Proceedings of the Annual Meeting of the Cognitive Science Society

Humans regularly perform tasks that require combining infor-mation across several sources of information to learn, reason,and make decisions. Bayesian models provide a computa-tional framework, and a normative account, for how humanscarry out these tasks. However, exact inference is intractablein most real-world situations, and extensive empirical workshows that human behavior often deviates significantly fromthe Bayesian optimum. A promising possibility is that peopleinstead approximate rational solutions using bounded avail-able resources. In this workshop, we bring together lead-ing researchers from cognitive science, neuroscience and ma-chine learning to build a better understanding of bounded op-timality in how humans learn, reason and make decisions.

Several of the central questions in language, social cognition,and developmental research focus on the roles of input, out-put, and interaction on learning and communication. While ithas become easy to collect long-form recordings, getting use-ful data out of them is a more daunting task. Across four mini-sessions, this tutorial aims to address pre- and post-data collec-tion concerns, and provide a hands-on introduction to manualand automated annotation techniques. Attendees will leave thistutorial with resources and concrete experience for collecting,annotating, and sharing/archiving naturalistic recordings, in-cluding specific open-science practices relevant for these data.

computational cognition; Bayesian activelearning; autonomous experimentation; adaptive designoptimization; Python software package

Navigation is a well-specified computational problem, and solving it is vital for survival. Given these constraints, we might expect that humans differ minimally in their wayfinding capabilities. Indeed, a lack of variation is often implicitly assumed when cognitive scientists debate the existence of cognitive maps or when cognitive neuroscientists search for the neural substrates of navigation. However, in everyday life, we frequently discuss how some people get lost with some frequency, or how women ask for directions while men use maps. Indeed, it is increasingly apparent in the scientific data on navigation (and other cognitive domains) that the study of normative functioning needs to be integrated with the study of human variation, with its attendant challenges regarding experimental design and use of psychometrics. The four papers in this symposium gather together current work in cognitive science and neuroscience that aim to integrate the study of variation into the more common normative approach.

Daily function depends on an ability to mentally map our environment. Environmental visibility and complexity can increase this challenge. Importantly, people vary dramatically in their ability to navigate flexibly and overcome such environmental challenges. In this paper, we will present experimental work targeting the mechanisms that underlie different navigational abilities, and how objective and introspective measures of ability interact to influence navigational strategy use. Using virtual reality, we manipulated environmental visibility and complexity. Participants then performed wayfinding, pointing, and route following tasks to probe cognitive map memory and navigational flexibility. Our findings reveal that individual differences in metacognition - such as perceived sense of direction - and in navigational strategy preference powerfully impact how environmental features affect spatial memory. We also gathered data on the neurocognitive foundations of these differences. Importantly, our methods highlight individualized interventions that can improve spatial learning and specify the mechanisms through which they operate.

Popular sources often assume the existence of a male advantage in navigation, but the scientific data are inconsistent. This meta-analysis evaluates the literature on behavioral sex differences in human navigation. We quantify the overall magnitude of sex differences in a variety of paradigms and populations and examine potential moderators in large-scale navigation skills, using 694 effect sizes from 266 studies and a multilevel linear modeling approach. Overall, we found that male participants outperform female participants, with a small to medium effect size (d= 0.34 to 0.38). The type of task, the type of dependent variable and the testing environment significantly contribute to variability in effect sizes. Pointing and recall tasks show larger sex differences than distance estimation tasks or learning to criterion; among the dependent variables, the deviation scores associated with pointing tasks show larger effect sizes. The largest estimate was d = .55 for tasks than required coordinating indoor and outdoor views. Interestingly, studies with children younger than 13 years showed very small effect sizes (d = .15) as compared to older age groups. We discuss the implications of these findings for the study of sex differences and identify avenues for future navigation research.

Research on spatial thinking needs reliable and valid measures of individual differences in skills. Visuospatial Perspective Taking (PT)—the ability to mentally maintain and transform spatial relationships between objects within an environment—is one kind of spatial skill that is especially relevant to navigation and building cognitive maps. However, the psychometric properties of various PT tasks have yet to be examined. The present study examines three main psychometric properties of PT tasks: 1) the reliability of two tasks developed for children but adapted in difficulty level for use in adult populations, 2) item difficulty and discriminability within and between four tasks using item response theory, and 3) relation of scores with general intelligence, working memory, and mental rotation. Results showed that two of the four PT tasks have promising psychometric properties for measuring a wide range of PT ability based on item difficulty, discriminability, and efficiency of a test information function.

Different memory systems, dependent on separate parts of the brain, can sustain successful navigation. The hippocampus is implicated in spatial memory strategies used when finding one’s way in the environment, i.e. it is allocentric and involves remembering the relationship between landmarks. On the other hand, another strategy dependent on the caudate nucleus can also be used, i.e. the response strategy, which relies on making a series of stimulus-response associations (e.g. right and left turns from given positions). Participants who use the response strategy are faster at learning navigation tasks lending themselves to using a single specified route. Young adult response learners have increased fMRI activity and grey matter in the caudate nucleus, but decreased fMRI activity and grey matter in the hippocampus. Research in my laboratory has shown that specific navigation strategies are associated with several genes, such as BDNF and ApoE, as well as hormones, such as cortisol and progesterone, but not estrogen and progesterone. Experiences dependent modulators such as age, habit, stress and rewards also modulate strategies dependent on the hippocampus and caudate nucleus. These results have important translational implications because a larger hippocampus has been associated with healthy cognition in normal aging and with a reduced risk of numerous neurological and psychiatric disorders such as Alzheimer’s disease, Schizophrenia, Post-Traumatic Stress disorder and Depression.

