize intensities of cooperations. We design a novel temporal convolutional network in the temporal dimension to extract effective features for the multi-agent reinforcement learning. Extensive experiments show that our method significantly improves the performance of multi-agent reinforcement learning on the public benchmark of micromanagement tasks in StarCraft II.

Dementia is a common and debilitating condition that typically gives rise to increasing language impairment. There is a need to understand the nature of this impairment further so that therapies may be developed, particularly in the case of bilinguals. This paper extends BiLex, an existing computational model of bilingual lexical access, to simulate language decline in dementia. Six lesion types are evaluated for their ability to reproduce the pattern of decline in the semantic variant primary progressive aphasia (svPPA) subtype of dementia. Semantic memory lesions reproduce this pattern of decline best in monolinguals, and further suggest patterns that are likely to be found in longitudinal data from bilingual dementia patients in the future.

This work studies publications in cognitive science and utilizes mathematical techniques to connect the analysis of the papers' content (abstracts) to the context (citation, journals). We apply topic modeling on the abstracts and community detection algorithms on the citation network, and measure content-context discrepancy to find academic communities that study similar topics but do not cite each other or publish in the same venues. These results show a promising, systematic framework to identify opportunities for scientific collaboration in a highly interdisciplinary and diverse field such as cognitive science.

Early language learning relies on statistical regularities that exist across timescales in infants’ lives. Two types of these statistical regularities are the routine activities that make up their day, such as mealtime and play, and the real-time repeated behaviors that make up the moment-by-moment dynamics of those routines. These two types of regularities are different in nature and are embedded at two different temporal scales, which led to divergent research in the literature – those who collect long-form recordings and observations of at-home behavior and those who use eye trackers and micro-level analyses to quantify real-time behavior in laboratories. The goal of present paper is to jointly examine and connect the statistical regularities at these two timescales. Towards this goal, we brought wearable eye trackers to English- and Spanish-speaking families’ homes to record parent and toddler visual attention during daily routines. We transcribed parent speech during object play and mealtime and coded toddler visual attention during naming moments. We found that parents and toddlers jointly interacted with the unique vocabularies of the two activities. Although naming and attention were more coordinated during object play, mealtime still afforded opportunities for high-quality naming moments. Our results lay the building blocks for connecting these two lines of research and demonstrate the feasibility of at-home data collection with eye trackers.

When deciding whether and how to punish, people consider not only the potential direct consequences, but also, how their choice will affect observers’ judgements about the values and motives underlying the choice. We formalize the decision to punish as a rational communicative social action (RCSA). The model generates rational decisions to punish, incorporating anticipated observers’ judgements obtained from a recursive model of inference using an intuitive theory of mind. Using this model, we synthesize patterns of human punishment from recently published papers. RCSA thus offers a formal model of the cognitive process that humans use to balance preferences for how they are perceived, with other goals for punishing.

Elucidating the mechanisms that underlie clustering and switching behavior is essential to understanding semantic memory search and retrieval. Hills, Jones, and Todd (2012) proposed a model of semantic foraging based on the observation that statistical signatures in memory search resemble optimal foraging in animal behavior. However, the original model was postdictive in explaining when a switch would occur, as opposed to predictive, and was agnostic as to the cues used by humans to make a decision to switch from local to global information. In this paper, we proposed a switching mechanism, \textit{Semantic Scent}, as a predictive model underlying such behavior. Semantic Scent extends optimal foraging theory, reproducing the same switch behavior observed animal foraging behavior in memory search. We evaluated Semantic Scent against competing models including \textit{Random Walk} and \textit{Fixed Count} to determine its effectiveness in classifying switches made in fluency tasks. A quantitative model comparison between the switch models demonstrated Semantic Scent's superior performance in fitting human data. These results provide further evidence of the importance of optimal foraging theory to semantic memory search.

