Proceedings of the Annual Meeting of the Cognitive Science Society

The study of how people make judgments and decisions

sion-making (JDM) has moved apart from its cognitive roots,

research could yield benefits to both from cross-pollination

plore some productive areas of overlap between cognitive

This full-day workshop focuses on Massive Online Experiments (MOEs). MOEs have transformative potential, as they effectively allow researchers to run hundreds of experiments simultaneously (cf. Hartshorne & Germine, 2015; Reinecke & Gajos, 2014). The goal of the workshop is to help a broad cross-section of cognitive scientists begin to incorporate MOEs into their research.

This half day tutorial will introduce participants to breath counting, a recent innovation in mindfulness research methodology that offers the first opportunity to collect a relatively direct behavioral measure of mindfulness (Levinson et al., 2014). This tutorial will review the basics of mindfulness practice before providing attendees with the knowledge, skills, and tools required to implement breath counting procedures in their own research. The widespread adoption of this quantitative measure could positively impact the field by offering a more straightforward and transparent methodology that would improve the overall consistency of mindfulness research across studies and between investigators.

Probability theory is the “logic of science” (Jaynes, 2003) and Bayesian data analysis (BDA) is the glue that brings that logic to data. BDA is a general, flexible alternative to standard statis- tical approaches (e.g., NHST) that provides the scientist with clarity and ease to address their personal scientific questions. Doing BDA in a probabilistic programming language (PPL) af- fords several additional advantages: a compositional approach to writing models, separation of model specification from al- gorithmic implementation (a la lm() in R), and continuity from articulating data analytic models to Bayesian cognitive mod- els. Furthermore, specifying one’s model and data analysis in a PPL allows you to search for “optimal experiments” for free. This tutorial will walk the participant through the basics of BDA to state-of-the-art applications, using an interactive on- line web-book and tools for integrating BDA into their existing workflow.

This symposium integrates findings across studies conducted in both laboratory and classroom contexts to draw attention to the relationships between Executive Functions (EFs) and feelings of anxiety in a context with educational consequences: Mathematics. EFs, the cognitive resources including working memory and inhibitory control that enable attentional control, manipulation of mental representations, and task switching (Miyake et al, 2000), powerfully predict mathematics achievement (Bull & Lee, 2014). Mathematics is also a domain in which anxiety and performance pressure are often heightened, which can result in worry ideation and load to EF resources (Foley et al, 2017; Schmader & Beilock, 2012). However, despite these relationships, mathematics cognition under pressure remains under-considered.

Although we possess intuitions about pedagogy from early in life, adults commonly fail to teach effectively in real-world situations. Why might adults struggle in more complex teaching tasks? Here we develop a simple teaching task where adults fail to teach naïve learners, despite reporting high confidence that they taught effectively. Using a formal model of a rational teacher, we analyze the sources of our adult teachers’ failures. Our model-based analyses reveal that teachers successfully provided high-quality examples, but failed to address hypotheses that naïve learners find plausible. We validate these results in a second experiment, where we find that constraining learners’ hypothesis space increases their performance in the task. Our findings help bridge the gap between children’s teaching proficiency in constrained tasks, and adults’ teaching failures in more naturalistic tasks.

Much research has investigated how children track and reason about accuracy when deciding who to trust. The majority of this work assumes a static link between accuracy and knowledge; that is, children are expected to attribute greater knowledge to accurate agents. But while accuracy often reveals knowledge, the two are not deterministically related. Ignorant agents can be accurate (for example, one could take a lucky guess), and knowledgeable agents can be inaccurate (for example, one could accidentally err). Given this, how do children reason about the relation between knowledge and accuracy? Across three experiments, we show that four- and five-year-olds are sensitive to the distinction between knowledge and accuracy. Specifically, children judge that an agent who accurately predicts an outcome is knowledgeable, but an agent who merely observes and then accurately describes the same outcome is not. Our findings show that when children gauge agents’ knowledge, they do not rely on accuracy alone; they infer knowledge only when an agent is right in the right kind of way

Past research has found that interruptions change the relationship between confidence and accuracy. However, it is unclear how interruptions affect confidence-accuracy calibration. In this study, we used a rule-based procedural task called UNRAVEL and compared confidence-accuracy calibration between interrupted and uninterrupted trials. Results showed that participants were better calibrated in the interruption condition than in the no interruption condition. We interpret this novel effect as a result of changes in the validity of internal cues for confidence between conditions. Specifically, we explore response time as one potential mediating factor.

Mechanistic complexity is an important property that affects how we interact with and learn from artifacts. Previous research finds that children successfully detect complexity contrasts when given information about the functions of simple and complex objects. However, do children spontaneously favor relevant information about an object’s causal mechanisms and functions when trying to determine an object’s complexity? In Study 1, 7–9-year-olds and adults, but not 5–6-year-olds, favored relevant information (e.g., the difficulty in fixing an object) over irrelevant information (e.g., the difficulty in spelling an object’s name) for making determinations of mechanistic complexity. Only in Study 2, in which the relevance contrasts were extreme, did the youngest age group favor relevant over irrelevant information. These results suggest that the ability to detect which object properties imply complexity emerges in the early school years; young children may be misled by features that are not truly diagnostic of mechanistic complexity.

A word’s predictability or surprisal, as determined by cloze probabilities or language models (e.g. Frank, Otten, Galli, & Vigliocco, 2015) is related to processing effort, in that less ex- pected words take more effort to process (e.g. Hale, 2001). A words surprisal, however, may also be influenced by the non- linguistic context, such as visual cues: In the visual world paradigm (VWP), for example, anticipatory eye movements suggest that comprehenders exploit the scene to predict what will be mentioned next (Altmann & Kamide, 1999). How vi- sual context affects word surprisal and processing effort, how- ever, remains unclear. Here, we present evidence that visually- determined probabilistic expectations for a spoken target word predict graded processing effort for that word, in both pupil- lometric (ICA) and ERP (N400) measures. These findings demonstrate that the non-linguistic context can immediately influence both lexical expectations, and surprisal-based pro- cessing effort.

We propose a computer-based testing environment, Reading Skill Test, to measure the effects of various types of systematic reading support systems. We prove its validity, reliability and one-dimensionality using 31,000 subjects. The effects of furigana system on the 5th to 8th grade students are analyzed using this environment. Furigana is a widely used Japanese reading support system that has been believed to be beneficial especially for pupils. Despite our expectation, we have to conclude that furigana failed to improve pupils’ reading significantly, and discuss why it did so.

We not only care about what others think of the world, but also about what others think of us. The ability to understand what others think of one’s competence is especially important for young children, as they are beginning to learn about them- selves and form new relationships with others. Here we ask whether young children can use others’ observations of their own failures and successes to infer others’ beliefs about the self’s competence, and would even forego an opportunity to teach new information in order to demonstrate their compe- tence. In Exp. 1 (3, 4, & 5-year-olds), when a confederate had observed the child initially fail but eventually succeed at op- erating a toy, children chose to teach her a new toy; however, when the confederate had observed the initial failures but not the final success, more children chose to show the familiar toy again to demonstrate their competence. In Exp. 2 (3- & 4- year-olds), we replicate this finding. Even in preschool years, children can reason about what others of their own competence and strategically decide whether to communicate information about the self or the world; these results are discussed in light of prior work on reputation management and Theory of Mind.

Praise is not only rewarding but also informative. It allows us to learn about our skills and competence even when we are uncertain or unable to judge for ourselves. Not all praise is equally meaningful, however: Praise from someone who praises indiscriminately is less informative than from some- one who praises selectively. Here we ask whether young chil- dren infer the informativeness of others’ praise based on the statistical dependence between praise and the quality of work. Exp. 1 shows that adults and 4-5 year-olds were more likely to trust praise from a teacher whose previous praise covaried with the quality of work (i.e., selective praise) than praise from a teacher who indiscriminately praised independent of the qual- ity of work (i.e., overpraise). Exp. 2 addressed the possibility that participants simply prefer a teacher who praises less often. Even for young children, praise is more than something nice. Rather, they can track the informativeness of others’ evalua- tive feedback and use it to learn about the quality of their own work.

Hills, Jones, and Todd (2012) observed that response patterns during the semantic fluency task (e.g., “name all the animals you can in a minute”) display statistical signatures of memory search that mirror optimal foraging in physical space. They proposed a model of memory search based on exploration- exploitation tradeoffs known to produce optimal foraging patterns when animals search for food resources, applied to a spatial model of semantic memory. However, Abbott, Austerweil, and Griffiths (2015) demonstrated that optimal foraging behavior could also naturally emerge from a random walk applied to a network representation of semantic memory, without reliance on a foraging process. Since then, this has been a very active are of debate in the literature, but core confounds have prevented any clear conclusions between the random walk and cue switching model. We control confounds here by using a fixed training corpus and learning model to create both spatial and network representations, and evaluate the ability of the cue switching model and several variants of the random walk model to produce the behavioral characteristics seen in human data. Further, we use BIC to quantitatively compare the models’ ability to fit the human data, an obvious comparison that has never before been undertaken. The results suggest a clear superiority of the Hills et al. cue switching model. The mechanism used to search memory in the fluency task is likely to have been exapted from mechanisms evolved for foraging in spatial environments.

In the course of sentence processing, comprehenders must identify relationships between sentence elements in accor- dance with the sentence’s syntactic structure. However, low- level associative processes, which may yield interpretations in- compatible with global sentence context, have also been sug- gested to be at play in the early moments of processing. In two experiments, we examine the influence of low-level associa- tive cues alongside combinatorial information in sentences of varying complexity. Verb-driven predictions are used as means to explore the use of these information sources in the earliest moments of processing. In addition, we explore effects of lis- tener age on processing, given past claims that older adults’ predictions are more shallow. However, results showed similar patterns across age groups, although we did find clear ways in which associative cues overshadowed combinatorial computa- tions when these cues occurred very close to the verb.

People can consciously think about only a few things at a time. But what determines the kind of things that come to mind, among a potentially infinite set of possibilities? Two experiments explored whether the things that come to mind are sampled from a probability distribution that combines what people think is statistically likely and what they think is prescriptively good. Experiment 1 found that when people are asked about the first quantities that come to mind for everyday behaviors and events (e.g., hours of TV that a person could watch in a day), they think of values that are proportional to, and intermediate between, what they think is average and what they think is ideal. Experiment 2 quantitatively manipulated distributions of times people devoted to engaging in a novel hobby (“flubbing”) and the corresponding distributions of goodness of doing this hobby for various amounts of time. The distribution of values that came to mind resembled the mathematical product of the statistical and prescriptive distributions we presented participants, suggesting that something must be both common and good to enter conscious awareness. These results provide insight into the algorithmic process generating people’s conscious thoughts and invite new questions about the adaptive value of thinking about things that are both common and good.

Interactive intelligent agents use cognitive models to antic- ipate and simulate human behavior, and a fundamental pil- lar of human cognition and interaction is narrative. As a result, agents need to understand human comprehension of various types of narratives. A key component of modeling comprehension is the perception of interestingness of con- stituent actions and events in the narrative. In this paper, we briefly review previous theories of interestingness, drawn from cognitive psychology and narratology. We propose expanded computationally amenable theory of interest which takes into account both cognitive and experiential aspects of perceived interest. To empirically validate the theory, we present a narrative generator for abstract animations inspired by Heider and Simmel’s experiments (Heider & Simmel, 1944). The generated animations are parameterized along the dimensions of our proposed theory. We present the results of a user study with this generative system and report on the effects of visual narrative parameters on perceived interest.

Recent empirical studies have affirmed the fundamental role of attention and memory processes in statistical word learning tasks. These processes interact in complex ways to guide spontaneous looking behaviors of learners as well as determine their overall learning performance. On the modelling side, studies have made it clear that computational models must provide process-based rather than only computational accounts of word learning, because these can connect to the empirically observed behaviors at a moment- to-moment timescale. Thus, here we present a neurally- grounded process model of word learning called WOLVES (Word-Object Learning Via Visual Exploration in Space) that integrates visual dynamics and word-object binding across multiple timescales. WOLVES integrates multiple established dynamic neural field models to allow fine-grained indexing of component processes driving the looking-learning loop. We report simulation results for three empirical cross-situational word learning experiments to validate the model.

Three experiments investigated the role of conceptual abstraction in category learning. We found that people in a low-level mindset over-weighted global features in classifying novel exemplars whereas those in a high-level mindset did not (Experiments 1 and 3). The effect was on the learning process, independent of perceptual response preference (Experiment 3) and occurred despite evidence of perceptual global dominance for all groups during learning (Experiments 2 and 3). We conclude that abstraction can subordinate perceptual salience to the larger goal, integrating discrete encounters into a comprehensive representation of the underlying structure.

Information sharing in competitive environments may seem counterintuitive, yet it is widely observed in humans and other animals. For instance, the open-source software movement has led to new and valuable technologies being released publicly to facilitate broader collaboration and further innovation. What drives this behavior and under which conditions can it be ben- eficial for an individual? Using simulations in both static and dynamic environments, we show that sharing information can lead to individual benefits through the mechanisms of pseudo- reciprocity, whereby shared information leads to by-product benefits for an individual without the need for explicit recipro- cation. Crucially, imitation with a certain level of innovation is required to avoid a tragedy of the commons, while the mecha- nism of a local visibility radius allows for the coordination of self-organizing collectives of agents. When these two mecha- nisms are present, we find robust evidence for the benefits of sharing—even when others do not reciprocate.

The explosion of data generated during human interactions on- line presents an opportunity for cognitive scientists to evaluate their models on popular real-world tasks outside the confines of the laboratory. We demonstrate this approach by evaluating two cognitive models of generalization against two machine learning approaches to recommendation on an online dataset of over 100K human playlist selections. Across two experiments we demonstrate that a model from cognitive science can both be efficiently implemented at scale and can capture generaliza- tion trends in human recommendation judgments which nei- ther machine learning model is capable of replicating. We use these results to illustrate the opportunity internet-scale datasets offer to cognitive scientists, as well as to underscore the impor- tance of using insights from cognitive modeling to supplement the standard predictive-analytic approach taken by many exist- ing machine learning approaches.

A plethora of research over the past two decades has demonstrated that citizens in countries around the world dramatically overestimate the size of minority demographic groups and underestimate the size of majority groups. Researchers have concluded that this misestimation is a result of characteristics of the group being estimated, such as level of threat the group poses and the amount of exposure someone has with to the group. However, explanations of this misestimation have largely ignored theoretical models of perception and measurement, such as those developed in classic psychophysics. This has led to interpretations that are at variance with modern theories of measurement. We present a model which combines an understanding of the nature of human estimations with a conceptualization of uncertainty, which extends to accommodate bias. We apply this model to three large-scale datasets collected by the Ipsos MORI research group. Model fits from our approach suggest that to a considerable degree, the errors people make are due to uncertainty rather than bias. These biases are quite different in character from those that other groups have reported. Many of the present biases, furthermore, are shared widely across different countries.

To explore whether current notions of statistically-based language learning could successfully scale to infants’ linguistic experiences “in the wild”, we implemented a statistical-clustering word-segmentation model (Saffran et al., 1997) and sent its outputs to an implementation of a “frame” based form class tagger (Mintz, 2003) and, separately, to a simple word-order heuristic parser (Gervain et al., 2008). We tested this pipeline model on various input types, ranging from quite idealized (orthographic words) to more naturalistic resyllabified corpora. We ask how these modeled capacities work together when they receive the noisy outputs of upstream word finding processes as input, which more closely resembles the scenario infants face in language acquisition.

People’s cognitive strategies are jointly shaped by function and computational constraints. Resource-rational analysis lever- ages these constraints to derive rational models of people’s cognitive strategies from the assumption that people make rational use of limited cognitive resources. We present a resource-rational analysis of planning and evaluate its predic- tions in a newly developed process tracing paradigm. In Ex- periment 1, we find that a resource-rational planning strategy predicts the process by which people plan more accurately than previous models of planning. Furthermore, in Experiment 2, we find that it also captures how people’s planning strategies adapt to the structure of the environment. In addition, our ap- proach allows us to quantify for the first time how close peo- ple’s planning strategies are to being resource-rational and to characterize in which ways they conform to and deviate from optimal planning.

The brain in conjunction with the body is able to adapt to new environments and perform multiple behaviors through reuse of neural resources and transfer of existing behavioral traits. Al- though mechanisms that underlie this ability are not well un- derstood, they are largely attributed to neuromodulation. In this work, we demonstrate that an agent can be multifunctional using the same sensory and motor systems across behaviors, in the absence of modulatory mechanisms. Further, we lay out the different levels at which neural reuse can occur through a dynamical filtering of the brain-body-environment system’s operation: structural network, autonomous dynamics, and tran- sient dynamics. Notably, transient dynamics reuse could only be explained by studying the brain-body-environment system as a whole and not just the brain. The multifunctional agent we present here demonstrates neural reuse at all three levels.

Shiffrin and Schneider (1977) demonstrated that highly efficient memory- and visual-search performance could be achieved through consistent item-to-response mapping (CM) training. It is theorized that subjects shifted from relying on working memory to learned item-response associations in long-term memory (Logan, 1988). The theory was tested and explored mostly through behavioral experiments and computational modeling. In a recent series of articles involving visual search (e.g. Woodman et al, 2013; Carlisle et al. 2011), Woodman and colleagues found that the contralateral-delay activity (CDA) of human event-related potentials is related to the maintenance of information in visual working memory and that the magnitude of the CDA decreases when target information is stored in long-term memory. We employed the CDA and other neural measures to study the nature of memory retrieval in CM memory search tasks. We observed a significant reduction in the magnitude of the CDA in CM training compared to a control condition in which item-response mappings varied from trial to trial (VM). The results provided converging evidence supporting the classic theoretical interpretation of the bases for CM and VM memory search. The results also raised interesting questions concerning the detailed interpretation of CDA.

An extensive body of work has documented the impact of analogous cases on transfer. However, far less work has explored the role of contrasting cases in facilitating transfer. We designed a novel contrasting cases activity to engage learners with center-of-mass concepts in an engineering design task – building a cantilever using Legos. Participants in three conditions analyzed either contrasting cases, single cases, or no cases in the midst of an engineering design activity. Contrasting cases facilitated near but not far transfer. However, all conditions built equally successful cantilevers and noticed the underlying structure of center-of-mass concepts to the same degree. Moreover, regardless of condition, participants who noticed the structure at a deeper level performed better on both the engineering task and the far transfer assessment. The work has implications for the design of science and engineering instruction, while expanding our understanding of the perceptual processes that underlie transfer.

Researchers exploring mathematical abilities have proposed that humans and animals possess an approximate number system (ANS) that enables them to estimate numerosities in visual displays. Experimental data shows that estimation responses exhibit a constant coefficient of variation (CV: ratio of variability of the estimates to their mean) for numerosities larger than four, and a constant CV has been taken as a signature characteristic of the innate ANS. For numerosities up to four, however, humans often produce error-free responses, suggesting the presence of estimation mechanisms distinct from the ANS specialized for this ‘subitizing range’. We explored whether a constant CV might arise from learning in generic neural networks using widely-used neural network learning procedures. We find that our networks exhibit a flat CV for numerosities larger than 4, but do not do so robustly for smaller numerosities. Our findings are consistent with the idea that estimation for numbers larger than 4 may not require innate specialization for number, while also supporting the view that a process different from the one we model may underlie estimation responses for the smallest numbers.

We propose a novel approach to identity-based choice that focuses on peoples’ representations of the cause-effect relationships that exist among features of their self-concepts. More specifically, we propose that people who believe that a specific aspect of identity, such as a social category, is causally central (linked to many other features of the self- concept) are more likely to engage in behaviors consistent with that aspect than those who believe that the same aspect is causally peripheral (linked to fewer other features). Across three studies, we provide evidence for our approach to identity-based choice. We demonstrate that among people who belong to the same social category, those who believe that the associated identity is more causally central are more likely to engage in behaviors consistent with the social category.

Preschool-age children show essentialism (Gelman, 2003), ascribing an essence to an object that includes its history, and which can determine behavior. While infants show the precursors of essentialism, such as maintaining object representations during naturalistic occlusion (6-month-olds; Kaufman, Csibra, & Johnson, 2005), and resisting individuating two disparate appearances of an object when shown that one can change into the other (14-month-olds; Cacchione, Schaub, & Rakoczy, 2013), the implicit precursors of essentialist reasoning in infants have not been directly studied. Here we tested whether young infants could use an object’s prior history to predict its behavior, even after it had changed into a novel shape. Critically, the object either smoothly morphed into the novel shape (facilitating an essentialist interpretation) or was replaced by a new shape (discouraging essentialist interpretation). Results showed that 9-month-old infants (N = 22) in the Morph condition predicted the novel object would have the same behavior as the pre- transformation object; an essentialist interpretation. However, in the Replace condition (N = 22), predictions for the novel object were at chance; infants seemed to have lost the link to the pre-transformation object. Furthermore, results from a group of 6-month-olds (N = 15) showed that they failed to maintain this link, even in the Morph condition (which may indicate a failure to apply essentialist reasoning, or, more likely, a failure to adequately remember the pre-transformation object and/or apply the matching rule to predict post- transformation behavior).

Question asking is a prevalent aspect of children’s speech, pro- viding a means by which young learners can rapidly gain infor- mation about the world. Although past work demonstrates that children are sensitive to the knowledge state of potential infor- mants (e.g., Koenig & Harris, 2005), less work has explored whether children spontaneously direct questions to adults over other children (who are less likely to be knowledgeable), and in particular if adult-directed questions focus on content that is more likely to support general learning. We recorded in- dividual children’s spontaneous speech in 40-minute sessions during their preschool day; for every production we coded whether the speech was directed towards an adult, another child, or was stated to self. Our results (N = 30, totaling 2,232 utterances) showed that questions took up a greater proportion of children’s adult-directed speech as compared to the pro- portion of questions in child-directed and self-directed speech. Furthermore, although children asked many kinds of questions (including conversational clarifications, specific information questions, and questions intended for general learning), chil- dren more frequently asked the questions intended for learning when they spoke to adults than to the other groups. Analysis revealed a developmental effect, with results strongest for the older preschoolers. Our findings suggest that children discrim- inately choose ”what” and ”whom” to ask in daily conversa- tions, and this ability improves over the course of development.

Abundant work has looked at children’s ability to appropri- ately reject testimonies and unverified claims (Butler et al, 2017; Frazier, Gelman, & Wellman, 2009; Koenig, Clement, & Harris, 2004). However, sometimes our current knowledge is insufficient for solving a problem. In these cases, we should reject unsatisfying facts and prefer satisfying, if speculative, conjectures. In two studies, we gave 4-7 year-old children (Study 1, N=66; Study 2, N=32) questions that either could or could not be answered with available information. For each question, children made a binary choice between a factual an- swer citing information from the story or a conjectural answer that made unverified claims. Across age groups, children suc- cessfully chose the more satisfying response regardless of its truth value: children chose facts for questions with known an- swers and conjectures for questions with unknown answers. These findings suggest that children will go beyond known in- formation to endorse unverified claims when they satisfy the question-under-discussion.

All agents must constantly learn from dynamic environments to optimize their behaviors. For instance, it is necessary in new environments to learn how to distribute attention – i.e., which stimuli are relevant, and thus should be selected for greater processing, and which are irrelevant, and should be suppressed. Despite this, many experiments implicitly assume that attentional control is a static process (by averaging performance over large blocks of trials). By developing and utilizing new statistical tools, here we demonstrate that the effect of flanking items on response times to a central item (often utilized as an index of attentional control) is systematically and continuously influenced through time by the statistics of the flanking items. We discuss the implications of this finding from the perspective of examining individual differences – where traditional data analysis approaches may confound the rate at which attentional filtering changes through time with the asymptotic ability to filter.

Units as they exist today are highly abstract. Meters, miles, and other modern measures have no obvious basis in concrete phenomena and can apply to anything, anywhere. We show here, however, that units have not always been this way. Focusing on length, we first analyze the origins of length units in the Oxford English Dictionary; next, we review ethnographic observations about length measurement in 111 cultures. Our survey shows that length units have overwhelmingly come from concrete sources—body parts, artifacts, and other tangible phenomena—and are often tied to particular contexts. We next propose a reconstruction of how abstract units might have emerged gradually over cultural time through processes of comparison. Evidence from how children understand length and measurement provides support for this account. The case of units offers a powerful illustration of how some of our most important, pervasive abstractions can arise from decidedly concrete, often embodied origins.

Block construction tasks are highly complex, yet even young children engage in these tasks in both informal and formal learning settings. In this paper, we ask whether the specific paths through which children build a structure are unique to the individual, or alternatively, constrained by similar principles across individuals and over age. Our results show that although children between 4 and 8 make frequent errors in copying model constructions, there is a striking amount of consistency in specific attributes of their paths of construction, and this consistency mirrors that of adults. The build paths suggest that although children sometimes build inefficiently, they tend to build layer-by-layer, consistent with a role for intuitive physics that enables the creation of stable structures.

When two events occur closely in time, an “association” exists between memories for those events. When a pair of associ- ated events is semantically similar, it is easier to recognize the complete pair and easier to tell the complete pair apart from pairs of events that did not co-occur; there is also, however, a bias to report that similar events had co-occurred, even when they had not. A new experiment shows that these phenomena occur whenever two events share features, whether those fea- tures are perceptual or conceptual in nature and whether the events themselves are verbal or non-verbal. We present a dy- namic model for storage and recognition of associations that shows how all these results can be explained by the princi- ple that shared features lead to correlated processing of similar events, which in turn increases capacity to process associative information.

To improve cyber defense, researchers have developed algorithms to allocate limited defense resources optimally. Through signaling theory, we have learned that it is possible to trick the human mind when using deceptive signals. The present work is an initial step towards developing a psychological theory of cyber deception. We use simulations to investigate how humans might make decisions under various conditions of deceptive signals in cyber-attack scenarios. We created an Instance-Based Learning (IBL) model of the attacker decisions using the ACT-R cognitive architecture. We ran simulations against the optimal deceptive signaling algorithm and against four alternative deceptive signal schemes. Our results show that the optimal deceptive algorithm is more effective at reducing the probability of attack and protecting assets compared to other signaling conditions, but it is not perfect. These results shed some light on the expected effectiveness of deceptive signals for defense. The implications of these findings are discussed.

Research has shown that estimation of correlation from scatter plots is done poorly by both novices and experts. We tested whether proficiency in correlation estimation could be improved by perceptual learning interventions, in the form of perceptual-adaptive learning modules (PALMs). We also tested learning effects of alternative category structures in perceptual learning. We organized the same set of 252 scatter plot displays either into a PALM that implemented spacing in learning by shape categories or one in which the categories were ranges of correlation strength. Both PALMs produced markedly reduced errors, and both led trained participants to classify near transfer items as accurately as trained items. Differences in category organization produced modest effects on learning; there was some indication of more consistent reduction of absolute error when learning categories were organized by shape, whereas average bias of judgments was best reduced by categories organized by different numerical ranges of correlation.

Previous research on acquisition of noun class systems, such as grammatical gender, has shown that child learners rely dispro- portionately on phonological cues to class, even when compet- ing semantic cues are more reliable. Culbertson, Gagliardi, and Smith (2017) use artificial language learning experiments with adults to argue that over-reliance on phonology may be due to the fact that phonological cues are available first; learners base early representations on surface phonological dependen- cies, only later integrating semantic cues from noun meanings. Here, we show that child learners (6-7 year-olds) show this same sensitivity to early availability. However, we also find intriguing evidence of developmental changes in sensitivity to semantics; when both cues are simultaneously available chil- dren are more likely to rely on a phonology cue than adults. Our results suggest that early availability and a bias in favor of phonological cues may both contribute to children’s over- reliance on phonology in natural language acquisition.

People can learn about the effects of their actions either by performing physical experiments or by running mental sim- ulations. Physical experiments are reliable but risky; mental simulations are unreliable but safe. We investigate how peo- ple negotiate the balance between these strategies. Participants attempted to shoot a ball at a target, and could pay to take practice shots (physical experiments). They could also simply think (run mental simulations), but were incentivized to act quickly by paying for time. We demonstrate that the amount of thinking time and physical experiments is sensitive to trial characteristics in a way that is consistent with a model that integrates information across simulation and experimentation and decides online when to perform each.

Acting effectively in the world requires learning and control- ling dynamic systems, that is, systems involving feedback re- lations among continuous variables that vary in real time. We introduce a novel class of dynamic control environments us- ing Ornstein-Uhlenbeck processes connected in causal Markov graphs that allow us to systematically test people’s ability to learn and control various dynamic systems. We find that per- formance varied across a range of test environments, roughly matching with complexity defined by a set of models trained on the task (an optimal model, a deep Reinforcement Learning agent, and a PID controller). The testbed of dynamic envi- ronments and class of models introduced in this paper lay the groundwork for the systematic study of people’s ability to con- trol complex dynamic systems.

Interventions, time, and continuous-valued variables are all potentially powerful cues to causation. Furthermore, when observed over time, causal processes can contain feedback and oscillatory dynamics that make inference hard. We present a generative model and framework for causal infer- ence over continuous variables in continuous time based on Ornstein-Uhlenbeck processes. Our generative model pro- duces a stochastic sequence of evolving variable values that manifest many dynamical properties depending on the nature of the causal relationships, and a learner’s interventions (man- ual changes to the values of variables during a trial). Our model is also invertible, allowing us to benchmark participant judgments against an optimal model. We find that when in- teracting with systems acting according to this formalism peo- ple directly compare relationships between individual variable pairs rather than considering the full space of possible models, in accordance with a local computations model of causal learn- ing (e.g., Fernbach & Sloman, 2009). The formalism presented here provides researchers in causal cognition with a powerful framework for studying dynamic systems and presents oppor- tunities for other areas in cognitive psychology such as control problems.

Immersive virtual reality (VR) offers important benefits over non-immersive displays, such as increased ecological validity and high experimental control. Studies in cognitive science using immersive VR are however still rather limited in number. The current paper illustrates the opportunities to apply VR in the cognitive sciences by using an immersive adaptation of a classic study by Godden and Baddeley (1975) on environmental context-dependent memory (ECDM). In this memory study, retrieval was facilitated when the context between learning and testing matched. In line with the literature showing small effects for context-dependent recall, the current study indicated a marginally significant ECDM effect for one virtual context, but when deep processing was controlled, a significant ECDM effect was obtained. In demonstrating the applicability and benefits of immersive VR, this study at last opens a doorway to the large-scale implementation of immersive VR for the cognitive sciences.

This paper investigates the manner in which path events are specified in Mandarin serial verb constructions (SVCs) and how such representations incorporate attentional processes, as reflected in Talmy’s (1996, 2000) theory of Windowing of Attention. Here we focus on the verbs laí (come) and qù (go). The results show that: (1) laí and qù in SVCs mainly represent open path, followed by fictive path and closed path respectively; (2) laí or qù in Mandarin SVCs tends to adopt final path windowing. Final windowing accounts for 60.3% for SVCs with laí and 65.7% for SVCs with qù. This suggests that Mandarin SVC with laí or qù profiles the final part of the construction, and the information at the end is the key information. The present study offers a new account for the information distribution of SVCs and sheds light on the event segmentation of SVCs.

This paper investigates a problem where the solver must firstly determine which of two possible causes are the source of an effect where one cause has a historically higher propensity to cause that effect. Secondly, they must update the propensity of the two causes to produce the effect in light of the observation. Firstly, we find an error commensurate with the ‘double updating’ error observed within the polarisation literature: individuals appear to first use their prior beliefs to interpret the evidence, then use the interpreted form of the evidence, rather than the raw form, when updating. Secondly, we find an error where individuals convert from a probabilistic representation of the evidence to a categorical one and use this representation when updating. Both errors have the effect of exaggerating the evidence in favour of the solver’s prior belief and could lead to confirmation bias and polarisation.

Flexibly adapting behavior to different contexts is a critical component of human intelligence. It requires knowledge to be structured as coherent, context-dependent action rules, or task-sets (TS). Nevertheless, inferring optimal TS is compu- tationally complex. This paper tests the key predictions of a neurally-inspired model that employs hierarchically-structured reinforcement learning (RL) to approximate optimal inference. The model proposes that RL acts at two levels of abstrac- tion: a high-level RL process learns context-TS values, which guide TS selection based on context; a low-level process learns stimulus-actions values within TS, which guide action selec- tion in response to stimuli. In our novel task paradigm, we found evidence that participants indeed learned values at both levels: not only stimulus-action values, but also context-TS values affected learning and TS reactivation, and TS values alone determined TS generalization. This supports the claim of two RL processes, and their importance in structuring our interactions with the world.

As compared to other animals, humans are particularly skilled at using and improving tools and other solutions to problems that were first discovered by other people. Although the human capacity for cumulative cultural evolution is well-known, the effectiveness of inheritance as a form of problem solving is an area in need of further research. We report an experiment designed to understand how effectively solutions to problems accumulate over generations of problem solving. Using a tool- discovery game, we found that participants were consistently able to discover more tools in a 25 minute session than their ancestors. Participants who inherited more tools required more time to recreate them, but their rate of new tool discovery was not slowed. In addition, we show that participants were able to recreate the tools they inherited more efficiently than their ancestors, but that inheritance did not confer any improvement in future problem solving. We discuss the limitations of this work, and motivate future directions.

Discovery and application of causal knowledge in novel problem contexts is a prime example of human intelligence. As new in- formation is obtained from the environment during interactions, people develop and refine causal schemas to establish a parsimo- nious explanation of underlying problem constraints. The aim of the current study is to systematically examine human abil- ity to discover causal schemas by exploring the environment and transferring knowledge to new situations with greater or differ- ent structural complexity. We developed a novel OpenLock task, in which participants explored a virtual “escape room” environ- ment by moving levers that served as “locks” to open a door. In each situation, the sequential movements of the levers that opened the door formed a branching causal sequence that began with either a common-cause (CC) or a common-effect (CE) struc- ture. Participants in a baseline condition completed five trials with high structural complexity (i.e., four active levers). Those in the transfer conditions completed six training trials with low structural complexity (i.e., three active levers) before completing a high-complexity transfer trial. The causal schema acquired in the transfer condition was either congruent or incongruent with that in the transfer condition. Baseline performance under the CC schema was superior to performance under the CE schema, and schema congruency facilitated transfer performance when the congruent schema was the less difficult CC schema. We com- pared between-subjects human performance to a deep reinforce- ment learning model and found that a standard deep reinforce- ment learning model (DDQN) is unable to capture the causal ab- straction presented between trials with the same causal schema and trials with a transfer of causal schema.

We are currently in the process of designing and implement- ing a computational cognitive system that combines percep- tion, memory, attention, and domain-specific semantic knowl- edge to perform data visualization tasks. While this work is still in early stages, we report here on one subset of this larger project that involves building a “visual long term memory” for the system. To constrain the problem, we assume a domain of astronomy, and we focus exclusively on scatterplot visual- izations. In this paper, we present three of our initial steps along this path. First, we collected and analyzed a catalog of 74 scatterplots from real astronomy sources (papers, books, etc.), which we consider to be typical data visualizations that astronomers would frequently encounter during their educa- tion. Second, we asked a team of human raters to rate all 74 scatterplots along nine dimensions describing shape cate- gories, taken from a computational approach originally sug- gested by John and Paul Tukey called scagnostics. Third, we calculated computer-based scagnostics for a subset of the scat- terplots. We measured inter-rater agreements among the hu- man raters and between the calculated and human ratings.

Reading is a critical skill as it provides a gateway for other learning within and outside of school. Many children struggle to acquire this fundamental skill. Suboptimal design of books for beginning readers may be one factor that contributes to the difficulties children experience. Specifically, extraneous details in illustrations (i.e., interesting but irrelevant to the story elements) could promote attentional competition and hamper emerging literacy skills. We used eye-tracking technology to examine this possibility. The results of this study indicated that excluding extraneous details from illustrations in a book for beginning readers reduced attentional competition (indexed by gaze shifts away from text) and improved children’s reading comprehension. This study suggests that design of reading materials for children learning to read can be optimized to promote literacy development in children.

Experiences of contingent interactions like referential cues (e.g., caregivers pointing to relevant text and pictures) during shared book reading predict better reading and language outcomes (Landry, & Smith, 2007). However, it is unclear whether contingent responsiveness in a digital book could provide similar support for children in the absence of contingent feedback from an adult. The effects on story comprehension using an interactive book with content-related animations that activated contingent on children’s vocalizations were investigated, with a focus on whether the interactive book might be especially useful for children with less developed attentional control. The present study used a within-subject design with data from 69 preschool-aged children. The use of the interactive book exhibited significantly increased comprehension, and was also found to be especially useful for children with less attentional control. Importantly, the associations between attention and comprehension gains were not entirely due to variance shared with verbal ability.

Here we explore whether drawing upon preschooler’s intuitive causal reasoning abilities may bolster their attention to the presence of conflicting data. Specifically, we examine whether prompting children to think counterfactually about alternative outcomes facilitates their anomaly detection in a causal reasoning task. The current task assesses whether children in two conditions successfully differentiate between potential causes: one that accounts for 100% of the data (no anomalies), and one that accounts for 75% of the data (anomalies observed). Results indicate that counterfactual prompts lead 5-year-olds to privilege the hypothesis that accounts for more of their observations, and also support transfer of this hypothesis to inform their inferences about novel cases. Findings suggest that counterfactual scaffolds may be beneficial in promoting causal reasoning in children.

Values drive our behavioral choices. Ample research has explored the cognitive and neural underpinnings of value- based computations related to decision-making. However, behaviorally relevant values that we associate with real-world objects are often not monetary. For instance, social values associated with specific people are crucial for social behaviors and interactions. Moreover, understanding and attributing social values allows for proper evaluations of potential interactions with others, and can lead to more beneficial social behaviors and relationships. Learning social values has been shown to recruit the same systems as reward values, however how they become associated with specific people remains to be established. The present study examined social value learning of other people using naturalistic face images. We found that before learning, distances between the faces in conceptual similarity spaces were organized corresponding to their perceptual similarity. However, after learning, faces were shifted in a manner that reflected similarity of their associated social values (generosity). Furthermore, distances were positively correlated with a post- learning index of preference to interact with a person in a future cooperative game. In other words, learned social values of the faces seemed to influence their representations in conceptual space, and such representational changes were related to propensities in future behavior.

We explore a recurrent neural network model of counting based on the differentiable recurrent attentional model of Gregor et al. (2015). Our results reveal that the model can learn to count the number of items in a display, pointing to each of the items in turn and producing the next item in the count sequence at each step, then saying ‘done’ when there are no more blobs to count. The model thus demonstrates that the ability to learn to count does not depend on special knowledge relevant to the counting task. We find that the model’s ability to count depends on how well it has learned to point to each successive item in the array, underscoring the importance of coordination of the visuospatial act of pointing with the recitation of the count list. The model learns to count items in a display more quickly if it has previously learned to touch all the items in such a display correctly, capturing the relationship between touching and counting noted by Alibali and DiRusso. In such cases it achieves performance sometimes thought to result from a semantic induction of the ‘cardinality principle’. Yet the errors that it makes have similarities with the patterns seen in human children’s counting errors, consistent with idea that children rely on graded and somewhat variable mechanisms similar to our neural networks.

We investigate a method for formulating context- and task- specific computational models of human performance in a con- strained semantic memory task. In particular, we assume that memory retrieval can only use a simple process – a random walk – and examine whether the effect of context and task specifications can be captured via a straightforward network estimation method that is sensitive to context and task. We find that a random walk model on the context-specific networks mimics aggregate human performance.

Metaphors can shape how people reason about complex issues,but most studies of metaphor framing rely exclusively onwritten materials. This is a significant limitation, as peopleregularly encounter linguistic metaphors in a variety ofdifferent communicative settings (e.g., read in the newspaper,heard on the radio, or viewed on television). Because researchfinds that variations in communication modality can influencemessage comprehension, retention, and persuasiveness, weexplored the relative power of metaphor framing in differentcommunication modalities. Across two experiments,participants read, heard, or watched a person describe fourdifferent metaphorically framed issues. They had to answer atarget question about each issue by selecting from two responseoptions, one of which was congruent with the metaphor frame.Results revealed a significant, similarly-sized effect ofmetaphor framing in every communication modality,suggesting that communication modality does not moderate theefficacy of metaphor framing.

Folk theories guide behavior and shape how people make sense oftheir environment. We investigated whether folk economic beliefswould moderate the widely publicized finding that people show aconservative shift in their politics when their majority status insociety is threatened. Across three experiments, participants readabout either projected demographic changes (threat) or changes inonline dating (control), indicated whether they viewed the economyas a zero- or non-zero-sum system, and responded to measures ofsociopolitical attitudes. Compared to controls, participants in thethreat condition who conceptualized the economy in zero-sum termssupported more conservative policies. However, those whoconceptualized the economy in non-zero-sum terms actuallyendorsed more liberal positions in this condition. These effectsobtained only when participants expressed their economic viewsbefore their political attitudes. This suggests folk economic beliefsshape how people respond to threats to their majority status,provided those beliefs are first made explicit.

Children tend to produce words earlier when they are connected toa variety of other words along both the phonological and semanticdimensions. Though this connectivity effect has been extensivelydocumented, little is known about the underlying developmentalmechanism. One view suggests that learning is primarily drivenby a network growth model where highly connected words in thechild’s early lexicon attract similar words. Another view suggeststhat learning is driven by highly connected words in the externallearning environment instead of highly connected words in the earlyinternal lexicon. The present study tests both scenarios system-atically in both the phonological and semantic domains, and across8 languages. We show that external connectivity in the learningenvironment drives growth in both the semantic and the phonolog-ical networks, and that this pattern is consistent cross-linguistically.The findings suggest a word learning mechanism where childrenharness their statistical learning abilities to (indirectly) detect andlearn highly connected words in the learning environment.

Language complexity seems to be influenced by populationcharacteristics such as the proportion of adult learners. Onepotential explanation for this link is that native speakers ac-commodate to non-native speakers, simplifying their languageuse during such interactions: learners may then acquire a lesscomplex language. We model accommodation in interaction ina Bayesian framework, where in order to accommodate appro-priately, an agent must first infer their interlocutor’s linguisticabilities. We find that when the agent consistently accommo-dates, learners end up with a simplified language, due to a rein-forcing effect between an initially underinformed learner andan accommodating native speaker.

The effects of feedback often depend on individual learnercharacteristics. In the current study, we experimentally testedwhether an individual’s task expectations influence learningfrom feedback on mathematics problems. Specifically, wemanipulated undergraduate students’ beliefs about thedifficulty of the task to influence their expectations forsuccess. Students (N = 160) were randomly assigned to one offour learning conditions based on a crossing of two factors:task expectations (easy or hard) and feedback during problemsolving (yes or no). On a final transfer test, feedback led tohigher scores than no feedback for those who expected thetask to be easy. But, feedback led to marginally lower scoresfor those who expected the task to be hard. Results suggestthat expecting the task to be hard and to experience failurecan lead to a self-fulfilling prophecy. When learning fromfeedback, students should set their expectations for success.

A wealth of evidence indicates the existence of a consolidationphase, triggered by and following a practice session, wherein newmemory traces relevant to task performance are transformed andhoned to represent new knowledge. But, the role of consolidation isnot well-understood in category learning and has not been studied atall under incidental category learning conditions. Here, weexamined the acquisition, consolidation and retention phases in avisuomotor task wherein auditory category information wasavailable, but not required, to guide detection of an above-thresholdvisual target across one of four spatial locations. We compared twotraining conditions: (1) Constant, whereby repeated instances of oneexemplar from an auditory category preceded a visual target,predicting its upcoming location; (2) Variable, whereby five distinctcategory exemplars predicted the visual target. Visual detectionspeed and accuracy, as well as the performance cost of randomizingthe association of auditory category to visual target location, wereassessed during online performance, again after a 24-hour delay toassess the expression of delayed gains, and after 10 days to assessretention. Results revealed delayed gains associated with incidentalauditory category learning and retention effects for both trainingconditions. Offline processes can be triggered even for incidentalauditory input and lead to category learning; variability of input canenhance the generation of incidental auditory category learning.

Meaning depends on context. This applies both in obviouscases like deictics or sarcasm as well as more subtle situationslike framing or persuasion. One key characteristic of context isthe identity of the participants in an interaction. Our interpre-tation of an utterance depends on a variety of factors such asour personal history, background knowledge, and our relation-ship to the source. While demographics allow us to capturesome of this variance, the relevance of specific demographicfactors varies across contexts. To address these challenges, weintroduce a method for combining demographics and contextinto situated demographic embeddings—mapping representa-tions onto a continuous space appropriate for the given domain.We further demonstrate how to make use of related externalresources so as to apply this approach in low-resource situa-tions. We show the resulting representations to be interpretableand consider domain-specific similarity. Finally, we show howthese representations can be incorporated to improve modelingof a real-world natural language understanding task.

We introduce a novel experimental paradigm for studying multi-modal integration in causal in-ference. Our experiments feature a physically realistic Plinko machine in which a ball is droppedthrough one of three holes and comes to rest at the bottom after colliding with a number of ob-stacles. We develop a hypothetical simulation model which postulates that people figure out whathappened by integrating visual and auditory evidence through mental simulation. We test themodel in a series of three experiments. In Experiment 1, participants only receive visual infor-mation and either predict where the ball will land, or infer in what hole it was dropped based onwhere it landed. In Experiment 2, participants receive both visual and auditory information – theyhear what sounds the dropped ball makes. We find that participants are capable of integratingboth sources of information, and that the sounds help them figure out what happened. In Exper-iment 3, we show strong cue integration: even when vision and sound are individually completelynon-diagnostic, participants succeed by combining both sources of evidence.