Explainability is assumed to be a key factor for theadoption of Artificial Intelligence systems in a wide rangeof contexts (Hoffman, Mueller, & Klein, 2017; Hoffman,Mueller, Klein, & Litman, 2018; Doran, Schulz, & Besold,2017; Lipton, 2018; Miller, 2017; Lombrozo, 2016).The use of AI components in self-driving cars, medicaldiagnosis, or insurance and financial services has shownthat when decisions are taken or suggested by automatedsystems it is essential for practical, social, and increasinglylegal reasons that an explanation can be provided tousers, developers or regulators.1Moreover, the reasons forequipping intelligent systems with explanation capabilitiesare not limited to user rights and acceptance. Explainabilityis also needed for designers and developers to enhancesystem robustness and enable diagnostics to prevent bias,unfairness and discrimination, as well as to increase trust byall users in why and how decisions are made. Against thatbackground, increased efforts are directed towards studyingand provisioning explainable intelligent systems, both inindustry and academia, sparked by initiatives like the DARPAExplainable Artificial Intelligence Program (DARPA, 2016).In parallel, scientific conferences and workshops dedicated toexplainability are now regularly organised, such as the ‘ACMConference on Fairness, Accountability, and Transparency(ACM FAT)’ (Friedler & Wilson, n.d.) or the ‘Workshop onExplainability in AI’ at the 2017 and 2018 editions of theInternational Joint Conference on Artificial Intelligence.However, one important question remains hithertounanswered: What are the criteria for a good explanation?

The past decade has seen remarkable progress in artificialintelligence, with such advances as self-driving cars, IBMWatson, AlphaGo, Google Translate, face recognition,speech recognition, virtual assistants, and recommendersystems. Ray Kurzweil and others think that it is only amatter of decades before AI surpasses human intelligence.This symposium will evaluate the extent to which AIcurrently approximates the full range of human intellectualabilities, and critically discuss the prospects for closing thegap between artificial and human intelligence. Participantswill combine the perspectives of computer science,psychology, and philosophy.

We bring together four researchers who study exper-tise in team or in solo (i.e., individual) performance. Teamresearch tends to either collect a lot of questionnaire dataafter performance or a little data, in real-time, by humanobservers. Studies of solo performers are often restrictedto convenience samples of task novices, who often spendless than an hour learning and performing the task. Incontrast, the research of all four of our panelists is no-table for using tasks which require days-to-years of prac-tice and for the quantities of data collected. Discussionswill emphasize the contributions these approaches aremaking to theoretical cognitive science.

Cognition is complex. This complexity is related tomultiple, distributed neurocognitive processes dynamicallyoperating across parallel scales, resulting in cognitiveprocessing. A major challenge in studying this complexity,relates to the abstractness of theoretical cognitive constructs,such as language, memory, or thinking in general. Suchabstractness is operationalized, indirectly, via behavioral,measures or in neural activity. In the past two decades, anincreasing number of studies have been applying networkscience methodologies across diverse scientific fields tostudy complex systems.Network science is based on mathematical graph theory,providing quantitative methods to investigate complexsystems as networks (Baronchelli, Ferrer-i-Cancho, Pastor-Satorras, Chater, & Christiansen, 2013; Siew, Wulff,Beckage, & Kenett, 2018). A network is comprised fromnodes, that represent the basic unit of the system (e.g.,concepts in semantic memory) and links, or edges, thatsignify the relations between them (e.g. semantic similarity).While the application of network science methodologies hasbecome an extremely popular approach to study brainstructure and function, it has been used to study cognitivephenomena to a much lesser extent. This, despite classiccognitive theory in language and memory being highlyrelated to a network perspective (Collins & Loftus, 1975;Siew et al., 2018). Already, network science in cognitivescience has enabled the direct examination of the theory thathigh creative individuals have a more flexible semanticmemory structure, identified mechanisms of languagedevelopment through preferential attachment, shed novellight on statistical learning, shown how specific semanticmemory network parameters influence memory retrieval,and provided new insight on the structure of semanticnetwork of second language in bilinguals (Siew et al.,2018).The aim of the current symposia is to demonstrate thepotential and strength of applying network sciencemethodologies to study cognition. This will be achieved bybringing together leading researchers that apply suchmethods to study various aspects of cognition, includinglanguage, learning, aging, and creativity. The presentationswill describe state-of-the-art progress and perspectives thatare achieved in applying these methods to study cognition.Importantly, these talks aim at stimulating discussion of thefruitfulness of such an approach and how such an approachcan powerfully and quantitatively study the complexity ofcognitive phenomena. Finally, this symposium aims todemonstrate how network science in cognitive science canbe used to quantitatively bridge across different levels ofanalysis, spanning the computational, behavioral, neural,and social.