Causal learning is shaped by people’s prior beliefs, including their expectations. In this paper, we specifically examine expectations of determinism: do they vary with perceptual features of physical causal events, and how do they influence subsequent causal learning from data? We show that perceptual features lead adults to different expectations of determinism for different causes of launching (Exps. 1A & 1B). Those expectations lead to significant differences in responses to causal “failures”; that is, we show a difference in violation-of-expectation effect after a failed launch (Exp. 2). Actual data can reduce or eliminate the impact of these expectations, but they do not override the effect of perceptual features (Exp. 3). Overall, spatiotemporal contiguity cues and expectation of determinism have similar effects on causal learning outcomes, but neither is fully reducible to the other.

Given considerable advancements in automated driving systems, the day when autonomous vehicles will be regularly present in our everyday life is impending. It is, therefore, very significant to put emphasis on the effect that giving up autonomy might have on an individual. We take into consideration an experimental data set regarding participants' reported discomfort levels to tackle the following questions: How can we represent a discomfort measurement in a meaningful way? Using this representation, can future discomfort reactions be predicted? We identify key features, identify baseline models, and develop a new approach based on the k-nearest neighbor model to considerably improve the prediction of individual user's discomfort measurements. A discussion of limits and potentials concludes the paper.

The Shapley value is one of the most important normative division schemes in cooperative game theory, satisfying basic axioms. However, some allocation according to the Shapley value may seem unfair to humans. In this paper, we develop an automatic method that generates intuitive explanations for a Shapley-based payoff allocation, which utilizes the basic axioms. Given any coalitional game, our method decomposes it to sub-games, for which it is easy to generate verbal explanations, and shows that the given game is composed of the sub-games. Since the payoff allocation for each sub-game is perceived as fair, the Shapley-based payoff allocation for the given game should seem fair as well. We run an experiment with 630 human participants and show that when applying our method, humans perceive the Shapley-based payoff allocation as more fair than the Shapley-based payoff allocation without any explanation or with explanations generated by other methods.

Cognitive theories for reasoning are about understanding how humans come to conclusions from a set of premises. Starting from hypothetical thoughts, we are interested which are the implications behind basic everyday language and how do we reason with them. A widely studied topic is whether cognitive theories can account for typical reasoning tasks and be confirmed by own empirical experiments. This paper takes a different view and we do not propose a theory, but instead take findings from the literature and show how these, formalized as cognitive principles within a logical framework, can establish a quantitative notion of reasoning, which we call plausibility. For this purpose, we employ techniques from non-monotonic reasoning and computer science, namely, a solving paradigm called answer set programming (ASP). Finally, we can fruitfully use plausibility reasoning in ASP to test the effects of an existing experiment and explain different majority responses.

Many models in Cognitive Science require data to calibrate parameters. Some modelers calibrate their models’ parameters for each individual in a data set, and others work at the aggregate level. Generally, the accuracy of a model is judged by the degree to which human data are replicated, and the model parameters are interpreted accordingly. It is not too surprising that models that are developed for a particular task and fit to each individual’s data in such a task replicate the human data well. The question is, however, whether those models can make predictions in the absence of human data. In this paper, we present a theory-driven model of a well-known sequential decision task (the Balloon Analog Risk Task, BART) which is able to make predictions in the absence of human data. The cognitive model is grounded on the processes and mechanisms of Instance-Based Learning (IBL) Theory of experiential choice. We demonstrate the simulation predictions from an IBL model and those of a well-known model of the BART, which depends on the fits to human data. We further show that when making predictions without data, the IBL model provides predictions that are both theoretically founded and accurate, while the Two-Parameter model performs much worse than when fit to data. We conclude with a discussion of the benefits of making theory-based predictions in the absence of human data for our community.

High-level cognition relies on the ability to determine what the relevant possibilities are in the face of incomplete information. For example, causal judgments require reasoning about the relevant counterfactual possibilities in the context in which the actual events occurred. Similarly, to judge that someone is morally responsible for a given action requires assessing what other actions were available to the agent in the context they acted within. Defining the set of relevant possibilities in a context---what we call the contextual 'modal space'---is computationally intractable both because such contexts are not well defined and because there are indefinitely many possibilities that one could consider. Interestingly though, high-level judgments that require reasoning over such possibilities are often made quickly and effortlessly, suggesting that they may recruit a heuristic or implicit representation of the contextually relevant modal space. In this paper we (1) introduce a sampling approach for empirically describing contextual modal spaces, (2) find that modal spaces are shaped by statistical and prescriptive normality, (3) demonstrate that the contextual modal spaces play a domain general role in high-level cognition in the sense that the same modal space is recruited across different domains of reasoning, and finally (4) we provide evidence that such judgments are made on the basis of heuristic representations of the contextual modal space rather than on-line sampling of relevant alternatives.