Understanding the nuanced relationship between attention andlearning in young children is difficult due to the lack ofdevelopmentally appropriate measures of attention. Youngchildren are in a measurement gap - they are too old formeasures typically employed with infants and toddlers andoften too young to produce useful data from more traditionalmeasures used with older children and adults. Due to thepaucity of developmentally appropriate measures it ischallenging to employ best practices and utilize convergingmeasures of attention. Additionally, existing behavioralobservation methods are time consuming and can suffer frompoor reliability due to their subjective nature. The presentstudy aims to address these limitations by leveragingaffordable technology to create a novel measure of attention,the Wiggleometer. The Wiggleometer is a custom chair thatcovertly measures body movement as an index of attention.The preliminary results help establish the concurrent validityof the measure and suggest the Wiggleometer can beemployed to better predict children’s learning outcomes.

Causal judgments are stubborn. If people learn about twocorrelated variables B and C, and judge that B causes C, theytypically stick to that judgment even when contradictoryevidence comes to light. One form of contradictory evidenceis that a third variable A causes both B and C, explaining thecorrelation. This paper extends prior work showing thatsimply presenting statistical evidence that A is the commoncause of both B and C does not lead to belief change about B.However, if first subjects learn to categorize phenomena bytheir underlying causal relationships (i.e., as exemplars of acommon cause category), then they can transfer their categoryknowledge to properly interpret the evidence. They recognizethat A is the common cause of B and C and revise their beliefabout B. These results suggest that teaching abstract causalcategories is a promising strategy to help revise false beliefs.

The ability to anticipate the future improves markedly acrossthe preschool years. One major area of improvement is in chil-dren’s ability to consider their future preferences. Whereas5-year-olds understand they will prefer adult items in the fu-ture, 3-year-olds indicate they will continue to prefer childitems. In the present research, we show that preschoolers(N=120) show an ownership-advantage in their future-orientedthinking—they are better able to indicate which objects theywill own as adults than to indicate what they will like. Thesefindings are informative about the basis for children’s difficultyanticipating their future preferences, and also reveal differ-ences between how children think about ownership and prefer-ences.

The goal of conjunctive visual search is to attentionally selecta location at which the visual array matches a set of cued fea-ture values. Here we present a neural dynamic architecturein which all neural processes operate in parallel in continu-ous time, but in which discrete sequences of processing stepsemerge from dynamic instabilities. When biased competitionselects an object location at which not all conjunctive featurevalues match the cue, the neural representation of a conditionof dissatisfaction is activated and induces an attentional shift.Successful match activates the neural representation of a con-dition of satisfaction that ends the search. The search takesplace in the current visual array but takes into account an au-tonomously acquired feature-space scene memory.

Inclusion of gesture – meaningful movements of the hands –during mathematics instruction is beneficial for teachingnaïve learners novel concepts, and it can affect a learner’sallocation of visual attention. Yet, it is unknown how childrenwith pre-existing knowledge of a math concept approachinstruction that includes gesture. Here, we examine howchildren’s prior knowledge and either the presence or absenceof gesture during instruction drive patterns in visual attentionduring a lesson. We find that prior knowledge does determinevisual attention patterns, independent of type of instruction(i.e. with or without gesture). These findings further ourunderstanding of the attentional mechanisms of gesture andhave implications for real-world classrooms, where levels ofprior knowledge are often mixed.

Infants are experts at playing, with an amazing ability to gen-erate novel structured behaviors in unstructured environmentsthat lack clear extrinsic reward signals. We seek to replicatesome of these abilities with a neural network that implementscuriosity-driven intrinsic motivation. Using a simple but ecolog-ically naturalistic simulated environment in which the agent canmove and interact with objects it sees, the agent learns a worldmodel predicting the dynamic consequences of its actions. Si-multaneously, the agent learns to take actions that adversariallychallenge the developing world model, pushing the agent toexplore novel and informative interactions with its environment.We demonstrate that this policy leads to the self-supervisedemergence of a spectrum of complex behaviors, including egomotion prediction, object attention, and object gathering. More-over, the world model that the agent learns supports improvedperformance on object dynamics prediction and localizationtasks. Our results are a proof-of-principle that computationalmodels of intrinsic motivation might account for key featuresof developmental visuomotor learning in infants.

We outline a new method to explore di↵erences in se-mantic representations between groups and apply it toa novel domain where we might expect to find such dif-ferences: politics. We hypothesize and find confirma-tory evidence that individuals of opposite partisanship,as measured by party identification, have di↵erent se-mantic representations. We further evaluate whetherdi↵erences in representations are predictive of attitudejudgments as long suggested by constructivist theoriesof attitudes from social psychology. We find di↵erencesare indeed predictive of attitudes even after controllingfor other strongly predictive covariates (party identifica-tion and ideology). In discussing our results we sketchout a broader theory of the role of semantic memory inattitude judgments.

In a world that relies increasingly on large amounts of data and on powerful Machine Learning (ML) models, the veracity of decisions made by these systems is essential. Adversarial samples are inputs that have been perturbed to mislead the in- terpretation of the ML and are a dangerous vulnerability. Our research takes a first step into what can be an important innova- tion in cognitive science: we analyzed human’s judgments and decisions when confronted with targeted (inputs constructed to make a ML model purposely misclassify an input as some- thing else) and non-targeted (a noisy perturbed input that tries to trick the ML model) adversarial samples. Our findings sug- gest that although ML models that produce non-targeted adver- sarial samples can be more efficient than targeted samples they result in more incorrect human classifications than those of tar- geted samples. In other words, non-targeted samples interfered more with human perception and categorization decisions than targeted samples.

We report a model for object identification based on an exper- iment that varies the arrival times of different features of the objects. A single object, a circle with four spokes extending in different directions, is presented and must be classified as either one of four well trained target stimuli, or one of four well trained foil stimuli. The features (spokes) are presented either simultaneously or successively at intervals of 16, 33, or 50 ms., with target diagnostic features arriving first or last. All durations are short enough that the display appears simultane- ous. The data show that individual decisions vary with both timing and diagnosticity. We apply a dynamic model based on one reported in (Cox & Shiffrin, 2017) for episodic recognition memory. Our model assumes features are perceived at vary- ing times following presentation, possibly in error. At each moment the current features are compared to the well learned memory representations of the eight stimuli, producing a like- lihood ratio for target vs foil. A decision is made when the log likelihood first exceeds a target decision boundary or falls be- low a foil decision boundary. The model implements a form of Bayesian optimal decision making given the assumptions con- cerning feature perception. It predicts the key findings quite well.

Words exist for referring at many levels of specificity: from the broadest (thing) to the most specific (Fido). What drives the emergence of these taxonomies of reference? Recent com- putational theories of language evolution suggest that commu- nicative demands of the environment may play a deciding role. Here, we investigate local pragmatic mechanisms of lexical adaptation that may undergird global emergence by manipulat- ing context in a repeated reference game where pairs of partic- ipants interactively coordinate on an artificial communication system. We hypothesize that pairs should converge on specific names (e.g. Fido) when the context requires frequently mak- ing fine distinctions between entities; conversely, they should converge on a more compressed system of conventions for ab- stract categories (e.g. dog) in coarser contexts, even if a finer mapping would be sufficient. We show differences in the lev- els of abstraction that emerged in different environments and introduce a statistical approach to probe the dynamics of emer- gence.

This study focuses on collaborative learning involving a knowledge integration activity, whereby learner dyads explain each other’s expert knowledge. It was hypothesized that learn- ing gain can be determined by the degree to which learn- ers synchronize their gaze (gaze recurrence) and use overlap- ping language (information overlap) during their interaction. Thirty-four learners participated in a laboratory-based eye- tracking experiment, wherein learners’ gazes and oral dialogs were analyzed. Multiple regression analysis was conducted, wherein learning performance was regressed on the two inde- pendent variables. Then, a simulation was conducted to view how the model predicts performance based on the collabora- tive process. The results showed that both gaze recurrence and lexical overlap significantly predicted learning performance in the current task. Furthermore, the suggested model success- fully predicted learning performance in the simulation. These results indicate that the two variables might be useful for de- veloping detection modules that enable a better understanding of learner-learner collaborative learning.

People’s desire to be patient or impatient can fluctuate from moment to moment, yet little is known about the effects of variability in time preference on intertemporal choice behavior. We examine this issue through the lens of an exponential discounting model with noisy discount factors. We show that such a model can generate decreasing patience over time, accounting for behavioral patterns typically attributed to hyperbolic discounting, while also making reasonable predictions regarding violations of intertemporal dominance. Additionally, two experiments reveal that many participants do display noise in their discount factors, and that a noisy discount factor model outperforms hyperbolic models in terms of quantitative fit. Ultimately the majority of participants are best described by some type of exponential discounting model (with or without noisy discount factors). These results indicate that it may not be necessary to assume alternate forms of non- exponential discounting, as long as the discount factors in an exponential model are permitted to vary at random. These results also highlight the importance of allowing for different sources of noise in choice modeling.People’s desire to be patient or impatient can fluctuate from moment to moment, yet little is known about the effects of variability in time preference on intertemporal choice behavior. We examine this issue through the lens of an exponential discounting model with noisy discount factors. We show that such a model can generate decreasing patience over time, accounting for behavioral patterns typically attributed to hyperbolic discounting, while also making reasonable predictions regarding violations of intertemporal dominance. Additionally, two experiments reveal that many participants do display noise in their discount factors, and that a noisy discount factor model outperforms hyperbolic models in terms of quantitative fit. Ultimately the majority of participants are best described by some type of exponential discounting model (with or without noisy discount factors). These results indicate that it may not be necessary to assume alternate forms of non- exponential discounting, as long as the discount factors in an exponential model are permitted to vary at random. These results also highlight the importance of allowing for different sources of noise in choice modeling.

When a speaker produces a referring expression, their overarching goal is to get the addressee to identify a particular object in the context. This goal leads to the expectation that speakers will use a referring expression tailored to the perspective of the addressee. While research in psycholinguistics has indeed found that speakers tailor their referring expressions to the addressee’s perspective, they also find egocentric tendencies; namely, a sensitivity to the speaker’s own perspective. Mozuraitis, Stevenson and Heller (2018) make the novel proposal that “mixing” perspectives is a design feature of the production system, modelling data from an experiment where knowledge mismatch concerned object function. Here we further test this model on the more common knowledge mismatch of visual perspective, modelling data from Vanlangendonck, Willems, Menenti and Hagoort (2016). The modelling results shed new light on concept of “referential success” that has been assumed to guide reference production.

Geminates (or double letters) are a feature of many languages, including English. Studies of the spelling errors produced by individuals with orthographic working memory deficits have provided evidence that geminates are not produced as two in- dependent instances of the same letter. Instead, there must be a special mechanism in the orthographic system that produces geminates. Several theories have attempted to model such mechanisms. However, in most cases, the predictions of such theories have been tested using data from single-case neuro- psychological studies. In the current study, we re-evaluate these theories using the largest corpus of geminate errors in typing collected to date, and show that no theory can explain all the findings. We then propose an enhanced model of gem- ination that can.

The classic Baddeley and Hitch (1974) model divides working memory into domain-specific subsystems and a shared, do- main-general central executive, which plays a role in allocating resources to items stored in the subsystems. The nature of this resource—in particular, its quantization (discrete vs. continu- ous) and the flexibility of its allocation—has been studied ex- tensively in the visual domain, with evidence from experiments using continuous response measures providing support for models with flexibly and continuously divisible resources. It remains unclear, however, whether similar mechanisms medi- ate the division of resources in phonological working memory. In this paper, we show that, despite representational differences between visual and auditory processing, continuous measures can also be employed for studying phonological working memory. Using such measures, we demonstrate that the prin- ciples of resource division in visual and phonological pro- cessing are indeed similar, providing evidence for a domain- general mechanism for allocating working memory resources.

Infants learn the sound categories of their language and adults successfully process the sounds they hear, even though sound categories often overlap in their acoustics. Most researchers agree that listeners use context to disambiguate overlapping cat- egories. However, they differ in their ideas about how context is used. One idea is that listeners normalize out the systematic effects of context from the acoustics of a sound. Another idea is that contextual information may itself be an informative cue to category membership, due to patterns in the types of contexts that particular sounds occur in. We directly contrast these two ways of using context by applying each one to the test case of Japanese vowel length. We find that normalizing out contextual variability from the acoustics does not improve categorization, but using context in a top-down fashion does so substantially. This reveals a limitation of normalization in phonetic acquisi- tion and processing and suggests that approaches that make use of top-down contextual information are promising to pursue.

When observing others’ behavior, people use Theory of Mind to infer unobservable beliefs, desires, and intentions. And when showing what activity one is doing, people will modify their behavior in order to facilitate more accurate interpretation and learning by an observer. Here, we present a novel model of how demonstrators act and observers interpret demonstrations corresponding to different levels of recursive social reasoning (i.e. a cognitive hierarchy) grounded in Theory of Mind. Our model can explain how demonstrators show others how to per- form a task and makes predictions about how sophisticated ob- servers can reason about communicative intentions. Addition- ally, we report an experiment that tests (1) how well an ob- server can learn from demonstrations that were produced with the intent to communicate, and (2) how an observer’s interpre- tation of demonstrations influences their judgments.

What makes some words harder to learn than others in a second language? Although some robust factors have been identified based on small scale experimental studies, many relevant factors are difficult to study in such experiments due to the amount of data necessary to test them. Here, we investigate what factors affect the ease of learning of a word in a second language using a large data set of users learning English as a second language through the Duolingo mobile app. In a regression analysis, we test and confirm the well-studied effect of cognate status on word learning accuracy. Furthermore, we find significant effects for both cross-linguistic semantic alignment and English semantic density, two novel predictors derived from large scale distributional models of lexical semantics. Finally, we provide data on several other psycholinguistically plausible word level predictors. We conclude with a discussion of the limits, benefits and future research potential of using big data for investigating second language learning.

People often engage in “offline simulation”, considering what would happen if they performed certain actions in the future, or had performed different actions in the past. Prior research shows that these simulations are biased towards actions a per- son considers to be good—i.e., likely to pay off. We ask whether, and why, this bias might be adaptive. Through com- putational experiments we compare five agents who differ only in the way they engage in offline simulation, across a variety of different environment types. Broadly speaking, our exper- iments reveal that simulating actions one already regards as good does in fact confer an advantage in downstream decision making, although this general pattern interacts with features of the environment in important ways. We contrast this bias with alternatives such as simulating actions whose outcomes are instead uncertain.

Research on arithmetic competence has emphasized the importance of arithmetic fluency – the use of efficient direct strategies when solving simple, conventional problems. Comparatively little attention has been focused on arithmetic sense, which we define as the adaptive use of direct and indirect strategies when solving complex, novel problems. The current study evaluates the new construct of arithmetic sense and investigates its predictive relationship to mathematical achievement. 302 students in 6 th grade completed a battery of tests of their cognitive and numerical abilities, arithmetic fluency, arithmetic sense, mathematics achievement, and pre-algebra skills. The central finding is that arithmetic sense is the best single predictor of mathematical achievement. In particular, it is better than arithmetic fluency. These findings open a new pathway for improving school-aged students’ algebraic thinking and mathematical achievement.

The present study explores the impact on memory and attitude change of social media messages that are incorporated into television programs, and the interaction of such messages with the viewer’s disposition to critical thinking. Sixty university students were allocated to one of two experimental conditions and viewed television content: social media messages were included in only one condition. The results showed a significant interaction between participants’ disposition (Objectiveness) and the experimental condition: participants with higher Objectiveness scores exhibited larger changes in their attitudes. An analysis of 10 participants’ eye fixations suggested participants’ tendency to change their allocation of attention to different types of message over time. Additionally, there was a significant correlation between the tendency to focus on these messages and scores for disposition to critical thinking (Objectiveness and Logical thinking). We discuss the possible conclusions on the impact of showing social media messages and the limitations of this study.

Predicting semantic similarity judgments is often modeled as a three-step process: collecting feature ratings along multiple dimensions (e.g., size, shape, color), computing similarities along each dimension, and combining the latter into an aggregate measure (Nosofsky, 1985). However, such models fail to account for over half of the variance in similarity judgments pertaining to complex, real-world objects (e.g., elephant and bear), even when taking into account their description along dozens of dimensions. To help explain this prediction gap, we propose a two-fold approach. First, we provide the first empirical evidence of a mismatch between similarity predicted by feature ratings and that reported by participants directly along individual dimensions. Second, we show that, surprisingly, separate sub-domains within directly reported dimension-specific similarities carry different amounts of information for predicting object-level similarity judgments. Accordingly, we show that differentially weighting directly reported dimension-specific similarity sub- domains significantly improves prediction of free (i.e., unconstrained) semantic similarity judgments.

We introduce a computational model capturing the high-level features of the complementary learning systems (CLS) frame- work. In particular, we model the integration of episodic mem- ory with statistical learning in an end-to-end trainable neural network architecture. We model episodic memory with a non- parametric module which can retrieve past observations in re- sponse to a given observation, and statistical learning with a parametric module which performs inference on the given ob- servation. We demonstrate on vision and control tasks that our model is able to leverage the respective advantages of nonpara- metric and parametric learning strategies, and that its behavior aligns with a variety of behavioral and neural data. In partic- ular, our model performs consistently with results indicating that episodic memory systems in the hippocampus aid early learning and transfer generalization. We also find qualitative results consistent with findings that neural traces of memories of similar events converge over time. Furthermore, without explicit instruction or incentive, the behavior of our model nat- urally aligns with results suggesting that the usage of episodic systems wanes over the course of learning. These results sug- gest that key features of the CLS framework emerge in a task- optimized model containing statistical and episodic learning components, supporting several hypotheses of the framework.

Abstraction to a single prototypical representation is a core principle of Distributional Semantic Models (DSMs) that learn semantic representations for words by applying dimension reduction to statistical redundancies in language. While the learning mechanisms for semantic abstraction vary widely across the many DSMs in the literature, they are essentially all prototype models in that they create a single abstract representation for a word’s meaning. The prototype method stands in stark contrast to work in the field of categorization that has converged on the importance of instance models. In comparison to the prototype method, instance-based models assume only an episodic store and, rather than applying abstraction mechanisms at learning, argue that meaning emerges in the act of retrieval. We cash this idea out by presenting and evaluating an instance theory of distributional semantics, and by demonstrating that it can explain diverging patterns of homonymous words that classic “abstraction-at-learning” models simply cannot as a consequence of their architectural assumptions.

Self-assessment, or the evaluation of one’s ability on a task, is widely perceived as a fundamental skill, yet in most studies, people are found to be poorly calibrated to their own abilities. Some results seem to show poorer calibration for low perform- ers than for high performers. This effect has been explained in multiple ways: it could indicate worse metacognitive abil- ity among the low performers (the “Dunning-Kruger” effect), or simply regression to the mean. To tease apart these expla- nations we develop a Bayesian model of self-assessment and evaluate its predictions in two experiments. Our results suggest that poor self-assessment is caused by the influence of prior be- liefs and imperfect skill at determining whether a problem was solved correctly or not, and offer only weak support for of a relationship between metacognitive ability and performance.

At first glance, children’s word learning appears to be mostly a problem of learning words like dog and run. However, it is small words like and and or that enable the construction of complex combinatorial language. How do children learn the meaning of these function words? Using transcripts of parent- child interactions, we investigate the cues in child-directed speech that can inform the interpretation and acquisition of the connective or which has a particularly challenging semantics. Study 1 finds that, despite its low overall frequency, children can use or close to parents’ rate by age 4, in some speech acts. Study 2 uses annotations of a subset of parent-child interac- tions to show that disjunctions in child-directed speech are ac- companied by reliable cues to the correct interpretation (ex- clusive vs. inclusive). We present a decision-tree model that learns from a handful of annotated examples to correctly pre- dict the interpretation of a disjunction. These studies suggest that conceptual and prosodic cues in child-directed speech can provide information for the acquisition of functional categories like disjunction.

When students solve problems with access to examples show- ing worked out solutions, they often resort to shallow methods like copying that do not result in learning. An open question is therefore how to encourage deeper processing in this type of instructional context. To address this question, in the present study, we investigate the impact of manipulating problem- example similarity over the course of a problem-solving ses- sion in several ways, including faded assistance (high to low similarity), reverse faded assistance (low to high similarity), and a control group with high, constant assistance. We found that the reverse faded assistance condition resulted in the great- est learning gains. We analyzed the gaze behaviours to shed light on this finding and found that participants in this condi- tion focused significantly more on the problem solution, sug- gesting more cognitive processing during problem solving than in the other conditions.

A core proposition in economics is that voluntary exchanges benefit both parties. We show that people often deny the mutually beneficial nature of exchange, instead using zero-sum thinking. Participants read about simple exchanges of goods and services, judging whether each party to the transaction was better off or worse off afterwards. These studies revealed that zero-sum beliefs are pervasive. These beliefs seem to arise in part due to intuitive mercantilist beliefs that money has value over- and-above what it can purchase, since buyers are seen as less likely to benefit than sellers, and barters are often seen as failing to benefit either party (Study 1). Zero-sum beliefs are greatly reduced by giving reasons for the exchange (Study 2), suggesting that a second mechanism underlying zero-sum thinking is a failure to spontaneously take the perspective of the buyer. Implications for politics and business are discussed.

Mathematicians often describe arguments as “beautiful” or “dull,” and famous scientists have claimed that mathematical beauty is a guide toward the truth. Do laypeople, like mathematicians and scientists, perceive mathematics through an aesthetic lens? We show here that they do. Two studies asked people to rate the similarity of simple mathematical arguments to pieces of classical piano music (Study 1) or to landscape paintings (Study 2). In both cases, there was internal consensus about the pairings of arguments and artworks at greater than chance levels, particularly for visual art. There was also some evidence for correspondence to the aesthetic ratings of undergraduate mathematics students (Study 1) and of professional mathematicians (Studies 1 and 2).

How do people actively explore to learn about functional relationships, that is, how continuous inputs map onto continuous outputs? We introduce a novel paradigm to investigate information search in continuous, multi-feature function learning scenarios. Participants either actively selected or passively observed information to learn about an underlying linear function. We develop and compare different variants of rule-based (linear regression) and non-parametric (Gaussian process regression) active learning approaches to model participants' active learning behavior. Our results show that participants' performance is best described by a rule-based model that attempts to efficiently learn linear functions with a focus on high and uncertain outcomes. These results advance our understanding of how people actively search for information to learn about functional relations in the environment.

Temporal difference (TD) learning models can perform poorlywhen optimal policy cannot be determined solely by sensoryinput. Converging evidence from studies of working memorysuggest that humans form abstract mental representations thatalign with significant features of a task, allowing such condi-tions to be overcome. The n-task learning algorithm (nTL) ex-tends TD models by utilizing abstract representations to formmultiple policies based around a common set of external in-puts. These external inputs are combined conjunctively withan abstract input that comes to represent attention to a task.nTL is used to solve a dynamic categorization problem that ismarked by frequently alternating tasks. The correct number oftasks is learned, as well as when to switch from one task repre-sentation to another, even when inputs are identical across alltasks. Task performance is shown to be optimal only when anappropriate number of abstract representations is used.

Two common visual representations of fractions are circulararea models and the number line. The present studyexamined effects of these visual representations onacquisition of fraction knowledge. In Experiment 1,elementary school students learned aspects of fractionarithmetic with a visual representation or with standardsymbolic notation alone. Results found no advantage for theinclusion of a visual representation. In Experiment 2,elementary and middle students were tested on their ability torecognize, discriminate, and construct area models offractions and number line representations of fractions. Theresults show higher accuracy for area model questions thanfor number line representation questions. Taken togetherthese findings suggest that for fractions less than 1, simplearea models may have advantages over the number line forrecognition and discrimination of fractions representations.However, the incorporation of area models into instruction onfractions arithmetic provided no benefit over instruction withsymbolic notation alone.

Trials in a two-alternative forced-choice (2AFC) recognition-memory task require individuals to choose the stimulus in apair that they deem as having been previously studied. Be-cause of the relative nature of the judgments made, 2AFC tri-als are typically considered to be free from response biasesconcerning the old/new status of stimuli. Recent studies havesuggested that this assumption is incorrect, and individuals of-ten resort to single-stimulus old-new (ON) judgments instead.The present study tests this claim by joint modeling 2AFCand ON judgments using extended SDT models that includethe possibility of ON contamination. Results show that therelative-judgment assumption provides an excellent account ofthe data, providing no support for the notion of ON contami-nation in typical experimental designs.

Discrete choice models (e.g. logistic regression) are popular models in the economics literature that describe choices between twoor more discrete alternatives. These models have been successfully used to model value-based decisions, e.g. decisions in moraldilemmas, although temporal components of a decision, such as reaction times and changes of mind are not included. In cognitivesciences, another class of decision models, namely sequential-sampling models, has gained popularity in modelling choice accuracy,reaction time and decision uncertainty (e.g. confidence judgments). Here, we model decisions in moral dilemmas using a variant ofa hierarchical drift-diffusion model, factor drift diffusion, that combines the value-based approach with that of evidence accumulationmechanism by sequential-sampling. Specifically, we model the evidence accumulation process as resulting from a subjective weightingof abstract moral dimensions (factors). We train our model on a data set of 6500 moral decisions by 500 respondents on a popularweb platform (MoralMachine.mit.edu) and separately infer different sources of uncertainty in moral decisions. We show that the modelsuccessfully predicts reaction times and choices in moral dilemmas, while also leading to unexpected results

Eye-tracking provides an opportunity to generate and analyzehigh-density data relevant to understanding cognition. How-ever, while objects in the real world are often dynamic, eye-tracking paradigms are typically limited to assessing gaze to-ward static objects. In this study, we propose a generativeframework, based on a hidden Markov model, for using eye-tracking data to analyze behavior in the context of multiplemoving objects of interest. We apply this framework to ana-lyze data from a recent visual object tracking task paradigm,TrackIt, for studying selective sustained attention in children.We also present a novel ‘supervised’ variant of TrackIt that weuse to tune and validate our model, while providing insightsinto the visual object tracking abilities of children and adults

Current causal theories argue that the statistical normality or abnormality of an action makes a difference to people’s causal judgements. In this paper, we present two experiments that explore the role of statistical norms in causal cognition. In our first experiment, we provide a preliminary test of two competing theories that aim to explain the effects of normality in causal cognition – the actual causal strength measure (Icard Kominsky & Knobe, 2017) and the correspondence hypothesis about causal judgements (Harinen, 2017). In addition, we control for an often neglected factor, the epistemic states of agents. Our second experiment investigates the effect of statistical normality in the same context, but with a probabilistic rather than deterministic causal structure. Our results favour Icard et al.’s (2017) model of causal strength, but show that the statistical normality of an action loses its influence when the occurrence of the outcome is probabilistic. We discuss the implications of our findings for current causal theories

If an object A moves until it is adjacent with a stationary objectB, at which point object A stops and object B begins moving,adults and infants 6 months of age and older perceive that Acaused B to move. These “launching” events correspond toreal-world collisions, which are governed by Newtonianmechanics. Previous work showed that infants were sensitiveto Newtonian constraints on relative speed. Here, we show thatinfant causal perception is sensitive to other physicalconstraints on collision events as well. Infants habituated to alaunching event will dishabituate to an event in which object Bmoves at a 90° angle relative to object A, but not to a rotatedversion of the launching event. This selective dishabituationwas not found for non-causal events. The results suggest thatearly-developing causal perception is sensitive to the manyphysical principles of real-world collision events

Certain events are irresistibly perceived as involving cause and effect. The prototypical exampleis the 'launching' effect, wherein one object (A) moves toward a stationary second object (B)until they are adjacent, at which point A stops and B starts moving in the same direction. Butthere are up to a dozen different events that have been studied under the umbrella of 'causalperception'. However, these events have typically been distinguished only using explicit self-report methods, and little work has explored whether these different event labels actually capturenatural "joints" in visual processing. Here, we use the psychophysical phenomenon ofretinotopically specific visual adaptation to demonstrate that launching events and 'triggering'events (in which B moves much faster than A) involve the same underlying form of 'causality' invisual processing, but launching events and 'entraining' events (in which A and B move togetherfollowing A's arrival) do not

Though communicative goals clearly drive word choice in language production, online demands suggest that accessibility might play a role, too. If the benefits of accessibility are important enough to communication, more accessible words (high-frequency words) might be chosen over more accurate, less accessible ones. We used a novel artificial language learning paradigm to test whether high-frequency words are preferred over low-frequency words at a cost of meaning accuracy. Participants learned eight words which corresponded to precise angles on a compass. On test trials, participants viewed angles lying in-between two trained angles and were asked to produce a word for the angle. Across two experiments, we showed that participants extended their use of high- frequency words to more distal angles compared to low- frequency words. In cases of competition between high- and low-frequency words, the former tended to win out even when less accurate, suggesting that accessibility can compromise some accuracy.

Certain generalizations are teleological, e.g., forks are foreating. But not all properties relevant to a particular conceptpermit teleological generalization. For instance, forks getwashed roughly as often as they’re used for eating, yet thegeneralization, forks are for washing, might strike reasoners asunacceptable. What explains the discrepancy? A recenttaxonomic theory of conceptual generalization (Prasada, 2017;Prasada & Dillingham, 2006; Prasada et al., 2013) argues thatcertain kinds of conceptual connections – known as“principled” connections – license generalizations, whereasassociative, “statistical” connections license only probabilisticexpectations. We apply this taxonomy to explain teleologicalgeneralization: it predicts that acceptable teleologicalgeneralizations concern concept-property pairs in which theconcept bears a principled connection to a property. Under thisanalysis, the concept fork bears a principled connection toeating and a statistical connection to washing. Twoexperiments and a regression analysis tested and corroboratedthe predictions of the theory.

Reasoning and other mental operations are believed to rely onmental models. Arguments have been made that mental mod-els share representational substrate with perception. Here, wedemonstrate that a neural dynamic architecture that perceptu-ally grounds language may also support the building of men-tal models. Supplied with a sequence of simple premises thatspecify the colors of object pairs as well as their spatial rela-tion, the architecture builds a mental model of the describedscene. We show how the neural processes of the architec-ture evolve in response to both determinate and indeterminatepremises. For indeterminate premises, we demonstrate thatthe preferred mental models observed in human participantsemerge from the underlying neural dynamics.

The present research infers aspects of spatial attention frommovement to targets (and preferably not to foils) of a mouse-controlled cursor on a computer monitor. The long-term goalis a data-rich and rapid assessment technique that can be usedto diagnose individual and clinical deficits of attention. Theaim of this present research is validating the approach usinga college population of subjects. In the experiment, partici-pants attempt to move a cursor toward three spatial positions atwhich targets appear rapidly but at irregular times, and attemptto inhibit movements toward foils appearing at those positions.We assume that cursor movements toward a position indicatesattention has been directed toward that position. A modifiedHidden Markov Model (HMM) uses five sources of evidence,all based on parameters to be estimated, to predict the timevarying movement of attention: four aspects of cursor move-ment and a probability that attention will move from one timeinterval to the next. Five minutes of data are used to estimateparameters for each subject that produce a predicted attentiontrajectory which best matches what the subject is instructed todo. These parameters are used to predict the attention trajec-tory for the remainder of the hour of testing. The predictionsof attention movements are then matched to the appearance oftargets and foils to infer such components of attention as abil-ity to respond to targets vs foils, times to do so, and changesin these components over time. The results illustrate a promis-ing approach to assessment of attention that could likely beemployed for both adults and children in clinical settings re-quiring short testing periods.

Most models of pricing embody a static, deterministic theoryof value where the monetary amount people assign to an itemis computed as a fixed function of its attributes. Preference re-versals — where prices assigned to gambles conflict with pref-erence orders elicited through binary choices – indicate thatthe response processes going into value assessments are impor-tant. In this paper, we additionally show that price responsesare sensitive to time pressure, suggesting a dynamic underlyingcognitive process. We also show that the elicited price distribu-tions can possess strong positive or negative skew, indicatingthat diverging information is used to generate buying versusselling and certainty equivalent prices. We develop a computa-tional cognitive model that predicts these continuous distribu-tions of price responses and how they change over time, show-ing that it can account for the major dynamic and distributionalproperties of prices and decisions.

Cognitive modelling uses computer models to investigate psy-chological theories. To conclude from executions of a cogni-tive model to the theory, the model needs to be a correct im-plementation of the theory since a defective cognitive modelmay yield wrong statistical figures. We consider three commonreasons for a model to be incorrect wrt. a theory: situationswhich unintentionally do not enable any production rule, ruleswhich erroneously construct undesired declarative knowledge,and wrongly chosen architecture parameters. Defects of thesekinds are hard to detect since repeated execution and observa-tion of the model does not guarantee to uncover these defects.In this work, we give formal definitions of the three kinds ofdefects in terms of an existing abstract formal semantics of thehybrid architecture ACT-R. We demonstrate the application offormal analysis techniques to ACT-R models to reliably detectthe considered defects and to thereby increase the confidencethat the model behaves according to the psychological theory.

Causal systems often include mutually exclusive events: events which cannot occur simultaneously. However, when events in a causal system are exclusive, the normative properties of the whole system change substantially. Are adults sensitive to the consequences of exclusivity for causal reasoning? Here, we systematically manipulated common-effect causal systems to have either exclusive or non-exclusive causes while holding all other factors constant. Adults showed a rich understanding of exclusive systems in making both predictive (Experiment 1) and diagnostic (Experiments 2 and 3) causal inferences. Adults’ success in these tasks suggests that exclusivity is an important dimension in human causal reasoning.

Cognitive control refers to the ability to adjust our thoughts andbehaviors in order to achieve internalized goals. In the past,researchers have proposed several models of cognitive controlto account for the characteristic patterns of response timesobserved in the tasks (e.g., Botvinick, Braver, Barch, Carter, &Cohen, 2001). The goal of this study is to evaluate empiricalvalidity of such models in an experiment. To that end, wecompared two models of cognitive control, the conflictmonitoring model and the expectancy-based model. Eachmodel was implemented in two different modelingframeworks, neural networks and simple linear models. Therelative fits of the four models were then evaluated andcompared based on observed data from a flanker taskexperiment. The model comparison results showed thatperformance of the simple linear models was entirelycomparable to that of the neural network models. We alsoconstructed and fitted hierarchical Bayesian latent mixtureversions of the linear models to investigate individualdifferences. The result suggests that no single model ofcognitive control, whether conflict monitoring or expectancy-based, would be able to account for individual performance onthe task.

We created practical moral dilemmas for which participantsrole-played witnessing a transgression by a target person. Theidentity of the transgressor was manipulated to be either astranger or the participant’s brother. Participants made factualand unethicality judgments regarding the incident andreported their willingness to report the transgressor to thepolice. When the factual situation was ambiguous,participants interpreted the facts in favor of the target personwhen that target was their brother. This family favoritism inturn led to partial moral judgments and decisions, whilecreating overall coherence. When it was made clear that theirbrother actually committed the transgression, partiality inunethicality judgment was reduced but partiality in thedecision to report persisted, even though overall coherencewas thereby reduced. Using path analyses, we show howstrong moral constraints such as family obligation can shiftmoral reasoning processes.

One major class of approaches to explaining the distribu-tion of linguistic forms is rooted in communicative effi-ciency. For theories in which an utterance’s communica-tive efficiency is itself dependent on the distribution oflinguistic forms in the language, however, it is less clearhow to make distributional predictions that escape circu-larity. I propose an approach for these cases that involvesiterating between speaker and listener in the RationalSpeech Act theory. Characteristics of the fixed points ofthis iterative process constitute the distributional predic-tions of the theory. Through computer simulation I applythis approach to the well-studied case of predictability-sensitive optional function word omission for the theoryof Uniform Information Density, and show that the ap-proach strongly predicts the empirically observed nega-tive correlation between phrase onset probability and rateof function word use.

Determining the correct form of a verb in context requires anunderstanding of the syntactic structure of the sentence. Re-current neural networks have been shown to perform this taskwith an error rate comparable to humans, despite the fact thatthey are not designed with explicit syntactic representations.To examine the extent to which the syntactic representationsof these networks are similar to those used by humans whenprocessing sentences, we compare the detailed pattern of er-rors that RNNs and humans make on this task. Despite signif-icant similarities (attraction errors, asymmetry between singu-lar and plural subjects), the error patterns differed in importantways. In particular, in complex sentences with relative clauseserror rates increased in RNNs but decreased in humans. Fur-thermore, RNNs showed a cumulative effect of attractors buthumans did not. We conclude that at least in some respects thesyntactic representations acquired by RNNs are fundamentallydifferent from those used by humans.

Much of human learning throughout the lifespan is achievedthrough seeking and generating explanations. However, verylittle is known about what triggers a learner to seek anexplanation. In two studies, we investigate what makes a givenevent or phenomenon stand in need of explanation. In Study 1,we show that a learner’s judgment of “need for explanation”for a given question predicts that learner’s likelihood ofseeking an answer to this question. In Study 2, we exploreseveral potential predictors of need for explanation. We findthat the need for explanation is greater for questions expectedto have useful answers that require expert understanding, andthat “need for explanation” can be differentiated from generalcuriosity.

Errors are common during learning, but what factorsinfluence whether those errors are corrected? Evidencesuggests error generation and memory for errors may be twoimportant factors. Middle-school children studied and weretested on their memory for math definitions. After receivingcorrect answer feedback, children recalled their initial testanswers before taking a final test. Memory for errors anderror correction rates were higher for errors that weregenerated compared to errors that were chosen from a list.Further, memory for errors was positively correlated witherror correction, even after controlling for age, grade, andmath and reading skills. However, this relationship was onlypresent for errors that were generated and not for errors thatwere chosen from a list. These findings suggest retrievalplays an important role in the relationship between memoryfor errors and error correction.

How do children’s representations of object categories changeas they grow older? As they learn about the world aroundthem, they also express what they know in the drawings theymake. Here, we examine drawings as a window into how chil-dren represent familiar object categories, and how this changesacross childhood. We asked children (age 3-10 years) to drawfamiliar object categories on an iPad. First, we analyzed theirsemantic content, finding large and consistent gains in howwell children could produce drawings that are recognizable toadults. Second, we quantified their perceptual similarity toadult drawings using a pre-trained deep convolutional neuralnetwork, allowing us to visualize the representational layoutof object categories across age groups using a common featurebasis. We found that the organization of object categories inolder children’s drawings were more similar to that of adultsthan younger children’s drawings. This correspondence wasstrong in the final layers of the neural network, showing thatolder children’s drawings tend to capture the perceptual fea-tures critical for adult recognition. We hypothesize that thisimprovement reflects increasing convergence between chil-dren’s representations of object categories and that of adults;future work will examine how these age-related changes re-late to children’s developing perceptual and motor capacities.Broadly, these findings point to drawing as a rich source ofinsight into how children represent object concepts.

Language comprehension in grounded, social contexts in-volves integrating information from both the visual and the lin-guistic signals. But how should listeners prioritize these differ-ent information sources? Here, we test the hypothesis that evenyoung listeners flexibly adapt the dynamics of their gaze toseek higher value visual information when the auditory signalis less reliable. We measured the timing and accuracy of adults(n=31) and 3-5 year-old children’s (n=39) eye movements dur-ing a real-time language comprehension task. Both age groupsdelayed the timing of gaze shifts away from a speaker’s facewhen processing speech in a noisy environment. This delayresulted in listeners gathering more information from the vi-sual signal, more accurate gaze shifts, and fewer random eyemovements to the rest of the visual world. These results pro-vide evidence that even young listeners adjust to the demandsof different processing contexts by seeking out visual informa-tion that supports language comprehension.

The paper explores how people revise their belief in ahypothesis and the reliability of sources given independenceof sources or partial dependence (e.g. the sources share abackground). Specifically, we test a formal model ofreliability revision.The study provides support for Bovens and Hartmann’smodel of reliability revision. If a source provides a positivereport for an unlikely hypothesis, participants initially revisethe reliability of the source negatively. However, as additionalpositive reports emerge, they increase their estimate of thereliability of the source. Further, if it becomes known thesources are partially dependent (here, taught the same schoolof thought), the reliability of the source decreases again. Bothof these findings are in line with the Bayesian approach toreliability revision.

This study investigated the delegation of tasks to a partner incooperation with a human partner and with an automation sys-tem as a system partner. In the experiment, a line-tracing taskwas used, in which the performance in the task of the partic-ipants and their partners was dynamically altered at multiplelevels. The participants were informed that their task partnerswere human (human condition) or automation system (systemcondition). However, in reality, all participants performed theirtask with an automation system. The results showed that a re-lationship between subjective trust in the partner and the per-centage of the task delegated to the partner was found only inthe system condition but not in the human condition. More-over, sensitivity to change in the task performance of the par-ticipants and their partners was higher, and the suitability oftask delegation was greater in the system condition than in thehuman condition. These results were discussed based on theprevious studies.

The Time-Invariant String Kernel (TISK) model of spokenword recognition (Hanngan et al., 2013) is an interactiveactivation model like TRACE (McClelland & Elman, 1986).However, it uses orders of magnitude fewer nodes andconnections because it replaces TRACE's time-specificduplicates of phoneme and word nodes with time-invariantnodes based on a string kernel representation (essentially aphoneme-by-phoneme matrix, where a word is encoded as byall ordered open diphones it contains; e.g., cat has /kæ/, /æt/,and /kt/). Hannagan et al. (2013) showed that TISK behavessimilarly to TRACE in the time course of phonologicalcompetition and even word-specific recognition times.However, the original implementation did not includefeedback from words to diphone nodes, precluding simulationof top-down effects. Here, we demonstrate that TISK can beeasily adapted to lexical feedback, affording simulation oftop-down effects as well as allowing the model todemonstrate graceful degradation given noisy inputs.

Large-scale societies are impossible to perceive directly.Unsurprisingly, lay demographic estimates are wildlyinaccurate. How should we interpret these errors? Mostaccounts assume these errors are evidence of topic-specificbiases and prejudices. (e.g., “People overestimateimmigration because immigrants threaten the status quo.”)But this glosses over the distortions that are introducedwhenever underlying perceptions are translated into explicitnumerical estimates. For instance, estimates are typicallyhedged, or ‘rescaled,’ toward an expected value — aperfectly rational strategy when information is uncertain.We show that uncertainty-based rescaling accounts for mosterror in individual demographic estimates. Residual errorswere not even always in the same direction; populations thatappeared to have been over-estimated (e.g., Asian-Americans) now appear to be under-estimated. The amountof rescaling engaged in by an individual was proportional totheir uncertainty (about politics or about numbers).Perceptions of society are surprisingly good; thepsychophysics of estimation gets in the way.

We present a general framework for capturing categorical cross-linguistic transfer effects – the influences of linguistic and con-ceptual categories in a bilingual speaker’s languages on eachother. By formulating the phenomenon as an instance of cogni-tive category shift, we achieve a general method for investigat-ing the extent and causes of crosslinguistic transfer in terms ofa category similarity space and a set of weighting factors. Weapply the model to the well-understood domain of color, formu-lating transfer as the modulation of conceptual color categoriesin one language on those of the other language. We analyze thecomponents of the model that predict salient aspects of humandata on an observed transfer effect in a range of languages.

Human free association (FA) norms are believed to reflect thestrength of links between words in the lexicon of an averagespeaker. Large-scale FA norms are commonly used as a datasource both in psycholinguistics and in computational mod-eling. However, few studies aim to analyze FA norms them-selves, and it is not known what are the most important factorsthat guide speakers’ lexical choices in the FA task. Here, wefirst provide a statistical analysis of a large-scale data set ofEnglish FA norms. Second, we argue that such analysis caninform existing computational models of semantic memory,and present a case study with the topic model to support thisclaim. Based on our analysis, we provide the topic model withdictionary-based knowledge about word synonymy/antonymy,and demonstrate that the resulting model predicts human FAresponses better than the topic model without this information.

Our actions and decisions are regularly influenced by the socialenvironment around us. Can social environment be leveragedto induce curiosity and facilitate subsequent learning? Acrosstwo experiments, we show that curiosity is contagious: socialenvironment can influence people’s curiosity about the answersto scientific questions. Our findings show that people are morelikely to become curious about the answers to more popularquestions, which in turn influences the information they chooseto reveal. Given that curiosity has been linked to better learning,these findings have important implications for education.

Like scientists, children have a sharp sense of when and howto seek evidence, but when it comes to generating causal in-terventions, their performance often falls short of normativeinformation-theoretic metrics such as the expected informationgain (EIG). We looked at whether this deviation resulted frommixing discriminatory strategies such as maximizing EIG withconfirmatory strategies such as the positive test strategy (PTS).Thirty-nine 5- to -7-year-olds solved 6 puzzles where they hadone opportunity to intervene on a three-node causal system toidentify the correct structure from two possibilities. Children’sintervention choices were better fit by a Bayesian model thatincorporated EIG and PTS compared to alternative models thatonly considered a single strategy or selected interventions atrandom. Our findings suggest that children’s intervention strat-egy may be a combination of discrimination and confirmation.

Recent work suggests that optimal spacing in learningrequires adaptive procedures (Mettler, Massey & Kellman,2016). Here, we studied how adaptive techniques might befurther enhanced by combining active and passive learningmodes. Participants learned geography facts that werescheduled using the ARTS (Adaptive Reaction-Time-basedScheduling) system under four conditions involving passiveand/or active trials. Conditions included: a) Passive Onlypresentations of learning items, b) Passive Initial Blocksfollowed by active adaptive scheduling, c) Passive InitialItems followed by active adaptive scheduling for each itemintroduced, or d) Active Only learning with no passivepresentations. We found an advantage for combinations ofactive and passive presentation (by blocks or items) overPassive Only or Active Only presentation. Passive trialspresented in blocks at the beginning of learning showed bestperformance. We discuss possible explanations for thesedifferences and suggest principles underlying optimalcombinations of active and passive modes in adaptivelearning.