We synoptically describe having achieved the unprecedentedlogicist cognitive computational simulation of quantified ver-sions of any n-level (FBTn, ∀n ∈ N) false-belief task, andhence of what we call the infinitary false-belief task (FBTω);the achievement is enabled by the automated reasoner Shad-owProver. Logicist cognitive computational simulation of thelevel-one (or, as it’s currently known, “first-order”) false-belieftask (FBT1) was achieved circa 2007 by Bringsjord et al. Butsubsequently cognitive science has seen the arrival such mod-eling and simulation successfully applied to the second-orderfalse-belief task (FBT2); see e.g. (Blackburn & Polyanskaya,forthcoming). (This is the level-two FBT in our hierarchy oftasks.) But now, courtesy of what we report, logicist cognitivecomputational simulation of any FBTn is accomplished for thefirst time, and hence the infinitary false-belief task (FBTω) isreached as well

DiPaola’s research endeavors to build top down Artificial Intelligence (AI) models of human creativity, empathy and expression for both use in new forms of computation systems as well as analysis of how the creative mind works. In doing so he has interviewed hundreds of artists, writers and musicians on how they perceive their creative talent and its originals. Combined with research from neuro-aesthetics and computer modelling, DiPaola notes that while many creative individuals report that they believe new insights as coming into them from an external source during creative flow, that evidence point to these new creative ideas and interpretations often more likely have internal roots from the individual’s, mid and long term past experiences and processes. DiPaola attempts to model this and other human creativity processes in computational form often as AI systems such as deep learning, reinforcement learning and evolution programming. Two efforts underway in DiPaola’s research lab are mapping out the creative process of a fine art portrait painter using 5 hierarchical AI systems, as well as modelling an empathetic embodied character agent who can understand emotions from those she talks with and construct creative narrative or quote like responses.

People spontaneously collaborate to solve a common goal.What factors affect whether teams are successful? Due tolack of large-scale naturalistic data and methods for investi-gating scientific questions on such data, previous work has ei-ther focused on very concrete cases, such as surveys of busi-ness teams, or abstract cases, such as GridWorld games, whereagents coordinate their movement so that each agent can get totheir own goal without obstructing other agents. We propose acomputational framework based on the multivariate Hawkesprocess and a novel algorithm for parameter estimation onlarge data sets. We demonstrate the potential of this methodby applying it to a large database of programming teams, pub-lic GitHub repositories. We analyze factors known to influenceteam performance, such as leader organization style and teamcognitive diversity, as well as other factors, such as the bursti-ness of effort, that are difficult to test using existing methods.Keywords: Collaborative cognition; Hawkes process; Organi-zational psychology; Bayesian nonparametrics

In a simple transcription task in which sections of Java programcode are copied by freehand writing, it is demonstrated thatchunk related temporal signals are sufficiently robust to permitthe measurement of programming competence. An experimentwith 24 participants revealed that the number of views of thestimulus per trial and the duration of writing per stimulus vieware both strongly correlated with independent measures of Javacompetence.

Proficient readers typically fixate near the center of a word,with a slight bias towards word onset. We explore a novelaccount of this phenomenon based on combining information-theory with perceptual constraints in a connectionist model ofvisual word recognition. This account posits that the amountof information-content available for word identification variesacross fixation locations and across languages. These differ-ences contribute to the overall fixation location bias in differ-ent languages, make the novel prediction that certain wordsare more readily identified when fixating at an atypical fixa-tion location, and predict specific cross-linguistic differences.We tested these predictions across several simulations in En-glish and Hebrew, and in a behavioral experiment. The resultsconfirmed that the bias to fixate closer to word onset alignswith reducing uncertainty in the visual signal, that some wordsare more readily identified at atypical fixation locations, andthat these effects vary across languages.

We introduce a new problem solving paradigm: solving physical puzzles by placing tool-like objects in a scene. Thepuzzles are designed to explicitly evoke different physical concepts such as support, blocking, tipping, and launching, andare typically solved in a handful of trials. We study human participants’ problem solving strategies, including what theytry first, how they update their actions based on failed attempts, and how many attempts they eventually take to solvethe puzzles. We introduce the ‘Sample, Simulate, Remember’ model that incorporates object-based priors to generatehypotheses, mental simulation to test hypotheses, and a memory and generalization system to update across simulationsand real-world trials, and show that all three components are needed to explain human performance. Further results canbe found at https://k-r-allen.github.io/tool-games/

The division of labor phenomenon has been observed with re-spect to both manual and cognitive labor, but there is no clearunderstanding of the intra- and inter-individual mechanismsthat allow for its emergence, especially when there are multipledivisions possible and communication is limited. Situationsfitting this description include individuals in a group splittinga geographical region for resource harvesting without explicitnegotiation, or a couple tacitly negotiating the hour of the dayfor each to shower so that there is sufficient hot water. We stud-ied this phenomenon by means of an iterative two-person gamewhere multiple divisions are possible, but no explicit commu-nication is allowed. Our results suggest that there are a lim-ited number of biases toward divisions of labor, which serveas attractors in the dynamics of dyadic coordination. How-ever, unlike Schelling’s focal points, these biases do not attractplayers’ attention at the onset of the interaction, but are onlyrevealed and consolidated by the in-game dynamics of dyadicinteraction.