A challenge in learning phonological grammars is learning how phonological processes interact. It has been argued that some process interactions are easier to learn than others. One basis for this argument is asymmetries observed in experimental settings: artificial languages generated from certain process interactions are more likely to be successfully reproduced by participants than others. In this paper, we argue that asymmetries in production do not necessarily provide direct support that some phonological interactions are easier to learn. Rather, we show that these asymmetries can instead emerge due to differences in the number of consistent or nearly-consistent grammars each pattern has. We present a noisy channel model of morpho-phonological learning and apply it to a recent behavioral study examining the learnability of phonological process interactions. We find that, due to the relative difference in the number of grammars that can exactly match or nearly match the observed data, the model achieves the same qualitative results as those observed in experimental settings.

Children are navigating a world with massive amounts of auditory input, sometimes relevant while other times purely noise, and must somehow make sense of it all. The early auditory environment is critical for speech perception and recognition, auditory discrimination, and word learning, all of which support language outcomes. What strategies do children use to learn in noisy environments? One potential strategy is environmental selection, which allows children to seek environments that align with particular goals. In the current paper, we examined whether children and adults make decisions about their environments by integrating auditory information and goal-states. While 3- and 4-year olds struggle with discriminating the level of noise in noisy speech streams (and likely do not use this information for environmental selection), 5-year-old children and adults can. Further, we show initial evidence that they can use this information to reason about acoustic environments that are consistent with specific goals.

Smartphones are now the most widely used devices in the world, and their usage monitoring applications have become a general interest topic. However, few experimental studies investigate the reflexive effects of this monitoring on users. To address this point, this paper presents a longitudinal experiment on the effects of monitoring on various variables (e.g. screen time, types of uses). Objective and subjective data from 60 participants, divided into treatment and control groups, were collected over a 3 weeks period. Both groups had to estimate their daily usages, but the treatment group subsequently had access to their real data. Results have shown a normalizing influence of monitoring on smartphone usage, by improving estimation of screen time, reducing time spent on some underestimated applications and increasing use of others overestimated applications. This research paves the way for public policies promoting mastery of its own technological uses and responsible digital usage.

This paper explores the value of skepticism towards the Approximate Number System (ANS). I sketch some of the main arguments levied against ANS-based interpretations of numerical cognition data and argue that there are empirical and conceptual reasons to reject wholesale replacement of the ANS with an Analog Magnitude System (AMS). To simplify the discussion, I focus for the most part on a recent critical review representative of this new wave of revisionist skepticism (Leibovich, T., Katzin, N., Harel, M., & Henik, A., 2017). I start with a brief review of some of the reasons offered to deny that experiments studying our numerical abilities reveal the presence of a system dedicated to representing quantities of discrete objects, before turning briefly to empirical responses to these worries. I then offer a few reflections on why even if the empirical rebuttal were to fail, there are conceptual reasons to doubt that we are only equipped with an AMS. While some of these reasons involve methodological implications of AMS-based theories, other conceptual reasons to doubt AMS skepticism revolve around how ANS-skepticism seems to go against the history of the relation between the continuous and the discrete, and how one cannot be derived from the other. I then end with a potential reply to my worries involving an appeal to the Object-File System (OFS) as a source of discrete content in our numerical abilities and find it wanting.