We introduce a new model of adaptive criterion setting withina signal detection framework, and show how this provides psy-chological insights that allow us to segregate causes of subop-timality in perceptual learning. We apply this to a perceptuallearning task for both neurotypical and autistic participants.The model parameters provide a bridge between the mecha-nisms of an aberrant precision account of autism and result-ing behavior that can be interpreted within a receiver operatingcharacteristic framework. The model makes superior out-of-sample predictions compared to standard signal detection the-ory, about how people adapt to different environmental manip-ulations when asked to categorize audio-spatial stimuli. Wefind that accuracy of participants is more strongly correlated tothe construct of persistence signals that inhibit response flexi-bility, than to the neuromodulatory gain. We also find evidencefor individual differences in persistence that are correlated toscores on the autistic traits questionnaire.

In this paper, we highlight the shortfall of conventionally de-scribed heuristics in multi-attribute decision theory, and pro-pose recasting these heuristics within a novel probabilisticframework. This framework is based on defining a psycho-logical feature space, with rule-based heuristics represented asprototypical representations within this space. We provide var-ious examples of meaningful heuristics that can be constructedunder this representation, including recasting probabilistic ver-sions of popular heuristics such as take-the-best. Next, wepropose an evaluation framework to measure the effectivenessof a consideration set of heuristics. This framework measureswhether the set of heuristics are sufficient to describe, predictand infer strategy selection and learning behavior. We proposethat this is a step towards a robust framework within whichmodels of strategy selection and learning should be evaluated.The framework aspires to develop a consideration set of heuris-tics that can be represented as a mathematically well-posed in-ference problem. We show that the heuristics redefined underour probabilistic framework generally perform better than con-ventional heuristics under this evaluation. We conclude with adiscussion on the possible applications of this framework.

Consensus seeking – abandoning one’s own judgment to alignwith a group majority – is a fundamental feature of humansocial interaction. Notably, such striving for majorityaffiliation often occurs in the absence of any apparenteconomic or social gain, suggesting that achieving consensusmight have intrinsic value. Here, we examine the affectiveproperties of consensus decisions by assessing the transfer ofvalence to concomitant stimuli. Specifically, in two studies,we show that contexts repeatedly paired with consensusdecisions are rated as more likable, and selected morefrequently in a two-alternative forced choice test, than arecontexts repeatedly paired with dissent from a unanimousmajority. In the second study, we rule out inferences aboutthe accuracy of the majority opinion as the basis for suchevaluative changes. Our results suggest that an intrinsic valueof consensus, or cost of dissent, may motivate and reinforcesocial conformity.

Real-word decision making often involves selecting a singlechoice from an arbitrarily large set of possible options. Giventhat it is typically not feasible to evaluate every possible op-tion in real world decision making, how are human decisionmakers able to efficiently make good decisions? We proposeand evaluate a two-step architecture according to which peoplefirst sample a small subset of options weighted by their previ-ously learned value, and then evaluate those options within thecurrent decision-making context. We demonstrate that a ver-sion of this model captures human decision making in prob-lems where time and resource constraints prevent the evalua-tion of every option, and connect this research to the growingliterature on the representation of non-actual possibilities.

Cognitive control is critical for accomplishing daily tasks andyet we experience it as effortful or costly. Researchers havebeen increasingly interested in estimating how costly cognitivecontrol is for a given individual, to better understand underly-ing mechanisms and predict motivational impairments outsidethe lab. Here we leverage a computational model of controlallocation to (a) demonstrate a procedure for estimating indi-vidual’s control costs from task performance and (b) highlightthe conditions under which estimated costs will be confoundedwith other motivational variables. We show that costs of cog-nitive control can be reliably estimated under perfect assump-tions about other motivational variables. However, our simu-lation results indicate that poorly calibrated estimates of thoseother variables can lead to potentially drastic misestimations ofsubjects’ control costs, compromising the validity of empiricalobservations. We conclude by discussing the implications ofthese analyses for assessing individual differences in the costsof cognitive control.

One of the most compelling characteristics of controlled pro-cessing is our limitation to exercise it. Theories of control allo-cation account for such limitations by assuming a cost of con-trol that constrains how much cognitive control is allocated toa task. However, this leaves open the question of why sucha cost would exist in the first place. Here, we use neural net-work simulations to test the hypothesis that constraints on cog-nitive control may reflect an optimal solution to the stability-flexibility dilemma: allocating more control to a task results ingreater activation of its neural representation but also in greaterpersistence of this activity upon switching to a new task, yield-ing switch costs. We demonstrate that constraints on controlimpair performance of any given task but reduce performancecosts associated with task switches. Critically, we show thatoptimal control constraints are higher in environments with ahigher probability of task switches.

Studies in distributed cognition (d-cog) almost exclusively focus on human-centered technological systems, such as ships, aircraft, automobiles, scientific and medical institutions, human-computer interfaces, and transactive memory systems. First, we review the literature and claim that d-cog is species-neutral. We then propose three experimentally operationalizable, necessary, and jointly- sufficient criteria for identifying d-cog: task orientation, interaction dominance, and agency. Here we build on previous research on nonhuman intraspecies d-cog by presenting human-dog systems as cases of interspecies d-cog. Domestic dogs’ (Canis familiaris) unique working relationships with humans allow for interspecies coordination and synchronization. Contrasting them with wolves (Canis lupus) and dingoes (Canis dingo), we suggest evolutionary history plays an important role in determining whether different species can form interspecies d-cog systems.

Students who explore a new concept prior to receiving direct instruction often demonstrate better conceptual understanding compared to traditional tell-then-practice methods. Often, exploratory learning activities have students invent solutions to a novel problem targeting the new concept. However, exploring prior to instruction is working memory demanding, inducing high cognitive load. The current experiments varied the guidance provided during exploration and examined subsequent learning. In Experiment 1, participants explored the procedures and concept of statistical variance prior to receiving instruction in one of three conditions: invention, completion problem, or worked example. Exploring using a worked example led to the highest learning outcomes and the least cognitive load. In Experiment 2, students in an undergraduate statistics class completed invention or worked example problems either before or after instruction. Learning was greater when problem solving preceded instruction. However, exploring using a worked example did not improve learning over the more cognitively-demanding invention problem. These findings demonstrate the benefits of exploratory learning in the classroom compared to more traditional tell-then-practice approaches. However, more research is needed to determine when and how guidance will enhance exploration.

Native adult speakers of a language can produce grammatical sentences fluently, effortlessly, and with relatively few errors. These characteristics make the highly-practiced task of speaking a viable candidate for an automatic process, i.e., one independent of cognitive control. However, recent studies have suggested that some aspects of production, such as lexical retrieval and tailoring speech to an addressee, may depend on the speaker’s inhibitory control abilities. Less clear is the dependence of syntactic operations on inhibitory control processes. Using both a direct manipulation of inhibitory control demands and an analysis of individual differences, we show that one of the most common syntactic operations, producing the correct subject-verb agreement, requires inhibitory control when a singular subject noun competes with a plural local noun as in “The snake next to the purple elephants is green.” This finding calls for the integration of inhibitory control mechanisms into models of agreement production, and more generally into theories of syntactic production.

We provide a new computational model of working memory inthe complex span task implemented in the ARCADIA cogni-tive framework. While there exist implementations of workingmemory successful enough to account for many of the bench-mark findings in the working memory literature, we demon-strate that further progress requires the integration of thesemodels with a rich conception of attention. ARCADIA pro-vides this intersection, allowing for precise control of the focusof attention on a time scale fine enough to begin to disentan-gle the overlapping effects of interference, temporal decay, andattentional refreshing.

When learning about the world, we develop mental represen-tations or concepts for things we have never seen. At the sametime, we also develop representations for things that are similarto what we have experienced. Traditionally, similarity-basedand rule-based systems have been used as distinct models tocapture conceptual representation. However, it seems implau-sible that we do not flexibly deploy both systems. Whetherboth systems can be used simultaneously to represent compo-nents of a single concept is an open empirical question. Oneexample suggesting that the use of both systems is possible isthe concept of a ZEBRA , which looks like a horse but striped.Using an artificial concept learning task, we test whether peo-ple can combine similarity and rules compositionally in orderto represent concepts. Our results suggest that people are ableto compose similarity and rules when mentally representing asingle concept.

Trust is central to social behavior. In interactions between strangers some information about group affiliation is almost always available. Despite this, how group information is utilized to promote trust in interactions between strangers is poorly understood. Here we addressed this through a two-stage experiment where participants interacted with randomly selected members of two arbitrary groups and learnt their relative trustworthiness. Next, they interacted with four novel individuals from these two groups. Two members, one from each group, acted congruently with their group’s previous behavior while the other two acted incongruently. While participants readily learnt the group-level information in the first phase, this was swiftly discounted in favor of information about each individual partner’s actual behavior. We fit a reinforcement learning model which included a bias term capturing propensity to trust to the data from the first phase. The bias term from the RL model predicted participants’ initial behavior better than their expectations based on group membership. Pro-social tendencies and individuating information can overcome knowledge about group belonging.

The quiet hum of interpersonal coordination that runs through-out social communication and interaction shows how individu-als can subtly influence one another’s behaviors, thoughts, andemotions over time. While the majority of research on co-ordination studies face-to-face interaction, recent advances incrowdsourcing afford the opportunity to conduct large-scale,real-time social interaction experiments. We take advantageof these tools to explore interpersonal coordination in a “min-imally interactive context,” distilling the richness of naturalcommunication into a tightly controlled setting to explore howpeople become coupled in their perceptual and memory sys-tems while performing a task together. Consistent with previ-ous work on postural sway and gaze, we found that individualsbecome coupled to one another’s cognitive processes withoutneeding to be co-located or fully interactive with their partner;interestingly, although participants had no information abouttheir partner and no means of direct communication, we alsofound hints that social forces can shape minimally interactivecontexts, similar to effects observed in face-to-face interaction.

Across four experiments, we look at whether adults andchildren can represent the amount of information needed todistinguish different populations in the context of an intuitivestatistical reasoning task requiring metacognitive monitoringand control. Consistent with a ground truth model ofinformation gain, adults (N=60) modulated their informationgathering with respect to the difficulty of the discriminationproblem. Adults also adjusted their confidence thresholddepending on task difficulty, allowing for more uncertainjudgments when the discrimination was more difficult orgathering data was more costly (Experiments 1 and 2). In asimplified version of the task, children (N = 42, M = 7.3years, range: 5.0-9.0) were also able to distinguish easy anddifficult discrimination problems and judge that they neededmore information to solve harder problems (Experiments 3and 4).

Much of what we know comes from other people, and thequantity of information provided is often constrained by timeor space. For a communicator, what information they chooseto convey depends not just on the nature of their topic, but alsoon the social inferences their listeners will make about thembased on what they say. For the listener, their interpretation ofinformation given to them depends not just on the informationitself, but also on what inferences they make about the bias andmotivations of the communicator they received it from. In thispaper we explore how and whether these social factors interactwith the “true” nature of the information being communicated.We find that stronger evidence does not always lead to strongerconclusions and often leads to increased perceived bias. Com-municators, perhaps for this reason and perhaps for others, of-ten modulate the evidence they present to be less unanimousthan warranted. This has implications for real-world situations,like communicating about climate change: in such situations,both communicators and listeners behave in what may be indi-vidually rational ways, but the end result is that the underlyingtruth gets distorted.

Real-world decisions often involve “black swan” choices withextremely low probability chances of catastrophic loss, likeriding a motorcycle or going on a dangerous trip. These haveseveral characteristics that make them especially difficult fromthe perspective of decision theory. How do people assign util-ities to losses like “go bankrupt” or “die”? Do people havethe representational resolution to encode differences betweenextremely tiny probabilities? We address these questions in twoexperiments in which people make decisions involving very lowprobabilities (as low as 1 in 10,000) of losing all of their points(and monetary bonus). Our results indicate that people mostlyappear not to encode differences between tiny probabilities andare indifferent to the magnitude of losses. These factors lead toa startling qualitative shift in behaviour between scenarios withthe same expected value and very similar absolute risk levels:people are risk averse when only one option is a black swan butbecome strongly risk seeking when both are.

Understanding how people represent categories is a core prob-lem in cognitive science. Decades of research have yieldeda variety of formal theories of categories, but validating themwith naturalistic stimuli is difficult. The challenge is that hu-man category representations cannot be directly observed andrunning informative experiments with naturalistic stimuli suchas images requires a workable representation of these stimuli.Deep neural networks have recently been successful in solvinga range of computer vision tasks and provide a way to com-pactly represent image features. Here, we introduce a methodto estimate the structure of human categories that combinesideas from cognitive science and machine learning, blendinghuman-based algorithms with state-of-the-art deep image gen-erators. We provide qualitative and quantitative results as aproof-of-concept for the method’s feasibility. Samples drawnfrom human distributions rival those from state-of-the-art gen-erative models in quality and outperform alternative methodsfor estimating the structure of human categories.

When reasoning about evidence under conditions ofuncertainty, one important consideration for accurate updatingis the presence (and influence) of dependencies. For instance,if considering whether a patient has a disease, the value oftwo doctors’ diagnoses indicating the presence of the diseasemay carry more value if such diagnoses were conductedindependently, rather than if, all else being equal, one doctorhas seen the other’s diagnosis before making their own. In thepresent paper, we demonstrate that lay reasoners prefer toavoid dependencies when considering evidential support.However, we additionally illustrate two cases in whichdependencies may carry evidential advantage: namely, wheninformation is partial or contradictory. Lay reasonerserroneously remain averse to dependencies even in suchcases, reflecting the difficulties inherent to considerations ofdependence.

People use space to conceptualize abstract domains like timeand number. This tendency may be a cognitive universal, butthe specifics of people’s implicit space-time and space-numberassociations vary across cultures. How does culture shape ourabstract concepts? In Western cultures, both time and numbersare arranged in people’s minds along an imaginary horizontalline, from left to right, but in other cultures the directions of themental timeline (MTL) and mental number line (MNL) arereversed. The directions of both the MTL and MNL have longbeen assumed to depend on the direction in which people readand write text. Here we argue that this assumption is false, andshow that the MTL and MNL are shaped by different aspectsof cultural experience. In a training experiment, participantsspatialized time and numbers in opposite directions across theirfingers. Training changed the MTL and MNL in oppositedirections, as predicted by a general principle called theCORrelations in Experience (CORE) principle: peoplespatialize abstract conceptual domains in their minds accordingto the ways these domains are spatialized in their experience.

The learning and generalization of patterns is an importantaspect of mathematical thinking, such that the ability toidentify and use patterns early in development predicts futuresuccess in algebra and math. Thus, understanding thecritical factors that facilitate this relational knowledge isimportant for the development of instructional materials andfor curriculum development. The aim of the present studywas to examine the factors that facilitate the learning andtransfer of pattern knowledge. In two experiments, 4- to 6-year-old children participated in a pre-post test design, inwhich they received training on novel patterns. Critically, wemanipulated (1) the language with which children wereexposed to novel patterns during training and (2) theperceptual format in which children were exposed to novelpatterns. We find that 4-6 year old children were able tolearn about novel patterns following this intervention, butfaired best when trained on abstract (“A-B-A”) rather thanconcrete (“red-blue-red”) labels. Furthermore, the extent towhich the training stimuli were grounded in visualrepresentations affected both learning and generalization ofthis newly acquired pattern knowledge. This work hasimplications for instructional design and curriculumdevelopment in the classroom.

In bilingual language environments, children learn twolanguages in the same amount of time that monolingualchildren learn one, and children do not learn their twolanguages at exactly the same rate. Furthermore, learning twolanguages requires children to deal with challenges not foundin monolingual input, notably the use of two languages withinone utterance (Do you like the perro?/¿Te gusta el doggy?).For bilinguals of all ages, switching between languages canimpede processing efficiency. But are all switches equallychallenging? We tested Spanish-English bilingual toddlers’processing of single-language and mixed-language sentencesin both languages. We found asymmetrical switch costs whentoddlers were tested in their dominant vs. non-dominantlanguage, and toddlers benefited from hearing nounsproduced in their dominant language. These results suggest animportant commonality between monolingualism andbilingualism: when toddlers have more robust representationsof a particular item, they can better recognize it in diversecontexts.

How can we leverage the cognitive science of lay theories toinform interventions aimed at correcting misconceptions andchanging behaviors? Focusing on the problem of vaccine skep-ticism, we identified a set of 14 beliefs we hypothesized wouldbe relevant to vaccination decisions. We developed reliablescales to measure these beliefs across a large sample of partici-pants (n = 1130) and employed state-of-the-art graphical struc-ture learning algorithms to uncover the relationships amongthese beliefs. This resulted in a graphical model describingthe system of beliefs relevant to childhood vaccinations, withbeliefs represented as nodes and their interconnections as di-rected edges. This model sheds light on how these beliefs re-late to one another and can be used to predict how interventionsaimed at specific beliefs will play out across the larger system.Moving forward, we hope this modeling approach will helpguide the development of effective, theory-based interventionspromoting childhood vaccination.

We evaluate how learners construct internal representationsof fraction values. Symbolic numbers written using fractionnotation are difficult for both children and adults to use.Errors made by learners suggest that even experienced adultscan lack fluency with fractions. One such error is the NaturalNumber bias phenomenon: when the relative size of fractionsvalues to be compared is incongruent with the relative size ofthe fraction components learners show a reaction time delayor decreased accuracy. For example, noting that 1/7 is smallerthan 1/5 may take longer that noting that 3/10 is smaller than5/10. We adjust the temporal dynamics of the fractioncomparison task to characterize how learners constructfraction values from the constituent parts. We also create amathematical model of the fraction value construction.

Language is often depicted as the sine qua non of mathematicalthinking, a view buttressed by findings of language-of-trainingeffects among bilinguals. These findings, however, have beenlimited to studies of arithmetic. Nothing is known about thepotential influence of language on the ability to learn rulesabout the relations among variables (e.g., algebra). To testwhether arithmetic and algebraic thinking differ, Chinese-English bilinguals were trained to solve arithmetic and algebraproblems in either Chinese or English and then tested on newand old problems in both languages. For arithmetic problems,solution times were always longer for English than Chinese; inboth languages, solution times dropped during training; aftertraining, solution times continued to drop for old problems, butreturned to pre-training levels for new problems. In contrast,for algebra problems, solution times did not differ acrosslanguage; solution times dropped during training; aftertraining, gains in speed were preserved for both old and newproblems. These findings suggest that the contribution oflanguage to mathematical thinking may be limited to the areasof mathematics that are learned by rote and not by rule.

Many psychologists have argued that language acquisition plays an important role in the development of Theory of Mind (ToM) reasoning in children. Several accounts of this interaction exist: some believe that language gives children the ability to express already formed ToM reasoning (e.g. He, Bolz, & Baillargeon, 2011), while others argue that learning specific grammatical structures engenders new reasoning abilities (e.g. de Villiers & Pyers, 1997). Questions remain about the mechanism by which this interaction occurs. In this paper, we show that the Analogical Theory of Mind (AToM; Rabkina et al., 2017) computational model can bootstrap aspects of ToM reasoning from sentential complement training, and that its performance matches improvement patterns of children who are trained using similar stimuli. This provides an implemented algorithmic account of bootstrapping ToM reasoning from language within a broader model of ToM development.

Human activity requires an ability to generalise beyond theavailable evidence, but when examples are limited – as theynearly always are – the problem of how to do so becomes par-ticularly acute. In addressing this problem, Shepard (1987)established the importance of representation, and subsequentwork explored how representations shift as new data is ob-served. A different strand of work extending the Bayesianframework of Tenenbaum and Griffiths (2001) established theimportance of sampling assumptions in generalisation as well.Here we present evidence to suggest that these two issuesshould be considered jointly. We report two experiments whichreveal replicable qualitative patterns of individual differencesin the representation of a single category, while also showingthat sampling assumptions interact with these to drive gener-alisation. Our results demonstrate that how people shift theircategory representation depends upon their sampling assump-tions, and that these representational shifts drive much of theobserved learning.

The robust and efficient recognition of visual relations in im-ages is a hallmark of biological vision. We argue that, de-spite recent progress in visual recognition, modern machinevision algorithms are severely limited in their ability to learnvisual relations. Through controlled experiments, we demon-strate that visual-relation problems strain convolutional neuralnetworks (CNNs). The networks eventually break altogetherwhen rote memorization becomes impossible, as when intra-class variability exceeds network capacity. Motivated by thecomparable success of biological vision, we argue that feed-back mechanisms including attention and perceptual groupingmay be the key computational components underlying abstractvisual reasoning.

Behavior that is contingent on conditional rules necessitatesan abstraction away from concrete stimulus-response identi-ties in order to form a rule template, but also a subsequenttransformation of representation back into sensorimotor for-mat in order to produce concrete behavior. Evidence suggeststhat dorsal premotor cortex (PMd) is well-positioned to me-diate such an operation. We utilized repetitive transcranialmagnetic stimulation, a non-invasive manner of perturbing thefunctioning of targeted cortical regions, to investigate the roleof dorsal premotor cortex during performance of a Rapid In-structed Task Learning paradigm. The task required partici-pants to form conditional associations between stimuli and re-sponses carrying varying levels of abstraction. Selective inter-ference of response times to stimuli presentation was observedonly when the task necessitated the participants to resolve aconditional response referring to an internally-produced repre-sentation of a rule element with relatively abstracted quality.We conclude that PMd specifically supports conditional rulebehavior through transformation of abstract representations toconcrete response, when the conditional rule necessary to re-solve includes abstract, internally-produced identities.

Recent research has placed episodic reinforcement learning(RL) alongside model-free and model-based RL on the list ofprocesses centrally involved in human reward-based learning.In the present work, we extend the unified account of model-free and model-based RL developed by Wang et al. (2017) tofurther integrate episodic learning. In this account, a genericmodel-free "meta-learner" learns to deploy and coordinate allof these RL algorithms. The meta-learner is trained on a broadset of novel tasks with limited exposure to each task, suchthat it learns to learn about new tasks. We show that whenequipped with an episodic memory system inspired by theoriesof reinstatement and gating, the meta-learner learns to use thesame pattern of episodic, model-free, and model-based RLobserved in humans in a task designed to dissociate among theinfluences of these learning algorithms. We discuss implicationsand predictions of the model.

Vowel systems exhibit organization, and several theoretical ac-counts have been proposed to explain this. A prominent ac-count explains organization in terms of maximizing the disper-sion of vowels, increasing acoustic perceptibility while reduc-ing articulatory effort. This implies modality-independence,but leaves open questions about the extent to which dispersionis driven by articulatory or acoustic pressures. We investigatedwhether vowel-like organization would emerge in a novel vi-sual communication system in the laboratory, in which partic-ipants took turns to send color signals to communicate a set ofanimal referents by moving their fingers around a color space.We manipulated the extent to which sender and receiver needswere aligned. Overall, systems exhibited significant levels ofdispersion; participants also took into account receiver needs,withconsequences for the structure of the resulting systems.

Spatial demonstratives, i.e. words like this and that, serve asimportant tools to establish joint attention, allowinginterlocutors to flexibly share spatial reference schemes.However, little experimental work has investigated whichperceptual and social factors drive speakers’ choices ofdemonstrative forms. We used a novel experimental paradigmto explore 1) the role of relative placement of competingreferents on the sagittal and lateral planes, 2) whether and howthe presence of an addressee modulates the speaker’s choice ofdemonstrative forms. We found that the choice ofdemonstratives is affected by the relative position ofcompeting referents both on the sagittal and lateral plane.Furthermore, we found that the presence of an interlocutorshifts attraction for proximal demonstratives towards theshared space of reference, but only in collaborative contexts.Together, these results suggest that spatial deixis is groundedin a contrastive organization of space tightly coupled to manualand social affordances.

Human behavior is often remarkably flexible, showing theability to quickly adapt to the statistical peculiarities of aparticular local context. When it comes to language, previ-ous work has shown that listeners’ anticipatory interpretationsof intonational cues are adapted dynamically when cues areobserved to be stochastically unreliable. This paper reportsnovel empirical data from manual response dynamics (mouse-tracking) on how listeners adapt their predictive interpretationwhen some intonational cues are occasionally unreliable whileothers are consistently reliable. A model of rational belief dy-namics predicts that listeners adapt differently to different un-reliable intonational cues, as a function of their initial eviden-tial strength. These predictions are borne out by our data.

After viewing Black males faces, US participants are typically faster to categorize weapons and slower to categorize tools than afterviewing White male faces, revealing the activation of implicit stereotypes linking Black males with violent crime. Here we testedwhether hearing Black male voices speaking in African American Vernacular English (AAVE) activates these same threat-relatedstereotypes. In a national US sample, participants were faster to categorize weapons compared to tools after hearing race-neutralnames spoken in AAVE than after hearing them spoken in Standard American English (SAE). Like Black faces, Black voices canactivate violent stereotypes, affecting visual discrimination of objects.

The role of intentions in motor planning is heavily weightedin classical psychological theories, but their role in generat-ing eye movements, and our awareness of these oculomotorintentions, has not been investigated explicitly. In this study,the extent to which we monitor oculomotor intentions, i.e.the intentions to shift one’s gaze towards a specific location,and whether they can be expressed in conscious experience,is investigated. A forced-choice decision task was developedwhere a pair of faces moved systematically across a screen.In some trials, the pair of faces moved additionally as soon asthe participants attempted to gaze at one of the faces, prevent-ing them from ever viewing it. The results of the experimentsuggest that humans in general do not monitor their eye move-ment intentions in a way that allows for mismatches betweenplanned gaze landing target and resulting gaze landing targetto be consciously experienced during decision-making.

What are the major topics of the Cognitive Science Societyconference? How have they changed over the years? To an-swer these questions, we applied an unsupervised learning al-gorithm known as dynamic topic modeling (Blei & Lafferty,2006) to the 2000–2017 Proceedings of the Cognitive Sci-ence Society. Unlike traditional topic models, a dynamic topicmodel is sensitive to the temporal context of documents andcan characterize the evolution of each topic across years. Us-ing this model, we identify historical trends in the popularity oftopics over time, and shifts in word use within topics indicativeof changing focuses within the field. We also measure the cor-relation across topics, and use the model to highlight the topicstructure of particular papers and labs. We believe dynamictopic models present an important tool towards understandingCognitive Science as it continues to grow and evolve over time

Humans often communicate through seemingly arbitraryactions, like winks, waves, and nods. While these non-iconicgestures derive their meanings from cultural consensus,people, and especially children, must be able to identify thesemovements as gestures. Here we propose that people expectthat communicative actions will be shaped to reveal that theyhave no external goal. In Experiment 1, we show that peoplejudge inefficient actions as more likely to be communicative.In Experiment 2, we show that these judgments are trulydriven by efficiency, rather than a movement’s visualcomplexity. Finally, in Experiment 3, we show that repetition– which unambiguously reveals that the goal of the action isthe movement itself – has a strong influence on inferencesabout communicativeness, independent of the motion’sefficiency. Our findings show how expectations about non-iconic communicative actions can be folded into a generalgoal inference framework structured around an expectationfor efficiency.

For almost two decades, the poor performance observed withthe so-called Director task has been interpreted as evidence oflimited use of Theory of Mind in communication. Here wepropose a probabilistic model of common ground in referentialcommunication that derives three inferences from an utterance:what the speaker is talking about in a visual context, what sheknows about the context, and what referential expressions sheprefers. We tested our model by comparing its inferences withthose made by human participants and found that it closelymirrors their judgments, whereas an alternative modelcompromising the hearer’s expectations of cooperativenessand efficiency reveals a worse fit to the human data. Ratherthan assuming that common ground is fixed in a givenexchange and may or may not constrain reference resolution,we show how common ground can be inferred as part of theprocess of reference assignment.

The literature has long emphasized the neocortex’s role in volitionalprocesses. In this work, we examined endogenous orienting in anevolutionarily older species, the archer fish, which lacks neocortex-like cells. We used Posner’s classic endogenous cuing task, in whicha centrally presented, spatially informative cue is followed by a tar-get. The fish responded to the target by shooting a stream of waterat it. Interestingly, the fish demonstrated a human-like “volitional”facilitation effect: their reaction times to targets that appeared onthe side indicated by the precue were faster than their reactiontimes to targets on the opposite side. The fish also exhibited inhi-bition of return, an aftermath of orienting that commonly emergesonly in reflexive orienting tasks in human participants. We believethat this pattern demonstrates the acquisition of an arbitrary con-nection between spatial orienting and a nonspatial feature of acentrally presented stimulus in nonprimate species. In the literatureon human attention, orienting in response to such contingencies hasbeen strongly associated with volitional control. We discuss theimplications of these results for the evolution of orienting, and forthe study of volitional processes in all species, including humans.

A striking limitation of human cognition is our inability to ex-ecute some tasks simultaneously. Recent work suggests thatsuch limitations can arise from a fundamental trade-off in net-work architectures that is driven by the sharing of representa-tions between tasks: sharing promotes quicker learning, at theexpense of interference while multitasking. From this perspec-tive, multitasking failures might reflect a preference for learn-ing efficiency over parallel processing capability. We explorethis hypothesis by formulating an ideal Bayesian agent thatmaximizes expected reward by learning either shared or sep-arate representations for a task set. We investigate the agent’sbehavior and show that over a large space of parameters theagent sacrifices long-run optimality (higher multitasking ca-pacity) for short-term reward (faster learning). Furthermore,we construct a general mathematical framework in which ratio-nal choices between learning speed and processing efficiencycan be examined for a variety of different task environments.

It is inconceivable how chaotic the world would look to hu-mans, faced with innumerable decisions a day to be made un-der uncertainty, had they been lacking the capacity to distin-guish the relevant from the irrelevant—a capacity which com-putationally amounts to handling probabilistic independencerelations. The highly parallel and distributed computationalmachinery of the brain suggests that a satisfying process-levelaccount of human independence judgment should also mimicthese features. In this work, we present the first rational, dis-tributed, message-passing, process-level account of indepen-dence judgment, called D∗. Interestingly, D∗ shows a curi-ous, but normatively justified tendency for quick detection ofdependencies, whenever they hold. Furthermore, D∗ outper-forms all the previously proposed algorithms in the AI litera-ture in terms of worst-case running time, and a salient aspectof it is supported by recent work in neuroscience investigatingpossible implementations of Bayes nets at the neural level. D∗exemplifies how the pursuit of cognitive plausibility can leadto the discovery of state-of-the-art algorithms with appealingproperties, and its simplicity makes D∗ potentially a good can-didate as a teaching tool.

What is ecological about Gibsonian Ecological Psychology?Well-known senses in which Gibson’s scientific program is‘ecological’ have to do with its theoretical, ontological andmethodological foundations. But, besides these, the Gibsonianframework is ‘ecological’ in an additional sense that has re-mained understudied and poorly understood—a sense of “eco-logical” that connects Gibson’s view to the environmentalismof environmental psychology and environmental ethics. Thispaper focuses on the latter sense of ‘ecological’, and exploresthe relevance of Gibson’s notion of “affordance” for thinkingabout environmental issues like deforestation, pollution andclimate change. One existing account is criticized and an al-ternative is proposed.

Two main views have informed the literature on the psy-chology of emotion in the past few decades. On one side,cognitivists identify emotions with processes such as judg-ments, evaluations and appraisals. On the other side, advo-cates of non-cognitive approaches leave the “intellectual” as-pects of emotional experience out of the emotion itself, in-stead identifying emotions with embodied processes involv-ing physiological changes. Virtually everyone on either sideof the cognitive/non-cognitive divide identify William James’view, also known as the James-Lange theory, fully on the non-cognitivist side. But this is a mistake. Re-interpreting James’writings in its scientific context, this paper argues that he actu-ally rejected the cognitive/non-cognitive divide, such that hisview of emotions did not fit either side—that is, James was nota “Jamesian” in the sense the term is used in the literature.

Tests of formal models of human categorization have traditionally been restricted to artificial categories because deriving psychological representations for large numbers of natural stimuli has been an intractable task. We show that deep learning may be used to solve this problem. We train an ensemble of convolutional neural networks (CNNs) to produce the multidimensional scaling (MDS) coordinates of images of rocks. We then show that not only are the CNNs able to predict the MDS coordinates of a held-out test set of rocks, but that the CNN-derived representations can be used in combination with a formal psychological model to predict human categorization behavior on a completely new set of rocks.

Is language production isolated from our experiences ofphysical events, or can physical motion affect the conceptualsaliency of the components of a to-be-described event, inways that affect its linguistic description? This studyexamined the influence of physical motion on theinterpretation and description of simple transitive events.More specifically, we investigated whether engagement innon-speech physical actions affects the relative location ofverbs versus arguments in sentence production, and therelative location and prominence of Agents, by testing nativespeakers of Truku, a language that allows flexibility in eachof these options and presents under-studied typologicalpatterns.

Human-made objects (artifacts) often provide rich social information about the people who created them. We explore how people reason about others from the objects they create, characterizing inferences about when social transmission of ideas (copying) has occurred. We test whether judgments are driven by perceptual heuristics, or structured explanation- based reasoning. We develop a Bayesian model of explanation-based inference from artifacts and a simpler model of perceptual heuristics, and ask which better predicts people’s judgments. Our artifact-building task involved two characters who built toy train tracks. Participants viewed pairs of tracks, and judged whether copying had occurred. Our explanation-based model accurately predicted on a trial-by- trial basis when participants inferred copying; the perceptual heuristics model was significantly less accurate. Efficient design ‘explained away’ similarity, making similarity weaker evidence of copying for efficient tracks. Overall, data show that like intuitive archeologists, people make rich explanation-based inferences about others from the objects they create.

Radiologists can discriminate between normal and abnormalbreast tissue at a glance, suggesting that radiologists might beusing some “global signal” of abnormality. Our study inves-tigated whether texture descriptions can be used to character-ize the global signal of abnormality and whether radiologistsuse this information during interpretation. Synthetic imageswere generated using a texture synthesis algorithm trained ontexture descriptions extracted from sections of mammograms.Radiologists completed a task that required rating the abnor-mality of briefly presented tissue sections. When the abnormaltissue had no visible lesion, radiologists seemed to use texturedescriptions; performance was similar across real and synthe-sized tissue sections. However, when the abnormal tissue had avisible lesion, radiologists seemed to rely on additional mech-anisms beyond the texture descriptions; performance increasedfor the real tissue sections. These findings suggest that radiol-ogists can use texture descriptions as global signals of abnor-mality in interpretation of breast tissue.

In STEM domains, students are expected to acquire domainknowledge from visual representations that they may not yetbe able to interpret. Such learning requires perceptual flu-ency, or the ability to intuitively and rapidly see the underlyingconcepts in visuals and to translate between them. Perceptualfluency is acquired via nonverbal, implicit learning processes.Thus far, we have lacked a principled approach for identify-ing a sequence of perceptual fluency problems that promoterobust learning. Here, we describe how a novel machine learn-ing technique can generate an optimal sequence of perceptualfluency problems. In a human experiment, we show that amachine-generated sequence outperforms both a random se-quence and a sequence generated by a human domain expert.Interestingly, the machine-generated sequence resulted in sig-nificantly lower accuracy during training, but higher posttestaccuracy. This suggests that the machine-generated sequenceinduced desirable difficulties. To our knowledge, our study isthe first to show that machine learning can yield desirable dif-ficulties for perceptual learning

In this paper we study the influence of personality traits in ne-gotiation by using a methodology for automated cognitive be-havior analysis (ACBA). This methodology uses genetic pro-gramming (GP) for hypothesis generation and testing of hu-man behavior with the goal of explaining the underlying men-tal structures guiding people’s actions during a task. ACBA it-eratively generates programs—the hypotheses—capable of ex-plaining the behavior exhibited by an individual during a multi-level, multi-issue, sequential bargaining task against an artifi-cial agent. Our study focuses on the influence of the personal-ity traits of social-value orientation (SVO) and Machiavellian-ism (Mach). The results show that by using ACBA, we are ableto identify differences in the outcomes of programs emergingfrom GP that are consistent with the influences that differentSVO and Mach profiles have in human negotiation behavior.

We examined how phonological competition effects inspoken word recognition change with word length. Cohorteffects (competition between words that overlap at onset) arestrong and easily replicated. Rhyme effects (competitionbetween words that mismatch at onset) are weaker, emergelater in the time course of spoken word recognition, and aremore difficult to replicate. We conducted a simple experimentto examine cohort and rhyme competition using monosyllabicvs. bisyllabic words. Degree of competition was predicted byproportion of phonological overlap. Longer rhymes, withgreater overlap in both number and proportion of sharedphonemes, compete more strongly (e.g., kettle-medal [0.8overlap] vs. cat-mat [0.67 overlap]). In contrast, long andshort cohort pairs constrained to have constant (2-phoneme)overlap vary in proportion of overlap. Longer cohort pairs(e.g., camera-candle) have lower proportion of overlap (inthis example, 0.33) than shorter cohorts (e.g., cat-can, with0.67 overlap) and compete more weakly. This finding hasmethodological implications (rhyme effects are less likely tobe observed with shorter words, while cohort effects arediminished for longer words), but also theoreticalimplications: degree of competition is not a simple functionof overlapping phonemes; degree of competition isconditioned on proportion of overlap. Simulations withTRACE help explicate how this result might emerge.

Cognitive measurement models decompose observed behav-ior into latent cognitive processes. For situations with morethan one condition, such models allow to test hypotheses onthe level of the latent processes. We propose a fully Bayesianensemble model approach to test hypotheses on the level ofthe latent processes in situations in which multiple measure-ment models or model classes exist. In the first step, one needsto perform a Bayesian model selection step comparing the hy-potheses within each model class. Aggregating the results ofthe first step yields ensemble posterior model probabilities. Weprovide an example for a working memory data set using anensemble of a resource model and a slots model.

We present a natural logic for reasoning with quanti-fiers that can predict human performance in appro-priate reasoning tasks. The model is an extension ofthat in (Geurts, 2003) but allows for better fit withdata on syllogistic reasoning and is extended to ac-count for reasoning with iterated quantifiers. Weassign weights to inference rules and operationalizethe complexity of a reasoning pattern as weightedlength of proof in our logic – this results in a measureof complexity that outperforms other models in theirpredictive capacity and allows for the derivation ofempirically testable hypotheses.

Singular causation queries require an assessment of whethera singular co-occurrence of two events c and e was causal orsimply coincidental. The current study builds on our previ-ous research (Stephan & Waldmann, 2018) in which we pro-posed a computational model of singular causation judgments.The model highlights that singular causation judgments needto take into account the power of the target cause C and ofalternative causes A, as well as the possibility of preemption.What was missing was a detailed model allowing us to esti-mate the probability of preemption of a target cause by thealternative causes. The present research fills this gap by elab-orating the temporal assumptions that might enter assessmentsof singular causation. We focus on assumptions about tempo-ral precedence between target and alternative causes, with aspecific focus on assumptions about causal latency. We reportthe results of two new experiments supporting the model.

We have previously shown that a biologically realistic spikingneuron implementation of an action selection/executionsystem (constrained by the neurological connectivity of thecortex, basal ganglia, and thalamus) is capable of performingcomplex tasks, such as the Tower of Hanoi, n-Back, andsemantic memory search. However, because the neuralimplementation approximates a strict rule-based structure of aproduction system, such models have involved hand-tweakingof multiple parameters to get the desired behaviour. Here, weshow that a simple, local, online learning rule can be used tolearn these parameters, resulting in neural models of cognitivebehaviours that are more reliable and easier to construct thanwith prior methods.

In a recent paper, Laird, Lebiere, and Rosenbloom (2017)highlight how 40 years of research on cognitive architectureshas begun to yield a dramatic convergence of different ap-proaches towards a set of basic assumptions that they calledthe “Standard Model of the Mind” (SMM), in analogy to theStandard Model of particle physics. The SMM was designedto capture a consensus view of “human-like minds”, whetherfrom AI or cognitive science, which if valid must also be trueof the human brain. Here, we provide a preliminary test ofthis hypothesis based on a re-analysis of fMRI data from fourtasks that span a wide range of cognitive functions and cog-nitive complexity, and are representative of the specific formof intelligence and flexibility that is associated with higher-level human cognition. Using an established method (DynamicCausal Modeling) to examine functional connectivity betweenbrain regions, the SMM was compared against two alternativemodels that violate either functional or structural assumptionsof the SMM. The results show that, in every dataset, the SMMsignificantly outperforms the other models, suggesting that theSMM best captures the functional requirements of brain dy-namics in fMRI data among these alternatives.

A gap currently exists in the literature regarding a quantitative exploration of the self-conscious emotions (i.e., pride, embarrassment, shame, and guilt). In order to address this gap, the present study sought to explore the possibility of systematically inducing one specific self-conscious emotion (shame). Various methods were explored to determine the most effective way to induce a sense of shame in an educational context. Results revealed significant differences in state shame as measured by the Experiential Shame Scale. However, this difference was related to a student’s proneness to shame, expectations of success, and perceptions of failure. Immediate implications for shame’s impact in a variety of educational contexts are discussed.

Previous research has shown that active engagement driveschildren’s remarkable learning capabilities. We investigatedwhether preschoolers are ecological learners, able to selectthose active learning strategies that are most informative in agiven task. Children (n = 114; 3 to 5 years old) chose betweentwo exploratory actions (opening vs. shaking) to find an eggshaker hidden in one of four small boxes, contained in twolarger boxes. Prior to this game, children learnt that the eggwas equally likely to be in any of the four small boxes(Uniform condition), or that it was most likely to be in oneparticular small box (Skewed condition). Results show that 3-and 4-year-olds, but not 5-year-olds, successfully tailored theirexploratory actions to these different likelihood distributions.We suggest that ecological learning may be a key mechanismexplaining how children can efficiently learn about the worldaround them.

Language provides multiple ways of conveying the opposite:A person not happy can be unhappy, sad, or perhaps neither,just not happy. Rather than being redundant, we hypothesizethat uncertainty about the meaning of negation markers allowslisteners to derive fine-grained distinctions among these vari-ous alternatives. We formalize this hypothesis in a probabilis-tic model of gradable adjectives (e.g., happy), and use this toaddress an outstanding puzzle: how to interpret double nega-tions (e.g., not unhappy). Our model makes surprising addi-tional predictions about a putative difference between morpho-logical antonyms (unhappy) and negated positives (not happy):Listeners should judge unhappy as more sad than not happyonly when confronted with alternatives in context; when inter-preted in isolation, we predict no difference in understanding.Two behavioral experiments confirm consistent orderings ofinterpretations that interact with the presentational context inthe way predicted. These findings support the hypothesis thatlisteners represent uncertainty even about the most logical ele-ments of language.

Relational categories are notoriously difficult to learn. We studied the impact of comparison on relational concept learning with a novel word learning task in 3- and 4-year olds. We contrasted a no-comparison (single) condition and two comparison conditions. In the latter case, the set of learning pairs was composed of either close or far pairs (e.g., close pair: knife1- watermelon, knife2-orange; far pair: ax- evergreen tree, saw-log, for the “cutter for” relation). We also manipulated the transfer stimuli semantic distance (near or distant semantic domain, e.g., a scissor for a piece of paper in the close case, and a shaver for a face in the far domain case). The no-comparison condition led to random generalizations in the younger group only. Overall the close learning condition and the near transfer condition led to good performance. We discuss these results in terms of the role of semantic distance and how participants integrate stimuli depending on distance.

We estimate lexical Concreteness for millions of wordsacross 77 languages. Using a simple regression framework,we combine vector-based models of lexical semantics withexperimental norms of Concreteness in English and Dutch.By applying techniques to align vector-based semantics acrossdistinct languages, we compute and release Concreteness esti-mates at scale in numerous languages for which experimentalnorms are not currently available. This paper lays out thetechnique and its efficacy. Although this is a difficult datasetto evaluate immediately, Concreteness estimates computedfrom English correlate with Dutch experimental norms at ρ= .75 in the vocabulary at large, increasing to ρ = .8 amongNouns. Our predictions also recapitulate attested relationshipswith word frequency. The approach we describe can be readilyapplied to numerous lexical measures beyond Concreteness.

One goal of cognitive science is to build theories of mentalfunction that predict individual behavior. In this project wefocus on predicting, for individual participants, which specificitems in a list will be remembered at some point in the future.If you want to know if an individual will remember something,one commonsense approach is to give them a quiz or test suchthat a correct answer likely indicates later memory for an item.In this project we attempt to predict later memory without ex-plicit assessments by jointly modeling both neural and behav-ioral data in a computational cognitive model which capturesthe dynamics of memory acquisition and decay. In this paper,we lay out a novel hierarchical Bayesian approach for com-bining neural and behavioral data and present results showinghow fMRI signals recorded during the study phase of a mem-ory task can improve our ability to predict (in held-out data)which items will be remembered or forgotten 72 hours later.

Clear speech is a vocal style used when a speaker wishes to im-prove comprehension, usually due to the presence of externalnoise, less-than-optimal listening conditions, or when they aresimply instructed to speak clearly. Clear speech has many dis-tinguishing features, including increased duration, pitch, andamplitude, as well as the exaggeration of articulatory move-ment. We use game theory to model the phenomenon of clearspeech, and make predictions of how it changes under differ-ent circumstances. We view the behaviours of speakers andhearers when communicating as optimal strategies in commu-nication games. When comprehension becomes more difficult,the optimal strategies of the games shift so that speakers exertmore energy to improve the likelihood of accurate communica-tion. We discuss how our models correspond to experimentalobservations and see what predictions are made for future ex-periments.