Others’ beliefs about the self can powerfully influence our everyday interactions with others. Recent work suggests that even preschool-aged children are sensitive to what others think of the self and actively attempt to manage these beliefs (Asaba & Gweon, 2018). What cognitive capacities underlie these early self-presentational behaviors, and in what contexts do these behaviors emerge? Here we show that preschoolers strategically demonstrate their competence to even a puppet, but only when an adult treats the puppet as an agent and specifically asks which toy the child wants to “show” to the puppet (Exp.1). However, they do not show such strategic demonstration of their competence when the same puppet is treated as an object (Exp.2). These results suggest that self- presentational behaviors can emerge even in the absence of any immediate prospect of social evaluation insofar as children consider the target entity as capable of holding beliefs. Furthermore, whether or not children ascribe a belief about the self to the target is heavily modulated by how an entity is treated by others. We discuss the relevance of these findings to early reputation management behaviors, and more broadly, the use of make-believe agents in developmental research.

At first glance, humans extract social judgments from faces, in-cluding how trustworthy, attractive, and aggressive they look.These impressions have profound social, economic, and polit-ical consequences, as they subconsciously influence decisionslike voting and criminal sentencing. Therefore, understand-ing human perception of these judgments is important for thesocial sciences. In this work, we present a modifying autoen-coder (ModifAE, pronounced “modify”) that can model andalter these facial impressions. We assemble a face impressiondataset large enough for training a generative model by ap-plying a state-of-the-art (SOTA) impression predictor to facesfrom CelebA. Then, we apply ModifAE to learn generalizablemodifications of these continuous-valued traits in faces (e.g.,make a face look slightly more intelligent or much less aggres-sive). ModifAE can modify face images to create controlledsocial science experimental datasets, and it can reveal datasetbiases by creating direct visualizations of what makes a facesalient in social dimensions. The ModifAE architecture is alsosmaller and faster than SOTA image-to-image translation mod-els, while outperforming SOTA in quantitative evaluations.

The Simple Recurrent Network (SRN) has a long tradition incognitive models of language processing. More recently, gatedrecurrent networks have been proposed that often outperformthe SRN on natural language processing tasks. Here, we in-vestigate whether two types of gated networks perform betteras cognitive models of sentence reading than SRNs, beyondtheir advantage as language models. This will reveal whetherthe filtering mechanism implemented in gated networks corre-sponds to an aspect of human sentence processing. We traina series of language models differing only in the cell types oftheir recurrent layers. We then compute word surprisal valuesfor stimuli used in self-paced reading, eye-tracking, and elec-troencephalography experiments, and quantify the surprisalvalues’ fit to experimental measures that indicate human sen-tence reading effort. While the gated networks provide betterlanguage models, they do not outperform their SRN counter-part as cognitive models when language model quality is equalacross network types. Our results suggest that the differentarchitectures are equally valid as models of human sentenceprocessing.

Using “A” in noun phrases such as “A father of the vic-tim” is odd, which is commonly explained by the princi-ple Maximize Presupposition, requiring speakers to usethe alternative with the strongest presupposition (here“The”, given its uniqueness presupposition). This re-sults in an anti-uniqueness inference for “A” (clashingwith stereotypical expectations here), sometimes labelledas an ‘anti-presupposition’ (Percus, 2006), as it derivesfrom reasoning over the presuppositions of alternativeforms. We compare these inferences to the uniquenessinferences associated with definites, while manipulatinguniqueness expectations in a picture manipulation taskusing visual world eye-tracking. This offers a minimalcomparison of uniqueness-based inferences that are lexi-cally encoded vs. pragmatically inferred, and furthermoretests the prediction that the accommodatability of the def-inite’s presupposition plays a role in the derivation of anti-uniqueness inferences (Rouillard & Schwarz, 2017).

Creative thinking has long been associated with spreading ofactivation through concepts within semantic networks. Herewe examine one potential influence on spreading activationknown as the fan effect: increasing concept knowledge leads toincreasing interference from related concepts. We testedwhether cue association size—an index of semantic richnessreflecting the average number of elements associated with aconcept—impacts the quantity and quality of responsesgenerated during the alternate uses task (AUT). Wehypothesized that low-association AUT cues should benefitquality at the cost of quantity because such cues are embeddedwithin a semantic network with fewer conceptual elements,thus yielding lesser interference from closely-related concepts.This hypothesis was confirmed in Study 1. Study 2 replicatedthe effect and found an interaction with fluid intelligence,indicating that cognitive control can overcome constraints ofsemantic knowledge. The findings thus highlight costs andbenefits of semantic knowledge for creative cognition.

How we mentally represent spatial relations is known to haveeffects on cognitive processes such as inferences, co-speechgesture, or memorizing. In addition, spatial positions oftenserve as metaphors that carry valence. For instance, “movingup the social ladder, “getting it right”, or being “in front” feelscertainly better than “moving down”, “having two left feet”,or “lagging behind”. Spatial position, however, depends onperspective, more concretely on which frame of reference(FoR) one adopts—and hence on cross-linguisticallydiverging preferences. What is conceptualized as “in front” inone variant of the relative FoR (e.g., translation) is “behind”under another variant (reflection), and vice versa. Do suchdiverging conceptualizations of an object’s location also leadto diverging evaluations? We tested this with speakers ofGerman, Chinese, and Japanese using an Implicit AssociationTest (IAT). Data from two studies suggest that acrosslanguages the object “in front of” another object is evaluatedmore positively than the one “behind”, and that both locationand evaluation depend on the adopted FoR. In other words:linguistically imparted FoR preferences appear to impact onevaluative processes.