Being able to learn from small amounts of data is a key characteristic of human intelligence, but exactly how small? In this paper, we introduce a novel experimental paradigm that allows us to examine classification in an extremely data-scarce setting, asking whether humans can learn more categories than they have exemplars (i.e., can humans do "less-than-one shot'' learning?). An experiment conducted using this paradigm reveals that people are capable of learning in such settings, and provides several insights into underlying mechanisms. First, people can accurately infer and represent high-dimensional feature spaces from very little data. Second, having inferred the relevant spaces, people use a form of prototype-based categorization (as opposed to exemplar-based) to make categorical inferences. Finally, systematic, machine-learnable patterns in responses indicate that people may have efficient inductive biases for dealing with this class of data-scarce problems.

Psychometric testing is a valuable educational tool for the assessment and monitoring of students’ abilities in core subjects. However, the manual development of these tests is a tedious process requiring test specialists to produce and curate large volumes of high-quality items. In this paper, we consider whether automating test item generation with modern machine learning methods is a feasible solution for obtaining strong psychometric test items at low cost in the domain of sentence reading fluency. We assess the ability of the large neural language model GPT-3 to produce items ``few-shot’’--- from a short prompt with only a handful of examples. Our results show that generated items closely resemble standardized test items in terms of their factual ambiguity, content appropriateness, and complexity. Furthermore, after filtering for correct answer-labeling these generated items possess similar latent psychometric properties to standardized test items, even capturing subtle grade-level variation.

The paper investigates whether English native speakers (NSs)’ and non-native speakers (NNSs)’ online processing of prepositional phrases were primed by schematic diagrams and whether the priming effects were modulated by different prepositions and senses. 125 adult NS and NNS participants judged on over and in phrases that encode spatial and extended senses in a semantic priming task. Results showed an inhibitory effect of diagrams on NNSs’ processing but a marginal facilitatory effect on NSs’ processing. In addition, the inhibitory effect of diagrams only applied to the processing of extended senses, but not spatial senses. The current findings indicate that NNSs’ processing of English PPs could be negatively influenced by schematic diagrams particularly when the target phrases are more complex (Boers, 2011), presumably due to NNSs’ more limited cognitive capacity of processing complex second language expressions under time pressure than NSs.

Subway systems in large cities witness high volumes of commuter traffic, with crowded coaches and limited seats. In such scenarios, commuters often carefully position themselves in strategic locations with the aim of maximizing their chances of getting a place to sit. While user behavior in subways around the world have been the focus of multiple studies in the past, these everyday acts of ‘optimal decision making’ is of particular interest to the cognitive scientist. This paper inquires into commuter positioning choices in simulated subway coaches, within the framework of Bayesian probabilistic modelling. Data on preferred standing positions were collected across 20 subjects for 30 co-passenger configurations, through an interactive computer game. A generative model based on a Bayesian network involving three key spatial parameters was constructed, and used for inferring preferred positions conditioned on the specific configurations. The model was able to accurately simulate the quick and intuitive decisions made by the players under constraints of time, and also effectively capture noise in responses across subjects

Speakers' referential expressions often depart from communicative ideals in ways that help illuminate the nature of pragmatic language use. Patterns of overmodification, in which a speaker uses a modifier that is redundant given their communicative goal, have proven especially informative in this regard. It seems likely that these patterns are shaped by the environment a speaker is exposed to in complex ways. Unfortunately, systematically manipulating these factors during human language acquisition is impossible. In this paper, we propose to address this limitation by adopting neural networks (NN) as learning agents. By systematically varying the environments in which these agents are trained, while keeping the NN architecture constant, we show that overmodification is more likely with environmental features that are infrequent or salient. We show that these findings emerge naturally in the context of a probabilistic model of pragmatic communication.

In the debate between single-route and dual-route models of verb inflection, the dissociation between regular and irregular verbs in the non-fluent variety of aphasia has been a key sticking point for the proponents of the single-route model. This paper adopts a state-of-the-art neural model which has previously been used to learn inflectional morphology, and shows that it can also be used to model data from non-fluent aphasia. This challenges the assumption that a dual-route model is necessary to capture apparent dissociations in aphasia data and encourages a reanalysis of the deficits involved in non-fluent aphasia.