While a large literature has studied how people make forecasts, less is known about how lay people process and interpret forecasts presented to them. We contrast two common ways of communicating an uncertain fore- cast, as either a chance (e.g., the probability of winning) or as an expected margin (e.g., the point spread). Across five studies, we find a robust chance-margin discrepan- cy: people tend to treat a chance forecast as conveying greater probability of the higher-likelihood outcome than the statistically equivalent margin forecast

The pervasive use of distributional semantic models or wordembeddings is due to their remarkable ability to represent themeanings of words for both practical application and cognitivemodeling. However, little has been known about what kind ofinformation is encoded in text-based word vectors. This lack ofunderstanding is particularly problematic when distributionalsemantics is regarded as a model of semantic representationfor abstract concepts. This paper attempts to reveal the internalknowledge encoded in distributional word vectors by the anal-ysis using Binder et al.’s (2016) brain-based vectors, explicitlystructured conceptual representations based on neurobiologi-cally motivated attributes. In the analysis, the mapping fromtext-based vectors to brain-based vectors is trained and predic-tion performance is evaluated by comparing the estimated andoriginal brain-based vectors. The analysis demonstrates thatsocial and cognitive information is predicted with the highestaccuracy by text-based vectors, but emotional information isnot predicted so accurately. This result is discussed in terms ofembodied theories for abstract concepts.

When asked to convey temporal concepts such as ‘yesterday’and ‘tomorrow’ via movements of a dot on a vertical bar,American undergraduates utilize analogical mappingsbetween spatial and temporal concepts. Previous work hasrevealed two different strategies, hypothesized to requirediffering amounts of artificial language exposure to learn.Different pairs of participants, when interacting about thesetime concepts, all settled on the same association betweenspatial magnitude and temporal duration, with largermovements used to convey temporal intervals of greaterduration. However, the association between particular spatiallocations and temporal concepts such as ‘past’ and ‘future’,elicited much more arbitrary solutions, where the mappingsdiffered across pairs of participants. These findings suggestedthat the duration mapping might be driven by mostly shared,initial cognitive biases, while contrasting mappings forpast/future result more clearly from extensive linguisticinteraction. Here we tested whether the brain respondsdifferently to duration mappings as compared to directionmappings by recording participants’ EEG as they learn amini-language that includes both kinds. ERPs time locked toEnglish words elicited larger amplitude N400 and P600 whenthey did not match the preceding signal than when they didmatch. The P600 results were larger and more robust for theduration than the direction stimuli, suggesting participantswere more sensitive to violations of the duration mappingscheme. These data support our hypothesis that people have acognitive bias for the duration mappings that supports theirearly emergence in the development of a semiotic system.

Before infants become mature speakers of their nativelanguage, they must acquire a robust word-recognition systemwhich allows them to strike the balance between allowingsome variation (mood, voice, accent) and recognizingvariability that potentially changes meaning (e.g. cat vs hat).The current meta-analysis quantifies how the latter, termedmispronunciation sensitivity, changes over infants’ first threeyears, testing competing predictions of mainstream languageacquisition theories. Our results show that infants weresensitive to mispronunciations, but accepted them as labelsfor target objects. Interestingly, and in contrast to predictionsof mainstream theories, mispronunciation sensitivity was notmodulated by infant age, suggesting that a sufficientlyflexible understanding of native language phonology is inplace at a young age.

We explored children’s developing understanding of mentallife using a novel approach to track changes in conceptualstructure from the bottom up by analyzing patterns of men-tal capacity attributions. US children (n=247) evaluated ele-phants, goats, mice, birds, beetles, teddy bears, dolls, robots,and computers on a range of mental capacities, allowing us toassess which attributions “go together” and how these concep-tual connections might develop over early and middle child-hood. Replicating previous studies with adults and older chil-dren, an exploratory factor analysis of older children’s (7-9y)responses revealed a three-way distinction between physiolog-ical abilities (e.g., hunger, smell), social-emotional abilities(e.g., guilt, embarrassment), and perceptual-cognitive abili-ties (e.g., choice, memory), corresponding to traditional no-tions of BODY, HEART, and MIND. Hints of this three-way distinction emerged among younger children (4-6y), butyounger children appeared to perceive markedly stronger con-nections among physiological and social-emotional abilities,while clearly distinguishing both from the MIND.

Belief bias in syllogistic reasoning refers to the finding thatindividuals are more likely to accept believable than unbeliev-able conclusions independent of their logical validity. Mosttheories argue that belief bias is driven by differences in rea-soning processes between believable and unbelievable syllo-gisms. In contrast, Dube, Rotello, and Heit (2010) proposedthat belief bias is solely an effect of response processes. Weinvestigated belief bias without having to rely on response biasmanipulations (Klauer, Musch, and Naumer, 2000) or confi-dence ratings (Dube et al., 2010). Instead, we added a thirdresponse (“I don’t know”) to the usual binary response set(“Yes”/“No”). This allowed us to test belief bias with a fullyidentified multinomial processing tree model, in a hierarchicalBayesian framework. We found evidence that the belief biasis driven by differences in response processes. Evidence for adifference in reasoning processes was inconclusive.

The Delta and Decay rules are two learning rules used to update expected values in reinforcement learning (RL) models. The delta rule learns average rewards, whereas the decay rule learns cumulative rewards for each option. Participants learned to select between pairs of options that had reward probabilities of .65 (option A) versus .35 (option B) or .75 (option C) versus .25 (option D) on separate trials in a binary-outcome choice task. Crucially, during training there were twice as AB trials as CD trials, therefore participants experienced more cumulative reward from option A even though option C had a higher average reward rate (.75 versus .65). Participants then decided between novel combinations of options (e.g, A versus C). The Decay model predicted more A choices, but the Delta model predicted more C choices, because those respective options had higher cumulative versus average reward values. Results were more in line with the Decay model’s predictions. This suggests that people may retrieve memories of cumulative reward to compute expected value instead of learning average rewards for each option.

The idea of a “cognitive map” was originally developed to ex-plain planning and generalization in spatial domains througha representation of inferred relationships between experiences.Recently, new research has suggested similar principles mayalso govern the representation of more abstract, conceptualknowledge in the brain. We test whether the search for rewardsin conceptual spaces follows similar computational principlesas in spatial environments. Using a within-subject design, par-ticipants searched for both spatially and conceptually corre-lated rewards in multi-armed bandit tasks. We use a GaussianProcess model combining generalization with an optimisticsampling strategy to capture human search decisions and judg-ments in both domains, and to simulate human-level perfor-mance when specified with participant parameter estimates. Inline with the notion of a domain-general generalization mecha-nism, parameter estimates correlate across spatial and concep-tual search, yet some differences also emerged, with partici-pants generalizing less and exploiting more in the conceptualdomain.

Multiple acoustic dimensions contribute to speechcategorization. Yet highly diagnostic dimensions contributegreater ‘perceptual weight’ in influencing speechcategorization than less diagnostic dimensions. Recentresearch demonstrates that perturbations in short-term inputregularities lead to rapid dynamic re-weighting of auditorydimensions. Here, we test the hypothesis that phonetic-category-level activation via a highly diagnostic acousticdimension is critical in driving this rapid tuning of how inputmaps to phonetic categories. To do so, we manipulate theinherent relative effectiveness, the perceptual weight, of twoacoustic dimensions in signaling English vowelcategorization using noise-vocoded versus clear speech. Weobserve that rapid tuning across statistical regularities isaffected by dimensions’ effectiveness in signaling the vowelcategories. These findings indicate that category activationvia a highly diagnostic dimension drives adaptive tuning inspeech perception, consistent with error-driven supervisedlearning.

In this study, we investigate whether emotional expressionsprovide cues to knowledge sufficient for predicting others’behavior based on their true and false beliefs. We adapted theclassic Sally-Anne task (Baron-Cohen, Leslie, & Frith, 1985)such that children (N = 62, mean: 5.58 years, range: 4.05-6.98years) were not told whether Sally saw Anne move the objector not. However, when Sally came back looking angry, evenfour-year-olds inferred that she had seen Anne move her toy;when she came back looking happy, children inferred that shehad not seen the transfer. Based on these inferences, five andsix-year-olds, although not four-year-olds, were able topredict where Sally would look for her toy.

The extent to which we generalize a novel property from asample of familiar instances to novel instances depends on thesample used. In these experiments, we are interested in twosample characteristics: number of types (discrete entities) andnumber of tokens (copies of the same entity) that share a novelproperty. Existing studies permit separate and conditionalhypotheses about the effects of adding types and tokens, but nostudy has examined the effects of both variables ongeneralization stimuli varying in similarity. We find thatadding types broadens generalization to similar stimuli, buttightens generalization to dissimilar stimuli. Adding tokensdoes not affect generalization, but adding repetitions that areframed as types produces some tightening. Implications formodels of inductive reasoning are discussed.

Word frequency effects have long served as an empirical andtheoretical test bed for theories of language processing. Anumber of recent studies have suggested that Contextual Di-versity (CD) is a better metric of retrieval processes than wordfrequency. Motivated by these findings, we sketch an activeaccount of lexical access during sentence processing: lan-guage users store statistics about contextualized lexical rep-resentations and use lexical-contextual relations to both con-struct context and predict words given the context. In linewith our account, we provide evidence from a frequency judg-ment experiment suggesting that words are not stored indepen-dently of their contexts of use. To further examine CD effectsin reading, we analyzed reading times in self-paced readingand eye-tracking corpora. We demonstrate that as context isconstructed, the role of CD in lexical retrieval is attenuated,reflecting a trade-off between context construction and contex-tualized word prediction.

Learning in the sciences is difficult for students from elementary school to university due to misconceptions, or incorrect prior knowledge, interfering with the acquisition of new knowledge. The process of replacing previously incorrect ideas with new and accurate ones is referred to as conceptual change. Which factors and to what extent they facilitate the conceptual change is debated. This study primarily investigates two key components to conceptual change in scientific knowledge: text style and epistemic beliefs. We also explored additional contributions of individual differences in prior knowledge, reading ability, and working memory. 157 college students completed a two-part, within subjects design study in which they completed pretests, read passages addressing a misconception, completed posttests, and were assessed on a battery of the individual difference measures. We noted conceptual change on the posttest, but individual readers appeared to respond to the text differently.

Our actions shape what we learn. Because of this dependency,learners are proficient at choosing their actions to maximizetheir information gain. However, learning often unfolds insocial contexts where learners have both informational goals(e.g., to learn how something works) but also social goals (e.g.,to appear competent and impress others). How do these goalsshape learners’ decisions? Here, we present a computationalmodel that integrates the value of social and informationalgoals to predict the decisions that people will make in a simpleactive causal learning task. We show that, in a context wherethe informational and social goals are in conflict, an empha-sis on performance or self-presentation goals leads to reducedchances of learning (Exp. 1) and that social context can pushlearners to pursue performance-oriented actions even when thelearning goal is highlighted (Exp. 2). Our formal model ofsocial-active learning successfully captures the empirical re-sults. These findings are first steps towards understanding therole of social reasoning in active learning contexts.

Allen Newell’s Knowledge Level theory is a philosophical position on the reality of knowledge that is best understood through the lens of Pragmatism--specifically, the view that the practical effects of general concepts are indelibly linked with the reality of those concepts. Consequently, the reality of the knowledge level is context-dependent. Newell’s theory reduces the complexity of analyzing every mechanism behind intelligence systems by abstracting away details irrelevant to predicting behavior and, as such, is more important than ever in light of current challenges in cognitive science.

The cognitive reflection test (CRT; Frederick, 2005) assesseshow well adults can reflect on the validity of their ownthinking, and it has been shown to predict several measures ofnormative reasoning. Here, we sought to create a version ofthe cognitive reflection test suitable for elementary-school-aged children, which could be used to study the emergence ofcognitive reflection as well as its role in the development ofother forms of higher-order cognition. We identified eightchild-friendly questions that elicit an incorrect, intuitiveresponse that must be inhibited in order to provide a correct,analytic response. We compared children’s and adults’performance on these questions (dubbed the CRT-D) toseveral measures of rational thinking (denominator neglect,base rate sensitivity, syllogistic reasoning, otherside thinking)and thinking dispositions (actively open-minded thinking,need for cognition). The CRT-D was a significant predictor ofrational thinking and normative thinking dispositions in bothchildren and adults. Moreover, performance on the CRT-Dcorrelated with performance on the original CRT in adults,and in children, it predicted rational thinking and normativethinking dispositions above and beyond age. These resultssuggest that the CRT-D is a valid measure of children’scognitive reflection and pave the way for future investigationsof its development and its developmental consequences.

Scientific ideas can be difficult to affirm if they contradictearlier-developed intuitive theories. Here, we investigatedhow instruction on counterintuitive scientific ideas affects theaccessibility of those ideas under time pressure. Participants(138 college undergraduates) verified, as quickly as possible,statements about life and matter before and after a tutorial onthe scientific properties of life or matter. Half the statementswere consistent with intuitive theories of the domain (e.g.,“zebras reproduce”) and half were inconsistent (e.g.,“mushrooms reproduce”). Participants verified the latter lessaccurately and more slowly than the former, both beforeinstruction and after. Instruction did, however, increaseaccuracy for counterintuitive statements within the domain ofinstruction, but changes in accuracy were not accompanied bychanges in speed. These results confirm the conclusion drawnfrom studies with professional scientists that scientific ideascan be prioritized over intuitive ones but the conflict betweenscience and intuition cannot be eliminated altogether.

As we navigate our information-rich world, we frequently interpret and integrate testimony from external sources (friends, teachers, books, internet articles, etc.) – deciding which pieces of information to believe, and which to discard. One cue to a statement’s trustworthiness is whether it comes from a consensus (i.e., when a majority of people agree). But what counts as consensus? When presented with a set of agreeing sources, do we evaluate the quality of consensus – for example, asking whether each source arrived at their conclusion by independent means? In a first experiment, we demonstrate that individuals are insensitive to the quality of a consensus, and are equally confident in conclusions drawn from a ‘true’ consensus (i.e., one derived from many primary sources) and those drawn from a ‘false’ consensus (i.e., one derived from many secondary sources but only a single primary source). In a second experiment, we find that this continues to be true even when the expertise of the secondary sources is minimized. Together, our experiments provide converging evidence that people do not properly discount (or discount at all) information from a ‘false’ consensus.

Gibson et al. (2017) argued that color naming is shaped bypatterns of communicative need. In support of this claim, theyshowed that color naming systems across languages supportmore precise communication about warm colors than cool col-ors, and that the objects we talk about tend to be warm-coloredrather than cool-colored. Here, we present new analyses thatalter this picture. We show that greater communicative preci-sion for warm than for cool colors, and greater communicativeneed, may both be explained by perceptual structure. How-ever, using an information-theoretic analysis, we also showthat color naming across languages bears signs of communica-tive need beyond what would be predicted by perceptual struc-ture alone. We conclude that color naming is shaped both byperceptual structure, as has traditionally been argued, and bypatterns of communicative need, as argued by Gibson et al. –although for reasons other than those they advanced.

Speakers often have more than one way to express the same meaning. What generalprinciples govern speaker choice in the face of optionality when near semanticallyinvariant alternation exists? Studies have shown that optional reduction in language issensitive to contextual predictability, where the more predictable a linguistic unit is, themore likely it gets reduced. Yet it is unclear whether speaker choice is geared towardaudience design, or toward facilitating production. Here we argue that for a differentoptionality phenomenon, namely classifier choice in Mandarin Chinese, UniformInformation Density and at least one plausible variant of availability-based productionmake opposite predictions regarding the relationship between the predictability of theupcoming material and speaker choices. In a corpus analysis of Mandarin Chinese, weshow that the distribution of speaker choices supports the availability-based productionaccount, and not Uniform Information Density.

We use hierarchical drift diffusion models to investigate theeffect of prominence in two-alternative multi-attributepreferential choice. We find that two types of prominenceeffects, option-based and attribute-based, both increase choiceprobabilities for options favored by prominence. However,model fits suggest that the two effects work through differentmechanisms. Altering choice option prominence leads to aresponse bias for the prominent option, whereas alteringattribute prominence leads to an evaluation bias for the optionthat is dominant on the prominent attribute. Our resultsillustrate how seemingly identical contextual factors can bedistinguished with the use of drift-diffusion modelling.

Research on intertemporal choice has suggested that decisionprocesses automatically favor immediate rewards. In thispaper, we use a drift diffusion model to conceptualize andempirically investigate the role of these biases. Our modelpermits automatic biases in the response process, automaticbiases in the evaluation process, as well as differentialweighting for monetary payoffs and time delays. We fit ourmodel to individual-level choice and response time data, andfind that automatic biases are prevalent in intertemporalchoice, but that the type, magnitude, and direction of thesebiases vary greatly across individuals. Our results pose newchallenges for theories of intertemporal choice behavior.

Risk-taking propensity is a general personality disposition. Survey, behavioral, and modeling approaches have been used to study it. We compared three behavioral tasks (BART, C-ART, S-ART) and corresponding computational models to learn which aspects of risky behavior they measure by correlating task performance and parameter estimates with survey responses (impulsivity, sensation seeking, drug use). Results indicated that the BART was not correlated with any of the above domains, whereas behavioral measure from the two ART tasks correlated with impulsivity and sensation seeking. The parameter estimates from the two ART tasks, while having some validity, were weaker indices than the traditional behavioral measure of these tasks. Our findings provide insight into the use and design of these behavioral tasks and their corresponding computational models.

In crowded and cluttered environments, infants can reduce visualclutter by using manual actions to bring objects closer to the eyes,what we refer to as hand-eye coordination. Hand-eye coordinationis therefore hypothesized to be an important ability for controllingand distributing attention. Little is known about how the emergingability to integrate both gaze and manual actions onto objectsimpacts how attention is distributed. Twenty-five infantsparticipated in a naturalistic toy play session that included 24 toys.Overall, infants generated distributions of attention that were right-skewed, reflecting coherence: a composition of selectivity of a fewhighly-frequent toys and exploration of many less-frequent toys. Weobserved that individual differences in hand-eye coordinationimpacted distributions of attention, with infants displaying lowhand-eye coordination having dramatically less coherentdistributions of visual attention during bouts of hand-eyecoordination. These results suggest that hand-eye coordination is acritical pathway for visual attention.

Parafoveal-on-foveal (POF) effects occur when reading time on a fixated word in the fovea is influenced by the upcoming word in the parafovea. Evidence for POF effects have been inconsistent and met with methodological scrutiny (Drieghe, 2011), but recent research suggests that skill differences in spelling may impact POF effects (Veldre & Andrews, 2014). To extend this literature, the current study examines the influence of spelling ability on POF effects by leveraging semantic ambiguity. Participants read sentences containing an ambiguous target immediately followed by a disambiguating word as their eye movements were recorded. Disambiguating words were manipulated to be either consistent or inconsistent with the likely interpretation of the ambiguous word. Results indicate that high-skilled spellers have longer reading times on the target word when the disambiguating word is inconsistent. These findings suggest that POF effects may be possible, particularly within a highly-skilled subset of skilled readers.

MILA-S is an interactive open learning environment for sci-entific modeling (Joyner, Goel, & Papin, 2014). It enablesstudents to build conceptual models of ecological phenom-ena, evaluate them through simulation, and revise the modelsas needed. MILA-S automatically spawns simulations fromthe conceptual models, making modeling easier for the stu-dent. Earlier work had described the use of MILA-S in middleschool. In this paper, we report an experiment on the use ofMILA-S in two college-level classes. In one class, we foundthat almost half of the students showed improved understand-ing of scientific modeling; in the other class, about two thirdsof the students showed enhanced understanding.

How does the mind sustain lengthy, continuous performances?Cognitive processes are continuous, dynamic and adaptive.However, until recently, we didn’t have the methodologicaltools to study these features. In this study, we use DetrendedFluctuation Analysis (DFA) and a sliding window, to analyzethe change in the fractal structure of body movement during thedelivery of an academic lecture. We show that fractal structurevaries widely during performance but also reveals a strongattraction towards 1/f noise. Our analysis also uncover ageneral inverted U pattern in the fractal organization of theperformance: speakers exhibit relatively low exponents (i.e.,less structure) at the beginning of their talk, that then increaseas they get into their performance, and then decrease again asthey finish their narration. This trajectory mirrors the familiaridea of academic lectures as performances in which we set upan argument, develop that argument, and conclude thatargument.

Certain concepts, words, and images are intuitively more sim-ilar than others (dog vs. cat, dog vs. spoon), though quantify-ing such similarity is notoriously difficult. Indeed, this kindof computation is likely a critical part of learning the categoryboundaries for words within a given language. Here, we usea set of 27 items (e.g. ‘dog’) that are highly common in in-fants’ input, and use both image- and word-based algorithmsto independently compute similarity among them. We findthree key results. First, the pairwise item similarities derivedwithin image-space and word-space are correlated, suggest-ing preserved structure among these extremely different rep-resentational formats. Second, the closest ‘neighbors’ for eachitem, within each space, showed significant overlap (e.g. bothfound ‘egg’ as a neighbor of ‘apple’). Third, items with themost overlapping neighbors are later-learned by infants andtoddlers. We conclude that this approach, which does not relyon human ratings of similarity, may nevertheless reflect stablewithin-class structure across these two spaces. We speculatethat such invariance might aid lexical acquisition, by servingas an informative marker of category boundaries.

The n-back task is one of the most popular methods forstudying working memory, and it is tested witheither auditory or visual stimuli. Previous researchcomparing stimulus modalities has demonstrated that auditoryand visual tasks often elicit differential responding and,potentially, different underlying cognitive processes. In thisstudy, performance accuracy and response time weremeasured during an n-back task that varied in termsof stimulus modality and difficulty. Findings demonstrate thatparticipants respond faster but less accurately during a visualas compared to an auditory condition where participants aremore accurate but slower to respond. These results arediscussed in terms of dual coding and feature binding.Implications for the presentation of n-back tasks in studies ofworking memory are discussed.

Deep learning networks have shown impressive performance in object recognition. We used the classification image method to probe whether a deep learning model employs the same features as humans in perceiving real and illusory contours. We adopted a deep learning network, pre-trained with natural images, and retrained the decision layer with laboratory stimuli to perform shape discrimination in the “fat/thin” task. We tested the network with real and illusory contour stimuli contaminated with luminance noise. We found that deep networks trained on natural images can be readily adapted to discriminate between psychophysical stimuli with an extremely high degree of accuracy. However, deep learning networks do not appear to represent illusory contours where they may aid performance in the fat/thin task, a process automatically performed in human vision. This divergence indicates an important difference between the kinds of visual representations formed by deep networks and by humans.

We investigate the effect of manipulating verb tense (e.g, getting $5 vs. will get $5) within a single language on in-tertemporal tradeoffs presented as written stimuli. Verb tense can significantly affect choices between options, with peoplepreferring present-tense options, due to inferences about timing. However, this occurs only in the complete absence ofother timing cues and is eliminated by introducing even vague or non-diagnostic time cues. Gricean maxims of conver-sational implicature say that people maximize relevance and minimize quantity in conversations. Our results suggest thatthat decision makers search across cues for the most relevant information. Tense is deemed to be such a cue in the absenceof other temporal information.

Exposure to mixed evidence can lead to polarization, oradopting a more extreme version of one’s initial attitude. Onepotential reason for this is attitude congruency bias, ratingevidence that supports one's attitude as stronger than evidencethat undermines it. Here we explore factors associated withthis bias and their relationship to attitude change followingexposure to mixed evidence. We conducted several tests,including an attitude survey on two controversial socialissues, a poll regarding participants’ affective involvement ineach issue, an argument rating task, and assessments ofknowledge about social issues and political sophistication.We replicated the attitude congruency bias. Ratings bias wasassociated with affective involvement, but not with measuresof topic knowledge or political sophistication. Attitudechange was predicted by a linear combination of objectiveargument strength and rating bias. Participants’ sensitivity toobjective argument strength suggests the attitude congruencybias does not inevitably lead to polarization.

The reaction time (RT) version of the Perruchet Effect is basedon a concurrent dissociation between RTs to respond andconscious expectancy of the outcome across runs of repeatedtrials. Consequently, the Perruchet Effect is considered strongevidence for multiple learning processes. This conclusion,however, relies on the RT trend being driven by associativelearning rather than, as some have argued, US recency orpriming mechanisms. Recent research examining themechanisms underlying the RT trend do so by examiningmotor activity associated with the response. With this aim inmind, the current study developed, and assessed the usefulnessof, a novel method to measure changes in the amount of forceapplied to the response button in an RT Perruchet paradigm.The results obtained could not be explained by a singlemechanism, but suggest multiple factors underlying the RTversion of the Perruchet effect.

Because learning a second language (L2) is difficult, manylearners start with easy words that look like their native lan-guage (L1) to jumpstart their vocabulary. However, this ap-proach may not be the most effective strategy in the long-term,compared to introducing difficult L2 vocabulary early on. Weexamined how L1 similarity affects pattern learning in L2 byteaching English monolinguals either an Englishlike or Non-Englishlike artificial language that contained repeated patterns.We found that the first words that individuals learned in anL2 influenced which words they acquired next. Specifically,learning a new word in one session made it easier to acquirea similar word in the next session. L2-similarity interactedwith L1-similarity, so that words that looked more like Englishwere easier to learn at first, but they were less effective at in-fluencing later word learning. This demonstrates that althoughnative language similarity has a beneficial effect early on, itmay hinder long-term learning by decreasing recognition of re-peated patterns within a second language. This surprising find-ing demonstrates that making early learning easier may not bethe most effective long-term strategy. Learning difficult vocab-ulary teaches the learner what makes the new language unique,and this general language knowledge about language structureis more valuable than the words themselves. We suggest thatdifficulties during learning are not always to be avoided, as ad-ditional effort early on can pay later dividends.

How do learners’ expectations about teachers’ informativenessshape subsequent learning? Here, we suggest that expecta-tions about teaching style may constrain learning through in-ferences over (1) the amount of information to be learned, and(2) the importance of the demonstrated information. Adult be-havioral data from two experiments conform with our predic-tions: Given a single pedagogical demonstration, as teacherswere expected to share less information, adults inferred thatthere should be more additional information to be learned, andgreater importance of the demonstrated information. Model-ing of these results sheds insight into how adults may be mak-ing these inferences, and provides a framework with which wemay predict future results of children’s exploration followingpedagogical demonstrations from different teachers.

In an optimal stopping problem, people encounter a sequenceof options and are tasked with choosing the best one; once anoption is rejected, it is no longer available. Recent studies ofoptimal stopping suggest that people compare the current op-tion with an internal threshold and accept it when the optionexceeds the threshold. In contrast, we propose that humans de-cide to accept or reject an option based on an estimate of theprobability that a better option will be observed in the future.We develop a computational model that formalizes this idea,and compare the model to the optimal policy in two experi-ments. Our model provides a better account of the data thanthe optimal model. In particular, our model explains how thedistributional structure of option values affects stopping behav-ior, providing a step towards a more complete psychologicaltheory of optimal stopping.

The diversity of a word’s contexts affects its acquisition andprocessing. Can differences between word types such asmonosemes (unambiguous words), polysemes (multiple relatedsenses), and homonyms (multiple unrelated meanings) be re-lated to distributional properties of these words? We tested fortraces of number and relatedness of meaning in vector repre-sentations by comparing the distance between words of eachtype and vector representations of various “contexts”: their dic-tionary definitions (an extreme disambiguating context), theiruse in film subtitles (a natural context), and their semanticneighbours in vector space (a vector-space-internal context).Whereas dictionary definitions reveal a three-way split betweenour word types, the other two contexts produced a two-way splitbetween ambiguous and unambiguous words. These inconsis-tencies align with some discrepancies in behavioural studiesand present a paradox regarding how models learn meaningrelatedness despite natural contexts seemingly lacking suchrelatedness. We argue that viewing ambiguity as a continuumcould resolve many of these issues.

Herein we describe CRAMM, a framework for Causal Reason-ing via Attention and Mental Models. CRAMM develops andextends assumptions made by a previously developed coun-terfactual simulation model of human causal judgment. Weimplement CRAMM computationally and demonstrate how itrobustly captures human causal judgments about simple two-object interactions at the level of underlying cognitive and per-ceptual processes, including data on eye-movements that serveas direct evidence for the role of counterfactuals in causal judg-ment.

The ability to reason about task difficulty is critical for manyreal-world decisions. Building on prior work on preschoolers’inferences about the difficulty of novel physical tasks (Gweonet al., 2017), here we ask whether this ability further supportsrational allocation of effort in collaborative and individual con-texts. When an agent could offer help to someone who had tocomplete a hard task versus someone who had to complete aneasy task, adults and preschoolers offered help with the hardertask (Collaborative Goal). When an agent could choose tocomplete a hard task or an easy task to achieve the same out-come, adults and preschoolers preferentially chose the easiertask (Individual Goal). In the absence of explicit informationabout the relative difficulty of tasks, even young children in-ferred the expected difficulty of tasks and appropriately allo-cated effort across agents and across tasks. Beyond expectingagents to choose actions that maximize their own utility in in-dividual contexts, our results show that even preschool-agedchildren readily understand how deviating from this choice canbe desirable in cooperative contexts.

Theories for visually guided action account for online con-trol in the presence of reliable sources of visual information,and predictive control to compensate for visuo-motor delayand temporary occlusion. In this study, we characterize thetemporal relationship between information integration windowand prediction distance using computational models. Subjectswere immersed in a simulated environment and attempted tocatch virtual balls that were transiently “blanked” during flight.Recurrent neural networks were trained to reproduce subjectsgaze and hand movements during blank. The models success-fully predict gaze behavior within 3◦, and hand movementswithin 8.5 cm as far as 500 ms in time, with integration windowas short as 27 ms. Furthermore, we quantified the contributionof each input source of information to motor output throughan ablation study. The model is a proof-of-concept for predic-tion as a discrete mapping between information integrated overtime and a temporally distant motor output.

Exploration is critical for discovering how the world works.Exploration should be particularly valuable for young children,who have little knowledge about the world. Theories ofdecision-making describe systematic exploration as beingprimarily sub-served by prefrontal cortex (PFC). Recentresearch suggests that systematic exploration predominates inyoung children’s choices, despite immature PFC, suggestingthat this systematic exploration may be driven by differentmechanisms. We hypothesize that young children’s tendencyto distribute attention widely promotes broad informationgathering, which in turn translates to exploratory choicebehavior, and that interrupting distributed attention allocationthrough bottom up attentional capture would also disruptsystematic exploration. We test this hypothesis using a simplechoice task in which saliency of the options was manipulated.Saliency disrupted systematic exploration. These resultssuggest that attentional mechanisms may drive systematicexploratory behavior, and may be part of a larger tendencytoward broad information gathering in young children.

Assessing the quality of a learner’s solution for a given task isan essential step in analyzing a learner’s performance. For awell-defined sequential problem, correctness and optimality ofthe solution as well as its length provide first simple and rea-sonable metrics. However, this ignores the fact that there areconceptually different errors that humans make when solving aproblem. This work proposes a rule-based system of error cat-egories which is able to classify conceptually different errorswith respect to their (assumed) motive. The principles the cat-egories are based on are valid for most well-defined sequentialproblems and can hence serve as a valuable tool in the analy-sis of human solutions for such a problem. In this work, theerror category system is adapted to the game Rush Hour. Weuse the category system as a tool for a detailed analysis of 115human solutions of a Rush Hour game. We found that the mostcommon error type is based on a simple solving heuristic, butmainly occurs in the first half of the solution process. Other er-ror types whose occurrence is numerically less dominant, arestill found in the majority of the solutions. However, they oc-cur in very specific game situations. As a first generalizationapproach of the category system, its application on a furtherdataset containing 56 different Rush Hour tasks and more than31, 000 human solutions yield promising results.

Communication accommodation is a phenomenon in social in-teractions in which people adjust their language to that of theirinterlocutor. A component of communication accommodationis research on power and dominance relations which suggestslanguage use is dependent on power position. There are differ-ent linguistic markers which imply power standing of people.For example, when high power individuals interact with peoplein low power positions, the language of the interaction tends tofollow the language of the high power individuals. While pre-vious studies have mostly focused on the word-level features,we show that not only people in low power mirror word usageof people in high power, but they also adjust their syntacticstructures to those in high power. Notably, we apply a compu-tational tool on two corpora and show that individuals in lowpower align their syntactic structures to those in high powerwhile people in high power do not.

Asking and answering questions is a pervasive activity. Over and above the survival benefits it provides, it is one thatcan be intrinsically pleasurable. Word puzzles provide a window into this process that allow us to go beyond laboratoryinvestigations to capture how question-asking functions in the real world. Analysis of New York Times crosswords,and quiz-show Jeopardy questions allow us to tease apart two phenomena that make for difficult questions: opacity (theindirectness of cues within a clue), and obscurity (the rarity of the answer). Vector-space models of natural language revealhow synergistic cues aid the puzzle-solver, overcoming obscurity in ways that contemplation of cues in isolation can not,and show how these effects compete with the obscurity of the answer itself. Our methods provide new ways to measurethese phenomena in question-asking, and show how they operate in this most basic of behaviors.

number of recent computational models treat concept learn-ing as a form of probabilistic rule induction in a space oflanguage-like, compositional concepts. Inference in such mod-els frequently requires repeatedly sampling from a (infinite)distribution over possible concept rules and comparing theirrelative likelihood in light of current data or evidence. How-ever, we argue that most existing algorithms for top-down sam-pling are inefficient and cognitively implausible accounts ofhuman hypothesis generation. As a result, we propose analternative, Instance Driven Generator (IDG), that constructsbottom-up hypotheses directly out of encountered positive in-stances of a concept. Using a novel rule induction task basedon the children’s game Zendo, we compare these “bottom-up” and “top-down” approaches to inference. We find thatthe bottom-up IDG model accounts better for human infer-ences and results in a computationally more tractable inferencemechanism for concept learning models based on a probabilis-tic language of thought.

Learning a language involves learning how to map specificforms onto their associated meanings. Such mappings canutilise arbitrariness and non-arbitrariness, yet, ourunderstanding of how these two systems operate at differentstages of vocabulary development is still not fully understood.The Sound-Symbolism Bootstrapping Hypothesis (SSBH)proposes that sound-symbolism is essential for word learningto commence, but empirical evidence of exactly how sound-symbolism influences language learning is still sparse. It maybe the case that sound-symbolism supports acquisition ofcategories of meaning, or that it enables acquisition ofindividualized word meanings. In two Experiments whereparticipants learned form-meaning mappings from eithersound-symbolic or arbitrary languages, we demonstrate thechanging roles of sound-symbolism and arbitrariness fordifferent vocabulary sizes, showing that sound-symbolismprovides an advantage for learning of broad categories, whichmay then transfer to support learning individual words,whereas an arbitrary language impedes acquisition ofcategories of sound to meaning.

Prior work suggests that young children consider others’knowledge and expertise to decide from whom to learn. Dochildren also consider how others came to know what theyknow? Here we investigate young children’s sensitivity to theprocess by which people have learned. In Exp.1, 3- to 6-year-olds preferentially sought help from an active learner, who hadfigured out how to solve a problem by herself, over learnerswho had learned through passive observation or direct instruc-tion. Yet, this preference emerged only when the problem chil-dren needed to solve was related to the one the learners hadpreviously solved (i.e., when they thought the active learner’scompetence would be relevant). These findings suggest chil-dren inferred competence from the process of active learning,but considered this competence to be constrained to a partic-ular task rather than more broadly generalizeable. The resultsof Exp.2 (3- to 7-year-olds) suggest that younger children’slearner preference might be driven by more superficial cues re-lated to active learning such as being alone and that a moreabstract understanding of the process of active learning mightdevelop with age.

Enumeration of canonical patterns (e.g., faces of six-sideddice) has generally been characterized by researchers as aholistic process, in which all items are perceived collectively.In previous work, based on a holistic processing view of enu-meration by pattern recognition, we predicted that enumera-tion of canonical forms would not be significantly affected byattentional load. In this paper, we present the results from twoexperiments designed to test this prediction using a divided-attention paradigm. In contrast to our predictions, enumerationof canonical patterns was disrupted by attentional load. Fur-thermore, enumeration of patterns under high attentional loadshowed evidence of conflation between patterns with similarcontours, providing evidence against a holistic processing ac-count of canonical pattern recognition.

What factors affect whether learners adopt a new problem-solving strategy? Potential factors include learners’evaluations of alternative strategies and the degree of similaritybetween their existing strategy and the alternatives. A first stepin answering this question is investigating how people evaluatestrategies. This exploratory study investigated how peopleevaluate strategies for solving algebraic word problems, andhow these evaluations vary as function of individualdifferences. Undergraduates rated three strategies on sixdimensions and judged each pair of strategies for similarity.Factor analysis showed that evaluations could be reduced totwo constructs: efficiency and intuitiveness. We calculatedfactor scores for each participant for each strategy. Efficiencydid not predict similarity ratings on its own, but it did interactwith Need for Cognition. These results suggest stable learnercharacteristics and moment-to-moment evaluations ofstrategies influence judgments about strategy similarity.

Research has examined the role of domain-general cognitive factors in second language (L2) acquisition, with emerging evidence implicating a role for procedural memory, a long- term memory system (e.g., Morgan-Short et al., 2014). Strong conclusions regarding the role of procedural memory are hindered by the lack of knowledge regarding the reliability and validity of procedural memory assessments. In this study, participants completed three assessments of procedural memory that have previously been used to study L2 learning, along with assessments of declarative memory, working memory, and an artificial L2 learning task. Results indicated that the procedural memory assessments generally showed evidence of reliability and discriminant validity, but, somewhat surprisingly, evidence for convergent validity was lacking. Finally, one procedural memory assessment showed predictive validity for the L2 learning task. Implications for future research on the role of procedural memory in L2 acquisition will be considered in light of these results.

The relationship between sentence production and comprehension is at the forefront of psycholinguistic research (e.g.Meyer et al., 2016). Psycholinguists are increasingly interested in cross-linguistic perspectives (e.g. Norcliffe et al., 2015).We report studies of the production and comprehension of variable number agreement in Yucatec Maya, an indigenouslanguage of Mexico. We examined the effects of numerosity through a picture description task involving sets of one,two and seven humans or animals depicting an intransitive action. In production more numerous sets led to higher rates ofplural production. In a timed acceptability decision task, number agreement rather than numerosity significantly facilitatedcomprehension. An interaction revealed that plural marking on the noun facilitated the comprehension of singleton versusnon-singleton sets. In contrast, plural marking on the verb facilitated comprehension of large versus small non-singletonsets. These results suggest divergent effects of numerosity in the nominal and verbal domains.

Motivated by the close relation of predictive coding and activeinference to cognition, we introduce a dynamic artificial neu-ral network-based (ANN) adaptation process, which we termREPRISE: REtrospective and PRospective Inference SchEme.REPRISE first executes a retrospective inference process, in-ferring the unobservable contextual state that best explains itsrecently encountered sensorimotor experiences. It then exe-cutes a prospective inference process, inferring upcoming mo-tor activities in the light of the inferred contextual state anda given goal state. First, the ANN – a recurrent neural net-work – is trained to learn one sensorimotor temporal forwardmodel, that is, the sensorimotor contingencies generated by thebehavior of three moving or flying vehicles. During training,additional three bits are provided as input, indicating whichmode currently applies. After training, goal-directed controland system state inference are activated: Given a goal state,the system imagines a motor command sequence optimizing itwith the prospective objective to minimize the distance to thegoal. Meanwhile, the system evaluates the encountered sen-sorimotor contingencies retrospectively, adapting its vehicleestimation activities and, in order to maintain coherence, theneural hidden states accordingly. This ANN’s ’mind’ is thuscontinuously imagining the future and reflecting on the past –showing superior performance on the posed control problems.The architecture effectively demonstrates that neural error sig-nals and neural activities can be projected into the past and intothe future, respectively, optimizing both neural context codesthat approximately generate the recent past and upcoming be-havior in the light of desired goal states.

Numerous studies have suggested a relationship between working memory and mathematical ability. However, despite theclear relationship between these two constructs, it is still unclear why working memory might be related to mathematicalability. In the current study, we tested three possible theories, the Positive Manifold, a mediation model, and a Transac-tional model. Using path analyses in a structural equation modeling (SEM) framework, fit indices indicated an excellent fitfor the Transactional model, while a poor fit was shown for the remaining models. This finding may suggest that workingmemory and mathematical ability interact in a recursive manner over time, and essentially influence one another over adevelopmental trajectory. Findings may demonstrate the continued importance of working memory early in developmentand understanding how improving working memory may help struggling students in mathematics.

Memory retrieval and probabilistic expectations arerecognized factors in sentence comprehension that capturetwo different critical aspects of processing difficulty: the costof retrieving and integrating previously processed elementswith the new input words and the cost of incorrect predictionsabout upcoming words or structures in a sentence. Althoughthese two factors have independently received substantialsupport from the extant literature, how they interact remainspoorly understood. The present study investigated memoryretrieval and expectation in a single experiment, pitting thesefactors against each other. Results showed a significantinterference effect in both response time to the comprehensionquestions and reading time at the last (spillover) sentenceregion. We also found that the interference effect on readingtime (but not on comprehension question response time) wascanceled when the word at the retrieval site was highlypredictable. Overall, our findings are consistent with thehypothesis of a modulatory effect of expectations on memoryretrieval and with the idea that expectation-based facilitationresults from pre-activation of the target word ahead of time.

Vowel harmony denotes a class of phonotactic constraintswhich limit which vowels can co-occur in words. The charac-teristics of harmony systems have been well-researched fromtheoretical, typological, and developmental perspectives. Chil-dren are sensitive to harmony very early in their development,as young as seven months, so the mechanisms responsible forharmony acquisition must be able to identify its presence aswell as the specifics of individual vowel harmony systemswith little input. Prior computational work has sought eitherto detect the presence of harmony without describing the spe-cific implementation or to describe a specific implementationwhen the general details are known beforehand. We presenta new computational acquisition approach inspired by phono-logical notions of restrictiveness which succeeds in automat-ically detecting harmony in some language and describes thegross characteristics of the underlying harmony grammar with-out prior knowledge about the type of system to expect.

A fundamental question in word learning is how, given onlyevidence about what objects a word has previously referred to,children are able to generalize the total class (Smith & Medin,1981; Xu & Tenenbaum, 2007). E.g. how a child ends upknowing that ‘poodle’ only picks out a specific subset of dogsrather than the whole class and vice versa. The Na ̈ıve Gen-eralization Model (NGM) presented in this paper offers an ex-planation of word learning phenomena grounded in categoryformation (Smith & Medin, 1981) The NGM captures a rangeof relevant experimental findings (Xu & Tenenbaum, 2007;Spencer, Perone, Smith, & Samuelson, 2011), including thosewhich are in conflict with a Bayesian inference theory (Xu &Tenenbaum, 2007).

A child’s perceived ability, over and above actual ability,matters for various behavioral outcomes, academic or personal.In the current paper, we looked at one type of self-efficacy:children’s perceived ability to regulate their own negativeemotions. Our question was whether regulatory emotional self-efficacy (RESE) affects math learning for children who arefaced with homelessness. The specific math enrichmentcentered on child-guided math practice: Children were given acommercially available app and encouraged to pick out theirown practice problems. Our thought was that RESE mightaffect children’s learning when they are given a chance todetermine their own math-practice path. The goal of the currentstudy was to establish this link empirically. The sampleincluded 5- to 12-year-olds who attended a summer programorganized for homeless children. Results confirmed ourhypothesis. Children who scored lowest on the RESE scales (N= 40) benefited less from the math practice than children whoscored highest (N = 46). Specifically, the improvement in mathwas correlated with number of practice sessions only for high-RESE children, not for low-RESE children. These resultssuggest that RESE is an important factor in learning math, tobe considered when developing student-centered pedagogy.

We compared students’ exam performance following one oftwo different types of active learning assignments. In oneversion students read text describing experimental evidence forthe principle being studied. In the other version, studentsinstead created a hypothesis and explanation, and then studiedand explained the results. The content was matched acrossconditions. Students performed better in exams requiringgeneralization to novel situations, after providing hypothesesand explanations than after reading the text and answeringquestions about it. These results suggest that prediction andexplanation cycles might be a better active learning approachto promote generalization and transfer than practice questions.

Some moral philosophers have suggested that a basicprohibition against intentional harm ought to be at the core ofmoral belief systems across human societies. Yet,experimental work suggests that not all harm is viewedequally—people often respond more negatively to harm thatoccurs among fellow social group members, rather thanbetween members of different groups. The present two studiesinvestigated how concerns about social group membershipfactor into the moral judgment system. Adults (N = 111, Study1) and children (N = 110, Study 2) evaluated instances ofinter- and intra-group harm under varying levels of cognitiveload. Both children and adults responded more slowly tointergroup harm than to intragroup harm. Furthermore, adultsunder cognitive load rated intergroup harm more lenientlythan intragroup harm, but adults who were not under loadrated the two types of behaviors similarly. These findingssuggest that across development, evaluations of intergroupharm rely more heavily on conscious deliberation thanevaluations of intragroup harm. Thus, people's evaluations ofharmful behaviors are made in light of information about thesocial category membership of the people involved.

Using an eye-tracking paradigm, we examined the strength and speed of access to lexical knowledge (e.g., our representation of the word dog in our mental vocabulary) and semantic knowledge (e.g., our knowledge that a dog is associated with a leash) via both spoken words (e.g., “dog”) and characteristic sounds (e.g., a dog’s bark). Results show that both spoken words and characteristic sounds activate lexical and semantic knowledge, but with different patterns. Spoken words activate lexical knowledge faster than characteristic sounds do, but with the same strength. In contrast, characteristic sounds access semantic knowledge stronger than spoken words do, but with the same speed. These findings reveal similarities and differences in the activation of conceptual knowledge by verbal and non-verbal means and advance our understanding of how auditory input is cognitively processed.