The associative theory of creativity suggests that creativeabilities rely on the organization of semantic associations inmemory. Recent research has demonstrated that semanticnetwork methods allow testing this hypothesis. The aim of thecurrent study was to investigate the properties of semanticnetworks at the individual level, in relation to creative abilities.Semantic judgement ratings were used to estimate individualsemantic networks, whose topological properties measured byseveral graph metrics were correlated with individual creativityscores. We found a correlation between the theoretical semanticdistance of our stimuli and the relatedness ratings given by theparticipants, demonstrating the validity of our approach.Importantly, we found a close relationship between creativeabilities assessed by an achievement questionnaire and divergentthinking tasks and individual semantic network metrics,replicating and extending previous similar findings.

Peak End Rule (Kahneman, 1993; 2011) suggests that the averageof the peak and end moments of an event disproportionately affectmemory and thus perception of the experience. We investigatePER’s application to the experience of reading fiction. GillianFlynn’s Gone Girl (2012) is an ideal case study because it iscommercially popular but, unlike most popular novels, has adistinctly amoral ending. We hypothesize that humans expect moralpayoffs at the end of narrative fiction, and that when theseexpectations are not met (i.e., pain at the end of the experience), asin the case of Gone Girl, readers’ perceptions of the story will beinfluenced by this pain and manifest as disappointment and dislike.We reference existing models in evolutionary psychology, whichseek to explain human altruism, and models in cognitive science,which seek to explain patterns in memory and assessment. Toquantify disappointment and dislike, we conduct a programmaticcorpus linguistic analysis of 40,000 web-scraped Amazon productreviews of Gone Girl, comparing them to reviews of eight othersimilarly popular novels from the same year. Results show thatreader sentiments about Gone Girl, both the overall review ratingsand analysis on a sentence-by-sentence basis, are more positive thanfor the comparison novels. When only reviews mentioning “end”are analyzed, however, the effect reverses, with a similar finding atthe more granular level of sentences mentioning “end.” Thesefindings support our hypothesis that moral endings, or lack thereof,significantly shape reader perceptions of a novel.

Pragmatic inferences are an integral part of language learn-ing and comprehension. To recover the intended meaning ofan utterance, listeners need to balance and integrate differentsources of contextual information. In a series of experiments,we studied how listeners integrate general expectations aboutspeakers with expectations specific to their interactional his-tory with a particular speaker. We used a Bayesian pragmaticsmodel to formalize the integration process. In Experiments1 and 2, we replicated previous findings showing that listenersmake inferences based on speaker-general and speaker-specificexpectations. We then used the empirical measurements fromthese experiments to generate model predictions about howthe two kinds of expectations should be integrated, which wetested in Experiment 3. Experiment 4 replicated and extendedExperiment 3 to a broader set of conditions. In both experi-ments, listeners based their inferences on both types of expec-tations. We found that model performance was also consistentwith this finding; with better fit for a model which incorporatedboth general and specific information compared to baselinesincorporating only one type. Listeners flexibly integrate dif-ferent forms of social expectations across a range of contexts,a process which can be described using Bayesian models ofpragmatic reasoning.

Fluent conversation is a marvel of multi-tasking within the language domain: listeners must simultaneously comprehend the speaker, predict a turn transition point, and plan a response. Experiment 1 used spontaneous conversation to investigate the apparent demands of conversation on working memory by manipulating the difficulty of a secondary task. The experiment found support for Load Theory's (e.g., Lavie et al. 2004) prediction that both conversational fluency and performance on a secondary task would decrease as working memory load increased. However, there was also some support for Pickering and Garrod's (2004, 2013) proposal that dialogue is facilitated by a collection of automatic cognitive operations when interlocutors are well-aligned (i.e., using the same words, phrases, and structures to discuss the same topics). Experiment 2 tested two claims motivated by this account: alignment is necessary for fluent turn transitions, and lexical repetition between speakers is an essential component of the alignment advantage. We found support for the former claim, but not the latter.

Understanding the processes leading to insight has remained one of psychology’s greatest challenges. In this study, we examined how different lexical properties affect cognitive processes involved in a popular class of insight problems: Compound Remote Associates (CRAs). These properties were familiarity, lexeme meaning dominance, and semantic transparency. We found that a higher proportion of problems were solved when they were presented beginning with the most familiar cues, but not when they began with right-headed dominant or the most semantically transparent cues. Further, we found that participants focused their efforts disproportionately on first and last cues, that subjective ratings of insight decreased as trial times elapsed, and that the magnitude of reported insight increased with the number of cues successfully solved. This suggests that participants can monitor their progress in such problems. These results contest longstanding assumptions of requisite periods of impasse and the absence of incremental progress in insightful problem solving.

Evidence exists that individuals prefer distinguishable strategies for self-organized task scheduling in multitasking. Theyeither prefer to work for long sequences on one task before switching to another (i.e., blocking), to switch repeatedly aftershort sequences (i.e., switching), or to process the current stimuli of two tasks before responding almost simultaneously(i.e., response grouping). We tested whether the strategies efficiency differs depending on the resource competition be-tween tasks in a free concurrent dual-tasking paradigm and whether individuals adapt their strategies accordingly. Ourresults show that switcher and response grouper are more efficient than blocker during low than high resource competitionbetween tasks. Comparably, more switchers shifted to a response grouping strategy than blockers towards a switchingstrategy. Overall, especially those individuals benefited from a lower resource competition, who already preferred a moreflexible approach in dealing with the multitasking demand during high resource competition.