Learning to establish joint reference is an important milestone of communicative and linguistic development. Pointing is one of the first entry points into this process, since gestures often precede verbal communication. During early development, as well as later language use, pointing and linguistic utterances interact in many ways, complementing each other. However, little is known about the development of this relationship during development. In this paper, we focus on the development of the co-occurrence of finger pointing and accompanying utterances in two different cultures: Russia and Chintang (Sino-Tibetan, Eastern Nepal). We show that despite the differences in environment, the development of finger pointing and accompanying language use show substantial similarities. Early on, a larger proportion of points is not accompanied by language. As the children's linguistic abilities develop, children first use language to specify what is being pointed at, and later elaborate on some aspect of the referent.

Query-Based Memory (QBM) models are heavily used in machine learning, though their relevance to human cognition is unclear. In this paper, we explore QBM models through both formal exploration and a simulation study to address this question. We found that QBM models are theoretically motivated, as they approximate rational induction with neurally-plausible mechanisms. Additionally, a simple implementation of the model could readily reproduce four benchmark findings in infant statistical learning. These results provide an encouraging starting point for further research using these formal tools to understand cognition across development.

Retrieval practice of information through testing has been shown to improve learning. So has studying examples. In this paper, we address inconsistencies in the literature concerning which of these two approaches is best. We test the hypothesis that learning depends on what is being learned; whereas practice emphasizes memorization, studying examples allows for selectivity of encoding, resulting in different information being learned. Accordingly, we predicted that practice will improve learning in situations that emphasize memorization (such as learning facts or simple associations), whereas studying examples will improve learning in situations where there are multiple pieces of information available and selectivity is necessary (such as when learning skills or procedures). We report evidence from a laboratory study using naturalistic materials showing results consistent with these predictions.

In many real-life situations, we make decisions between a defined set of options, which can be either discrete (as when deciding between going on driving and stopping the car) or continuous (as when stirring the wheel, the possible range of angles goes from $-30$ to $30$ degrees). However, most computational models for decision-making focus on decisions between a discrete set of options. While there are a few sequential sampling models that can explain behavioral patterns (i.e., choices and response times) of decisions in a continuous option space (i.e., the CDM and the SCDM), these models have a few limitations. For example, these models assume no leakage in the evidence accumulation process and no spatial inhibition (i.e., inhibition among different areas of the option space depending on their distance to each other). In this paper, we propose a novel sequential sampling model based on an existing computational model (i.e., the leaky competing accumulator model) for decisions in a continuous option space. Our proposed model includes leakage and spatial inhibition and is thus more biologically plausible.

Empirical studies have demonstrated that when comprehenders are faced with informationally redundant utterances, they may make pragmatic inferences to accommodate the informationally redundant utterance (Kravtchenko & Demberg, 2015. Previous work has also shown that the strength of these inferences depends on prominence of the redundant utterance – if it is stressed prosodically, marked with an exclamation mark, or introduced with a discourse marker such as “Oh yeah”, atypicality inferences are stronger (Kravtchenko & Demberg, 2015; 2022; Ryzhova & Demberg, 2020). The goal of the present paper is to demonstrate how both the atypicality inference and the effect of prominence can be modelled using the rational speech act (RSA) framework. We show that atypicality inferences can be captured by introducing joint reasoning about the habituality of events, following Degen, Tessler, and Goodman (2015); Goodman and Frank (2016). However, we find that joint reasoning models principally cannot account for the effect of differences in utterance prominence. This is because prominence markers do not contribute to the truth-conditional meaning. We then proceed to demonstrate that leveraging a noisy channel model, which has previously been used to model low-level acoustic perception (Bergen & Goodman, 2015), can successfully account for the empirically observed patterns of utterance prominence.

It has been conjectured that verb learning is hard because verb meanings are not readily "packaged'' from the physical world. To provide new empirical evidence on this account, we analyzed egocentric video collected from natural toy-play interaction and focused on the naming events when action verbs were uttered in parent speech. Using the Human Simulation Paradigm, we showed egocentric videos of those naming events to adult observers and asked them to guess the target verb in parent speech. We found that adult observers used many different verbs to describe the same visual event, and only one of them matched with the verb in parent speech. We analyzed mismatched verbs and identified several sources of mismatch, and found that all of the mismatched verbs are relevant to the target verb, but they capture different properties (temporal, semantic, etc) of the visual event. We also found that different naming events for the same verb also differ in terms of the degree of ambiguity. Taken together, the results in the present paper provided new evidence from the child’s view, showing that verb learning is hard not only because multiple possible meanings are embedded in each learning situation, but also because these candidate meanings expand across multiple dimensions of the physical world, overlap with each other, and relate to the target meaning in many different ways.