In face recognition, eye gaze to the eye region is reported to be associated with better performance than to the center of a face. Nevertheless, Caucasians and Asians differ in how much they look at the eyes when they scan a face, but have comparable identification performance. To resolve this issue, here we test the hypothesis that optimal face recognition performance involves a balance between global and local face processing. Thus, Asians may benefit from enhancement of local processing and vice versa for Caucasians. We showed that local attention priming using hierarchical letter stimuli led to more eye-focused eye movement patterns compared to global attention priming in both Asians and Caucasians. However, Asians had better performance after local priming than global priming, whereas Caucasian showed the opposite effect. These results suggest that engagement of global/local attention leads to face-center/eye biased eye movements respectively, and optimal recognition performance involves both global and local processing/gaze transitions between the face center and eyes.

Childrens poor emotional self-regulation is associated withpoor mental health outcomes. This study presents methods thatimprove prediction rates of polite and frustrated speech usinglinguistic cues. These improvements can be used to help auto-matically identify characteristics of poor self-regulation in fu-ture studies. This work adds to previous research by consider-ing existing computer science, psychology, and psycholinguis-tics methodologies and findings. More specifically, featuresassociated with childrens cognitive control capacities acrossage groups are considered to investigate acoustic, semantic,and syntactic features in speech. The current analyses indi-cate that the features most predictive for polite and frustratedspeech differ, a combination of features work best for predict-ing both speech types, and the predictive quality of featuresdo not vary substantially by age. Further work should be con-ducted to clarify how well these findings transfer to general andclinical populations as well as to consider the developmentalnorms of different age groups.

Research has shown that individual variation in our bodies,such as differential hand dominance, can influence the way thatwe interact with and perceive the world (Casasanto, 2009). Forexample, right-handed individuals are more likely to associatetheir right spatial plane as more positive than their left, an effectthat is switched in left-handed individuals. Here, we exploredwhether asking participants to use their dominant (“good”)versus nondominant (“bad”) hand on a motor task influencedsubsequent valanced face judgment. Results demonstrate thatsimply asking a participant to use their right or left hand tocomplete a task can have a significant effect on the perceivedvalence of neutral faces. These findings add to the evidencethat the way we physically interact with our world may haveimportant consequences for our perceptions of social stimuli.

Illusory causation is a consistent error in human learning inwhich people perceive two unrelated events as being causallyrelated. Causal illusions are greatly increased when the targetoutcome occurs frequently rather than rarely, a characteristicknown as the outcome density bias. Unlike most experimentaldesigns using binary outcomes, real-world problems to whichillusory causation is most applicable (e.g. beliefs aboutineffective health therapies) involve continuous and variableconsequences that are not readily classifiable as the presenceor absence of a salient event. This study used a causallearning task framed as a medical trial to investigate whetheroutcome density effects emerged when using a continuousand variable outcome that appeared on every trial.Experiment 1 compared the effects of using fixed outcomevalues (i.e. consistent low and high magnitudes) versusvariable outcome values (i.e. low and high magnitudesvarying around two means in a bimodal distribution).Experiment 2 compared positively skewed (low density) andnegatively skewed (high density) continuous distributions.These conditions yielded comparable outcome density effects,providing empirical support for the relevance of the outcomedensity bias to real-world situations in which outcomes arenot binary but occur to differing degrees.

Abstract: When inferring the extent of others’ knowledgefrom samples of what they know, certain kinds of samplesimply richer content. One candidate kind is knowledge ofcausal mechanism. In the current study, we investigatewhether children and adults think that knowledge aboutmechanism generalizes more broadly than non-mechanisticfactual knowledge. We find an early-emerging assumptionthat mechanistic knowledge about a basic level categoryimplies greater knowledge about a superordinate category,compared to factual knowledge about the same basic levelcategory. Even young children have a sophisticated sense ofhow causal mechanisms generalize across categories, despitepossessing little mechanistic knowledge themselves. Theseintuitions likely support the epistemic inferences we makefrom early childhood onward.

Anchoring is a well-known, robust effect causing estimates tobe biased towards previously seen values – regardless of theirrelevance. Reducing anchoring bias is important for optimizingestimation. Herein, we tested the MOLE (More-Or-LessElicitation) tool’s ability to limit the impact of anchors onestimates. In a direct elicitation task, 62 participants’ bestestimates correlated with anchor values at 0.27 whereas, whenusing the MOLE, this relationship disappeared (r = .02).Results also showed, however, that expertise reduces theimpact of anchoring (r = -0.46). We conclude that use of theMOLE assists in avoiding anchoring and that this will be mosthelpful in areas of high uncertainty.

Recent research has seen a surge in interest in the role of theindividual in sound change processes. Do fast speakers have aunique role in sound change processes? Fast speech leads togreater rates of lenition (reduction). But should it mean thatfast talkers would be more likely to lenite even when speak-ing slowly? In two corpus studies we show that even whenfast talkers speak more slowly they are (a) more likely to omitsegments and (b) more likely to perform variable reduction ofconsonants. This draws attention to habitual speech rate as alikely factor in the actuation of lenition processes.

We present data on convergence in the Switchboard corpus, ad-dressing differences across measures and across speakers. Wemeasured convergence in four characteristics, to test consis-tency in related and unrelated measures: F0 median, F0 vari-ance, speech rate, and odds of the fillers uh and um. Conver-gence was significant in all measures and exhibited variationboth between individuals and within individuals. Most notably,convergence in one measure was not predictive of convergencein other measures, except between closely related measures.The results demonstrate some of the limitations of generaliz-ing convergence results from one measure to other measures.

With the integration of emoji into digital keyboards, people areincreasingly using multimodal interactions between text andimage in real-time interactions. One technique of using emojiis to substitute them into sentences. We here investigate theonline processing of these interactions, by modulating eitherthe grammatical category of those substitutions (Experiment 1:nouns vs. verbs) or the type and location of substitutions(Experiment 2: emoji vs. logos, within sentences vs. at theirend). We found a processing cost for self-paced reading timesof images compared to words, which indeed extended past theemoji itself, but no difference in comprehensibility ratingsbetween word and congruent-image substitutions. Overall,these results suggest that, despite costs of switching modalities,text and images can be integrated into holistic multimodalexpressions.

This study reports one eye-tracking and one sentencecompletion study investigating the antecedent biases ofTurkish-speaking L2 speakers of English, for anaphor it anddeixis this. Our results show L2 speakers displayed native-like sensitivity to the type of antecedents while using it andthis in sentence completion, but this sensitivity was notreplicated in our online reading experiment. This showslimitations in L2 speakers’ use of information in onlinereading, and poor performance in making use of pragmaticchanges in context to track the antecedents of it and this.

Affective states, exploration, and learning are tightly inter-twined. For example, research has connected surprise to playand learning in early development (Stahl & Feigenson, 2015),but less is known about the potential impact of other affec-tive states and how they might influence exploration and sub-sequent discovery. Given that past research has suggested thatawe may increase feelings of uncertainty and lead to pursuitof cognitive accommodation in adults (Valdesolo & Graham,2014), we posit that awe-induced uncertainty may similarlylead children to think-outside-the-box and explore more duringplay. In Experiment 1, we modify emotion-inducing videos(Awe, Happy and Calm) and validate them on adult partic-ipants using the perceived self-size Circle Task (Bai et al.,2017). In Experiment 2, children were presented with one ofthe three videos and their exploratory play with a novel toy wasrecorded. Results revealed both a significant effect of the ma-nipulation (children associated with smaller selves in the Awecondition) and also an influence of the videos on play. Childrenin the Awe condition played more and explored more variablythan children in the control conditions. These results suggestthat awe influences motivation that increases variability anddiscovery in exploration.

Information initially presented as a likely cause of an eventbut turns out to be incorrect can affect people’s reasoningdespite being clearly corrected – a phenomenon known as thecontinued influence effect of misinformation. The presentwork extends previous findings showing that misinformationthat implies a likely cause of an adverse outcome is moreresistant to correction than misinformation that explicitlystates a likely cause. Participants either read a reportdescribing a fire or a crash. The difference between impliedand explicitly stated misinformation was replicated with thefire scenario, which has been commonly used in continuedinfluence research. There was little evidence of a continuedinfluence of misinformation for the (novel) crash scenario.The results constrain the generalizability of the continuedinfluence effect and suggest that corrections that clearlyinvalidate initial misinformation can be effective.

People’s reasoning about physical and social explanations iswell understood (Keil, 2008). However, less is known abouthow people reason about mathematical explanations (Johnsonet. al., 2017). Experiment 1 replicates the central result ofJohnson et. al (2017), that people impose order on simplearithmetic explanations, as well as sets the limits of thatpreference. Experiment 2 extends the results of a second factor,the character of the relationship between the operations relatedby the explanation.

The goal of this study is to explore the predictive capability ofseveral psychosocial variables, such as personality and groupcohesion, towards determining multiplayer online battle arenagame outcomes - namely diversity, cohesion, and resilience, oncollective performance. Our study finds that measures ofindividual and team perceptions of qualities provided a usefulprecursor for match victory. Using individual-level attributes,our cohesion survey questions provided the highest predictivevalue, and higher levels of perceived cohesion were associatedwith higher victory odds. In light of our results, we discuss theimplications of using behavioral data derived from onlinegames and opportunities for future large-scale game datacollection.

Frequency often predicts when children will acquire unitsof language such as words or phones. An additionalpredictor of language development may be a phone’sfunctional load (FL), or the contrastive work a soundperforms in a language. A higher FL may correlate withearlier phone emergence in child speech as childrenselectively converge upon the most meaningful contrasts intheir input. This hypothesis is tested across fivetypologically diverse languages that vary by phoneinventory size and structure as well as word composition.Consonant FL was calculated over more than 390,000words of child-directed speech. Results demonstrate thatthe relationship of frequency and FL to speechdevelopment is dependent upon the language of exposure.Models fit to bootstrapped corpus data suggest thatfrequency may be the stronger of the two parameters.

The semantic memory literature has recently seen the emergenceof predictive neural network models that use principles ofreinforcement learning to create a “neural embedding” of wordmeaning when trained on a language corpus. These models havetaken the field by storm, partially due to the resurgence ofconnectionist architectures, but also due to their remarkablesuccess at fitting human data. However, predictive embeddingmodels also inherit the weaknesses of their ancestors. In this paper,we explore the effect of catastrophic interference (CI), long knownto be a flaw with neural network models, on a modern neuralembedding model of semantic representation (word2vec). We usehomonyms as an index of bias depending on the order in which acorpus is learned. If the corpus is learned in random order, the finalrepresentation will tend towards the dominant sense of the word(bankà money) as opposed to the subordinate sense (bankàriver). However, if the subordinate sense is presented to thenetwork after learning the dominant sense, CI produces profoundforgetting of the dominant sense and the final representationstrongly tends towards the more recent subordinate sense. Wedemonstrate the impact of CI and sequence of learning on the finalneural embeddings learned by word2vec in both an artificiallanguage and in an English corpus. Embedding models show astrong CI bias that is not shared by their algebraic cousins.

Two experiments examined whether an explicit attention toanother’s perspective fosters perspective-taking. The firstexperiment attempted to replicate Todd et al.’s (2010) findingsthat a mind-set focusing on self-other differences incitesrespondents to adopt another person’s perspective in asubsequent task. Results showed that perceivers focusing onself-other differences were just as likely to describe an object’slocation from their egocentric perspective as perceiversfocusing on self-other similarities. The second experimentintensified perceivers’ awareness of self-other differences byallocating them to one of the perspective-settings (none, self-focus, other-focus). Participants in the perspective-settingsreceived explicit instructions to regard their own (self-focus)or another person’s (other-focus) viewpoint during theperspective-taking task. Findings revealed that other-focusedrespondents were more likely to adopt another person’sperspective than self-focused respondents. Compared to thebaseline, however, an explicit self- or other-focus did not fosterperspective-taking. Our findings indicate the robustness ofrespondents’ egocentric bias.

This study investigated whether an explicit and stimulatedattention to the mental states of an uninformed other fostersperspective-taking. The experimental aim of this study wastwofold. First, we aimed to replicate Keysar’s (1994) curse ofknowledge effect, indicating how privileged information biasescorrect perspective-judgments. The second aim was toinvestigate whether this curse of knowledge effect diminishesby explicit instructions to become aware of another person’sperspective. Findings showed that we replicated Keysar’s(1994) curse of knowledge effect. Perceivers were more likelyto impute their perception of speaker’s sarcasm onto anuninformed addressee when their privileged informationsuggested that the speaker was being sarcastic rather than beingsincere. Findings further revealed that perceivers were just aslikely to overestimate the extent to which their privateperspective was shared by an uninformed addressee, regardlessof their explicit and stimulated attention to this addressee’sperspective.

To this day, a great variety of psychological theories of reason-ing exist aimed at explaining the underlying cognitive mecha-nisms. The high number of different theories makes a rigorouscomparison of cognitive theories necessary. The present articleproposes to use Multinomial Processing Trees to compare twoof the most prominent theories of syllogistic reasoning: theMental Models Theory and the Probability Heuristics Model.For this, we reanalyzed data from a meta-analysis on six stud-ies about syllogistic reasoning. We evaluate both models withrespect to their overall fit to the data by means of G2, AIC,BIC, and FIA, and on a parametric level. Our comparison in-dicates that a MMT-variant, though having more parameters, isslightly better on all criteria except of the BIC. Yet, none of thetwo models, realized as MPTs, is clearly superior. We outlinethe impact of the different theoretical principles and discussimplications for modeling syllogistic reasoning.

How may controlled breathing be beneficial, or detrimental to behavior? Computational process models are useful to specify the potential mechanisms that lead to behavioral adaptation during different breathing exercises. We present a physio-cognitive model of slow breathing implemented within a hybrid cognitive architecture, ACT-R/Φ. Comparisons to data from an experiment indicate that the physiological mechanisms are operating in a manner that is consistent with actual human function. The presented computational model provides predictions of ways that controlled breathing interacts with mechanisms of arousal to mediate cognitive behavior. The increasing use of breathing techniques to counteract effects of stressors makes it more important to have a detailed mechanistic account of how these techniques may affect behavior, both in ways that are beneficial and detrimental. This multi-level understanding is useful for adapting to changes in our physical and social environment, not only for performance, but for physical and mental health.

An important challenge for human-like AI is compositional se-mantics. Recent research has attempted to address this by us-ing deep neural networks to learn vector space embeddings ofsentences, which then serve as input to other tasks. We presenta new dataset for one such task, “natural language inference”(NLI), that cannot be solved using only word-level knowledgeand requires some compositionality. We find that the perfor-mance of state of the art sentence embeddings (InferSent; Con-neau et al., 2017) on our new dataset is poor. We analyzethe decision rules learned by InferSent and find that they arelargely driven by simple heuristics that are ecologically validin its training dataset. Further, we find that augmenting train-ing with our dataset improves test performance on our datasetwithout loss of performance on the original training dataset.This highlights the importance of structured datasets in betterunderstanding and improving AI systems.

High immersion in virtual reality is often hypothesized to improve learning and memory. This immersion benefit is frequently attributed to presence, the user’s feeling of being present inside the computer-generated environment. Obtaining learning gains due to high immersion may however be difficult, as is evidenced by the null results of multiple studies in this area. In the current study we investigated the role of presence in low- and high-immersion virtual reality settings. No differences in performance in object location and spatial memory were found between low- and high-immersive conditions. Yet, when considering self-report measures of presence, performance improvements in the high immersive condition did become apparent. The finding of the importance to consider the role of presence in virtual reality highlights the complexity of immersion effects in simulated environments

To what extent can the similarity structure of categories beinferred based on paradigmatic vs syntagmatic information?We explore this question in two studies that aim to captureparadigmatic information directly: first by having participantsgenerate near-neighbors to exemplars from 15 basic categories,and second by having them partially rank the most similar ex-emplars. After constructing neighborhood graphs of the itemsin each category, we derived a local measure (based on di-rect neighbors) and a global measure (including indirect pathsas well) of paradigmatic information. Both measures predictindependently-obtained human pairwise similarities for eachcategory, but incorporating indirect information substantiallyimproves this prediction. In a third study, we contrast thesemeasures with syntagmatic information obtained from a vast se-mantic network derived from 3 million judgments. The paradig-matic graphs are better predictors of similarity despite onlyencoding a fraction of these data. Broad implications for wordlearning and meaning are discussed.

How people learn and produce sequential actions (e.g., making coffee) has been the subject of empirical and theoreticalscrutiny, for it covers most human activities. One useful distinction is between stimulus-based control, in which actionselection is driven largely by the environment, and plan-based control, which assumes learning of structured sequences ofactions and effects. Task demands, instructions, and participants’ individual abilities and inclinations can all modulate thecontrol mode used. We investigate two sequence learning tasks, with one key difference: in the cued task either controlmode is possible, while learning in the reinforcement task requires plan-based control. Using measures of visuospatialworking memory (VWM) capacity, locus of control, need for structure, and IQ, we seek to explain individual differencesin choice of control mode and task performance, establishing a link between VWM capacity and performance, as well asexplicit knowledge evidencing plan-based control.

People learn and use complex sequential actions on a daily basis, despite living in a high-dimensional environment andbody. Sequential action learning is sometimes studied in cognitive psychology using button-pressing tasks such as Nissenand Bullemers (1987) serial respone time (SRT) task. However, the SRT task only measures the speed of button presses,neglecting the richand difficult to controltrajectory of the arm, which can show predictive movements and other contextualeffects. In this study, we evolve neural networks to carry out a mouse-based SRT task under conditions of differingprediction uncertainty. We replicate behaviors found in a recent human experiment, and explore ramifications for humansequence learning.

In this paper we demonstrate how to use graph matching touncover heterogeneity in the structure of preferences acrossa population of decision-makers. We propose a novel non-parametric approach to formally capture the concept of pref-erence structure using preference graphs, thereafter clusteringdecision-makers based on graph embedding methods. We ex-plore the approach with simulated choice and empirical datafrom the most common classes of economic and psychologicalmodels. The approach uncovers heterogeneity in preferencestructure across a variety of dimensions, without requiring anyprior knowledge of those structures.

People reliably make two errors when making inferences aboutthree-variable causal structures: they violate what is known asthe Markov assumption (mediation) on causal chains andcommon cause structures, and fail to sufficiently ‘explainaway’ on common effect structures. Our goal for the presentstudy was to quantitatively compare these two errors aftersubjects have learned the statistical relations between threevariables using procedures designed to maximize the accuracyof their learning and inferences. Aligning with prior research,we found that subjects violated the Markov assumption, anddid not sufficiently explain away. We also found judgmentsabout mediation were worse than judgments about explainingaway for one inference, but better for another, suggesting thatpeople are not uniquely worse at reasoning about one structurethan another. We discuss the results in terms of a theory of cueconsistency.

The ability to infer other people’s emotions is an important aspect of children’s social cognition. Here, we examined whether 4- to 6-year-olds use probability to infer other people’s happiness. Children saw a scenario where a girl receives two desired and two undesired gumballs from a gumball machine and were asked to rate how the girl feels about this outcome. Children either saw the gumballs come from a machine that had mostly desired gumballs or a machine that had mostly undesired gumballs. Five- and 6- year-olds rated the girl as being happier when the gumballs came from a machine that had mostly undesired gumballs. Four-year-olds, on the other hand, rated the girl’s happiness similarly regardless of whether the machine held mostly desired or undesired gumballs. These findings show that by the age of 5, children use probability to infer happiness. Further, they demonstrate that children understand that our happiness with an outcome depends on whether a better or worse outcome was initially more likely.

Cue competition effects in human contingency learningappear to be sensitive to the causal nature of cue-outcomerelationships. While blocking effects are reliablydemonstrated in scenarios where cues are presented as causesof outcomes, several studies have failed to find blocking inscenarios where cues are presented as effects of outcomes, afinding that is typically taken as evidence for the involvementof controlled reasoning processes in cue competition. Thesestudies typically measure blocking with continuous causalratings about individual cues. Previous studies have foundthat sensitivity to causal model may depend on how the testquestion is phrased. In contrast, the current study tested thesensitivity of blocking to causal scenarios across differentformats of the same test question. Participants completed acausal learning task with instructions suggesting either apredictive (i.e. cue causes outcome) or diagnostic (cue iscaused by outcome) cue-outcome relationship. Participantswere then asked about the likelihood of outcomes occurringby either giving a continuous rating of each outcome or adiscrete choice about the most likely outcome. Whenmeasured by continuous ratings of individual cues, blockingwas evident in predictive, but not diagnostic scenarios.However, when measured by discrete choice or using acompound negation test, blocking was robust and insensitiveto causal scenario. The results suggest that contributions ofpredictive memory and causal reasoning to cue competitioneffects may depend substantially on the type of measure used.

In the last decades there was a shift from more logically in-spired theories describing human reasoning towards the newparadigm of probabilistic approaches. One of the most promi-nent models for syllogistic reasoning is the Probability Heuris-tics Model (PHM) which has been formulated based on fiveheuristics. The contribution of this article is: (i) to provide ananalysis of different formalizations of the PHM, (ii) to exam-ine the impact of each heuristic, and (iii) to identify possibleviolations of underlying assumptions in present implementa-tions. A systematic analysis of the model parameters shows asurprising variation in parameter values across experiments. ABayesian modeling approach explains this variance of param-eters. Implications for probabilistic approaches are discussed.

This paper presents a motivation-based model in order to explore crowd behavior. The case study is about what motivates the decision processes of passengers about choice of location on the station platform for ingressing and egressing trains. The goal of the research is twofold: to establish a cognitive generic crowd behavior modeling method and to respond to a major challenge of public transportation: to reduce dwell time to ensure a high level of service. We first introduce motivation-based modeling for the simulation of the dynamics of numerous cognitive agents and report the collection of passengers’ dynamics that was done through an extensive survey observation. Most significant variables were then extracted from factor analysis to compose and distinguish six main motivation based strategies that are to be used for the simulation of crowd behavior in the train station. Discussion is about the advantages of motivation-based simulation in terms of robustness and adaptability and conclusion about how Artificial Intelligence, Cognitive Psychology and Data Science operate together to model such complex systems.

This study is an exploratory analysis of young children’s representation of five spatial terms: above, under, by, next to, and between. Children (n = 76) and adults (n = 11) indicated the spatial extent of a grid they thought each term indicated. Qualitative analyses were used to categorize responses, separately for each word, and showed more agreement among adults than children. Furthermore, children who showed adult- like representations were generally older than those who showed unsystematic responses. Quantitative analyses, using a median split in age to create two groups of children, compared representational sizes and distances from the referent(s). For above, under, and between, adults had larger representations than children; the trend was reversed but not significant for by and next to. Furthermore, representation size was correlated for above and under, but not for by and next to. Analyses of distances showed a predicted reversal in the vertical dimension of above and under that interacted with age. There were no differences across age groups or terms for by and next to, but between showed a decrease in horizontal distance over development. These results suggest that children may initially understand words differently than adults do.

People use rich prior knowledge about the world in order toefficiently learn new concepts. These priors–also known as“inductive biases”–pertain to the space of internal models con-sidered by a learner, and they help the learner make inferencesthat go beyond the observed data. A recent study found thatdeep neural networks optimized for object recognition developthe shape bias (Ritter et al., 2017), an inductive bias possessedby children that plays an important role in early word learning.However, these networks use unrealistically large quantities oftraining data, and the conditions required for these biases to de-velop are not well understood. Moreover, it is unclear how thelearning dynamics of these networks relate to developmentalprocesses in childhood. We investigate the development andinfluence of the shape bias in neural networks using controlleddatasets of abstract patterns and synthetic images, allowing usto systematically vary the quantity and form of the experienceprovided to the learning algorithms. We find that simple neuralnetworks develop a shape bias after seeing as few as 3 exam-ples of 4 object categories. The development of these biasespredicts the onset of vocabulary acceleration in our networks,consistent with the developmental process in children.

Fairness depends on the principles that people use to justifytheir actions, and on the outcomes that they produce. Here wepropose that, from early in childhood, we also judge fairnessbased on whether we believe the resulting outcomes werecaused by the underlying principles. In Experiment 1 we showthat four- five- and six-year-olds believe that an agent who paidattention when distributing resources is more fair than an agentwho was distracted when distributing resources, even whenthey both produce identical outcomes. In Experiment 2 weshow that children of the same ages believe that an agent whocounts when distributing resources is more fair than an agentwho does not count, even when both agents attend to how theydistribute their resources and produce identical outcomes.Together, our findings suggest that children do not judgefairness based on the outcome alone, and they add to a growingbody of work suggesting that, from early childhood, ourintuitions about fairness are tightly linked with intuitions aboutexactness.

People frequently offload cognitive tasks onto the environment by,for example, digitally storing information they want to rememberlater. This frees up cognitive resources, leading to an increasedability to learn new information (the “Saving-Enhanced MemoryEffect”). We tested whether this effect would generalize beyondthe digital realm. On every trial, participants studied two printedlists of words before being tested on their memory for both lists.For half the trials, participants shredded the first list beforeattempting to learn the second one. For the remaining trials, theysaved the first word list in a folder before learning the second list.Results revealed a robust Saving-Enhanced Memory Effect, aspeople remembered more words on average from the second listwhen they had saved the initial word list. These findings suggestthat the effects of offloading memories onto the external world aresimilar for information stored in digital and physical formats.

While statistical learning (SL) has long been described as alearning mechanism that operates automatically across agesand modalities, there are a growing number of cases in whichstatistical regularities are not learned automatically, and inwhich attention seems to impact learning. We examined therole of attentional instruction on adults’ ability to learn twostatistical patterns simultaneously. Results suggest that evenwithout explicit instruction to attend to either pattern,participants automatically learn both patterns, and thatexplicit instruction to attend to one or both streams improveslearning, but only for the attended stream(s). In addition,when attention is directed at only one stream, the learningbenefit for that stream is coupled with a learning cost for theunattended stream. This adds to our understanding of thenuanced relationship between attention and SL, by suggestingthat when more than one structure is present attentionselectively improves SL of attended information in adults, butat the cost of unattended information.

Scientific communication includes primary scientificliterature written by and for scientists, as well as pressreleases written about these scientific articles that are used toinform the popular press. By the time new scientific findingsare reported by the press, the reporting can often reflect 'spin',or reporting that minimizes uncertainties and exaggeratesimpact, as compared to the original study. In this work, weexamine the role that the press release may play incommunicative change, in particular with respect todifferences in portrayed confidence between abstracts ofscientific articles and press releases. We examine a largecorpus of over 15,000 documents collected from onlinedatabases covering a range of scientific topics, leveragingautomated analysis tools from natural language processing toexamine how the readability, sentiment, subjectivity, andportrayed confidence varies between primary literature andpress releases. We find that press releases are often easier toread, portray more positive sentiment, use language thatimplies greater objectivity, and demonstrate higher confidencein the findings. Future work should focus on examining ifthese differences between press releases and primary articlesdo indeed engender different perceptions in readers.

Are the perceptual and cognitive changes associated with expertise due to improvements in domain-general abilities orchanges to domain-specific representations? Elmer et al. (2014) measured how controls, language experts, and musicianscategorized perceptually ambiguous sounds (blends of speech and music) and concluded domain-general changes underlieexpertise. Acoustic and perceptual analyses of their stimuli suggested their stimulus creation methodology might havedistorted the results. An experiment replicated and extended their findings with revised stimuli. Results suggest thatexpertise leads instead to domain-specific changes in representational weighting or selective attention.

We present a computational model of the acquisition of German case that is evaluated against empirical data obtained from naturalistic speech. The model substitutes nouns into existing contexts, and proceeds through a number of stages that reflect increasing knowledge on the part of a child, both of the determiner-noun sequences that are legal in German, and of the determiner-noun sequences that are appropriate in specific sentential contexts (cases). The model provides a natural account of gender and case errors, the two most common error types produced by children, and shows the highest error rates in dative contexts and lowest error rates in nominative contexts, as is true of children learning German. However, the model’s error rates in the early stages are considerably higher than those shown by children, suggesting that children possess a fairly sophisticated representation of how lexical contexts assign case from a relatively early age.

Language that describes people in a concise manner may con-flict with social norms (e.g., referring to people by their race),presenting a conflict between transferring information effi-ciently and avoiding offensive language. When a speakeris describing others, we propose that listeners consider thespeaker’s use or absence of potentially offensive language toreason about the speaker’s goals. We formalize this hypothe-sis in a probabilistic model of polite pragmatic language un-derstanding, and use it to generate predictions about interpre-tations of utterances in ambiguous contexts, which we testempirically. We find that participants are sensitive to poten-tially offensive language when resolving ambiguity in refer-ence. These results support the idea that listeners representconflicts in speakers’ goals and use that uncertainty to inter-pret otherwise underspecified utterances.

Children learning their first language face multiple problemsof induction: how to learn the meanings of words, and howto build meaningful phrases from those words according tosyntactic rules. We consider how children might solve theseproblems efficiently by solving them jointly, via a computa-tional model that learns the syntax and semantics of multi-word utterances in a grounded reference game. We select awell-studied empirical case in which children are aware of pat-terns linking the syntactic and semantic properties of words –that the properties picked out by base nouns tend to be relatedto shape, while prenominal adjectives tend to refer to otherproperties such as color. We show that children applying suchinductive biases are accurately reflecting the statistics of child-directed speech, and that inducing similar biases in our compu-tational model captures children’s behavior in a classic adjec-tive learning experiment. Our model incorporating such biasesalso demonstrates a clear data efficiency in learning, relative toa baseline model that learns without forming syntax-sensitiveoverhypotheses of word meaning. Thus solving a more com-plex joint inference problem may make the full problem of lan-guage acquisition easier, not harder.

Do students learn better with material that is perceptually harder-to-process? Previous research has been equivocal concerning thisquestion. To clarify these discrepancies, the present studyexamined two potential boundary conditions to determine whendisfluent text is, and is not, beneficial to learning. The twoboundary conditions examined were: type of judgement oflearning (JOLs) and testing expectancy. Boundary conditionswere examined in separate Group (incidental aggregate JOLs vs.intentional aggregate JOLs vs. item-by-item JOLs) by Disfluency(Masked vs. Nonmasked) mixed ANOVAs. Results revealed thattype of JOL did not moderate the disfluency effect, but testingexpectancy did. These results bring forth questions pertaining tothe utility of disfluency on learning.

Although children become adept problem-solvers early in life,creating tools to solve novel problems remains challengingthroughout the early school years. To explore this problem,we gave one group of 4-7-year-old children (N = 25) theopportunity to compare multiple materials with matchingfunctional properties across three trials. A second group (N =26) saw the same materials in each trial. We consideredwhether children improved across trials and whether theytransferred any learning to a new exemplar that required adifferent functional technique. Although children learnedequally well across the first three trials regardless ofcondition, children who had the chance to compare materialswere more likely to improve from the initial trial to thetransfer trial. We discuss the implications for identifying theorigins of innovative problem-solving.

Research on the development of future hypothetical andcounterfactual thinking suggests that children as old as fivemay be unable to consider multiple, equally probablepossibilities simultaneously. Yet, a large literature on thedevelopment of causal reasoning suggests that much youngerchildren are able to generate, evaluate, and test causalhypotheses, often by integrating information about severalcandidate causes at once. The current research seeks to bridgethese two bodies of research. In three experiments, adults andtoddlers (18–30 months) observe a sequence of evidence thatis equally consistent with two hypotheses, each occupying adifferent level of abstraction (individual vs. relational).Results suggest that learners generate more than one potentialcause, hold both in mind, and flexibly apply the appropriatehypothesis to inform their inferences at test. Findingschallenge previous suggestions that much older children failto consider multiple, equally probable possibilities.

Books designed for beginning readers typically intermix textwith illustrations in close proximity. Prior research suggeststhis standard layout may reduce literacy skills due toincreased attentional competition between text andillustrations. The current study extends this work byexamining whether manipulations to the book layout canenhance reading performance and explores whether individualdifferences in executive function and processing speed arerelated to children’s decoding and reading comprehensionwhen reading books which utilize the standard layout.Separating text and illustrations improved readingcomprehension. Preliminary results also suggest workingmemory, inhibitory control, and processing speed are relatedto reading performance.

Attitudes toward statistics influence the way students engagein learning statistics. This study examined how attitudestoward statistics were related to the comprehension of astatistics text and the accuracy with which learners judgedtheir comprehension. Results showed that more negativeattitudes were associated with lower performance onprocedural comprehension questions, but not on conceptualcomprehension questions. At the same time, more negativeattitudes resulted in overestimations of proceduralcomprehension when making prospective and retrospectivejudgments of comprehension. To explain the findings, wedraw on theoretical models that assume that learners usedifferent types of cues to make comprehension judgments.

The present study was conducted to determine the effect ofinhibition and working memory training on analogicalgreasoning and theory of mind in young children. We presentthe results of 58 4-year-old children who were given a pre-test and post-test with analogical reasoning tasks and falsebeliefs tasks. Between the pre-test and the post-test a specifictraining was provided. Children were divided in three groupsaccording to the type of the training: a) group with inhibitiontraining; b) group with working memory training; c) controlgroup with conservation tasks training. Each training was 7days long, 25 minutes per child every day. The resultsshowed a significant increase in the post-test results of thegroups undergoing inhibition and working memory trainings.The performance of the children tested was significantlybetter on the post-test in comparison to both the pre-test andthe control group. The results clearly indicated the relation ofinhibition and working memory to analogical reasoning andfalse belief understanding, and also the importance of trainingsuch executive functions in order to increase other cognitiveabilities.

Relational processing has been linked to cognitive capacitymeasures, such as working memory and fluid intelligence.Sufficient capacity, however, does not ensure attention torelational structure, as propensity for relational processing mayalso be driven by an individual’s cognitive style. The currentstudy took an individual-differences approach to investigate theprerequisites for relational processing. College studentscompleted a battery of standardized tests of individualdifferences related to fluid intelligence and cognitive style, aswell as a series of experimental tasks that require relationalreasoning. Moderate correlations were obtained betweenrelational processing and measures of cognitive capacity, whilethe influence of cognitive style was restricted to individuals withgreater cognitive capacity. These results support the hypothesisthat a capacity threshold exists, above which cognitive styleimpacts relational processing.

Travel behavior research shows that the disutility of waiting times looms larger than the disutility of in-vehicle times.However, little has been said about the plausibility of the assumption of compensatory behavior in the preferences forwaiting and traveling. Another open question is whether the variability in waiting and in-vehicle times affects transportdecisions in the same way. To answer these research questions, we conducted a lab experiment with university studentsfrom London, UK and Santiago, Chile. Participants were presented with 14 decisions scenarios that manipulated theaverage and the variability of waiting and in-vehicle times in two bus routes under the choice paradigms of decisions fromdescription and from experience. We found that participants did not compensate waiting and in-vehicle times; rather, theysought to minimize overall journey times. In addition, participants disliked more variability while waiting than traveling.Interestingly, both behaviors were only observed in the experiential choices.

Learning from expository science texts is challenging. These studies explore whether difficulties can be attributed to poor memory or poor reasoning. To eliminate the need for memory during testing, some students took the tests with the texts available. To test for the effects of reasoning on performance, some students were prompted to engage in explanation activities during or after reading. The effects of these manipulations were tested on text-based and inference questions. Allowing the reader access to the texts during testing improved performance for text-based questions. In contrast, engaging in explanation activities during reading improved performance on inference questions. These results suggest that achieving a better understanding from expository texts depends on engaging in constructive reasoning processes, and not simply improving memory for the texts.

The folk theory of gender seems to involve two contradictorybeliefs that people can hold simultaneously. One belief is thatgender is biologically determined and immutable, and theother is that one has to earn gender membership by followinggender norms or otherwise risk disqualifying oneself as a realmember of the gender category. To explain this contradiction,as Leslie (2015) suggested, we turned to the dual-characterconcept framework proposed by Knobe, Prasada, andNewman (2013). Within this framework, we examinedwhether gender has two separate, parallel dimensions forevaluating category membership such that one can be amember in one sense but not the other. We found that genderconcepts appeared dual-character-like in metalinguisticjudgments but not in judgments of specific individuals whoviolate prescriptive gender norms identified by previousresearch. We might be witnessing a historical change wheregender categories remain dual-character-like, but adherenceto specific gender norms is no longer seen as definitional.

With the advent of social media, the last decade has seen pro-found changes to the way people receive information. This hasfueled debate about the ways (if any) changes to the nature ofour information networks might be affecting voters’ beliefsabout the world, voting results, and, ultimately, democracy. Atthe same time, much discussion in the public arena in recentyears has concerned the notion that ill-informed voters havebeen voting against their own self-interest. The research report-ed here brings these two strands together: simulations involvingagent-based models, interpreted through the formal frameworkof Condorcet’s (1785) Jury Theorem, demonstrate how changesto information networks may make voter error more likely eventhough individual competence has largely remained unchanged .

Across languages, adjectives are subject to ordering restric-tions. Recent research shows that these are predicted by ad-jective subjectivity, but the question remains open why this isthe case. We first conduct a corpus study and not only replicatethe subjectivity effect, but also find a previously undocumentedeffect of mutual information between adjectives and nouns.We then describe a rational model of adjective use in whichlisteners explicitly reason about judgments made by differentspeakers, formalizing the notion of subjectivity as agreementbetween speakers. We show that, once incremental process-ing is combined with memory limitations, our model predictseffects both of subjectivity and mutual information. We con-firm the adequacy of our model by evaluating it on corpus data,finding that it correctly predicts ordering in unseen data withan accuracy of 96.2 %. This suggests that adjective orderingcan be explained by general principles of human communica-tion and language processing.

A prominent theory posits that time and number are processed by a common magnitude system (CMS). Yet, recent studieshave revealed inconsistencies in quantity processing. For example, identical emotional stimuli evoke temporal overestima-tion, but numerical underestimation. These data discount the CMS and have led researchers to speculate about the distinctmechanisms that underlie these unique biases. In particular, differences in arousal have been posited to evoke temporaloverestimation, whereas altered attention results in numerical underestimation. In the current study, we explored adulttemporal and numerical processing under cognitive load, a task that compromises attention. Inconsistent with a CMS,baseline performance on the temporal and numerical tasks was not correlated. Similar to the work with emotional stimuli,cognitive load resulted in numerical underestimation, yet marginal temporal overestimation. Together, our data challengethe CMS, while also providing support for the role of attentional processes involved in numerical underestimation.

While current deep learning systems excel at tasks such asobject classification, language processing, and gameplay, fewcan construct or modify a complex system such as a tower ofblocks. We hypothesize that what these systems lack is a “re-lational inductive bias”: a capacity for reasoning about inter-object relations and making choices over a structured descrip-tion of a scene. To test this hypothesis, we focus on a task thatinvolves gluing pairs of blocks together to stabilize a tower,and quantify how well humans perform. We then introducea deep reinforcement learning agent which uses object- andrelation-centric scene and policy representations and apply itto the task. Our results show that these structured represen-tations allow the agent to outperform both humans and morena ̈ıve approaches, suggesting that relational inductive bias isan important component in solving structured reasoning prob-lems and for building more intelligent, flexible machines.

Distributed cognition and decision making has been a topic ofintense research in the recent years. In this paper, a computa-tional model of distributed decision making using a commu-nity of predictive coding agents is developed. The agents areembodied multimodal entities and situated in a shared envi-ronment. They have different visibility of the environment dueto unique sensory and generative models. We show that com-munication between agents helps each of them reach a shareddecision in a way that cannot be reached by brain processes ina single agent. Using a simulated environment, we show thatsensory limitations may lead to incorrect or delayed causal in-ferences giving rise to conflicts in the mind of a predictive cod-ing agent, and communication helps to resolve such conflictsand overcome the limitations.

Graded effects in morphological processing have beenshown in lexical decision tasks in English (e.g.,Gonnerman et al., 2007; Quémart et al., 2017).However, most studies in other languages support adecomposition view of the processing of complexwords (e.g., Longtin and Meunier, 2005). To determinewhether graded priming effects for morphologicallycomplex words can be found in other languages,Quebec French speakers participated in a cross-modallexical decision task in which auditory primes varied indegree of semantic similarity with visual targets (e.g.,bergerie-berge; infirmerie-infirme; fromagerie-fromage). Results indicate that morphological primingrequires the prime and target to be both semanticallyand phonologically similar, with semantic similaritymodulating priming effects in morphologically relatedwords. This pattern of results is similar to gradedmorphological priming previously reported for Englishand supports an emergentist view of morphologicalprocessing (Gonnerman et al., 2007).

There are cross-linguistics differences in the type of verb used to describe placement events. Dutch uses semantically specific placement verbs (zetten, leggen), whereas English uses a semantically general placement verb (put). This semantic focus is reflected in speaker’s gestures, which can be specific and object-focused by showing object-incorporating handshapes, or not. This study investigates the semantic placement event focus of Dutch L2 speakers of English, by investigating verb use and gesture production in placement event descriptions. Results showed that placement verb production was native- like, with a majority correct usage of put. However, gesture production showed many object-incorporating handshapes, similar to L1 Dutch gesture production. These results suggest that although the Dutch L2 speakers of English sounded native- like in speech, they were still trying to express Dutch-like placement verb meaning, by showing a continued focus on the object, as expressed in their gesture production.

Previous research has demonstrated that speakers changephonetic forms in response to variability in their immediatelinguistic milieu, such that they converge with an interlocutor.While much is known about the impact of social dynamics onthis process, the impact of individual variability in cognitionand perception is less well-explored. The present study seeksto examine the impact of these individual differences onphonetic convergence during a naturalistic conversation,comparing convergence in autism spectrum disorder (ASD)and typical development. Results showed a small effect oftemporal convergence within typically developing dyads,compared with evidence of divergence within ASD dyads.While preliminary, this pattern of results suggests that socialmotivation may play a more important role in phoneticconvergence than sensory accounts (such as self-monitoring).

Little is known about how categories are learned incidentallywithout instructions to group objects, overt decisions aboutcategory identity, or feedback about these decisions. Here weinvestigate how category learning may occur based on theassociation of categories with behaviorally-relevant events andactions. Previous research developed the SystematicMultimodal Associations Reaction Time (SMART) task inwhich participants report the location of a visual target with akeypress. The location of an upcoming visual target ispredicted by the identity of a novel sound category, exemplarsof which precede appearance of the visual target. Thiscategory-to-location mapping supports incidental learning ofauditory categories, with generalization to novel exemplars.Here, we examined whether this learning is driven by thecategory-to-location relationship, or instead by the associationwith distinct response alternatives. Across two experiments, weobserve that both a covert, reaction time measure of categorylearning and an overt labeling task testing generalization oflearning converge to indicate that the category-to-responserelationship drives incidental learning in the SMART task.

Previous studies have shown that when choosing one of thelogically equivalent frames (e.g., “half full” or “half empty”),speakers tend to choose one based on a reference point. Forexample, when the amount of water in a glass with 500ml ca-pacity was originally 0ml (or 500ml) and then increased (ordecreased) to 250ml, speakers tended to express the contentof water in the glass as “half full” (or “half empty”). We ex-amined why speakers chose one of the logically equivalentframes. In addition, we examined whether an irrelevant refer-ence point affected speakers’ choice of frame. In order to ex-amine these two issues, we conducted three behavioral exper-iments using a frame choice task. Specifically, participantswere presented with a task-relevant (story-based) or task-irrelevant (prime-based) reference point and then asked tochoose a frame. Following this, they were asked to reveal thereason for the frame choice. Our findings were summarizedwith the following two points. First, when reference pointswere task-relevant, many participants chose a frame based onthe reference point with explicit reason. Second, even whenreference points were task-irrelevant, they affected framechoices and almost all of our participants did not report the ef-fect of the irrelevant reference point. These results indicatethat the effect of reference points on frame choices is robustand that people do not always notice the effect.

When reasoners explain everyday patterns and observations,they tend to generate explanations based on inherent propertiesof the observations (Cimpian & Salomon, 2014). Cimpian(2015) and his colleagues hypothesized that inherent propertiespermit rapid explanation, but the mechanism by whichreasoners rapidly build explanations remains unclear. Anygiven concept may relate to innumerable inherent properties,and no theory explains how reasoners avoid protractedsearches through semantic memory. Prasada and colleagues(2013) describe a novel conceptual framework thatdistinguishes between principled and statistical inherentproperties. Here, we argue that the framework can resolve thepredicted link between rapid explanation and the inherencebias. Two studies provide evidence that people systematicallyprefer principled inherent explanations. The finding allows foran integrated, mechanistic account of how reasoners generateexplanations in which a preference for inherent explanationsemerges from a preference for principled connections.

Speech-accompanying gestures vary depending on features ofthe communicative situation. In the present study, weexamined whether they might also be affected by extraneousactivity in the speaker’s sensorimotor system. We askedparticipants to describe short animations that involved verticalmotion while simultaneously watching a display that depictedvertical motion in either a congruent or an incongruentdirection. Speakers produced gestures depicting verticalmotion at a higher rate when describing the target motionevents when they were simultaneously watching a display thatdepicted motion in the same direction than when watchingmotion in the opposite direction. These results suggest thatthe cognitive basis of gesture lies in the sensorimotor system.

In uncertain environments we must balance our need to gather information with our desire to exploit current knowledge. This is further complicated in reactive environments where actions produce long-lasting change. In three experiments, we investigate how people learn to make effective decisions from experience in a dynamic four-armed bandit task. In contrast to the diminishing rewards found in most previous studies, options were framed as skills that developed greater rewards when chosen. We find that most individuals learn effective strategies for coping with reactive environments. We present a psychological model positing that decision makers move through three distinct processing phases, and show that it accounts for key behavioral patterns across experiments.

We propose a formal model of humans’ intuitive theories of others’ emotions. From a single choice in a social interaction(e.g. the choice to cooperate in a Prisoner’s Dilemma game), human observers can infer a player’s complex values, such asprosocial preferences and reputational concerns. When the player then experiences a new situation (the game’s outcome),observers infer the player’s reaction to the event based on the mental state likely to have produced the player’s action. Herewe capture this process by inverting a richly structured generative model of social gameplay, including social equity andreputational dimensions, and translate players’ subjective motivations, expectations, and prediction errors into forwardpredictions of the emotional experiences of the players. Our model infers players’ values and expectations, generatespatterns of play that match observers’ intuitions, and supports formally generated emotion predictions with substantiallyextended breadth and nuance.