People increasingly turn to online social networks forinformation and debate. This means that the structures andproperties of these networks, and the information theypropagate, play crucial roles in the development of socialbeliefs, attitudes, and morals. Recently, research has shownthat the presence of specific language drives the diffusion ofmoral messages, regardless of the informational quality, in aphenomenon dubbed moral contagion (Brady et al., 2017).Due to the widespread attention and implications of suchfindings for science and society, we investigate the presenceof moral contagion across six sets of data that capture thecommunications of naturally-occurring networks on Twitter.Across a large corpus of diverse tweets (n = 525,229), we findmoral contagion to be an inconsistent and often absentphenomenon that does not effectively predict messagediffusion. The implications and reasons for this finding arediscussed.

The current study seeks to extend research on motivated reasoning by examining how prior beliefs influence the interpretation of objective graphs displaying quantitative information. The day before the 2018 midterm election, conservatives and liberals made judgments about four economic indicators displaying real-world data of the US economy. Half of the participants were placed in an 'alien cover story' condition where prior beliefs were reduced under the guise of evaluating a fictional society. The other half of participants in the 'authentic condition' were aware they were being shown real-world data. Despite being shown identical data, participants in the Authentic condition differed in their judgments of the graphs along party lines. The participants in the Alien condition interpreted the data similarly, regardless of politics. There was no evidence of a „backfire‟ effect, and there was some evidence of belief updating when shown objective data.

What happens to the acoustic signal after it enters the mind of a listener during real-time speech processing? Sinceprocessing involves extracting linguistic evidence from multiple, temporally distinct sources of information, successfulcommunication relies on a listeners ability to combine these potentially disparate signals. Previous work has shown thatlisteners are able to maintain, and rationally update, some type of intermediate representations over time. However, exactlywhat type of information is being maintainedbe it acoustic-phonetic or rather a probability distribution over phonemeshasbeen underspecified. In this paper we present a perception experiment aimed at identifying the internal contents of in-termediate representations in speech processing. Using an accent-adaptation paradigm, we find that listeners adapt tomodulated acoustic signal when the corresponding orthography is provided before the audio, but not when audio followsthe orthography. This supports the position that intermediate representations are uncertainty-distributions over discreteunits (e.g. phonemes) and that, by default, speech processing involves no maintenance of the acoustic-phonetic signal.

It has long been noted that the best examples, or foci, ofcolor categories tend to align across diverse languages (Berlin& Kay, 1969)—but there is limited documentation of suchuniversal foci in other semantic domains. Here, we explorewhether spatial topological categories, such as “in” and “on”in English, have focal members comparable to those in color.We document names and best examples of topological spatialrelations in Dutch, English, French, Japanese, Korean, Man-darin Chinese, and Spanish, and find substantial consensus,both within and across languages, on the best examples of suchspatial categories. Our results provide empirical evidence forfocal best examples in the spatial domain and contribute fur-ther support for a theory of “natural concepts” in this domain.Keywords: Language and thought; spatial cognition; cate-gories; semantic universals.The central role of fociFor decades, discussions of natural language categories suchas “dog” or “blue” have emphasized prototypes, family re-semblance, and fuzzy sets—all notions specifying relationsbetween central cases and boundaries, and recognizing gra-dation in category membership. An especially well-studiedand debated case is that of focal colors, or best examplesof color categories (e.g. Berlin & Kay, 1969; Heider, 1972;Kay & McDaniel, 1978; Roberson et al., 2000; Regier etal., 2005; Abbott et al., 2016). Despite the ongoing debate,there is broad consensus that such best examples of color cat-egories often (but not always) align across languages, andthat languages sometimes have composite categories appar-ently organized around multiple foci—for example a com-posite green-blue or “grue” category.Despite the attention given to focal colors, studies of cate-gorization and semantic typology in many other semantic do-mains have not emphasized category best examples as promi-nently, but have instead tended to characterize categories assets, such that an exemplar may simply be a member of thecategory or not. Within the domain of spatial topological re-lations, previous work has drawn on extensional patterns innaming as evidence for central exemplars and core meaningsof categories like “in” and “on” (e.g., Levinson et al., 2003;Johannes, Wang, Papafragou, & Landau, 2015; Johannes,Wilson, & Landau, 2016; Landau, Johannes, Skordos, & Pa-pafragou, 2017), but without directly querying speakers aboutbest examples per se. Here, we employ empirical best ex-ample data to provide a long-overdue response to a call byFeist (2000: 236) to determine whether spatial relational cat-egories, like colors, have focal members.In what follows, we review key findings on focal colorsand their relationship to color category semantics. We thendescribe parallels to color in the domain of spatial topologicalrelations, and summarize an account (Levinson et al., 2003)of focal spatial relations that was developed and evaluatedon the basis of spatial naming data, but without groundingin empirical best examples. We then present our study, whichreexamines the hypotheses of this previous account using em-pirical best example data from seven languages. We explore

This study sketches the language environments of children ages 0;03;0 growing up in two traditional, indigenous com-munities: one Tseltal (Mayan) and the other Yl (Papuan). Past ethnographic work has suggested that caregivers’ ideasabout talking to young children differ greatly between these two communities. However, the present daylong recordinganalyses suggest that, in fact, children are rarely directly addressed in both places, with no age-related increase and withmost child-directed speech coming from adults. Children’s manual activities also suggest that child-carrying practicesand cultural context moderate the extent to which children might use co-occurrence between held objects and ambientlanguage to learn words.