Integrating visual representations in an interactive learning activity effectively scaffolds performance and learning. However, it is unclear whether and how sustaining or interleaving visual scaffolding helps learners solve problems efficiently and learn from problem solving. We conducted a classroom study with 63 middle-school students in which we tested whether sustaining or interleaving a particular form of visual scaffolding, called anticipatory diagrammatic self-explanation in an Intelligent Tutoring System, helps students’ learning and performance in the domain of early algebra. Sustaining visual scaffolding during problem solving helped students solve problems efficiently with no negative effects on learning. However, in-depth log data analyses suggest that interleaving visual scaffolding allowed students to practice important skills that may help them in later phases of algebra learning. This paper extends scientific understanding that sustaining visual scaffold does not over-scaffold student learning in the early phase of skill acquisition in algebra.

Adaptive scheduling systems aim to estimate the ability of an individual in order to prescribe a personalized training schedule. These adaptive systems are often founded on regularities of human memory such as a learning, forgetting, and the spacing effect. One such model which has been developed to both account for regularities of memory and be used in applied contexts is the Predictive Performance Equation (PPE). One limitation of the PPE is that it is only able to account for and incorporate information about a participant’s accuracy on a task and cannot take into account additional performance measures such as reaction time. To expand the PPE, we propose a simple extension to the model, allowing it to account for both accuracy and reaction time measures. Our paper reports the extension to the PPE as well as a formal model comparison to another model of learning and retention (Pavlik and Anderson, 2005). The results of our model comparison reveal that the extended PPE can both better account and predict an individual’s performance than Pavlik and Anderson (2005) model.

Children start to communicate and use language in social interactions from the very early stages in development. This allows them to experiment with their current linguistic knowledge and receive valuable feedback from their interlocutors. We conducted a large-scale corpus study to examine the quality of positive and negative Communicative Feedback signals that caregivers provide in terms of time-contingent responses and clarification requests. We found evidence for the effect of such feedback in supporting children’s production of intelligible speech. The broad impact of this paper is to highlight how general social feedback mechanisms that govern human communication can also support child language acquisition.

Diversity of social identities can improve the performance of groups through varied cognitive and communicative pathways. Recently, research efforts have focused on identifying when we should expect to see these potential benefits in real-world settings. While most research to date has studied this topic at individual and interpersonal levels, in this paper, we develop an agent-based model to explore how various aspects of homophily, the tendency of individuals to associate with similar others, affects performance at a larger scale. Study 1 examines how two types of homophily---identity-driven and opinion-driven---impact collective performance on a sequential decision-making task via modulating network formation and trust relations. Study 2 considers how the presence of identity-based conformity pressure can affect the findings from the first study. Overall, we find that the effect of homophily on performance is complex, depending on the operative dimensions of similarity, mediating pathways, and the specific outcome of interest. Finally, we discuss the implications of our results for policy interventions aiming to improve group performance.

Politeness is a social linguistic phenomenon. Modeling polite language production and understanding is difficult, as it may contradict conversational maxims and is shaped by extralinguistic social influences, such as the speaker-hearer relationship. This paper extends Yoon et al.'s (2016) Rational Speech Act-based model of politeness by mapping speaker-hearer relationship influences to the utility weights of the model and instantiates it in German. Three online experiments, for empirical analysis and collection of behavioural data for model training and evaluation, are presented. These confirm the influence of the speaker-hearer relations on indirect politeness. Furthermore, two versions of the model are trained and evaluated to find out which part of the model is better suited for the integration of social influences. Overall, both model versions yielded similar results and were able to predict the meaning of polite speech acts.