Some researchers claim that language acquisition is critically dependent on experiencing linguistic input in order of in-creasing complexity. We tested this hypothesis using a simple recurrent neural network (SRN) trained to predict wordsequences in CHILDES, a 5-million-word corpus of speech directed to children. First, we demonstrated that age-orderedCHILDES exhibits a gradual increase in linguistic complexity. Next, we compared the performance of two groups ofSRNs trained on CHILDES which had either been age-ordered or not. Specifically, we assessed learning of grammaticaland semantic structure and showed that training on age-ordered input facilitates learning of semantic, but not of sequentialstructure. Follow-up analyses suggest that higher noun-density in speech to younger children combined with weight en-trenchment could account for this effect. The persistent learning improvement is consistent with the neural commitmenthypothesis in the second language acquisition literature, which asserts that L1 representation reduces neural resourcesavailable for L2 learning. Similarly, exposure to noun-rich input first but not last (age-ordered CHILDES), may induce arepresentational advantage for lexical semantic acquisition.

The Iowa Gambling Task (IGT) is an established toolused for evaluating the role of emotional learning underconditions of uncertainty. To date, however, themajority of studies have not explicitly manipulated theemotional content within the IGT or examined the effectof doing so on different populations. We address thisgap in the present study, focusing our analysis on twogroups: low vs. high psychopathy individuals insubclinical populations. Our findings demonstrate thatemotional content boosted learning for the high but notthe low psychopathy group.

A surge of recent research on fraction representation hasprovided substantial insight into how people think aboutproportional information in written, symbolic form and invisual, non-symbolic form. However, how fractions anddecimals are verbally labeled is an often-overlooked aspect ofproportion representation. In the current study, weinvestigated how adults label fractions and decimals (Study 1)and how children in a range of grades label fractions (Study2), using a novel web-based platform for accessing studentdata from real classrooms (ASSISTments). In both studies,children and adults showed remarkable consistency in thekinds of labels they used. However, there were somedifferences in label preferences across notation and grade-level. Although the relations between fraction labeling andfraction ability remain unclear, these studies provide a firstlook at the kinds of labels that people typically use andprovide some initial hypotheses for future research intosymbolic representations of proportion.

Among the numerous formal and informal theories of the at-tentional blink, the common theoretical thread is that the deficitstems from selective attention and working memory processesbeing tied up in processing the first target (T1) when the sec-ond target (T2) appears. Rusconi & Huber (2017) challengedthis view by proposing the ’perceptual wink’ model of the AB,which posits that for categorical AB tasks (e.g., number/letter)the deficit reflects a failure to perceive that T2 belonged to thetarget category. The model makes the assumption that percep-tion is ’multi-faceted’; that is, there are separate, independentperceptual representations for an item’s identity and its cate-gory, and that either representation can be used to drive per-formance (e.g., trigger attentional encoding) depending on thetask demands. To differentiate between attention versus per-ceptual accounts of the AB, we used a stripped down RSVPtask where participants were asked to either report the iden-tity or category of the third item in a sequence of characters.In support of the perceptual account, we found priming foridentity or category depending on the task. Furthermore, wefound that the category results were analogous to the AB andthe spread of sparing even though the first character was nota target and there was no need to selectively filter items intoworking memory.

This research investigated the acquisition process of the color lexicon, specifically how color words are initially grounded and develop into the lexical system possessed by the adults in the ambient language. We conducted a longitudinal study in which Japanese-learning 2-year- olds were tested every month on their understanding of basic words denoting 8 chromatic colors, continuing until they were able to map these words onto their referents consistently. The results strongly endorse the view that acquisition of the color lexicon should be characterized as a process of system construction, through which children reorganize prelinguistic color categories onto the linguistic categories of the ambient language, thereby representations of individual words are continuously refined along with the refinement of the representation of the system as a whole.

A common assumption in the cognitive sciences is thatartificial and natural language learning rely on sharedmechanisms. However, attempts to bridge the two haveyielded ambiguous results. We suggest that an empiricaldisconnect between the computations employed duringlearning and the methods employed at test may explain thesemixed results. Further, we propose statistically-basedchunking as a potential computational link between artificialand natural language learning. We compare the acquisition ofnon-adjacent dependencies to that of natural languagestructure using two types of tasks: reflection-based 2AFCmeasures, and processing-based recall measures, the latterbeing more computationally analogous to the processes usedduring language acquisition. Our results demonstrate thattask-type significantly influences the correlations observedbetween artificial and natural language acquisition, withreflection-based and processing-based measures correlatingwithin – but not across – task-type. These findings havefundamental implications for artificial-to-natural languagecomparisons, both methodologically and theoretically.

How does listeners’ perceptual bias influence their interpreta-tion of an ambiguous multiword utterance? We address thisquestion by investigating the relationship between word or-der in a binomial (an expression of type “A and B”) and vi-sual properties of image pairs serving as its potential referents.We found that listeners’ choices were strongly influenced byiconicity and relative salience of images within the pair: par-ticipants preferred referents where the first mentioned imagewas located on the left, as well as pairs where the first im-age was larger than the second image. The effect of image or-der tended to be stronger than the influence of image size, andboth were modulated by participants’ general visual field pref-erences (determined in a separate experimental condition). Wefurther show that binomial phrase interpretation can be simu-lated by a Rational Speech Act model that includes both wordorder effects and utterance-independent preferences of the par-ticipants.

A pressing issue in both psychology and agent-modeling com-munities is the inability to account for the wide variance in hu-man variability and individual differences. Added to this is thefurther complexity of changing goals and social meaning in adynamic, sequential interaction. While prior work on artificialagent design has prominently addressed physical cues and non-verbal behavior, there is a lack of emphasis on (1) examiningcues in combination, and (2) assessing judgments of social sit-uational meaning. In the current work, we present an ontologyof physical behavior (Social Kinematics) that accounts for thecombinatorial effects of multiple cues, as well as the changingsocial meaning associated with these different combinations ofcues. Here, we assess individuals social situational judgmentsof multiple combinations of ambiguously-defined virtual agentanimations. Ultimately, this paper provides a potentially usefulframework that has relevance for researchers in social robotics,agent modeling, and cognitive science.

Events unfold over time, i.e., they have a beginning and endpoint. Previous studies have illustrated the importance of endpoints for event perception and memory (Lakusta & Landau, 2005, 2012; Papafragou, 2010; Strickland & Keil, 2011; Zacks & Swallow, 2007). However, this work has not compared endpoints to other potentially salient points in the internal temporal profile of events (e.g., midpoints) and has only discussed events with a self-evident endpoint. In the present study, we explored sensitivity to event endpoints and midpoints in events of different types. Our results show that people are more disturbed by interruptions at the end compared to interruptions in the middle of an event – but only when perceiving a bounded event (i.e., an event with an inherent endpoint). This finding reveals complex tracking of the abstract internal temporal structure of events during event perception.

Story-telling helps to define the human experience. Donarratives also inform our predictions and choices? Thecurrent study provides evidence that they do, using financialdecision-making as an example of a domain where,normatively, publicly available information (about the pastor the future) is irrelevant. Despite this, participants usedpast company performance information to project futureprice trends, as though using affectively laden informationto predict the ending of a story. Critically, these projectionswere stronger when information concerned predictionsabout a company’s future performance rather than actualdata about its past performance, suggesting that people notonly rely on financially irrelevant (but narratively relevant)information for making predictions, but erroneously imposetemporal order on that information.

When given the chance to choose between two tasks, one willmore likely choose the easier, less demanding task. Thiseffect has been shown in various domains and referred to asthe law of minimum effort or demand avoidance. Themeasure of demand avoidance that is currently used is theproportion of low-demand choices. We show that the currentmeasure is not appropriate for accurately assessing individualdifferences in demand avoidance, because the process ofdemand selection is contingent upon the process of demanddetection. Subsequently, we suggest a new measure ofdemand avoidance that combines demand detection anddemand selection. We show that the new measure of demandavoidance correlates in the expected direction (i.e.,negatively) with established measures of willingness andability to carry out cognitively demanding tasks. We proposea novel, performance-based measure of cognitive effortavoidance that can be used to enhance the validity of researchin cognition, perception, and neurosciences.

As observed in the World Color Survey (WCS), some univer-sal properties can be identified in color naming schemes overa large number of languages. For example, Regier, Kay, andKhetrapal (2007) and Regier, Kemp, and Kay (2015); Gib-son et al. (2017) recently explained these universal patterns interms of near optimal color partitions and information theoreticmeasures of efficiency of communication. Here, we introducea computational learning framework with multi-agent systemstrained by reinforcement learning to investigate these universalproperties. We compare the results with Regier et al. (2007,2015) and show that our model achieves excellent quantitativeagreement. This work introduces a multi-agent reinforcementlearning framework as a powerful and versatile tool to investi-gate such semantic universals in many domains and contributesignificantly to central questions in cognitive science.

One of the driving forces of language evolution is the selection of variants that suit the communicative needs of its users.Crucially, fitness of linguistic variants may largely depend on the structure of the environment in which language is learned,transmitted, and used. This hypothesis has gained support in various domains. We apply it in the context of scalar termswith a major focus on quantifiers, such as ’most’. Based on a model that combines logic and evolutionary game theory, weargue that such signals might have evolved as stable semantic units through adaptation to general communicative principlesand distributional properties of the environment such as normality.

How does the presence of background noise affect thecognitive processes underlying spoken-word recognition? Andhow do these effects differ in native and non-native languagelisteners? We addressed these questions using artificial neural-network modelling. We trained a deep auto-encoderarchitecture on binary phonological and semanticrepresentations of 121 English and Dutch translationequivalents. We also varied exposure to the two languages togenerate ‘native English’ and ‘non-native English’ trainednetworks. These networks captured key effects in theperformance (accuracy rates and the number of erroneousresponses per word stimulus) of English and Dutch listeners inan offline English spoken-word identification experiment(Scharenborg et al., 2017), which considered clean and noisylistening conditions and three intensities of speech-shapednoise, applied word-initially or word-finally. Our simulationssuggested that the effects of noise on native and non-nativelistening are comparable and can be accounted for within thesame cognitive architecture for spoken-word recognition.

Dimensional adjective interpretation is dependent on the com-parison class – the set of object representations – against whichthe object being modified by the adjective is judged. This paperexplores the factors determining the composition of the com-parison class, arguing that real world size information and pro-totypicality play crucial parts in its determination. Researchersoften implicitly assume that only the objects in immediate vi-sual context constitute the comparison class. However, Exp.1 shows that this information from the visual context is inte-grated with knowledge of real world size and category proper-ties to form the comparison class. Exp. 2 shows that prototypeinformation is utilized when making size judgments of cartoonimages, while size judgments of objects in photographs drawmore heavily on a speaker’s prior knowledge about the actualsize of the objects in the world. Exp. 3 demonstrates that theeffects observed in Exp. 1 and 2 were not caused by the adjec-tives used, but rather reflect differences between the size of theobjects depicted in the images.

The current research explores the role of two different motives underlying volunteering (or defecting) in a simple economic game. We find in Study 1 that in a symmetric Volunteer’s Dilemma (VoD) the willingness to volunteer is reduced more strongly by an increase in the payoff for unilateral defection (suggesting more greed) than by an increase in the payoff for mutual defection (suggesting less fear). In Study 2, we replicate this finding when only the participants’ own payoffs are varied, but not when only the other player’s payoffs are varied. These findings are inconsistent with standard (i.e., Nash) game-theoretic predictions and Schelling’s focal-point hypothesis. Instead, the empirical patterns suggest that participants approach the VoD using egocentric decision heuristics.

In numerosity comparison, performance is faster, more accurate, and less noisy with the ratio of compared numbers.Whereas the ratio-dependency has been intensively studied in relation to internal noise, processes of numerosity com-parison that may increase internal noise have not been fully understood. In this paper, we propose a process theory thataccounts for non-numerical, visuo-spatial processes in numerosity comparison. Consistent with the theory, we found thatas required processes decreased, performance improved significantly, to the extent that there were no differences betweennon-symbolic and symbolic number comparison in reaction time, accuracy, and internal noise. The findings suggest thatcomparing numerosities requires multiple processes homogenizing ancillary stimulus dimensions and that the homoge-nization processes are the major source of fuzziness in approximate number comparison.

Resting state functional connectivity (rsFC) can provide awindow into the neural architecture of functional networks inthe brain. Functional networks measured both during task andduring “resting” (task-absent) state are correlated withcognitive function, and much development of these networksoccurs between infancy and adulthood. However, rsFC studyin young children has been sparse, mainly due to a paucity ofchild-appropriate neural measures and behavioral paradigms.We present a new paradigm to measure rsFC in children,utilizing functional near-infrared spectroscopy (fNIRS) andFreeplay, a behavioral setup designed to approximate restingstate in children. Results suggest this paradigm is practicaland has good construct validity and test-retest reliability.

The ability to plan, inhibit, and execute motor movements are all necessary for achieving goal-directed behavior. These processes are closely related to memory, as perceptual input and memory of that input often recruit motor movements. Unknown, however, is how the engagement of perception- action processes impact the memory of objects. One such interaction suggests that participants have worse memory recall for stimuli which elicit inhibition of a motor response than stimuli which afford the execution of a motor response (Chiu & Egner, 2015). This effect has been explained through competition for common neural resources: allocation of resources toward response inhibition reduces the amount of resources available for memory. Alternatively, this effect could be driven at the level of perception-action coupling: engaging and pairing the motor system with visual perception enhances the memory of stimuli which elicited the motor preparation or response. To test these hypotheses, we first replicated Chiu and Egner (2015). In Experiment 2, we included neutral stimuli that did not necessitate motor preparation processes. Memory was enhanced for stimuli presented in conjunction with motor engagement, providing evidence for an account of memory that is facilitated when coupled with the motor system.

Human infants have the remarkable ability to learn any hu-man language. One proposed mechanism for this ability isdistributional learning, where learners infer the underlyingcluster structure from unlabeled input. Computational mod-els of distributional learning have historically been principledbut psychologically-implausible computational-level models,or ad hoc but psychologically plausible algorithmic-level mod-els. Approximate rational models like particle filters can po-tentially bridge this divide, and allow principled, but psycho-logically plausible models of distributional learning to be spec-ified and evaluated. As a proof of concept, I evaluate one suchparticle filter model, applied to learning English voicing cate-gories from distributions of voice-onset times (VOTs). I findthat this model learns well, but behaves somewhat differentlyfrom the standard, unconstrained Gibbs sampler implementa-tion of the underlying rational model.

Research has found that when children or adults attemptto explain novel observations in the course of learning,they are more likely to discover patterns that support idealexplanations: explanations that are maximally simple andbroad. However, not all learning contexts support suchexplanations. Can explaining facilitate discoverynonetheless? We present a study in which participantswere tasked with discovering a rule governing theclassification of items, where the items were consistenttwo non-ideal rules: one correctly classified 66% of cases,the other 83%. We find that when there is no ideal rule tobe discovered (i.e., no 100% rule), participants promptedto explain are better than control participants atdiscovering the best available rule (i.e., the 83% rule).This supports the idea that seeking ideal explanations canbe beneficial in a non-ideal world because the pursuit ofan ideal explanation can facilitate the discovery ofimperfect patterns along the way.

Leading theories of risky choice predict that decision makersare sensitive to the variability of payoff distributions. Yet, lit-tle is known about how experience affects perceived variabil-ity. Existing empirical research on risky choice provides onlyinconclusive evidence about this issue because choices are notonly affected by perceived variability but also perceived valueand (unobserved) risk preferences. In re-analyses of experi-mental data and survey data from two nationally representativepanels, we show that perceived variability strongly depends onsample variability. In a new experiment, we also demonstratethat perceived variability systematically depends on samplesize, a result consistent with the predictions of a recent the-oretical paper by the authors (Konovalova & Le Mens, 2017).

In theory, there exists no bound to a causal explanation – every explanation can be elaborated further. But reasoners rate some explanations as more complete than others. To account for this behavior, we developed a novel theory of the detection of explanatory incompleteness. The theory is based on the idea that reasoners construct mental models of causal explanations. By default, each causal relation refers to a single mental model. Reasoners should consider an explanation complete when they can construct a single mental model, but incomplete when they must consider multiple models. Reasoners should thus rate causal chains, e.g., A causes B and B causes C, as more complete than “common cause” explanations (e.g., A causes B and A causes C) or “common effect” explanations (e.g., A causes C and B causes C). Two experiments validate the theory's prediction. The data suggest that reasoners construct mental models when generating explanations.

Human activity generates dynamic, multi-modal sensorystreams. Effectively processing this complex flow ofinformation on-the-fly is essential if one is to remember andrespond to others’ action, anticipate what they might do next,and learn how to perform new actions. Selectively attending toinformation-rich regions of activity seems key to fluentprocessing. However, what counts as information-rich likelydepends on numerous factors including relevance to the causalstructure of the activity, local opportunity for repeated viewing,and processing goals of the observer. We explored theinfluence of these factors on observers’ attention to a dynamic,novel activity sequence. A performance context elicitednuanced differences in processing in contrast to a remembercontext. Specifically, individuals given a perform contexttuned in to causally distinct regions of the action stream andfine-level event details. These findings provide altogether newinformation regarding how processing rapidly reorganizesaround novel activity and responds to the processing task athand.

Marr’s levels of analysis constitute one influential approach tothe central program of cognitive science—the multilevel anal-ysis of cognition as information processing. The distinctiveaspects of Marr’s framework are an emphasis on identifyingthe computational problems and constraints faced in cognition,and conceptual machinery to relate cognitive mechanisms tothat computational level of analysis. Although related ideashave been explored in a range of social science disciplines,Marr’s framework, and particularly its notion of the preciseformulation of computational problems and solutions, has yetto be applied widely in social analysis. In the present workwe develop a formulation of Marr’s levels for social systems,provide examples of this approach, and address potential criti-cisms. The consequence is a computational perspective on thesociological school of structural functionalism, and an appara-tus for conducting multiscale analysis of social systems.

German 5-year-olds are able to rapidly recruit depicted ac-tions to assign thematic roles in unambiguous sentences whenthese actions can be inspected throughout sentence presenta-tion (Münster, 2016; Zhang & Knoeferle, 2012). In two visual-world eye tracking studies, we investigated whether these find-ings extend to locally structurally ambiguous utterances andto short-lived action presentation. In addition, we comparedthe action depiction to a character’s wiggling motion. The ac-tion and the wiggle served as cues to the agent (subject) indifficult-to-understand OVS sentences. Participants listenedto structurally ambiguous object-verb-subject (OVS) sentencesabout, for instance, a bug being pushed by a bull while in-specting a bull, a bug, and a worm. We manipulated the sceneat verb-onset such that either a) no action no wiggle, b) noaction one wiggle, c) one action no wiggle, or d) one actionone wiggle appeared. Both of these animations caused theadults and the children to visually anticipate the agent rolefiller (corresponding to the subject in the OVS sentence) be-fore its mention. However, in answering post-trial who-does-what-to-whom comprehension questions, the children did not(unlike suggested by previous findings) benefit from the actiondepictions. Together the eye-gaze and post-trial comprehen-sion results suggest that the nature of cue presentation (e.g.,the abrupt onset of an action or a wiggle and limitations on cuepresence) plays an important role in both the immediate visualattention and somewhat later interpretation effects of such vi-sual cues during children’s language comprehension.

Model-free (MF) and model-based (MB) reinforcement learn-ing (RL) have provided a successful framework for under-standing both human behavior and neural data. These two sys-tems are usually thought to compete for control of behavior.However, it has also been proposed that they can be integratedin a cooperative manner. For example, the Dyna algorithm usesMB replay of past experience to train the MF system, and hasinspired research examining whether human learners do some-thing similar. Here we introduce an approach that links MFand MB learning in a new way: via the reward function. Givena model of the learning environment, dynamic programmingis used to iteratively approximate state values that monotoni-cally converge to the state values under the optimal decisionpolicy. Pseudorewards are calculated from these values andused to shape the reward function of a MF learner in a waythat is guaranteed not to change the optimal policy. We showthat this method offers computational advantages over Dyna intwo classic problems. It also offers a new way to think aboutintegrating MF and MB RL: that our knowledge of the worlddoesn’t just provide a source of simulated experience for train-ing our instincts, but that it shapes the rewards that those in-stincts latch onto. We discuss psychological phenomena thatthis theory could apply to, including moral emotions.

How do people plan ahead when searching for rewards? Weinvestigate planning in a foraging task in which participantssearch for rewards on an infinite two-dimensional grid. Ourresults show that their search is best-described by a modelwhich searches at least 3 steps ahead. Furthermore, partici-pants do not seem to update their beliefs during planning, butrather treat their initial beliefs as given, a strategy similar to aheuristic called root-sampling. This planning algorithm corre-sponds well with participants’ behavior in test problems withrestricted movement and varying degrees of information, out-performing more complex models. These results enrich ourunderstanding of adaptive planning in complex environments

Human judgments about the physical attributes of—and causalrelationship between—two colliding objects have been stud-ied extensively over the past seventy years. Recent computa-tional evidence suggests that judgments about the mass ratioof two colliding objects, as well as their perceived causal re-lation, can be explained by a coherent framework based on aNewtonian physical model and probabilistic inference result-ing from noisy observations of object movements. However,it remains unclear how the physical and causal reasoning sys-tems interact with the motion perception system when formingthese judgments. The current study aims to examine whetherhigh-level judgments are guided by object motion representedas relative motion with reference to a moving background, oras absolute motion with reference to a stationary position inthe world. Both experimental evidence and model simulationresults support the notion that physical and causal inference inobject collisions depend on relative motion rather than abso-lute motion.

I motivate and articulate a dispositional account of aversiveracism. By conceptualizing and measuring attitudes in termsof their full distribution, rather than in terms of their mode ormean preference, my account of dispositional attitudes givesambivalent attitudes (qua attitude) the ability to predictaggregate behavior. This account can be distinguished fromother dispositional accounts of attitude by its ability tocharacterize ambivalent attitudes such as aversive racism atthe attitudinal rather than the sub-attitudinal level and itsdeeper appreciation of the analogy between traits andattitudes.

Psychologists and philosophers have not yet resolved whatthey take implicit attitudes to be; and, some, concerned aboutlimitations in the psychometric evidence, have evenchallenged the predictive and theoretical value of positingimplicit attitudes in explanations for social behavior. In themidst of this debate, prominent stakeholders in science havecalled for scientific communities to recognize andcountenance implicit bias in STEM fields. In this paper, Istake out a stakeholder conception of implicit bias thatresponds to these challenges in ways that are responsive to thepsychometric evidence, while also being resilient to the sortsof disagreements and scientific progress that would notundermine the soundness of this call. Along the way, myaccount advocates for attributing collective (group-level)implicit attitudes rather than individual-level implicitattitudes. This position raises new puzzles for future researchon the relationship (metaphysical, epistemic, and ethical)between collective implicit attitudes and individual-levelattitudes.

Most previous studies on time perception have examinedtemporal order and distance judgments in isolation usingcontrolled stimuli. However, in real life, these two elementarytemporal experiences are related. Here, we examine the effectsof age and event structure on temporal estimation and introducea novel timeline estimation paradigm comprising temporalorder and distance judgments with naturalistic stimuli. In twoexperiments, we asked participants to view a three-minute-longvideo clip and mark the temporal order and distance of aspecific scene of the video on a horizontal timeline. In the firstexperiment, we conducted the timeline estimation task withthree different age groups – 6-8-year-olds, 9-11-year-olds andadults – and found age-related differences in the participants’accuracy and variability of temporal estimation. Thenonlinearity between their estimates and stimulus distancedecreased as their ages increased. In Experiment 2, we testedthe effect of event structure on participants’ timeline estimationand observed that more complicated video resulted in moredistorted temporal estimation. In sum, the current studycorroborated the timeline estimation task to be a valuable toolfor assessing temporal judgments across development.

Communication is a powerful means to disseminate socialinformation, and gossip is an effective way of obtainingupdated information about others. However, without acomprehensive theoretical framework of socialcommunication, it is difficult to predict a priori when and whysocial information will be disseminated. There are generaltheories of human social interaction, however, they do notsufficiently capture the sociocognitive components underlyinghuman decision-making in social settings. Therefore, we havedeveloped a model of social communication, enabling thecharacterization of specific conditions under which socialinformation will be spread: for example, when an agent shoulddirectly communicate with the target of the information, gossipit to others, or simply do nothing. We describe the model, themethods used to generate model predictions, and then list ninepredictions derived from it as the current results. We next planto test the predictions empirically and develop the modelcomputationally.

In function learning experiments, where participants learnrelationships from sequentially-presented examples, peopleshow a strong tacit expectation that most relationships are lin-ear, and struggle to learn and extrapolate from non-linear rela-tionships. In contrast, experiments with similar tasks wheredata are presented simultaneously – typically using scatterplots – have shown that human learners can discover and ex-trapolate from complex non-linear trends. Do people have dif-ferent expectations in these task types, or can the results beattributed to effects of memory and data availability? In a di-rect comparison of both paradigms, we found that differencesbetween task types can be attributed to data availability. Weshow that a simple memory-limited Bayesian model is consis-tent with human extrapolations for linear data for both highand low data availability. However, our model underestimatesthe participants’ ability to infer non-monotonic functions, es-pecially when data is sparse. This suggest that people trackhigher-order properties of functions when learning and gen-eralizing.

The mind adeptly registers statistical regularities inexperience, often incidentally. We use a visual statisticallearning paradigm to study incidental learning of predictiverelations among animated events. We ask what kinds ofstatistics participants automatically compute, even whentracking such statistics is task-irrelevant and largely implicit.We find that participants are sensitive to a quantity governingassociative learning, DP, independently of conditionalprobabilities and chunk frequencies, as previously considered.DP specifically reflects the uniqueness, as well as strength, ofconditional probabilities; we find that uniqueness is equallyaffected by a single strong alternative predictor as by severalweak predictors. Performance is well captured with anadapted version of the Rescorla-Wagner delta learning rule(Rescorla & Wagner, 1972). We conclude that incidentalpredictive learning is governed by considerations ofuniqueness, and that this is computed by normalizingconditional probabilities by events’ base-rates. This opens thepossibility of common mechanisms between statisticallearning, associative learning, and causal inference.

Prior research shows that languages balance syntactic complexityagainst morphological complexity. We explore this relationshipusing a new measure of syntactic complexity, functionalindeterminacy, which measures the aggregate uncertainty ofmapping from lexical items to syntactic function. We predict thatgreater functional indeterminacy for nouns will correlate withlanguages having case systems, and for those with case systems,increased number of cases. We operationalize indeterminacy as thesimple and normalized conditional entropies of the summedfrequency distributions of nouns across syntactic dependencies. Wecompute these measures for 44 languages. We then correlate themeasures with presence and number of cases in two regressionanalyses, controlling for genetic affiliation between languages.Results show that as the functional indeterminacy of nounsincreases, languages are more likely to have case systems, and ifso, to have more cases. These data provide new support for thefunctionally motivated relationship between morphological andsyntactic complexity.

We explore whether de-lexicalized syntactic information impactsthe phonetic duration of nouns. The motivating expectation is thatnouns that carry more syntactic information will be more difficultto produce in situ, leading to longer durations. We approach thisquestion from two perspectives: pure diversity of a noun'sdistribution across its available syntactic relations, and distance ofthis distribution from the average distribution of nouns in thelanguage at large. The former measure is designed to capture theinterconnectivity between the lexical and syntactic tiers oflinguistic representation. The latter measure targets how well anindividual noun fits the behavior expected for the noun class. Wefind that durations are sensitive to both measures incomplementary fashion: nouns with more diverse syntacticdistributions are produced with longer durations, and nouns thathave distinctive (non-prototypical) distributions have shorterdurations.

Cultural norms vary dramatically across social groups. Herewe use large scale data to examine the extent to which languageplays a role in shaping one such norm—the gender norm to as-sociate men with careers and women with family. We measurecross-cultural variability in this gender bias using previously-collected estimates from the Implicit Association Task (IAT; N= 663,709). We then try to predict bias variability by the waythat gender is encoded in language semantics and grammar.We quantify gender bias in semantics using word-embeddingmodels trained on different languages. Our data suggest thatthe linguistic encoding of gender predicts the degree of speak-ers’ gender bias in the IAT, pointing to a causal role for lan-guage in shaping gender norms.

Recent research suggests that people do not perform wellon some of the most crucial components of causal reason-ing: probabilistic independence, diagnostic reasoning, and ex-plaining away. Despite this, it remains unclear what con-texts would affect people’s reasoning in these domains. Inthe present study we investigated the influence of manipulatingpriors of causes and structural complexity of Causal BayesianNetworks (CBNs) on the above components. Overall we foundthat participants largely accepted the priors and understoodprobabilistic independence, but engaged in inaccurate diagnos-tic reasoning and insufficient explaining away behavior. More-over, the effect of manipulating priors on participants’ perfor-mance in diagnostic reasoning and explaining away was sig-nificantly larger in a structurally less complex CBN than in astructurally more complex CBN.

A large literature has demonstrated an abnormal sense ofagency (SOA) in schizophrenic individuals. One limitation ofsuch studies is that they focus exclusively on cognitive orperceptual judgments, thus failing to address affective aspectsof SOA. In our recent work, we have used instrumentaldivergence – the distance between outcome probabilitydistributions associated with available actions – as a formalmeasure of agency, demonstrating an influence of this noveldecision variable on behavioral choice preferences andassociated neural computations in neurotypical adults. Here,we show that the preference for high instrumental divergence(i.e., for high-agency environments) is significantlymodulated by individual differences in positive and negativeschizotypy dimensions. Implications for future assessmentsof clinical populations are discussed.

Previous studies investigating effects of language comprehension on spatial processing have used existing words with pre-existing spatial associations. Here participants learnt novel words and novel objects with spatial associations. Following training, participants had to judge whether a visual object matched a word. Objects could match in identity or in spatial location. In Experiment 1, participants learnt just novel words and objects; Experiment 2 compared performance with existing objects with pre-existing spatial associations. We found mismatching (but task irrelevant) spatial information interfered with judgements of object identity, but only for novel words. In Experiment 3, we altered correspondence between visual targets and semantics using a target discrimination task, where the target had no relationship to the verbal cue. We found the opposite results to the previous two studies, as responses to spatially matching targets were slower than spatially mismatching targets. We discuss implications for embodied and non-embodied accounts of these findings.

Both science and religion offer explanations for everydayevents, but they differ with respect to their tolerance formysteries. In the present research, we investigate laypeople’sperceptions about the extent to which religious and scientificquestions demand an explanation and the extent to which anappeal to mystery can satisfy that demand. In Study 1, wedocument a large domain difference between science andreligion: scientific questions are judged to be more in need ofexplanation and less appropriately answered by appeal tomystery than religious questions. In Study 2, we demonstratethat these differences are not driven by differing levels of beliefin the content of these domains. While the source of thesedomain differences remains unclear, we propose severalhypotheses in the General Discussion.

Theories of event-predictive, anticipatory behavior state that action planning, decision making, and control are realizedby activating future goal states. That is, anticipated and desired final event boundaries as well as sensorimotor-groundedevent codes are activated before actual motor control unfolds. The involved active inference process thereby focuses sen-sorimotor processing on those upcoming events and event boundaries, in which expected uncertainties need to be resolved.Here, we investigated anticipatory behavior during object interactions, that is, grasping and placing bottles. We investi-gated whether peripersonal hand space is remapped onto the to-be grasped bottle during action preparation and whetherthis remapping depends on (i) the bottle’s orientation and (ii) the certainty about upcoming sensorimotor contingencies.To do so, we conducted two experiments in an immersive virtual reality, combining the crossmodal congruency paradigm,which has been used to study selective interactions between vision and touch within peripersonal space, with a graspingtask. In both experiments, we observed anticipatory crossmodal congruency between vision and touch at the future fingerposition on the bottle. Moreover, in the second experiment, a manipulation of the visuo-motor mapping of the partici-pants’ virtual hand while approaching the bottle selectively reduced crossmodal congruency at movement onset. Thus, theexpected movement uncertainty decreased the anticipatory remapping of peripersonal space. Our results support theoriesof event-predictive cognition and show how expected uncertainties influence anticipatory, active inference processes.

The Dimensional Change Card Sort Task (DCCS) is a measure of the developmental status of early childhood EF. In this task, children use verbal rules regarding the features and dimensions of objects to sort cards by shape or color. A recent dynamic neural field model explains development in the DCCS task based on the strength of associations between labels and visual features. In this project, we explored the role of dimensional label learning (DLL) in the development of flexibility in the DCCS task. Three- and 4-year-olds were given DLL tasks along with the DCCS task. We measured hemodynamic activity as children performed these tasks using fNIRS. Results showed that color label production produced activation throughout frontal and left temporal areas. Importantly, hemodynamic activation during the DLL tasks predicted performance in the DCCS. These results suggest that the neural systems involved in DLL influences children’s ability to flexibly switch between rules.

Words show complex dynamics of meaning change. In somecases, a word may acquire novel senses. In other cases, ex-isting senses of a word may become obsolete. The rates atwhich words gain and lose senses may vary, but it is an openquestion which factors might account for this variation. Build-ing on work in computational linguistics and cognitive science,we develop a computational approach that explores this ques-tion by leveraging word sense records from a large histori-cal database of English. Our results suggest that polysemouswords tend to gain and lose senses more than words with fewersenses, and that these effects are robust when word frequencyand length are both controlled for. These results are consis-tent with recent findings on the mechanisms of emergent wordmeanings and they further suggest stability in the temporal dy-namics of word meanings.

In visual word recognition, having more orthographicneighbors (words that differ by a single letter) generallyspeeds access to a target word. But neighbors can mismatch atany letter position. In light of evidence that informationcontent varies between letter positions, we consider howneighbor effects might vary across letter positions. Resultsfrom a word naming task indicate that response latencies arebetter predicted by the relative number of positional friendsand enemies (respectively, neighbors that match the target at agiven letter position and those that mismatch) at some letterpositions than at others. In particular, benefits from friendsare most pronounced at positions associated with low a prioriuncertainty (positional entropy). We consider how theseresults relate to previous accounts of position-specific effectsand how such effects might emerge in serial and parallelprocessing systems.

While explanations provide the power to understand the worldaround us, people are often overconfident about their ownunderstanding. We explored how people’s perceptions of theirunderstanding of phenomena is related to endorsement ofconspiracy theories. We first tested people’s perceptions oftheir understanding of the 2016 Presidential electoral processand then measured their beliefs that the election itself wasillegitimate, a form of conspiratorial belief. We found thatparticipants who still endorsed high levels of understandingafter generating an explanation for the 2016 election were alsomore likely to endorse the election was illegitimate. However,this finding only obtained for participants who voted for thelosing candidate. These results suggest interesting avenues forexploring individual differences that may be related to theillusion of explanatory depth.

Syntactic rules in natural language typically need to make ref-erence to hierarchical sentence structure. However, the simpleexamples that language learners receive are often equally com-patible with linear rules. Children consistently ignore theselinear explanations and settle instead on the correct hierarchi-cal one. This fact has motivated the proposal that the learner’shypothesis space is constrained to include only hierarchicalrules. We examine this proposal using recurrent neural net-works (RNNs), which are not constrained in such a way. Wesimulate the acquisition of question formation, a hierarchicaltransformation, in a fragment of English. We find that someRNN architectures tend to learn the hierarchical rule, suggest-ing that hierarchical cues within the language, combined withthe implicit architectural biases inherent in certain RNNs, maybe sufficient to induce hierarchical generalizations. The like-lihood of acquiring the hierarchical generalization increasedwhen the language included an additional cue to hierarchy inthe form of subject-verb agreement, underscoring the role ofcues to hierarchy in the learner’s input.

Prior research on biological concepts suggests that peopleunderestimate within-species variability and rejectmetamorphosis as a possible change for unfamiliar organisms.This may be due to psychological essentialism. This studyinvestigated whether manipulating perceptions of biologicalvariability (both within species and between species) led toincreases in endorsement of metamorphosis amongundergraduate students. We manipulated perceptions ofvariability by priming students before a lesson and byhighlighting variability in the diagrams used during the lesson.Priming led to more endorsement of metamorphosis, but onlyamong those with high prior knowledge. Our results suggestthat manipulating perceptions of variability is not only possiblebut might be beneficial for those who have strong priorknowledge about biology.

Fraction knowledge is not only important for later STEM related education achievement but also crucial for employmentand health. However, many children have great difficulties learning fractions. The present study examined predictorsof fraction knowledge performance in second and fifth grade, including both math-specific skills and general cognitiveabilities. Individual differences in non-symbolic ratio acuity, whole number line estimation, and auditory working memorywere significant predictors of symbolic fraction knowledge performance. Non-symbolic ratio acuity made the largestcontribution to symbolic fraction performance compared to other predictors. The implications of these findings for theoriesof numerical development and for improving mathematics learning are discussed.

Phonological characteristics of a voice, such as th-stopping(pronouncing them as “dem”) associated with AfricanAmerican English (AAE), provide indexical sociolinguisticinformation about the speaker. Word usage also signals thissocial dialect, i.e. usage of crib to mean house. The currentstudy examines the effect of these sociolinguisticcharacteristics on word recall, as well as the interactionbetween the phonological and the lexical levels of variation. Ina modified word recognition task, listeners displayed moreaccurate veridical word recall of AAE lexical items and voices.Furthermore, there was an interaction between phonologicaland lexical variation: listeners were even more accurate atrecognizing AAE-specific lexical items heard in an AAE voice.This study adds to a growing body of work finding thatsociolinguistic information influences word memory.

Textbooks commonly present scientific results with graphs. However, high school students struggle with interpreting them,in part because they do not focus on the most relevant comparisons among data values. Using a pre-/posttest design, weasked whether using visuospatial cues to teach this skill to high school students could improve graph comprehension. Halfof the students were randomly assigned to complete a visuospatial learning module, and the other half completed a non-visual control learning module. Data comparison performance increased significantly between pretest and posttest for thevisuospatial group, but decreased for the control group. Teaching students how to perceptually judge relevant comparisonscan thus improve graph comprehension.

How much does children’s performance on analogy tasks reflect general analogical reasoning versus specific knowledge? We asked this by comparing young children’s performance on conceptual (e.g., whole, broken) versus spatial (e.g., above, overlapping) analogies. We asked two primary research questions. First, does children’s performance correlate across tasks that depict conceptual versus spatial analogies? Second, if children complete the easier analogical task first, does that experience boost performance on the second, harder task? Successfully solving analogy problems in one domain could provide insights to children that may carry over to a new domain. However, if poor performance reflects an underlying lack of knowledge, rather than weak analogical reasoning, then additional analogy experience will not be beneficial. Results showed that children performed significantly better on conceptual than spatial analogies, and that the order of tasks did not influence performance. Furthermore, performance was not correlated across domains. These results suggest that performance on these two tasks primarily reflects children’s understanding of the concepts and relations needed to complete the analogies, rather than analogical reasoning.

Listener gaze can predict reference resolution as it reflectslisteners’ understanding. Further, speakers commonly referin installments to co-present objects by providing a descrip-tion incrementally. Here, we investigate whether listener gazecould be utilized to refer incrementally, in spoken installments.Specifically, we implemented a system that generates instruc-tions, describes objects, and reacts to listener gaze with verbalfeedback. We compared unambiguous vs. ambiguous instruc-tions supplemented by two levels of feedback specificity: ei-ther underspecified (“No, not that one!”) or more informative,contrastive responses (“Further left!”). Our findings show thatambiguous instructions with underspecified feedback did notbenefit task performance. In contrast, ambiguous instructionswith contrastive feedback (referring in installments) resulted inmore efficient interactions. Moreover, this strategy even out-performed the one providing unambiguous instructions.

This study investigated how beat gesture and pitch accentaffect the cognitive load of listeners during languagecomprehension. Evidence from pupillometry and dwell timeindicated that more cognitive resources were required toprocess the combination of these cues than their absence, andthey suggest that beat gesture may have required morecognitive resources to process than pitch accent. Additionally,pupil size positively correlated with reaction time anddecreased as the task progressed, demonstrating its usefulnessas a measure of cognitive processing. These results indicatethat viewing gesture in conjunction with speech may increasecognitive load during language processing, and that thisincreased load may result in enriched representations.

Word games such as crossword puzzles are widely used in ed-ucation to help familiarize students with technical vocabulary.Despite an extensive literature discussing their use, few pub-lished research articles have established their effectiveness onmemory retention and retrieval, especially in comparison tocontrol study methods. We report two experiments in whichuniversity introductory psychology students studied materialsrelevant to their coursework using an on-line interactive wordgame. Results showed that word games improved later tests ofthe material, both on immediate test and after a delay, and re-tention was most enhanced in comparison to control when theclues were solved repeatedly and given with difficult ortho-graphic hints. Importantly, easy clues were not retained overtime even with multiple repetitions. Results suggest that wordgames can be effective in this domain, but it depends on howthey are implemented, and several factors predicted by existingcognitive theory can guide implementation choices.

Computer-based navigation aids support navigation by foregrounding route instructions. Using those aids can have detri-mental effects on the ability to orient oneself without assistance. Rather than providing allocentric spatial informationin the form of conventional maps, navigation aids may support another form of human spatial memory that is essen-tially egocentric. In the present study, visualizations for navigation assistance systems were experimentally varied usinga virtual environment. Effects on wayfinding success and orientation after navigation were investigated. Results showthat dynamic, user-aligned views during navigation (providing spatial directional information to unseen targets) reducedthe risk of making erroneous turning decision during navigation and improved orientation after navigation, in contrast toview-independent representations with a stable north-aligned coordinate system. It is concluded that user-aligned viewscan facilitate the acquisition of egocentric survey knowledge while avoiding representational conflicts during navigation.

Storing knowledge of an agent’s environment in the form of aprobabilistic generative model has been established as a cru-cial ingredient in a multitude of cognitive tasks. Perceptionhas been formalised as probabilistic inference over the state oflatent variables, whereas in decision making the model of theenvironment is used to predict likely consequences of actions.Such generative models have earlier been proposed to under-lie semantic memory but it remained unclear if this model alsounderlies the efficient storage of experiences in episodic mem-ory. We formalise the compression of episodes in the norma-tive framework of information theory and argue that seman-tic memory provides the distortion function for compressionof experiences. Recent advances and insights from machinelearning allow us to approximate semantic compression in nat-uralistic domains and contrast the resulting deviations in com-pressed episodes with memory errors observed in the experi-mental literature on human memory.

A recent approach based on Bayesian inverse planning for the“theory of mind” has shown good performance in modelinghuman cognition. However, perfect inverse planning differsfrom human cognition during one kind of complex tasks dueto human bounded rationality. One example is an environmentin which there are many available plans for achieving a specificgoal. We propose a “plan predictability oriented model” as amodel of inferring other peoples’ goals in complex environ-ments. This model adds the bias that people prefer predictableplans. This bias is calculated with simple plan prediction. Wetested this model with a behavioral experiment in which hu-mans observed the partial path of goal-directed actions. Ourmodel had a higher correlation with human inference. We alsoconfirmed the robustness of our model with complex tasks anddetermined that it can be improved by taking account of indi-vidual differences in “bounded rationality”.

Gain scores are obtained as the difference between two consecutive measurements of knowledge. Although they arewidely acknowledged as a measure of change, they have been harshly criticized since empirical research has shown itsserious conceptual problems. To gain insight on the nature of these problems, I developed a model for the gains ofknowledge in the setting of a pretest-posttest instructional intervention. The model explains seemingly odd phenomenaassociated to gain scores: (a) negative gain-pretest correlations, and (b) lack of correlations between gain scores andlearner’s cognitive abilities. This highlights the potential of the proposed model for investigating the change of knowledgein a pretest-posttest educational setting and emphasizes the importance of modelling change by using information providedby specific application areas. Further work may lead to developing novel statistical methods for analysing educational dataand for estimating the change of knowledge in diverse educational contexts.

We gesture when we talk. Nonetheless, our speech is disfluent at times. The present study investigated whether ges-tures accompanying disfluencies may facilitate speech production by shortening the duration of disfluencies. FourteenEnglish-speaking adults were presented with educational videos and told to teach others after seeing these videos. Alltheir disfluencies and gestures were coded. Results reveal that disfluencies accompanied by representational gestures aresignificantly shorter as compared to if they had been accompanied by non-representational gestures or no gestures at all.There was no significant difference in duration between disfluencies accompanied by the latter two. This suggests thatrepresentational gestures may play a role in aiding speakers in the resumption of their speech. Implications for models ofhow gestures may help speech production are discussed.

To improve road safety, it is important to understand the impact that the contingencies around traffic lights have upon drivers’ behavior. There are formal rules that govern behavior at UK traffic lights (see The Highway Code, 2015), but what does experience of the contingencies do to us? While a green light always cues a go response and a singleton red a stop, the behavior linked to amber is ambiguous; in the presence of red it cues readiness to start, while on its own it cues "preparation" to stop. Could it be that the contingencies between stimuli and responses lead to implicit learning of responses that differ from those suggested by the rules of the road? This study used an incidental go/no-go task in which colored shapes were stochastically predictive of whether a response was required. The stimuli encoded the contingencies between traffic lights and their appropriate responses, for example, stimulus G was a go cue, mimicking the response to a green light. Evidence was found to indicate that G was a go cue, while A (which had the same contingencies as an amber light) was a weak go cue, and that R (a stop cue) was surprisingly responded to as a neutral cue.