The current study aimed to elucidate the contributions of thesubcortical basal ganglia to human language by adopting theview that these structures engage in a basic neurocomputationthat may account for its involvement across a wide range of lin-guistic phenomena. Specifically, we tested the hypothesis thatbasal ganglia reinforcement learning mechanisms may accountfor variability in semantic selection processes necessary forambiguity resolution. To test this, we used a biased homographlexical ambiguity priming task that allowed us to measure au-tomatic processes for resolving ambiguity towards high fre-quency word meanings. Individual differences in task perfor-mance were then related to indices of basal ganglia function-ing and reinforcement learning, which were used to group sub-jects by learning style: primarily from choosing positive feed-back (Choosers), primarily from avoiding negative feedback(Avoiders), and balanced participants who learned equally wellfrom both (Balanced). The pattern of results suggests that bal-anced individuals, whom learn from both positive and negativereward equally well, had significantly lower access to the sub-ordinate homograph word meaning. Choosers and Avoiders,on the other hand, had higher access to the subordinate wordmeaning even after a long delay between prime and target. Ex-perimental findings were then tested using an ACT-R compu-tational model of reinforcement learning that learns from bothpositive and negative feedback. Results from the computa-tional model confirm and extend the pattern of behavioral find-ings, and provide a reinforcement learning account of lexicalpriming processes in human linguistic abilities, where a dual-path reinforcement learning system is necessary for preciselymapping out word co-occurrence probabilities.

How infants determine correct word-referent pairings withincomplex environments is not yet fully understood. Thecombination of multiple cues, including gestures, may guidelearning as part of a communicative exchange between parentand child. Gesture use and word learning are interlinked, withearly child gesture predicting later vocabulary size, andparental gesture predicting child gesture. However, the extentto which parents alter gesture cues during word learningaccording to referential uncertainty is not known. In this study,we manipulated the number of potential referents acrossconditions during a word learning task with 18–24-month-olds,and explored how changes in parental gesture use translatedinto infant word learning. We demonstrate that parents altertheir gesture use according to the presence, but not the degree,of referential uncertainty. We further demonstrate that a degreeof variability in the number of potential referents appears tobenefit word learning.

The purpose of this study is to explore the role goals play inconcept acquisition. Goals motivate and shape ourinteractions with items, so it stands to reason that they alsoimpact the learning that occurs as a result of thoseinteractions. There is abundant evidence that goals orient usto particular information about the items we encounter. Amore speculative claim is that goals play a more integral rolein the acquired concept in that they also help to structure andcohere the acquired conceptual knowledge. Using a novelconcept learning paradigm, we examined participantknowledge of attributes of the items they interacted with inan experimental task. We found evidence that the interactionof the goal with the learning situation impacted the centralityof the attribute information within their conceptualknowledge. These results support the idea that conceptualknowledge is organized in terms of goals active duringlearning.

Human knowledge accumulates over generations, amplifyingour individual learning abilities. What is the mechanism ofthis accumulation? Here, we explore how language allows ac-curate transmission of conceptual knowledge. We introduce anovel experimental paradigm that allows direct comparison oflearning from examples and learning from language. In ourexperiment, a teacher first learns a Boolean concept from ex-amples; they then communicate this concept to a student in afree conversation; finally, we test both teacher and student onthe same transfer items. We find that learning from languageis both sufficient and efficient: Students achieve accuracy veryclose to their teachers, while studying for less time. We thenexplore the language used by teachers and find heavy relianceon generics and quantifiers. Taken together, these results sug-gest that cultural accumulation of conceptual knowledge arisesfrom the ability of language to directly convey generalizations.

Cognitive science is an archipelago of concepts and models,with cross-pollination between topics of interest often prohib-ited by incompatible approaches. Despite this, behavioral per-formance universally depends on information transmission be-tween brain regions and is limited by physical and biologicalconstraints. These constraints can be formalized as informa-tion theoretic constraints on transmission, which provide nor-mative predictions across a surprising range of cognitive do-mains. To illustrate this, we describe a simple variable-lengthrate coding model built with Poisson processes, Bayesian in-ference, and an entropy-based decision threshold. This modelreplicates features of human task performance and provides aprincipled connection between a high-level normative frame-work and neural rate codes. We thereby integrate several dis-joint ideas in cognitive science by translating plausible con-straints into information theoretic terms. Such efforts to trans-late concepts, paradigms and models into common theoreti-cal languages are essential for synthesizing our rich but frag-mented understanding of cognitive systems.

A key question in early word learning is how infants learn their first object names despite a natural environment thought to provide messy data for linking object names to their referents. Using head cameras worn by 7 to 11-month-old infants in the home, we document the statistics of visual objects, spoken object names, and their co-occurrence in everyday meal time events. We show that the extremely right skewed frequency distribution of visual objects underlies word-referent co- occurrence statistics that set up a clear signal in the noise upon which infants could capitalize to learn their first object names.