This study examined the differences among individuals in theperformance of insight problem solving. The problem-solvingcharacteristics of an individual seemed to be dependent onwhat and how they had learned. Thus, we compared theperformances of insight problem solving betweenreinforcement and supervised learners. The results showed thatthe performances of reinforcement learners were better thanthose of supervised learners, although the non-insight problemsolving performance of both learner types was comparable.This result suggests that insight might be supported by thecognitive mechanisms underlying reinforcement learning. Inparticular, we speculate that the degree of exploration, bywhich reinforcement learning is characterized, might have animpact on the performance of insight problem solving

The ability to express negation is an important part of early lan-guage. Despite the fact that negation is a complex and abstractconcept, “No” is one of the first words that children produce.Past analyses have found that children’s early negations tendto express concepts like refusal (negations expressing that achild does not want to do something) rather than denial (nega-tions expressing that something is false). Does this mean thatyoung children are incapable of expressing denial? In Study1, we examine children’s spontaneous production of negationand find that some children produce denial negation earlier andmore frequently than past literature suggests. In Study 2, weexamine one possible explanation for individual variation inchildren’s negation production: differences in the joint activ-ities that they engage in with their caregivers. A comparisonof two children suggests that reading may be associated withthe production of denial negation. We discuss our data in lightof previous findings, and suggest that certain communicativecontexts are more likely to elicit different types of negation.

We hypothesized that college biology students difficulty interpreting relationships depicted in evolutionary trees (clado-grams) at least partly reflects their responding based on Gestalt grouping principles. Students from non-majors introduc-tory, majors introductory, and upper-level biology classes (N = 310) evaluated two pairs of cladograms after classroominstruction on evolutionary trees. The cladograms in each pair depicted the same evolutionary relationships among threetarget taxa but grouping of those taxa differed due to Gestalt principles. Students were asked which cladogram best repre-sents the specified relationships among the target taxa or whether both cladograms are equally good (the correct answer).As predicted, for all three biology groups, students responses most often were consistent with the Gestalt principles ofgrouping rather than with the pattern of evolutionary relationships (M = 1.28 out of 2; t(309) = 13.55, p ¡ .001). Clearly,biology instruction needs to address the potentially interfering role of Gestalt grouping.

Simple probability judgments pervade human experience.Decades of research have revealed a pattern of heuristic errorsin simple random draw predictions of both children and adults.Participants often make their choice based on the magnitude ofthe target or the non-target set without relating the two quanti-ties. In a series of experiments, we demonstrate that this biasis robust in both timed and untimed tasks (Experiment 1) andmay be overcome when the child is given the adequate amountand type of feedback (Experiment 2).

Obrecht and Chesney (2016) contend that deliberation supports greater base rate use. In line with this, they found that prompting deliberation by evaluating arguments about the usefulness of base rate and/or stereotype data increased subsequent use of base rates in judgment tasks. However, an alternative account of these results is that the intervention increased base rate use merely by increasing the salience of base rate information, rather than by increasing deliberation. Here we examine these accounts in two experiments. Experiment 1 showed that participants prompted to deliberate by evaluating arguments used base rates more in subsequent judgements, compared to participants who were merely reminded of relevant information. Experiment 2 showed that participants prompted to deliberate by completing math problems prior to the judgment task also increased their base rate use. Taken together, these results support the theory that tasks that prompt deliberative processes increase normative use of base rates.

Anchoring effect, the effect of precedent stimuli on subsequentnumerical estimation, is one of the most studied topics injudgment and decision making. Many researchers haveexamined its psychological processes from many perspectives.However, few studies have directly compared strength ofanchoring effects generated by different anchor types. Thepresent study involved a behavioral experiment (numericalestimation task after presenting an anchor) and compared theeffect size of anchoring effect on numerical estimations amongdifferent five anchors. We found that significant anchoringeffect occurred only in two types of anchor. Common twofeatures of these two anchors were representation of specificnumber and the dimensional equivalence between an anchorand a target in the numerical estimation task. Thus, thesefindings indicated that presentation of a specific number withdimensional equivalence as in the target of a numericalestimation task plays an important role in the generation ofrobust anchoring effect. Psychological mechanisms ongeneration of anchoring effect are discussed.

An essential part of cognitive science is designing experimentsthat distinguish competing models. This requires patience andingenuity—there is often a large space of possible experimentsone could run but only a small subset that might yield informa-tive results. We need not comb this space by hand: If we useformal models and explicitly declare the space of experiments,we can automate the search for good experiments, looking forthose with high expected information gain. Here, we presentan automated system for experiment design called webppl-oed.In our system, users simply declare their models and experi-ment space; in return, they receive a list of experiments rankedby their expected information gain. We demonstrate our sys-tem in two case studies, where we use it to design experimentsin studies of sequence prediction and categorization. We findstrong empirical validation that our automatically designed ex-periments were indeed optimal.

Making analogies is considered to depend on executivefunctions. We examined the role of the central executive insolving pictorial cross-mapping problems while generatingrandom digits ranging 1-3 for one group of subjects, and 1-9for another. We used three indices assessing different aspectsof randomness and a self-report measure to evaluate the effectof the concurrent task. Subjects who had to generate digitsbetween 1 and 9 perceived the task to be harder but stillproduced more random sequences than those in the smaller-range condition. Although the manipulation of cognitive loadwas successful, no difference was observed in the proportionof relational responses to the cross-mapping task, suggestingthat analogies may not be as cognitively demanding asotherwise assumed. We also provide correlational support forthe influence of individual differences in fluid intelligence onrelational mapping abilities.

A growing number of studies has revealed that humans and nonhuman animals have the ability to process magnitudes ofnonsymbolic ratios. Lewis, Mathews & Hubbard (2015) hypothesized that this ability may depend on a ratio processingsystem (RPS) that may help acquire symbolic fractions knowledge. The present study investigated ratio processing in2nd and 5th graders using functional MRI. In the scanner, children decided which of two ratios was numerically larger.The stimuli were constructed as pairs of nonsymbolic line ratios, symbolic fractions, and mixed notations. Both 2ndand 5th graders showed the distance effect the behavioral performance and the neural activation were modulated by thenumerical distance between two ratios. Notably, 5th graders showed greater neural distance effect and more overlapsin activation across notations when compared to 2nd graders. These results suggest that educational experience mightpromote recruitment of the RPS for processing symbolic fractions as well.

The Serial Reaction Time (SRT) and the Alternating Serial Reaction Time (ASRT) tasks are widely used assessments of sequence learning (SL) wherein repetitive patterning of visual- spatial elements leads participants to anticipate locations of subsequent elements in the series. In the SRT task, the predictive dependencies involve adjacent elements whereas in the ASRT task they involve nonadjacent elements, due to the insertion of random elements into the pattern. We tested college students (N = 74) to explore whether the SRT and the ASRT tasks relied on similar underlying learning mechanisms while also examining associations between task performance and nonverbal fluid intelligence, visual-spatial working memory, and sentence processing ability. There was no correlation in performance across the two SL tasks (r = –.18), suggesting distinct learning mechanisms. Whereas 95.9% of participants demonstrated sequence-specific learning in the SRT task, only 64.9% demonstrated learning in the ASRT task. SL in the ASRT but not the SRT task was associated with nonverbal intelligence, visual-spatial working memory, and sentence comprehension. The observed results run counter to the claim that the ASRT relies only on implicit learning mechanisms presumed to be unrelated to executive functioning or general intelligence.

Research on semi-supervised category learning has beensparse despite its representativeness of naturalistic categorylearning and potential applications. Most of the semi-supervised literature to date has focused on establishing thephenomenon. These efforts have produced mixed results andhave explored a relatively limited set of learningcircumstances. In the current work, we contribute a novelinvestigation of semi-supervised learning by extending theparadigm to relational category learning and evaluating therole that item similarity plays in the effectiveness ofunsupervised learning opportunities. Our results show first-ever evidence of semi-supervised learning in the induction ofrelational categories and, further, that the similarity betweensupervised and unsupervised examples critically dictateswhether benefits of unsupervised exposures accrue. Weconclude with implications and future directions.

The framework of cognitively bounded rationality treats prob-lem solving as fundamentally rational, but emphasises that itis constrained by cognitive architecture and the task environ-ment. This paper investigates a simple decision making heuris-tic, Take The Best (TTB), within that framework. We formu-late TTB as a likelihood-based probabilistic model, where thedecision strategy arises by probabilistic inference based on thetraining data and the model constraints. The strengths of theprobabilistic formulation, in addition to providing a boundedrational account of the learning of the heuristic, include naturalextensibility with additional cognitively plausible constraintsand prior information, and the possibility to embed the heuris-tic as a subpart of a larger probabilistic model. We extend themodel to learn cue discrimination thresholds for continuous-valued cues and experiment with using the model to accountfor biased preference feedback from a bounded rational agentin a simulated interactive machine learning task.

Human understanding of relations between objects depends onthe ability to code meaningful role bindings. Computationalmodels of relational reasoning have proposed that neuraloscillations provide a basic mechanism enabling workingmemory to code the bindings of objects into relational roles.We adapted a behavioral oscillation paradigm to investigatemoment-to-moment changes in representations of semanticroles. On each trial, a picture was presented showing an action(chasing) relating two animals, one animal playing an agentrole (chaser) and the other playing a patient role (chased). Afterthe picture disappeared, the inter-stimulus interval (ISI) wasvaried in densely-sampled increments followed by a verbalprobe indicating an animal in a role. Reaction time (RT) todecide the validity of the verbal probe was recorded. We foundthat RTs varied systematically with ISI in an oscillatoryfashion. A task that required memory for a relational roleevoked stronger theta- and alpha-band oscillations than did amemory task not involving relational roles. The behavioraloscillation patterns in the role-identification task revealed aphase shift between the two semantic roles in the alpha band.

Modern convolutional neural networks (CNNs) are able toachieve human-level object classification accuracy on specifictasks, and currently outperform competing models in explain-ing complex human visual representations. However, the cate-gorization problem is posed differently for these networks thanfor humans: the accuracy of these networks is evaluated bytheir ability to identify single labels assigned to each image.These labels often cut arbitrarily across natural psychologi-cal taxonomies (e.g., dogs are separated into breeds, but neverjointly categorized as “dogs”), and bias the resulting represen-tations. By contrast, it is common for children to hear bothdog and Dalmatian to describe the same stimulus, helping togroup perceptually disparate objects (e.g., breeds) into a com-mon mental class. In this work, we train CNN classifiers withmultiple labels for each image that correspond to different lev-els of abstraction, and use this framework to reproduce classicpatterns that appear in human generalization behavior.

People increasingly interact with different types ofautonomous robotic systems, ranging from humanoid socialrobots to driverless vehicles. But little is known about howpeople interpret the behavior of such systems, and inparticular if and how they attribute cognitive capacities andmental states to them. In a study concerning people’sinterpretations of autonomous car behavior, building on ourprevious research on human-robot interaction, participantswere presented with (1) images of cars – either with orwithout a driver – exhibiting various goal-directed trafficbehaviors, and (2) brief verbal descriptions of that behavior.They were asked to rate the extent to which these behaviorswere intentional and judge the plausibility of different typesof causal explanations. The results indicate that people (a)view autonomous car behavior as goal-directed, (b)discriminate between intentional and unintentionalautonomous car behaviors, and (c) view the causes ofautonomous and human traffic behaviors similarly, in termsof both intentionality ascriptions and behavior explanations.However, there was considerably lower agreement inparticipant ratings of the driverless behaviors, which mightindicate an increased difficulty in interpreting goal-directedbehavior of autonomous systems.

A new Bayesian model of visual word recognition is used tosimulate neighborhood frequency effects in lexical decision.These effects have been reported as being either facilitatory orinhibitory in behavioral experiments. Our model manages tosimulate the apparently contradictory findings. Indeed, study-ing the dynamic time course of information accumulation inthe model shows that effects are facilitatory early, and becomeinhibitory at later stages. The model provides new insights onthe mechanisms at play and their dynamics, leading to betterunderstand the experimental conditions that should yield a fa-cilitatory or an inhibitory neighborhood frequency effect.

Testimony is a fundamental feature of human life: typically, wereceive testimonial evidence from others multiple times each day.Often, we have more than one source attesting to a particularclaim. This paper examines the way people integrate testimonialevidence from multiple sources. We find evidence that participantsdeviate substantially from the normative expectation. Instead,results seem indicative of the operation of simple, non-compensatory heuristics, at least some of the time.

Despite a general recognition that there are two styles of think-ing: fast, reflexive and relatively effortless (Type 1) versus slow,reflective and effortful (Type 2), dual-process theories of cogni-tion remain controversial, in particular, for their vagueness. Toaddress this lack of formal precision, we take a mathematicalcategory theory approach towards understanding what under-lies the relationship between dual cognitive processes. Fromour category theory perspective, we show that distinguishingfeatures of Type 1 versus Type 2 processes are exhibited viaadjoint and representable functors. These results suggest thatcategory theory provides a useful formal framework for devel-oping dual-process theories of cognition.

We investigate the relationship between evaluations of knowl-edge and belief in human adult theory of mind, and provideevidence that evaluations of knowledge are made without priorevaluations of belief. Our studies find that (1) people can ac-curately evaluate others’ knowledge before they evaluate theirbeliefs; (2) this pattern cannot be not explained by pragmaticdifferences; (3) it occurs cross-linguistically and unlikely tobe accounted for by differences in word frequency, and (4) italso generalizes to the larger class of factive and non-factiveattitudes (to which knowledge and belief respectively belong).Together, these studies demonstrate that human adults can as-cribe knowledge without first ascribing a belief state. Moregenerally, they lend support to the view that knowledge repre-sentations are a distinctive and basic way in which we makesense of others’ minds.

Infancy is the foundational period for learning from adults, and the dynamics of the social environment have long beenproposed as central to childrens development. Here we reveal a novel, naturalistic approach for studying live interactionsbetween infants and adults. Using functional near-infrared spectroscopy (fNIRS), we simultaneously and continuouslymeasured the brains of infants (9-15 months) and an adult while they communicated and played with each other in realtime. We found that time-locked neural synchrony within dyads was significantly greater when they interacted witheach other than with control individuals. In addition, we found that both infant and adult brains continuously trackedthe moment-to-moment fluctuations of mutual gaze and infant emotion with high temporal precision. This investigationmarks a new means of understanding how the brains and behaviors of infants both shape and reflect those of their caregiversduring real-life communication.

Linguistic devices that mark confidence (uncertainty) havebeen well documented (e.g., choice of modals, hedges, etc),yet there has been surprisingly little empirical work thatexplicitly measures how uncertainty is signaled andinterpreted. We present an initial report on a project designedto investigate how interlocutors communicate uncertainty anduse that information in acquiring new information andintegrating interlocutor based input with their prior beliefs.Experiment 1 establishes that speakers and listeners agree onthe relative degree of uncertainty for a set of phrases.Experiment 2 manipulated how likely it was that a participantwould recognize an object using images that varied inrecoverability, finding that recoverability mapped ontocertainty. Experiment 3 used a word-learning paradigm toestablish that learners take into account the certainty withwhich a speaker labels uses a novel word to label a novelshape.

We tested whether relational knowledge is represented as a set of relations among entities or as a set of relational roles towhich entities are bound. Participants performed four relational processing tasks with the same set of word-pair stimuli:relational exemplar generation; similarity ranking; analogical verification; and a paired-associate learning task. In thesimilarity ranking task, we gathered separate rankings for relational, role and semantic similarity between word pairs.Relational similarity predicted exemplar generation frequencies, analogical verification accuracy and RTs, and relationalluring in associative memory. Role similarity predicted exemplar generation frequency, and, weakly, analogical verificationRTs. Semantic similarity did not predict any of the tasks, after controlling for the other two factors. Contrary to currenttheories which posit that semantic similarity is more important for retrieving relevant analogues, and that analogicalmapping is based on role-filler bindings, relational similarity was the strongest predictor across all tasks.

People’s expectations play an important role in their evalua-tions and reactions to events. There is often disappointmentwhen events fail to meet expectations—sometimes even whenthe events are still positive overall—and there is a special thrillto having one’s expectations exceeded. In four studies, weexamined how expectations influence people’s judgments ofevents where another person or people were harmed. Partici-pants judged pairs of events where a victim experienced a sim-ilar harm, but where victims were at different prior risk of be-ing harmed. We found that people judged these events as beingworse when they were less expected–that is, when the victimswere initially at lower risk of being harmed. We argue that thisbias has pernicious moral consequences.

People generally ignore evidence that is contrary to theirbeliefs (Nickerson, 1998). To examine the factors thatpromotes attitude change with a new perspective, this studyexamined how people change their beliefs on a range of topicsfrom gender identity to gun control on the Redditforum Change My View. Specifically, we examined howpeople on Change My View cite evidence to change otherpeople’s minds. As prior work suggests, we find that peopleare not easily convinced to change their beliefs about socialand moral issues, and this occurs even though people citeconsiderably more evidence while discussing theseissues. However, our data provides one source of optimism:We found that the amount of evidence provided in adiscussion predicts attitude change, suggesting that whileattitude change is hard-won, providing facts and evidencemay nonetheless be an effective persuasive tactic.

The paper investigates the blame blocking effect with respect to assignments of punishments and pursues the question of whether the effect generalizes to people with legal education. The blame blocking effect predicts that an agent is punished more severely when an intendedly harmful action does not lead to harm, compared to the case in which the harm results but is caused independently of the agent (Cushman, 2008). Firstly, we replicate the blame blocking effect for people without legal education. Secondly, our findings indicate that this effect is not present in people with a sufficient degree of legal training: In contrast to first-year students – who still seem to exhibit blame blocking – the effect was not observed for people with more than one year of legal education.

The debate about semantic reference between Frege’s (1948) descriptivism and Kripke’s (1972) causal theory of reference has recently been approached through experimental psychology. However, no consensus has been reached on the direction of the results. While some studies face clear methodological charges, even those that are currently uncontested do not reach a mutual conclusion. We propose a novel experimental paradigm with methodology designed to evade the problems of previous studies. Contrary to the past literature, we find a prevalence of descriptivists under lenient criteria for consistency across trials, while under strict criteria we find an equal amount of descriptivists and hybrids, with low numbers of referentialists (causal theory of reference) under both criteria. We suggest an interpretation of this result, and where future research might head.

A common dependent measure used in visual-world eye-tracking experiments is the proportion of looks to a visuallydepicted object in a certain time window after the onset of thecritical stimulus. When interpreting such data, a common as-sumption is that looks to the object reflect the listener’s beliefthat the object is the intended target referent. While this isintuitively plausible (at least for paradigms in which the taskrequires selecting a referent), relatively little is known abouthow exactly the proportion of looks to an object is related toa listener’s current belief about that object. Here, we test asimple, explicit linking hypothesis: the proportion of looks toan object correlates with the probability that the listener as-signs to the object being the target. To test this hypothesis,we supplement the eye-tracking data from Leffel, Xiang, andKennedy (2016) with an offline incremental decision task tomeasure participants’ beliefs about the intended referent at var-ious points in the unfolding sentence, and assess the extent towhich these beliefs predict the eye-tracking data. The resultssuggest that the degree to which an object is believed to be thereferent is only one factor that affects eye movements in ref-erential tasks. Preliminary free production data we have col-lected for the scenes suggests that utterance expectations alsoplay a role. We discuss methodological implications of theseresults for experimental linguistics.

We address competing perspectives on how social-indexicalmeaning is learned in language, using data from artificial lan-guage learning experiments and two studies in small-scalesocieties. Our results indicate that learning social-indexicalmeaning is primarily allocentric as opposed to egocentric:speaker success in learning a social-indexical meaning patterndepends on overall exposure to the pattern more than the pat-tern’s relative importance to the speaker. We base these claimson data from American English-speaking adults, Datooga-speaking children, as well as adults and children speakingMurrinhpatha. The results highlight the importance of widen-ing the sample of methods and data sources in studying howvariation in language is learned and maintained.

Reasoning about possibilities is fundamental in daily life and inartificial intelligence. It is formalized in modal logics, of whichthere are infinitely many. Two experiments showed thatindividuals make inferences that are parsimonious aboutpossibilities, and that they reject conclusions referring topossibilities that the premises do not support. Both sorts ofinference contravene modal logics, i.e., the simplest system ofmodal logic and the infinite number of systems based on it.

Many theories of reasoning and many experiments presupposethat human ability is stable over time, and so people usuallydraw the same conclusion from the same premises. The as-sumption has hitherto had little or no empirical investigation.We therefore analyzed a study in which 20 participants drewtheir own conclusions to the 64 sorts of syllogisms on twooccasions separated by roughly a week. We report the na-ture of the changes in the participants’ conclusions includingtheir spontaneous improvement in logical accuracy, and use amodel-based program, mReasoner, to explain the results.

Cognitive architectures are an advantageous tool for creatingcognitive models. They provide a framework integrating generalcognitive structures and assumptions about the mind as forexample the working memory, structural modularity or theirinterconnections. A vast number of cognitive architectures havebeen developed in the last decades. While the architectures realizethe cognitive perspective, the formal foundation and similaritiesof cognitive architectures remain open. To identify the cognitivesubstrate of the architectures, we propose a generalized cognitiveframework allowing to embed different cognitive architecturesto analyze their properties and to have a common and formalground for comparisons. We demonstrate our approach – as proof-of-concept – by embedding the two most popular architectures,ACT-R and SOAR, and evaluate cognitive models for recognitionmemory in our approach. Potentials and limitations are discussed.

In this study, we test the possibility that real-life events induce anabstract category activation in a way that permits structurally-basedretrievals. We used a free-recall reminding paradigm whereparticipants had to report any memory that come to mind when facedwith a target cue embodying a familiar concept. This methodallowed us to consider the retrievals of any analog that shares ameaningful structural similarity in the participants’ own eyes.Results revealed that most participants predominantly retrievedSuperficially Dissimilar Analogs (SDAs) rather than SuperficiallySimilar Disanalogs (SSDs). Interestingly, retrievals of SDAs werepreponderant over retrievals of Superficially Similar Analogs(SSAs). These data suggest that familiar abstract knowledge mayhave a more important role in promoting abstract encoding andstructurally-based retrievals than it was supposed to.

Perceptual aspects of events, such as the visual prominence of event participants, have been shown to influence how people describe events. We investigate the relationship between such perceptual effects and patterns of eyegaze, focusing on a little- explored perceptual manipulation: the extent to which an event participant is wholly or partially visible. Using an eyetracking method, we found a correlation between this perceptual contrast and patterns of eyegaze at the beginning of the event but not the end. This finding supports the view that early visual attention to events has important downstream consequences for event conceptualization and linguistic description.

Fractions processing is a topic of major interest both innumerical cognition and mathematics education. Theliterature on the processing of common fractions has focusedon whether fractions are compared by their magnitude orthrough their components. Only a few neuroimaging studieshave looked at this question. The N400 component,traditionally seen in linguistic semantic congruency event-related-potential (ERPs) experimental designs, has beenadapted to study arithmetic processing. Observing the N400,allows the study of how different arithmetic componentsaffect overall processing. In this study, an N400 paradigm isused to investigate semantic congruency during a fractionmagnitude comparison task (Match/Mismatch) in 24 adults.Behavioral results reveal interference by shared componentsacross the compared fractions. EEG analysis results show anN400-like difference wave between Match and Mismatchconditions. Shared components modulate the latency of thisN400 effect. These results show the N400 as a viable methodfor studying fractions.

Our central question is: how accurate are laypeople’sstatistical intuitions about probability distributions within thedomain of health? Specifically, can participants produceentire probability distributions for the duration of illnesses?While a large body of decision making research has suggestedthat people use a flawed process to arrive at decisions, weposit that participants may be using an optimal process, butwith flawed information. To this end, we assess accuracy interms of both the mean and form of distributions for bothacute illnesses for which people might have experience, andchronic conditions for which people are less likely to haveexperience. We find that participants can accurately estimatethe mean and form of distributions for acute illnesses.

It is believed that decisions about what information needs additional study before an upcoming exam are dependent upon metacognitive processes. While a great deal of research has explored these processes, far less work has explored how to optimize restudy decisions. In the present study we examined both what cues are most predictive of future retrieval and test two potential ways of nudging learners to use these cues when making their restudy decisions. All methods and analyses were pre-registered on the Open Science Framework. Assessment of cue-utilization revealed that pre-judgment recall accuracy and pre-judgment retrieval latency, but not stimulus font size, predicted future retrieval. Additionally, both feedback about pre-judgment retrieval accuracy and having participants make retrospective confidence judgments led learners to more heavily weigh prejudgment retrieval accuracy when making their restudy decisions. This increase in relevant cue use, however, did not carry over into more accurate restudy decisions. These findings suggest that subtle manipulations can push learners to utilize more appropriate cues when making their restudy decisions.

Many people report experiencing their thoughts in the form ofnatural language, i.e., they experience ‘inner speech’. Atpresent, there exist few ways of quantifying this tendency,making it difficult to investigate whether the propensity toexperience verbalize predicts objective cognitive function orwhether it is merely epiphenomenal. We present a newinstrument —The Internal Representation Questionnaire(IRQ) —for quantifying the subjective format of internalthoughts. The primary goal of the IRQ is to assess whetherpeople vary in their stated use of visual and verbal strategies intheir internal representations. Exploratory analyses revealedfour factors: Propensity to form visual images, verbal images,a general mental manipulation factor, and an orthographicimagery factor. Here, we describe the properties of the IRQ andreport an initial test of its predictive validity by relating it to aspeeded picture/word verification task involving pictorial,written, and auditory verbal cues.

How can social learning influence children’s inclinationstoward equality-based or merit-based fairness? To investigatethis question, six- and seven-year-olds were first presentedwith a pre-test distribution task in which they divided eightstickers between two hypothetical children, one of whom wasa more productive worker. Participants were then given brief,direct testimony that advocated either equality- or merit-basedfairness (whichever was not preferred at pre-test), and thatappealed either to emotions or reason. A novel experimenterthen presented participants with a post-test distribution task.The results indicated that a majority of children changed theirdistribution patterns from pre-test to post-test after beingprovided with direct testimony. These changes in resourcedistribution were accompanied by marked changes in thekinds of explanations that children provided. This researchindicates that children’s preferences for different forms of justresource distribution can be heavily influenced by socialcommunication.

Humans master complex systems of interrelated concepts likemathematics and natural language. Previous work suggestslearning these systems relies on iteratively and directly re-vising a language-like conceptual representation. We intro-duce and assess a novel concept learning paradigm calledMartha’s Magical Machines that captures complex relation-ships between concepts. We model human concept learning inthis paradigm as a search in the space of term rewriting sys-tems, previously developed as an abstract model of compu-tation. Our model accurately predicts that participants learnsome transformations more easily than others and that theylearn harder concepts more easily using a bootstrapping cur-riculum focused on their compositional parts. Our results sug-gest that term rewriting systems may be a useful model of hu-man conceptual representations.

In this pre-registered study, we investigated whether facial expressions were implicitly encoded when forming impressions of others, and whether differences between people in their encoding of angry and happy facial expressionswere related to depressive symptoms. These questions were addressed using the category confusion or Who Said What (WSW) paradigm. Results indicated that both angry and happy emotional expressions from human faces were encoded when forming impressions of others, with no difference in strength of encoding between both.We observed no evidence for associations between encoding of angry or happy facial expressions and depressive symptoms.

Redemptive narratives are stories of challenge, failure, or adversity that in some way acknowledge the goodness or personal growth that came of the recounted difficult event. In this paper we use a corpus-statistic based approach to explore the role of cognition and emotion in these narrative arcs. In particular, we trace the shift from negative to positive sentiment (a change in the emotional valence) and vice to virtue (evidence of cognitive, moral processing) within the narrative. Our results suggest that cognitive processes, more than emotion, drive the shift to goodness and growth that is at the core of redemptive narratives. We discuss the implications of these results to both narrative psychology and cognitive psychology.

Humans not only can effortlessly imagine a wide range ofnovel instances and scenarios when prompted (e.g., a newshirt), but more remarkably, they can adequately generate ex-amples which satisfy a given set of constraints (e.g., a new,dotted, pink shirt). Recently, Nobandegani and Shultz (2017)proposed a framework which permits converting deterministic,discriminative neural nets into probabilistic generative models.In this work, we formally show that an extension of this frame-work allows for generating examples under a wide range ofconstraints. Furthermore, we show that this framework is con-sistent with developmental findings on children’s generativeabilities, and can account for a developmental shift in infants’probabilistic learning and reasoning. We discuss the impor-tance of integrating Bayesian and connectionist approaches tocomputational developmental psychology, and how our workcontributes to that research.

The Availability bias, manifested in the over-representation ofextreme eventualities, is a well-known cognitive bias, and isgenerally taken as evidence of human irrationality. In thiswork, we present the first rational, metacognitive account ofthe Availability bias, formally articulated at Marr’s algorith-mic level of analysis. Concretely, we present a normative,metacognitive model of how a cognitive system should over-represent extreme eventualities, depending on the amount oftime available for decision-making. Our Sample-based Ex-pected Utility model also accounts for two well-known fram-ing effects in human decision-making under risk—the fourfoldpattern of risk preferences in outcome probability (Tversky& Kahneman, 1992) and in outcome magnitude (Markovitz,1952)—thereby providing the first metacognitively-rationalbasis for the aforementioned effects. Empirical evidence con-firms an important prediction of our model. Surprisingly, ourmodel is strikingly robust with respect to its focal parameter.We discuss the implications of our work for studies on hu-man decision-making, and conclude by presenting a counter-intuitive prediction of our model, which, if confirmed, wouldhave intriguing implications for human decision-making un-der risk. To our knowledge, our model is the first metacog-nitive, resource-rational process model of cognitive biases indecision-making. Notably, our work also contributes to thefields of artificial intelligence and computational statistics, bypresenting a previously unknown proposal distribution, withfirm rational grounds, broadly applicable to the influential sub-field of importance sampling Monte Carlo methods.

The Availability bias, manifested in the over-representation ofextreme eventualities, is a well-known cognitive bias, and isgenerally taken as evidence of human irrationality. In thiswork, we present the first rational, metacognitive account ofthe Availability bias, formally articulated at Marr’s algorith-mic level of analysis. Concretely, we present a normative,metacognitive model of how a cognitive system should over-represent extreme eventualities, depending on the amount oftime available for decision-making. Our Sample-based Ex-pected Utility model also accounts for two well-known fram-ing effects in human decision-making under risk—the fourfoldpattern of risk preferences in outcome probability (Tversky& Kahneman, 1992) and in outcome magnitude (Markovitz,1952)—thereby providing the first metacognitively-rationalbasis for the aforementioned effects. Empirical evidence con-firms an important prediction of our model. Surprisingly, ourmodel is strikingly robust with respect to its focal parameter.We discuss the implications of our work for studies on hu-man decision-making, and conclude by presenting a counter-intuitive prediction of our model, which, if confirmed, wouldhave intriguing implications for human decision-making un-der risk. To our knowledge, our model is the first metacog-nitive, resource-rational process model of cognitive biases indecision-making. Notably, our work also contributes to thefields of artificial intelligence and computational statistics, bypresenting a previously unknown proposal distribution, withfirm rational grounds, broadly applicable to the influential sub-field of importance sampling Monte Carlo methods.

The importance of hierarchically structured representations fortractable planning has long been acknowledged. However, thequestions of how people discover such abstractions and how todefine a set of optimal abstractions remain open. This problemhas been explored in cognitive science in the problem solvingliterature and in computer science in hierarchical reinforce-ment learning. Here, we emphasize an algorithmic perspec-tive on learning hierarchical representations in which the ob-jective is to efficiently encode the structure of the problem, or,equivalently, to learn an algorithm with minimal length. Weintroduce a novel problem-solving paradigm that links prob-lem solving and program induction under the Markov Deci-sion Process (MDP) framework. Using this task, we target thequestion of whether humans discover hierarchical solutions bymaximizing efficiency in number of actions they generate or byminimizing the complexity of the resulting representation andfind evidence for the primacy of representational efficiency.

Folk psychology takes perception and cognition to be two distinct processes. It seems that when we perceive the world we are engaged in one kind of activity and when we think about it we are engaged in a different one. This conception underlies various discussions within the cognitive sciences, such as on the architecture and modularity of the mind, and the cognitive penetrability of perception. But is the distinction justified? This paper looks for an answer in two opposing paradigms in the sciences of the mind: traditional cognitivism and ecological psychology. Even though cognitivism is the dominant paradigm, we argue that it has thus far failed to give a definite account of the relation between perception and cognition, and to support or to deny their separation. Ecological psychology, on the other hand, rejects the distinction and integrates cognition with perception. We discuss previous work within the ecological view and sketch directions for future research.

The ability to process social information is a critical compo-nent of children’s early language and cognitive development.However, as children reach their first birthday, they begin tolocomote themselves, dramatically affecting their visual ac-cess to this information. How do these postural and locomotorchanges affect children’s access to the social information rele-vant for word-learning? Here, we explore this question by us-ing head-mounted cameras to record 36 infants’ (8-16 monthsof age) egocentric visual perspective and use computer visionalgorithms to estimate the proportion of faces and hands in in-fants’ environments. We find that infants’ posture and orienta-tion to their caregiver modulates their access to social informa-tion, confirming previous work that suggests motoric develop-ments play a significant role in the emergence of children’s lin-guistic and social capacities. We suggest that the combined useof head-mounted cameras and the application of new computervision techniques is a promising avenue for understanding thestatistics of infants’ visual and linguistic experience.

Research in psychology on reasoning has often been restrictedto relatively inexpressive statements involving quantifiers.This is limited to situations that typically do not arise inpractical settings, such as ontology engineering. In orderto provide an analysis of inference, we focus on reasoningtasks presented in external graphic representations wherestatements correspond to those involving multiple quantifiersand unary and binary relations. Our experiment measuredparticipants’ performance when reasoning with two notations.The first used topology to convey information via node-linkdiagrams (i.e. graphs). The second used topological andspatial constraints to convey information (Euler diagrams withadditional graph-like syntax). We found that topological-spatial representations were more effective than topologicalrepresentations. Unlike topological-spatial representations,reasoning with topological representations was harder wheninvolving multiple quantifiers and binary relations than singlequantifiers and unary relations. These findings are comparedto those for sentential reasoning tasks.

In this paper we explored how people represent categories that include exceptions by examining contributions that features of regular and exception items make to determining category membership. We examined performance of 4-year-old children and adults and found significant developmental differences. While for 4-year-olds, deterministic features of regular items and exceptions contributed comparably to determining category membership, an asymmetry was found in adults. For adults, deterministic features of regular items contributed more to determining category membership than features of exceptions. The results are discussed in relation to the SUSTAIN clustering model of category learning (Love, Medin, & Gureckis, 2004).

People often believe that orderly structures were created by agents. We examine the cognitive basis of this tendency, asking if learned associations or causal reasoning drives us to link order with agents. Causal reasoning predicts that knowledge of an alternative physical-mechanical cause should ‘explain away’ orderliness, weakening the link with agents. In a preregistered experiment, we manipulated the context to provide (or not provide) a physical-mechanical explanation for orderly outcomes, and participants judged if an object or agent had been present. We compared outcomes differing in (a)levels of orderliness and (b)whether context provided an alternative explanation. We found that environmental context ‘explained away’ orderliness, such that participants observing order inferred agency only when there was no alternative explanation. The link between order and agents is moderated by causal reasoning, and is malleable: It can be weakened by understanding alternative causal mechanisms by which order could arise.

In this paper, we introduce CALM, a process model that is de-signed to abstract solutions in simple and complex categorylearning tasks. The model includes strong assumptions aboutthe interaction of processes driving learning behavior, typicallyaddressed in terms of feature attention, stimulus generaliza-tion, rule abstraction and knowledge partitioning. We presentsimulations of CALM, showing that the model can accountboth for systematic variations in Type II category difficulty,and for individual differences in extrapolation of an XOR cat-egory structure.

Complexity matching—converging temporal correlations measured by correlating the slopes of power spectra—is a new measure of coordination based on information exchange between complex networks. To date, studies have focused on the dyadic case, but complexity matching may generalize to interacting complex networks in the left and right hemispheres of a single brain. We examined complexity matching in a perceptual-motor task between individuals and dyads. Participants alternated hitting targets in a Fitts-like task with the left and right hands of one individual, or analogously between two people. Response coupling was manipulated by making targets drift randomly (decoupled) or contingently (coupled). Results showed long-range correlations in time series of inter-response intervals exhibited complexity matching for both individuals and dyads, but only when responses were coupled via contingent drift. We conclude that complexity matching observed between individuals can similarly occur within one individual, suggesting a general principle of interaction at work.

In the process of collectively inventing new words for new con-cepts in a population, conflicts can quickly become numerous,in the form of synonymy and homonymy. Remembering all ofthem could cost too much memory, and remembering too fewmay slow down the overall process. Is there an efficient be-havior that could help balance the two? The Naming Game isa multi-agent computational model for the emergence of lan-guage, focusing on the negotiation of new lexical conventions,where a common lexicon self-organizes but going through aphase of high complexity. Previous work has been done onthe control of complexity growth in this particular model, byallowing agents to actively choose what they talk about. How-ever, those strategies were relying on ad hoc heuristics highlydependent on fine-tuning of parameters. We define here a newprincipled measure and a new strategy, based on the beliefsof each agent on the global state of the population. The mea-sure does not rely on heavy computation, and is cognitivelyplausible. The new strategy yields an efficient control of com-plexity growth, along with a faster agreement process. Also,we show that short-term memory is enough to build relevantbeliefs about the global lexicon.

Humans demonstrate a remarkable ability to infer physical properties of objects and predict physical events in dynamicscenes. These abilities have been modeled as probabilistic simulations of a mental physics engine akin to 3D physicsengines used in computer simulations and video games (Battaglia, Hamrick & Tenenbaum 2013; Sanborn, Mansinghka &Griffiths 2013), but it is unknown if and how such a physics engine is implemented in the brain. Does the brain representquantities corresponding to the key latent variables of physical objects that contribute to their dynamics? To find out,we used multivariate pattern classification analyses of fMRI data from subjects viewing videos of dynamic objects. Themass of depicted objects could be decoded, across physical scenarios and object materials, from brain regions previouslyimplicated in intuitive physics. This invariant representation of mass may serve as a key variable in a generalized enginefor intuitive physics.

According to the honing theory of creativity, the iterativeprocess culminating in a creative work is made possible by theself-organizing nature of a conceptual network, or worldview,and its innate holistic tendency to minimize inconsistency. Assuch, the creative process is not limited to the problem domain,and influences on creativity from domains other than that of thefinal product are predicted to be widespread. We conducted astudy in which participants with varying levels of creativeexperience listed their creative outputs, as well as influences(sources of inspiration) on these outputs. Of the 758 creativeinfluences, 13% were within-domain narrow, 13% within-domain broad, 67% cross-domain, and 6% unclear. Thesefindings support the hypothesis that to trace the inspirationalsources or ‘conceptual parents’ of a creative output, and thustrack its cultural lineage, one must look beyond the problemdomain to the creators’ self-organizing, inconsistency-minimizing worldview at large.

A proper representation of space and a joint attention mecha-nism are indispensable for an effective deictic communicationwith embodied agents. Taking inspiration from developmen-tal psychology may help us to tackle computational challengesfor robots. Although some developmental joint attention mod-els for robots have already been proposed, to the best of ourknowledge, there is no such model that can stand for the ef-fects of pointing gestures on covert attention in infants. Thuswe have designed and implemented a developmental roboticsmodel for joint spatial attention combining connectionist anddynamical approaches. The hybrid architecture was struc-tured over two existing computational models: a connectionistmodel of gesture comprehension and a Dynamic Field (DF)model of spatial attention in infants. These models were ex-tended with various perceptual modules and dynamical neu-ral fields, and implemented on the state-of-art iCub humanoidrobot. In this paper, the computational architecture is intro-duced with some preliminary results that show the model’s ca-pability of representing deixis and perceived objects, and theireffects on attention over space and time.

High-level cognition is often accomplished not byindividuals working in isolation, but by distributed, complexcognitive systems. Examples include teams of scientists orcollaboratively improvising musicians. These distributedsystems can undergo critical transitions, suddenly movingfrom one stable pattern of activity to another. For instance,in ‘free jazz,’ where musicians improvise without apredetermined plan or a central leader, the performance willoften settle into a particular texture or style beforetransitioning to something entirely new, often quitesuddenly. When do these transitions occur? Are theyforeseeable? Inspired by suggestions that cognitive systemsare, in some sense, a kind of ‘ecosystem,’ we draw on recentwork in quantitative ecology that has begun to describegeneric early warning signals of impending criticaltransitions in ecosystems. We apply these techniques to acorpus of audio recordings of professional jazz quartetsplaying improvised music. We find that the same genericmeasures that have been used successfully to predict criticaltransitions in natural ecosystems describe the complexdynamics of improvised musical performance in the lead-upto transitions. By taking seriously the metaphor thatcognition occurs in ‘ecosystems,’ we gain new insights intohow stable patterns of thought can emerge suddenly incomplex cognitive systems.

In many daily life situations, people face decisions involving atrade-off between exploring new options and exploiting knownones. In these situations, observing the decisions of others caninfluence people’s decisions. Whereas social information oftenhelps making better decisions, research has suggested that undercertain conditions it can be detrimental. How precisely socialinformation influences decision strategies and impacts perfor-mance is, however, disputed. Here we study how social informa-tion influences individuals’ exploration-exploitation trade-offand show that this adaptation can undermine their performance.Using a minimal experimental paradigm, we find that partici-pants tend to copy the solution of other individuals too rapidly,thus decreasing the likelihood of discovering a better solution.Approximating this behavior with a simple model suggests, thatindividuals’ willingness to explore only depends on the value ofknown existing solutions. Our results allow for a better under-standing of the interplay between social and individual factorsin individual decision-making.

n the current study we present an air traffic control (ATC)task in which we measured pupil dilation to automaticallydetermine high and low workload periods. We manipulatedworking memory (WM) requirements across three conditions:a no WM condition, a passive WM condition in whichinformation was accumulated, and an active WM condition inwhich information had to be added to and removed from WM.Results showed that no WM resulted in the least dilation, butthat passive WM and active WM did not differ. Next, we usedthe pupil data to train a range of classifiers to differentiatebetween high and low workload periods with the ultimategoal to create an online task-independent interruptionmanagement system (IMS). The best predicting features werethe median and a second-order polynomial fit, going back 12seconds from the to-be-predicted moment. Using thesefeatures, our classifier was able to predict workload at highaccuracy (77%). We conclude that pupil dilation can be usedto create a reliable IMS.

Reducing cognitive phenomena to neural activity is seen by many as lacking in scientific utility. The conceptual chasm between electrochemical activity and the act of making a choice is too broad to span in a single step. Instead, we adopt a multi-scale approach to cognitive neuroscience by constructing a conceptual ladder that incrementally climbs from neuronal spikes to cognitive processes with each step offering theoretic reductions. Here we propose a sequence of intermediate neurocomputational processes that are promising for understanding an array of cognitive phenomena. We illustrate this approach in the context of the dynamics of choice. These dynamics emerge from serial evaluation mediated by systems in frontal cortex and the basal ganglia. The effect is to promote neural oscillations that provide a substrate for communication through coherence. Both empirical and simulation studies are described to support this view of emergent behavior.

Several computational methods have been proposed toevaluate the relevance of an instantiated cause to anobserved consequence. The paper reports on an ex-periment to investigate the adequacy of some of thesemethods as descriptors of human judgments aboutcausal relevance.

Previous research shows that performance is better when a highpitch is responded with up or right responses and a low pitch isresponded with down or left responses, called the spatial-pitchassociation of response codes (SPARC) effect. Despite the in-tuitive coupling of perception-action, studies investigating theSPARC effect have, however, used feedback to manipulate thestimulus-response mapping. Feedback contradicts the purposeof intuitive stimulus-response mapping by enabling short-termlearning. This study primarily investigates the role of feedbackon SPARC effect. We believe that feedback can facilitate in-congruent mapping and can, therefore, reduce the cost betweenincongruent and congruent mapping resulting in a diminishedSPARC effect. Our results, however, show that feedback hasno influence on the SPARC effect indicating that long-termassociations can not be overcome by short-term learning dueto robust perception-action coupling. Further, unlike previousstudies, we observed a strong horizontal SPARC effect in non-musicians as well.

We combine motion captured data with linguistic notions in a game-like intelligent tutoring system, in order to helpelementary school students to better differentiate literal from metaphorical uses of motion verbs, based on embodied in-formation. In addition to the thematic goal, we intend to improve young students attention and spatiotemporal memory, bypresenting sensorimotor data experimentally collected in our motion capturing labs. Furthermore, we examine the accom-plishment of games goals and compare it to curriculums approach. Sixty nine elementary school students were randomlydivided in two experimental groups (game and traditional) and one control group. Two way analysis of variance suggeststhat the experimental groups showed progress in posttests, with game group showing remarkable progress especially inthe verbs/actions presented during the intervention. This finding was considered as a first indication of attentional andspatiotemporal memorys improvement, while the games assistance features cultivated students metacognitive perception.

Many words have multiple yet predictably related meanings. For example, in English and in other languages, the sameroot morphemes can be used flexibly, to label an action and the instrument used to perform the action (e.g., we hammerwith a hammer and mix with a mixer). Previous findings indicate that four- and five-year-olds have formed abstractgeneralizations about these patterns and use them to infer new word meanings, such that they expect a word that haslabeled an action to also label its instrument. But how do these generalizations develop? Across five experiments witha large sample of English-speaking children, we show that in the third year of life, children begin to generalize wordsbetween actions and instruments: e.g., they expect that if an action involving an instrument and patient has been calledpabbing, then a pab (or a pabber) will refer to the instrument. Additionally, we find that children of the same age alsospontaneously extend words between actions and instruments: e.g., if an action has been called pabbing, children indicatethat the instrument cannot be a neefoo, presumably because they think it should instead be called a pab or a pabber.Critically, we show that these results do not depend on whether the new word labels an event for which children knowa word (e.g., hammering) or instead labels a novel event involving a novel instrument. These findings suggest that byage three, children’s knowledge of lexical flexibility is generative and abstract, and may not be constructed through item-specific learning.

Children with autism have impairments in communication and social interactions. Past studies have shown that robotbasedinterventions are effective in improving their gestural use. The present study asked whether or not children with autismcould meet the level of gestural production found in age-matched children with typical development after intervention.Four- to six-year-old children with autism in the intervention group (N = 15) took four training sessions in which theyimitated the gestures demonstrated by a social robot in various narratives. Age-matched children with autism in the wait-list control group (N = 15) and children with typical development (N = 15) received the training after the completion of theresearch. Children with autism in the intervention condition produced gestures more accurately in the training and novelstories than those in the wait-list control group in the posttests. Even more promising, the level of gestural productionaccuracy in children with autism in the delayed posttest of novel stories was comparable to that in children with typicaldevelopment, suggesting that children with autism could catch up to the level of gestural production found in children withtypical development.

Psychopaths have long been associated with a unique abilityto manipulate others (Hare, 1999). According to the“bottleneck” hypothesis of psychopathy (Newman & Baskin-Sommers, 2012), psychopaths’ cognitive abilities are directlyrelated to goal-directed behavior. To shed more light onlanguage production in psychopathy, two languageproduction studies were completed contrasting content andfluency under different motivational and difficulty conditions.Individuals high in psychopathy (HP) were less fluent butmaintained a more complex lexicon than their lowpsychopathy (LP) counterparts when under high cognitiveload and low motivation. Yet when HP individuals were underlow cognitive load and high motivation, they were morefluent, but used a less complex lexicon. Furthermore, the HPgroup produced more emotional language in both conditions.The results suggest that HP individuals’ language productionis inherently related to motivation and they attempt to balancefluency and complexity when cognitive load is increased.

As demonstrated in previous studies, Alzheimer’s diseaseleads to a degradation of vocabulary and communication skills.Novels by writers who are known to have suffered from thisdisease were compared with respect to their lexical richnessand syntactic complexity. Those written after the break-outof the disease have shown to use a considerably smaller lex-icon and a reduced syntactic complexity of the sentences.This makes us assume that writings of individual authors canbe classified automatically into “pre-Alzheimer’s period” and“Alzheimer’s period”. But the writing style of an author ishighly individual. Can we still detect whether any given novelis written by an author who suffers from Alzheimer’s? To as-sess this, we use a corpus of novels by three well-known writ-ers who were diagnosed with Alzheimer’s: Iris Murdoch, TerryPratchett and Agatha Christie. Using a mostly stylistic set offeatures we are able to distinguish between novels written un-der the influence of the disease and novels written by healthywriters with more than 82% accuracy. The classification ofthe novels of a given author into “pre-Alzheimer’s period” and“Alzheimer’s period” is accomplished with more than 86% ac-curacy. We also prove that our feature set is versatile enoughto be able to distinguish between authors in general and bookswith high precision.

A single concept can manifest in many varied forms,depending on the context in which it is activated. That is,concepts appear to be flexible rather than static. Here weimplement a compositional model of conceptual knowledge inwhich basic-level concepts are represented as graphtheoretical networks, with the specific goal of quantifyingconceptual flexibility. We collect within-concept statisticsusing online participants, construct network models, andvalidate these models in a classification analysis. We thenextract network measures and find that network diversity andcore-periphery structure correspond to conceptual flexibilityand stability, respectively. These results suggest that acompositional network model can be used to extract formalmeasures that are interpretable and useful in the study ofconceptual knowledge.

Two studies investigated how people learn the strength of therelation between a cause and an effect in a time series settingin which both variables exhibit temporal trends. In priorresearch, we found that people control for temporal trends byfocusing on transitions, how variables change from oneobservation to the next in a trial-by-trial presentation (Soo &Rottman, 2018). In Experiment 1, we replicated this effect,and found further evidence that people rely on transitionswhen there are extremely strong temporal trends. InExperiment 2, we investigated how people infer causalrelations from time series data when presented as time seriesgraphs. Though people were often able to control for thetemporal trends, they had difficulty primarily when the causeand effect exhibited trends in opposite directions and therewas a positive causal relationship. These findings shed lighton when people can and can’t accurately learn causal relationsin time-series settings.

We report an individual with music-odor synaesthesia who experiences automatic and vivid odor sensations when she hears music. S’s odor associations were recorded on two days, and compared with those of two control participants. Overall, S produced longer descriptions, and her associations were of multiple odors at once, in comparison to controls who typically reported a single odor. Although odor associations were qualitatively different between S and controls, ratings of the consistency of their descriptions did not differ. This demonstrates that crossmodal associations between music and odor exist in non-synaesthetes too. We also found that S is better at discriminating between odors than control participants, and is more likely to experience emotion, memories and evaluations triggered by odors, demonstrating the broader impact of her synaesthesia.

Comprehending action language recruits the action system. To what extent do action simulations reflect the fine-grained parameters of real world action? We investigate whether action simulations are sensitive to speed of an action. In two experiments participants completed a motor task where they moved slowly or quickly, followed by a sentence sensibility task. We found an overall action effect for sentences describing hand actions: moving slowly increased accuracy (Exp.1) and reduced response time (Exp. 2). For sentences describing full-body actions, responses were more accurate when movement speed matched the speed implied in the sentence, in Experiment 1 only. This study demonstrates online action simulation and provides evidence that speed of action can be simulated during sentence comprehension.

The nature of the mental processes involved in metaphorcomprehension has been the focus of debate. Research related tothis debate has mainly examined the comprehension of simplenominal metaphors. Here we take an individual-differencesapproach to examine the comprehension of slightly morecomplex metaphors, some taken from literary sources, using twotypes of comprehension tests (selecting an overall interpretationor else selecting a completion). In a series of metaphor-comprehension experiments with college students, we measuredboth fluid intelligence (using the non-verbal Raven’s ProgressiveMatrices test) and crystalized verbal intelligence (using a newSemantic Similarities Test). Each measure had a dissociablepredictive relationship to metaphor comprehension, at least forthose of the more complex literary variety. The pattern ofindividual differences suggests that metaphor comprehensionbroadly depends on both crystalized and fluid intelligence, withthe latter less important for relatively simple metaphors.

Generative models of human identity and appearance havebroad applicability to behavioral science and technology, butthe exquisite sensitivity of human face perception means thattheir utility hinges on the alignment of the model’s representa-tion to human psychological representations and the photoreal-ism of the generated images. Meeting these requirements is anexacting task, and existing models of human identity and ap-pearance are often unworkably abstract, artificial, uncanny, orbiased. Here, we use a variational autoencoder with an autore-gressive decoder to learn a face space from a uniquely diversedataset of portraits that control much of the variation irrele-vant to human identity and appearance. Our method generatesphotorealistic portraits of fictive identities with a smooth, navi-gable latent space. We validate our model’s alignment with hu-man sensitivities by introducing a psychophysical Turing testfor images, which humans mostly fail. Lastly, we demonstratean initial application of our model to the problem of fast searchin mental space to obtain detailed “police sketches” in a smallnumber of trials.

Do all languages convey semantic knowledge in the same way?If language simply mirrors the structure of the world, theanswer should be a qualified “yes”. If, however, languagesimpose structure as much as reflecting it, then even ostensiblythe “same” word in different languages may mean quitedifferent things. We provide a first pass at a large-scalequantification of cross-linguistic semantic alignment ofapproximately 1000 meanings in 55 languages. We find thatthe translation equivalents in some domains (e.g., Time,Quantity, and Kinship) exhibit high alignment acrosslanguages while the structure of other domains (e.g., Politics,Food, Emotions, and Animals) exhibits substantial cross-linguistic variability. Our measure of semantic alignmentcorrelates with known phylogenetic distances betweenlanguages: more phylogenetically distant languages have lesssemantic alignment. We also find semantic alignment tocorrelate with cultural distances between societies speakingthe languages, suggesting a rich co-adaptation of language andculture even in domains of experience that appear mostconstrained by the natural world.

Conceptual frameworks such as religion or politics may play a pervasive role in people’s interpretation of experience, but the empirical evidence for such effects is limited. To the extent that conceptual frameworks are real, they should have a pervasive impact on how people talk about the world. Such an influence may be detected in people’s everyday language. In a series of studies, text from the social media platform Reddit was used to train machine learning classifiers to identify people’s association with a particular religion or mental disorder. Impressively, classifiers trained on text focusing on religion and mental disorders could be used to identify people’s association with a particular religion or mental disorder even when the text was not explicitly about these topics, such as when it was about buying a car or playing tennis. Not only could the classifiers predict people’s religion or mental illness in the present, they could also do so prospectively, indicating that people’s everyday language gives away information about the kinds of conceptual frameworks they may hold in the future. An analysis of the features learned by the classifier suggested that they learned features with high face validity for the underlying conceptual framework. Together, the results provide evidence for the existence of conceptual frameworks by virtue of the imprint they leave across a wide range of language contexts.

Children learn number words slowly, acquiring exactmeanings for their first words in sequence, with many monthsin between words. The long delays are surprising in light ofevidence that infants can discriminate, e.g., sets of 2 from 3.Here, we test the hypothesis that, rather than facing aperceptual problem, children have difficulty identifyingnumber as the dimension of meaning encoded by an adjectivelike “three.” We trained children on an unknown number wordin the context of a proper noun (a giraffe named “Mr. 3” withthree spots), and found that 1- and 2-knowers were later betterat identifying the giraffe from a lineup, relative to children whohad heard the same giraffe described with an adjective (“withthree spots”). These results support the hypothesis thatidentifying number as a dimension of meaning, rather thanvisual discriminability or salience, is a bottleneck on earlynumber word learning.

To help others, we need to infer one’s goal and intention and make an action which complements one’s action yet to meet the underlying goal. In this study, we consider the computational mechanism how a person can infer the other’s intention and goal from his or her action, which is not completed or fails to meet the goal. As a minimal motor control task toward a goal, we analyzed single-link pendulum control tasks and its variation. By analyzing two types of pendulum control tasks, we show that a sort of fractal dimension of movements is characteristic of the difference in the underlying motor controllers. Further, using the fractal dimension as a criterion of similarity between movements, we show that the simulated pendulum controller can make an action toward the goal, toward which other’s incomplete action was made, but was not observable in behavior due to its failure.

Across two experiments, we use ordinal ranking to examinethe processing and representations involved in the estimationof large-scale, real-world proportions. Specifically, in twoexperiments people estimated two kinds of important real-world proportions: the demographic makeup of theircommunities, and spending by the U.S. Federal government.Our goal was to assess the metric scaling properties thatcharacterize perceptions of these quantities. In particular,previous work in numerical proportions has positedlogarithmic or linear representations (Opfer & Siegler, 2007),or linear representations with task-dependent rescaling (Barth& Paladino, 2011; Cohen & Blanc-Goldhammer, 2011). Thecurrent context differs markedly from this prior work in thatthe values we are examining are not explicitly presented toparticipants, nor directly experienced, but must be estimatedon the basis of masses of complex experiences. Ordinalranking of the quantities, combined with a Thurstonianmodeling approach, allows a unique means for estimating theinternal scale properties of numerical structures. We find thatpeople largely rely on mixed representations that emphasizelog-odds transformations of these vaguely known, butsocially important values. While the budget data explored inExperiment 1 were unable to distinguish log and log-oddstransformed internal models, the demographic proportionsexplored in Experiment 2 favored log-odds models.

Recent research using attention-monitoring techniques and fMRI has revealed that a shared neurocognitive mechanismunderlies both social decision making concerning the welfare of others and purely economic decision making for oneself.Such commonalities have been demonstrated mainly in isolated contexts, and it remains to be seen whether they extendto settings involving interactions with fellow decision makers. Using a behavioral study of distributive choices for othersand gambling decisions for self, we investigated how self-censorship in social contexts may mitigate the cognitive com-monalities demonstrated in isolated contexts. Results showed that, in both tasks, individual participants took more timeto respond when they expected subsequent discussion with another participant about reaching a consensus. In addition,we found a cognitive pattern unique to distributive choices for others only: participants expecting social interaction madetheir distributive choices in a more cognitively consistent manner, aligning with a rationale that they thought would bedefensible in subsequent discussion. No such systematic pattern was observed in gambling choices for self. These resultsindicate that anticipation of subsequent social interaction triggers self-censoring processes for some (but not all) tasks,whereby participants pre-edit their individual decisions systematically to prepare for social interaction.

Sensitivity to statistical co-occurrence regularities is present from infancy. This sensitivity may contribute to learning in many domains, including category learning. However, prior research has not examined whether everyday input conveys category-relevant statistical regularities. This study assessed whether statistical regularities relevant to real-world categories are present in a commonly experienced source input – children’s picture books. We focused on animal categories because this is a domain in which children receive much exposure from an early age, while simultaneously holding persistent misconceptions about category membership beyond preschool years. Analysis of 80 books revealed that they: 1) Were likely to contain regularities from which individual species categories (e.g., “chicken”) might be learned, but 2) Were unlikely to contain regularities from which broader taxonomic categories (e.g., “bird”) might be learned. These findings point to a paucity of taxonomically-relevant statistical regularities that may contribute to persistent taxonomic misconceptions.

Prior evidence suggests that category learning can occur implicitly by detecting regular co-occurrences of features within categories. Less studied is whether regularities wherein category membership predicts other events or actions also foster category learning. Moreover, we know little about whether, and to what degree, exposure to these regularities facilitates subsequent supervised learning. Here, participants were pre-exposed to exemplars from two categories during a cover task, while uninformed of their category membership. Pre-exposure occurred under conditions in which category membership did (Predictive Mapping) or did not (Mere Exposure) predict task events to which participants responded. Baseline participants completed the same task with category- irrelevant stimuli. Subsequently, all participants were taught the categories (using pre-exposure exemplars) under explicit supervision. Whereas neither Predictive Mapping nor Mere Exposure influenced cover task performance (vs. Baseline), Predictive Mapping substantially improved subsequent supervised category learning. These findings point to latent category learning given pre-exposure to Predictive Mapping regularities.

Knowledge about categories supports learning andgeneralization, and this knowledge is particularly importantearly in development. Although most theories of categoryknowledge posit a role for experience in acquiring thisknowledge, the current evidence for the presumed role ofexperience in category knowledge acquisition remains limitedto correlational evidence, indirect measures of categoryknowledge, and computational studies. Here we providedirect evidence that repeated experience with a biologicaldomain in an ecologically-valid setting changed children’scategory representations, with increased differentiation ofitems within that domain and relative to a second domain. Theimplications of these results for understanding the role ofexperience in category acquisition, and the contribution ofenrichment experiences to school readiness are discussed.

The disambiguation of syntactically ambiguous sentences canlead to reading difficulty, often referred to as a garden path ef-fect. The surprisal hypothesis suggests that this difficulty canbe accounted for using word predictability. We tested this hy-pothesis using predictability estimates derived from two fam-ilies of language models: grammar-based models, which ex-plicitly encode the syntax of the language; and recurrent neuralnetwork (RNN) models, which do not. Both classes of mod-els correctly predicted increased difficulty in ambiguous sen-tences compared to controls, suggesting that the syntactic rep-resentations induced by RNNs are sufficient for this purpose.At the same time, surprisal estimates derived from all mod-els systematically underestimated the magnitude of the effect,and failed to predict the difference between easier (NP/S) andharder (NP/Z) ambiguities. This suggests that it may not bepossible to reduce garden path effects to predictability

Explanations often highlight inductively rich relationships thatsupport further generalizations: learning that the knife is sharpbecause it is for cutting, we correspondingly infer that other thingsfor cutting might also be sharp. When do children appreciate thatexplanations are good guides to generalization? We report a study inwhich 108 4- to 7-year-old children evaluated mechanistic,functional, and categorical explanations for the properties of objects,and subsequently generalized those properties to novel objects onthe basis of shared mechanisms, functions, or category membership.Older children, but not younger children, were significantly morelikely to generalize when the explanation they had received matchedthe subsequent basis for generalization (e.g., generalizing on thebasis of a shared mechanism after hearing a mechanisticexplanation). These findings shed light on how explanation andgeneralization are coordinated over development, as well as the roleof explanations in young children’s learning

Behavioral studies often warrant the inclusion of computa-tional participants in addition to humans. However, connect-ing computational cognitive or AI frameworks to GUI-basedsoftware developed for human use is extremely difficult. Thisresults in researchers either (1) diving into software code toappend an API for computational participants, (2) developingtwo separate versions of task code – one for human and one forcomputational participants, (3) cherry-picking research tasksthat already include both a GUI and an API, or (4) finding away to publish the research “as is” without the potentially use-ful results from running simulated participants on task. Theseemingly minor nuisance of the API-GUI dichotomy in to-day’s world of software development is, in fact, responsiblefor reduction in scientific progress. This work proposes afunctional-essence approach to software development, and theuse of STAP (Simple Task-Actor Protocol) as a standard UIinteraction language, for overcoming the API-GUI dichotomyand enabling access to the same software for both human andcomputational participants. We envision the adaptation of theproposed methodology to enable selection of off-the-shelf be-havioral tasks, decorative templates, and cognitive/AI frame-works for a more efficient path to research results.

Social groups highlight latent structure in the social worldand support inductive inferences about individuals. In thepresent work, we examined children and adults’ intuitionsabout shared preferences within social groups. In Exp.1, 3- to5-year-old children treated preferences as a consistent propertyof social groups; that is, children expected members of a so-cial group to like the same toys that other members have liked.However, they did not treat preferences as diagnostic of socialgroups; they did not expect individuals to belong to a groupthat shares their preferences. By contrast, in Exp.2, adultsreadily treated preferences as both a consistent and diagnos-tic property of social groups. These results suggest that chil-dren’s inferences about social groups are asymmetric: Chil-dren readily infer preferences based on group membership, butnot group membership based on preferences.

Humans learn from other knowledgeable informants whochoose data to foster learning. Mathematical models of teach-ing and learning have formalized this process of learning fromhelpful others. While these approaches have been successful incapturing teaching and learning in a variety of contexts, theyhave been limited to relatively simple domains. One of theopen questions regarding Bayesian teaching is whether it canscale to teach from naturalistic domains with more interestingdatasets. In this work, we show how to apply Bayesian teach-ing to teach human participants categories learned by a super-vised machine learning model. The effectiveness of teaching ismeasured by how well the participants can predict the behaviorof the target machine learning model. Our results demonstratethat Bayesian teaching can be applied to naturalistic domains,show that the best sets of examples according to the modelyield better learning, and suggest avenues for improving ourability to automate teaching of image categories.

This paper explores two experiment designs that seek to determine the extent to which, if at all, observation can be free from theory. The two designs are compared and found to be similar in certain ways. One particular feature critical to both is that they seek to create conditions that compel test subjects with diverse theoretical backgrounds to resort to bare observational skills. If judgments made on the basis of these skills converge, such convergence would provide support for the view that theory-neutral observations can be had.

Reports of the relationship between aspects of cognitivecontrol and bilingual language experience in youngadults have been inconsistent. This study comparedperformance between monolingual and bilingual youngadults on working memory (WM) updating as ameasure of cognitive control and examined howdifferences in bilingual language experience manifestin updating performance. A letter N-back task with setsize and lure manipulations was used to measureupdating processes in the presence of increasedmemory load and interference. We expected to see aneffect of the bilingual experience on WM updating, aswell as within task variations related to the use ofdifferent updating mechanisms. While the monolingualand bilingual groups did not perform significantlydifferently, high non-English reading proficiencysignificantly predicted accuracy and reaction timewithin the bilingual group, particularly in high load,interference conditions. Results showed that youngadults categorized as bilingual in a broadly definedgroup may be difficult to uniformly compare to amonolingual group as they show large variations inperformance depending on their individual languageexperience.

Risk and delay co-occur. Intertemporal choices are rarely certain; risky choices are rarely atemporal. Behavioral evidence suggests that risk and time are entangled: time discounting is different for risky outcomes than for riskless outcomes. A prominent model of risky intertemporal choice (Baucells & Heukamp, 2012) combines risk and delay into psychological distance. It predicts that risk and time will be entangled for outcome risk (risk with one zero outcome and at least one positive outcome) but not for amount risk (risk with three or more positive outcomes) unless assuming non- cumulative probability weights. We show that BH does not quantitatively fit risky intertemporal choices better than a model assuming risk and time are independent. Many participants were best fit by a random response model. The functional form for risky intertemporal choices is difficult to detect. While risk and time are entangled, they do not seem to be evaluated as psychological distance.

We investigated imagery by making participants (n=530)draw stick-figure drawings of sentences containing atransitive action ("She kisses him"). Previous findings showthat prominent features of meaning and sentence structure areplaced to the left in drawings, according to reading direction.We replicated three findings: the first mentioned element isplaced on the left more often, the agent is placed on the left,and the grammatical subject is placed on the left. We furthertested hypotheses related to deixis and gender. By addingadverbs (here and there), that work as demonstratives inDanish, we tested whether deictic proximity is translated intoa leftward bias. This hypothesis was not supported. Analysesof gender tested the presence of a gender identification and agender stereotype bias, where either own or male gender isgiven prominence and thus placed on the left. We wereunable to support for either of the gender hypotheses.

Previous research has shown that secondary tasks sometimesinterfere with the perception of time. In this study, we look atthe impact of background music, and the familiarity of musicon the reproduction of a time interval. We hypothesize thatboth music listening and attending to time require declarativememory access, and that conflicts between the two can explainwhy the reproduced intervals are longer when participants lis-ten to music. A cognitive model based on the PRIMs architec-ture, but built from two existing models can explain the data,including the effect of music familiarity. The model is a com-bination of two existing models: one of time perception, whichrequires occasional memory access to check whether the inter-val is already over, and one of music perception, which triesto predict the next musical phrase based on the one currentlyperceived. The memory conflict between the two models re-produces the effects found in the data.

Children develop a sense of joint commitment and sharedintentionality during collaborative activities, which mayproduce prosocial effects in social coordinative activities.Past studies have found mixed results on the prosocial effectof shared intentionality. We hypothesized that it is the degreeof coordination and not simply shared intentionality thatfacilitates social bonding. In a block-assembly task with 4-year-old children, we manipulated degree of coordination.Children in the continuous high-level coordination conditionwere more generous in a Dictator Game and more willing tohelp their partner complete a task, compared with childrenwho engaged in a task with the same end-product thatrequired less coordination. Surprisingly, we also found thatboys shared more resources than girls, a result that weattributed to the emphasis on the importance of generosity formales in Chinese culture.

Do people have dispositions towards visual or verbal think-ing styles, i.e., a tendency towards one default representationalmodality versus the other? The problem in trying to answerthis question is that visual/verbal thinking styles are challeng-ing to measure. Subjective, introspective measures are themost common but often show poor reliability and validity; neu-roimaging studies can provide objective evidence but are in-trusive and resource-intensive. In previous work, we observedthat in order for a purely behavioral testing method to be ableto objectively evaluate a person’s visual/verbal thinking style,1) the task must be solvable equally well using either visualor verbal mental representations, and 2) it must offer a sec-ondary behavioral marker, in addition to primary performancemeasures, that indicates which modality is being used. Wecollected four such tasks from the psychology literature andconducted a small pilot study with adult participants to see theextent to which visual/verbal thinking styles can be differenti-ated using an individual’s results on these tasks.

According to mental model theory, spatial reasoning is basedon the construction and variation of mental modelsrepresenting spatial arrangements. Several effects in humanspatial reasoning are known to support this theory, forexample the ordering effect. Yet, reasoning effects have beenobserved for which the cognitive mechanisms are not entirelyexplained. To investigate how these effects can be attributedto neural computation, we modeled spatial reasoning in theNeural Engineering Framework.We selected three experiments to simulate tasks in a cognitivemodel based on an internal display. In our model,performance declines with an increase of objects which isexplained by the neural drift over time. We replicated effectsfrom the studies which we have found to be due to continuouspremise integration. By modeling and simulating spatialreasoning tasks, we showed that effects reported inpsychological studies can be explained by the emergentproperties of neural computation.

One of the principle tenets of modern behavioral ethics is thathuman morality is dynamic and malleable. Recent work intechnology ethics has highlighted the role technologies canplay in this process. As such, it is the responsibility oftechnology designers to actively identify and address possi-ble negative consequences of such technological mediation. Inthis work, we examine dialogue systems employed by currentrobotic agents, arguing that they can have deleterious effectson both the human moral ecosystem and human perception ofthe robots, regardless of the robots’ actual ethical competence.We present a preliminary Bayesian analysis of empirical datasuggesting that the architectural status quo of clarification re-quest generation systems may (1) cause robots to unintention-ally miscommunicate their ethical intentions (our two tests forthis yielded Bayes factors of 1319 and 1099) and (2) weakenhumans’ contextual application of moral norms (Bayes fac-tor of 1069).

The brain’s working memory system relies heavily on themesolimbic dopamine system and the delivery of reward sig-nals. The interaction between the prefrontal cortex (PFC) andthe basal ganglia are the main components simulated in work-ing memory models. The Working Memory Toolkit (WMtk) isa framework that allows the incorporation of working memoryinto robotic/artificial systems. The HWMtk is built on top ofWMtk by using holographic reduced representations for con-cept encoding. This allows end users to adopt the frameworkwithout the need to understand details of the algorithms in-volved. While the HWMtk captures human and animal per-formance on some cognitive tasks, tasks with multiple con-text layers are still problematic. We extended the HWMtkframework by adding a multilayer context reasoning work-ing memory system. We tested our system on the AX-CPTtask, 1-2-AX-CPT task and a 2-layer context task that is par-tially observable. Our results show that our model is capableof learning after a reasonable number of trials, thus making itamenable for comparison with human and animal performancedata.

This paper proposes a recurrent neural network model thatlearns to perform morphological affixation, a fundamental op-eration of linguistic cognition, and has interpretable relationsto descriptions of morphology at the computational and algo-rithmic levels. The model represents morphological sequences(stems and affixes) with distributed representations that sup-port binding of symbols to ordinal positions and position-basedunbinding. Construction of an affixed form is controlled at theimplementation level by shifting attention between morphemesand across positions within each morpheme. The model suc-cessfully learns patterns of prefixation, suffixation, and infixa-tion, unifying these at all levels of description around the theo-retical notion of a pivot. Connections of the present proposal toneural coding of ordinal position, and to computational modelsof serial recall, are noted.

Recent studies suggest that multisensory linkages betweenspeech and vision are implicated in the development ofdyslexia. Current data only address a relationship in adultswith existing diagnoses, but do not inform us about thedevelopmental trajectory of the association. We conducted apre-registered study of multisensory matching in 388 pre-readers in Singapore (Age: 5y 10m) using an adaptation of thebouba-kiki task (the Alien Zoo), and compared children’sperformance on this task to their earlier scores on measuresknown to predict dyslexia: phonological awareness,vocabulary size and letter knowledge. As reported elsewhere,children’s Alien Zoo scores were lower than adults’(Woon &Styles, 2017a). The language measures were strongly inter-correlated, suggesting persistent language skills acrossmultiple domains, However we found no significantrelationship between performance on the Alien Zoo task andany of the predictors of dyslexia. This may mean that therelationship is yet to emerge in this population. The childrenin this cohort will be tracked and tested at a later time point toestablish the developmental trajectory of this relationship.

Metacognitive monitoring plays an important role in self- regulated learning. Accurate metacognitive monitoring facilitates effective control, which affects learning outcomes. Most studies that explore metacognitive monitoring have investigated learners’ monitoring abilities when learners are explicitly cued to monitor. However, in real-world educational settings, learners are more commonly cued to control their learning. The primary goal of the current study was to investigate whether learners monitor their learning processes using retrieval when explicitly cued to control. Two experiments were conducted in pursuit of this goal. In the experiments, participants were instructed to learn Swahili-English word-pairs. Their learning performance was tested in subsequent cued-recall tests. Results suggest retrieval is likely practiced when learners are explicitly cued to control, but at a lower frequency or a more shallow level than when learners are explicitly cued to retrieve. In addition, the current study reported attempts to measure retrieval-based metacognitive monitoring using objective and online methods.

Emotional expressions are typically transient; while we mayreact emotionally to a new event, we are unlikely to respondwith the same emotion once the event becomes familiar. Herewe look at whether toddlers understand the relationshipbetween people’s epistemic states and their emotionalresponses. Younger (12-17-month) and older (18-24-month)toddlers were familiarized with a movie in which an observerwas knowledgeable or ignorant about a recurring event. On thetest trial, the observer saw the event and either remained neutralor changed to a valenced emotional reaction (positive ornegative). We predicted that the change from a neutral to avalenced expression would be more surprising if the event wasfamiliar to the observer than if the event was novel. We foundan interaction between epistemic state and emotion for olderbut not younger toddlers. These results suggest that before agetwo, children begin to understand the transient nature ofemotional reactions and their dependence on people’sepistemic states.

This paper aims to answer the fundamental but still unan-swered question: how can brains represent 3D objects? Ratherthan building a model of visual processing, we focus on mod-eling the haptic sensorimotor processes through which objectsare explored by touch. This idea is inspired from two mainfacts: 1) in developmental terms, tactile exploration is the pri-mary means by which infants learn to represent object shapes;2) blind people can also represent and distinguish objects justby haptic exploration. Therefore, in this paper, we firstly es-tablish the relationship between the geometric properties of anobject and constrained navigation action sequences for tactileexploration. Then, a neural network model is proposed to rep-resent 3D objects from these experiences, using a mechanismthat is computationally similar to that used by hippocampalplace cells. Simulation results based on a 2 × 2 × 2 cube anda 3 × 2 × 1 cuboid show that the proposed model is effectivefor representing 3D objects via tactile exploration and compar-ative results suggest that the model is more efficient and accu-rate when learning a representation of the 3×2×1 cuboid withan asymmetrical geometrical structure than the 2 × 2 × 2 cubewith a symmetrical geometrical structure.

We examined the effect of instructor gesture and distractorpresence on students’ encoding of slope and intercept ingraphs of linear functions. In Experiment 1, participantswatched an instructor avatar introduce a linear graph whileeither pointing to the intercept, tracing the over-and-upincrease for slope, or not gesturing (i.e., gaze only). They thenreconstructed the graph on paper. Participants weresignificantly more successful at encoding slope after watchingthe slope gesture than after watching no gesture. InExperiment 2, participants watched the avatar either point tothe intercept or trace the slope, each either in the presence orabsence of a visual distractor. Participants were significantlymore successful at encoding slope after watching the tracinggesture than after watching the pointing gesture. Distractorpresence did not affect performance. Taken together, theseresults suggest that teachers’ gestures promote students’encoding of relevant information and could help explain whyteachers’ gestures often benefit students’ learning.

We investigated whether memories of different time scales(i.e., week, day, hour) are used independently (i.e., indepen-dence of scales). To overcome the limitations of previousstudies that have low ecological validity in selecting the teststimuli, we used experience sampling technology. Participantswore a smartphone around their neck for two weeks, whichwas equipped with an app.that automatically collected time,images, GPS, audio and accelerometry. After a one-week re-tention interval, participants were presented with an image thatwas captured during their data collection phase, and tested ontheir memory of when the event happened (i.e., week, day ofweek, and hour). We find that, in contrast to previous studies,memories of different time scales were not retrieved indepen-dently in everyday life. Additionally, we replicated previouslaboratory findings such as correlations between confidencerating and memory performance, and patterns found betweenvalence rating and memory accuracy.

Social gaze-following consists of both reflexive and volitionalcontrol mechanisms of saccades, similar to those evaluated inthe antisaccade task. This similarity makes gaze-following anideal medium for studying attention in a social context. Thepresent study seeks to utilize reflexive gaze-following to de-velop a social paradigm for measuring attention control. Weevaluate two gaze-following variations of the antisaccade task.In version 1, participants are cued with still images of a socialpartner looking either left or right. In version 2, participantsare cued with videos of a social partner shifting their gaze tothe left or right. As with the traditional antisaccade task, par-ticipants were required to look in the opposite direction of thetarget stimuli (i.e., gaze cues). Performance on the new gaze-following antisaccade tasks are compared to the traditional an-tisaccade task and the highly related ability of working mem-ory.

People do not think in isolation. Whether purchasing a new product on Amazon, deciding what movie to watch, or evaluating scientific evidence, we often rely on aggregated sources of information (e.g., product ratings or reviews) to make decisions. Indeed, the internet has given rise to unprecedented levels of aggregated information, to the extent that it is difficult to imagine anything for which we might not be able to find summary information. In other words, what we know (or think we know) is constrained not just by our own knowledge, but by the knowledge of our community (Sloman & Rabb, 2016). Yet this raises a question: what happens when a community of knowledge is not in agreement? Here, we assess this question by pitting cases of high confidence against cases of high consensus. Results from two experiments show that 1) individuals are sensitive to both confidence and consensus; 2) individuals utilize such information in a predictable but context-dependent manner; and 3) perceptions of confidence and consensus influence judgments and decisions in a substantial way, even when individuals are not aware of the contrast between them. Taken together, the findings suggest that individuals are highly sensitive to variability in aggregated information – rather than merely an average – and that these ‘summary statistics’ of aggregated information have a substantial, reliable impact on decision-making.

Despite its omnipresence in this information-laden society, statistics is hard. The present study explored the applicability of a grounded cognition approach to learning basic statistical concepts. Participants in 2 experiments interacted with perceptually rich computer simulations designed to foster understanding of the relations between fundamental statistical concepts and to promote the ability to reason with statistics. During training, participants were asked to estimate the probability of two samples coming from the same population, with sample size, variability, and difference between means independently manipulated. The amount of learning during training was measured by the difference between participants’ confidence judgments and those of an Ideal Observer. The amount of transfer was assessed by the increase in accuracy from a pretest to a posttest. Learning and transfer were observed when tailored guidance was given along with the perceptually salient properties. Implications of our quantitative measures of human sensitivity to statistical concepts were discussed.

We investigated whether, and in what, ways people use visual structures to evaluate mathematical expressions. We also explored the relationship between strategy use and other common measures in mathematics education. Participants organized long sum/products when visual structure was available in algebraic expressions. Two experiments showed a similar pattern: One group of participants primarily calculated from left to right, or combined identical numbers together. A second group calculated adjacent pairs. A third group tended to group terms which either produced easy sums (e.g., 6+4), or participated in a global structure. These different strategies were associated with different levels of success on the task, and, in Experiment 2, with differential math anxiety and mathematical skill. Specifically, problem solvers with lower math anxiety and higher math ability tend to group by chunks and easy calculation. These results identify an important role for the perception of coherent structure and pattern identification in mathematical reasoning.

The Rational Speech Act (RSA) model, which proposesthat probabilistic speakers and listeners recursively reasonabout each other’s mental states to communicate, hasbeen successful in explaining many pragmatic reasoningphenomena. However, several theoretical questions remainunanswered. First, will such a pragmatic speaker–listenerpair always outperform their literal counterparts who donot reason about each others mental states? Second, howdoes communication effectiveness change with the number ofrecursions? Third, when exact inference cannot be performed,how does limiting the computational resources of the speakerand listener affect these results? We systematically analyzedthe RSA model and found that in Monte Carlo simulationspragmatic listeners and speakers always outperform theirliteral counterparts and the expected accuracy increases asthe number of recursions increases. Furthermore, limitingthe computation resources of the speaker and listener so theysample only the top k most likely options leads to higherexpected accuracy. We verified these results on a previouslycollected natural language dataset in color reference games.The current work supplements the existing RSA literature andcould guide future modeling work.

Human conversation is marked by alternation–partners takingturns speaking and listening. Consequently, language produc-tion happens under time pressure; speakers who cannot gettheir message out quickly enough lose their turn. When adultshave struggle to retrieve the words they want to say, they canchoose alternatives. But children just beginning to learn lan-guage may solve this problem with gesture. If young children’sproduction systems reflect a sensitivity to communicative pres-sure, they should use deictic gesture to refer when they cannotretrieve a lexical label quickly enough. We confirm this pre-diction in a longitudinal corpus of naturalistic parent-child in-teractions, showing that the frequency and recency of a wordin children’s input predict the probability that they will refer toits referent with gesture, even for words they know.

Non-adjacent dependencies are ubiquitous in language, butdifficult to learn. Previous research has shown that the presenceof high variability between dependent items facilitateslearning. Yet what allows learning of non-adjacentdependencies even without high variability in interveningelements? One possibility is that learning non-adjacentdependencies highlights similar structures, allowing people tolearn new non-adjacent dependencies that are otherwisedifficult. In two studies, we show how being exposed tolearnable non-adjacent dependencies can change learners’sensitivity to novel non-adjacent regularities that are moredifficult to detect. These findings demonstrate a new way inwhich learning can build on and shape later learning aboutcomplex linguistic structure.

Many everyday decisions are based not only on memories ofdirect experiences, but on memories that are integrated acrossmultiple distinct experiences. Sometimes memory integrationbetween existing memories and newly learnt informationoccurs rapidly, without requiring inference during thedecision. It is known that prior knowledge (i.e. schema)affects the initial acquisition, and consolidation, of memories.In this study, we explore the effect of schema on theintegration of acquired memories between paired associates(e.g. integrating A-B and B-C into A-B-C) that were schemaconsistent or inconsistent, as confirmed with a latent semanticanalysis of text corpora. We find that enabling fast learning,by using material that is consistent with a schema, allows forfast memory integration. These behavioral results areconsistent with predictions generated from neuroscientifichypotheses suggesting that an existing schema might enableneocortical learning that is distinct from a more explicithippocampus-mediated integration of new information.

The present study seeks to substantiate a cognitively-groundedmodel of synchronic meaning variation and diachronic mean-ing change. We propose that inter-comprehender vari-ability in CONTEXT-SENSITIVITY drives variation in word-meanings along conceptual structure pathways; we test thismodel through English have and its underlying LOCATION-POSSESSION conceptual structure. Through acceptability rat-ings, self-paced reading times, and ERPs, we show thatrelevant context can facilitate the dispreferred but plausibleLOC interpretation of a have-sentence–the degree of facil-itation is predicted by individual differences in CONTEXT-SENSITIVITY, indexed here by gender and Autism Quotient.Altogether, our results suggest that the variation of have-sentences’ meanings is principled due to its unified concep-tual structure, and that conceptual structure together with con-text cooperate in guiding comprehension by modulating thesalience of competing variants in real-time. Ultimately, di-achronic change is naturally emergent from this model of nor-mal language processing.

Deductive Mastermind is a deductive reasoning game that isimplemented in the online educational game system Math Gar-den. A good understanding of the difficulty of Deductive Mas-termind game instances is essential for optimizing the learningexperience of players. The available empirical difficulty rat-ings, based on speed and accuracy, provide robust estimationsbut do not explain why certain game instances are easy or hard.In previous work a logic-based model was proposed that suc-cessfully predicted these difficulty ratings. We add to this workby providing a model based on a different logical principle—that of eliminating hypotheses (dynamic epistemic logic) in-stead of reasoning by cases (analytical tableaux system)—thatcan predict the empirical difficulty ratings equally well. Weshow that the informational content of the different feedbacksgiven in game instances is a core predictor for cognitive dif-ficulty ratings and that this is irrespective of the specific logicused to formalize the game.

Drift diffusion models of decision-making offer some of the most robust predictions of response time for a number of memory manipulations. Some drift diffusion models have been extended to explain confidence judgments. Many of these models assume that confidence judgments are independent and are not systematically related to other task items. In this paper the authors report a relationship between confidence judgments in procedural tasks and how the Memory for Goals model would explain this relationship.

250 million children worldwide lack basic literacy and numeracy skills, many of whom have no access to regular schooling.Inexpensive tablet computers have the potential to scale up the distribution of intelligent tutoring systems to children inneed. We introduce a collection of tablet games presenting core literacy and numeracy concepts in a way that enablesself-directed learning, reinforced by a shared content engine with an adaptive algorithm that re-prioritizes content basedon the accuracy and timing of the learner’s responses to effectively space and distribute practice. The difficulty of eachgame’s dynamics adjust to the learner over time. We analyze response data from school children in Tanzania, examininghow they distribute their attention across the games and as a function of performance within each game. We also evaluatedifferent methods for determining their knowledge state and learning progress based on their responses, and examine howself-direction influences stimulus spacing.

Metacognitive monitoring, like error detection, is crucial for appropriate self-regulating processes. Some researchersargue that metacognitive monitoring automatically occurs (Spehn & Reder, 2000). Whether the automatic monitoringprocesses exist or not and what tasks are needed to investigate the processes have been topics of considerable discussion.We attempted to observe these automatic metacognitive monitoring processes. Two calculus equations were verticallypresented on a computer screen for 50ms, followed by an auditory cue to indicate one of the two equations. Twenty-sevenuniversity students were asked to judge whether the cued equation was correct or incorrect. The result showed that RTwas longer when the distractor, non-cued equation, was incorrect than when it was correct, although the distractor couldn’thave been intentionally calculated. This finding suggests that errors in equations were rapidly and automatically detected.We discuss whether automatic metacognitive monitoring processes are observed in our task.

The Erikson flanker task is a well-established measure of selective attention for adults. In this task, participants judgethe direction a central target points with flanking distractors that are neutral (no direction), congruent (same direction astarget), or incongruent (opposite direction of target). This task has recently been modified for use with young children,but it is unclear whether developmental differences in childrens spatial skills and language limit its appropriateness. Thecurrent study tested preschool-aged children in both the classic directional version and new color version (i.e., blue and redtargets, with blue, red, or white flankers). Results showed significantly better performance on the color versus directionalversion, with trial types showing the same pattern in both tasks: worst performance on incongruent trials, comparableperformance on congruent and neutral. Ongoing work is comparing the same tasks in adults to see if this difference islimited to early childhood.

Learned categorical perception (CP) occurs when judgments of stimulus similarity or discriminability are altered as a resultof learning to categorize the stimuli; for example, there may be enhanced discrimination of items straddling a categoryboundary or of differences anywhere along a category-relevant dimension. Typical visual learned CP experiments donot test for both kinds of effects or employ control groups receiving exposure to the stimuli comparable to that receivedby category learning groups, rendering the results ambiguous in multiple ways. We will present results from a newexperimental paradigm that is designed to achieve the following important goals: (a) test for and clearly distinguishall known types of boundary and dimension-wide effects considered variants of learned CP and (b) determine whetherobserved changes in performance are actually due to categorization training by comparing them to the changes caused bycomparable stimulus exposure in the absence of category learning.

Many children struggle to comprehend basic space science, including the scientific explanations of the day/night cycle andseasonal change (e.g., Plummer, 2014; Vosniadou & Brewer, 1994). With notable exceptions (e.g., Samarapungavan, Vos-niadou, & Brewer, 1994), prior research has focused on Westerners’ ideas and experiences. Using structured interviews,we explored U.S. and Indonesian 3rd graders’ conceptions of the day/night cycle and seasonal change. Children fromboth communities had similar explanations of the day/night cycle, often confusing the Sun’s apparent movement as actualmotion. Cross-cultural differences emerged in children’s explanations of seasons: U.S. children were more likely to usechanges in Sun-Earth proximity, whereas Indonesian children tended to provide Earth-centric, geographical explanations(e.g., ”America gets snow because it is near the North Pole”). These findings reveal an interesting interplay betweenchildren’s geographically limited observations of the sky, the seasons, and their ideas about invisible causal forces in thesolar system.

We explore how attributes and relations contribute to generalization of a property across stimuli for ecologically validstimuli used often to communicate: emojis. We use the Bayesian Generalization Framework to model generalizationjudgments from given triplets of emojis to new triplets of emojis that share either a common relation, common attribute,both, or neither. Based on the model predictions, we conducted a behavioral experiment investigating the strength ofattributes and relations when generalizing across emojis. The model learned to use attributes or relations appropriately;however when given triplets that share both a common attribute and relation, it gave more weight to the common attributesthan human participants did. This suggests that people are strongly, but not completely, biased towards using relationswhen generalizing a novel property across triplets of emojis.

Bilinguals are purported to be more creative than monolinguals, but the mechanism for this bilingual advantage is stillunresolved, with several different accounts proposed. Others have challenged the existence of bilingual advantages ingeneral. We examine existence as well as hypothesized semantic network difference based mechanisms for the relationshipbetween bilingualism and creativity here by measuring creativity and fluency for monolinguals and bilinguals. The fluencymeasure allowed us to analyze the structure of individuals semantic networks (average shortest path length, clusteringcoefficient, and modularity). We found no differences in creativity between monolingual and bilingual participants, with aBayesian test showing substantial evidence for the null hypothesis. We did find that aspects of semantic network structurepredicted creativity. These findings suggest that, contrary to previous work, the bilingual advantage does not exist in therealm of creativity.

Previous studies show that auditory intervals are, in general, more accurately discriminated than visual or tactile intervals(Grondin, 2003). Also for discrimination tasks, when the markers of brief empty intervals are delivered from differentsensory modalities, sensitivity to time is much lower than it is when the markers are delivered from the same modality(Grondin & Rousseau, 1991). The purpose of this study was to evaluate the effect of intermodality on the temporaldiscrimination. Twelve participants (mean = 25.33, SD = 5.12) performed a bisection temporal task. During eight sessions,three conditions were manipulated: certainty about the origin of the second marker (certainty, uncertainty), standardduration (300ms, 900ms), and modality (auditory- auditory, auditory- tactile, auditory-visual). Results showed intramodalintervals are better discriminated than intermodal intervals. In both 300ms and 900ms, intervals were better discriminatedwhen the second modality was auditory than when it was tactile or visual.