Frequency of reward and average reward value are two types of reward information we utilize when making decisions between two alternative options. Often, these two pieces of information coincide with the highest value option, however, when a slightly less valuable option is presented more frequently, standard reinforcement learning models such as the Delta model can make incorrect predictions. This paper explores the discrepancy in these predictions by way of simulating relevant behavioral tasks with the Delta model, the Decay model, and a novel Bayesian model based on the Dirichlet distribution. We then compare model predictions to behavioral data from some of the same tasks that were simulated. The Delta model provides a poor fit to the data for each of the three presented tasks when compared to the Decay model and the two Bayesian learning models, because it predicts a bias toward options with higher average reward, while the Decay and Bayesian models predict a bias toward reward frequency. The Decay and Bayesian models show a distinct similarity in prediction and fits to the data for most of the tasks. This is because both models predict a bias toward reward frequency rather than average reward magnitude, despite different computational formalisms. However, we also note some interesting discrepancies between the Decay and Bayesian models which will show that in some cases, the frequency of reward may be more important than the reward value.

One of the central goals of language evolution research is toexplain how systematic structure emerges. A culturalevolutionary approach proposes that the systematic structure oflanguage arises from the use and transmission of language.Motamedi and colleagues (2016) investigated the influences ofthese forces on the evolution of language by generating anartificial sign language in the lab. Over several generations ofnew learners and their interactions, an initially unsystematic setof silent gestures developed markers for functional categoriesof person, location, object, and action. Here we describe resultsof two studies that compared the learnability of solo-producedpantomimes versus signals that had been transmitted and usedby interlocutors. In these studies, participants saw an artificialsign and judged whether an English translation matched ormismatched the meaning of the sign. In an event-relatedpotential (ERP) study, we found that mismatches elicited largernegativities in the ERP than matches. However, those effectswere most reminiscent of the classic N400 response in theevolved signs. This study provides a clearer view on how themechanisms that drive language evolution change language toadapt to a learner’s brain.

Do people adjust their conversational strategies to the specificcontextual demands of a given situation? Prior studies haveyielded conflicting results, making it unclear how strategiesvary with demands. We combine insights from qualitative andquantitative approaches in a within-participant experimentaldesign involving two different contexts: spontaneouslyoccurring conversations (SOC) and task-oriented conversations(TOC). We systematically assess backchanneling, other-initiated repair and linguistic alignment. We find that SOCexhibit a higher number of backchannels, a reduced and moregeneric repair format and higher rates of lexical and syntacticalignment. TOC are characterized by a high number of specificrepairs and a lower rate of lexical and syntactic alignment.However, when alignment occurs, more linguistic forms arealigned. The findings show that conversational strategies adaptto contextual demands.

In language, speakers are more likely to mention the goals, orendpoints, of motion events than they are to mention sources,or starting points (e.g. Lakusta & Landau, 2005). Thisphenomenon has been explained in cognitive terms, but mayalso be affected by discourse-communicative factors: Forparticipants in prior work, sources can be characterized asgiven, already-known information, while goals are new,relevant information to communicate. We investigate to whatextent the goal bias in language (and memory) is affected whenthe source is or is not in common ground between speaker andhearer, and thus whether it is discourse-given or -new. We findthat the goal bias in language is severely diminished whensource and goal are discourse-new. We suggest that the goalbias in language can be attributed to discourse-communicativefactors in addition to any cognitive goal bias. Discourse factorscannot fully account for the bias in memory.

During reading, readers intentionally do not fixate a wordwhen highly confident in its identity. In a rational model ofreading, word skipping decisions should be complex functionsof the particular word, linguistic context, and visual informa-tion available. In contrast, simple heuristic of reading onlypredicts additive effects of word and context features. Here wetest these predictions by implementing a rational model withBayesian inference, and predicting human skipping with theentropy of this model’s posterior distribution. Results showeda significant effect of the entropy in predicting skipping abovea strong baseline model including word and context features.This pattern held for entropy measures from rational modelswith a frequency prior but not from ones with a 5-gram prior.These results suggest complex interactions between visual in-put and linguistic knowledge as predicted by the rational modelof reading, and a dominant role of frequency in making skip-ping decisions.

Curiosity is considered essential for learning and sustained en-gagement, yet stimulating curiosity in educational contexts re-mains a challenge. Can people’s curiosity about a topic bestimulated by evidence that the topic has potential value? Intwo experiments we show that increasing people’s perceptionsabout the usefulness of a scientific topic also influences theircuriosity and subsequent information search. Our results alsoshow that simply presenting interesting facts is not enough toinfluence curiosity, and that people are more likely to be curi-ous about a topic if they perceive it to be directly valuable tothem. Given the link between curiosity and learning, these re-sults have important implications for science communicationand education more broadly.

A consistent finding in research on conditional reasoning isthat individuals are more likely to endorse the valid modus po-nens (MP) inference than the equally valid modus tollens (MT)inference. This pattern holds for both abstract task and prob-abilistic task. The existing explanation for this phenomenonwithin a Bayesian framework (e.g., Oaksford & Chater, 2008)accounts for this asymmetry by assuming separate probabil-ity distributions for both MP and MT. We propose a novelexplanation within a computational-level Bayesian account ofreasoning according to which “argumentation is learning”.We show that the asymmetry must appear for certain priorprobability distributions, under the assumption that the condi-tional inference provides the agent with new information thatis integrated into the existing knowledge by minimizing theKullback-Leibler divergence between the posterior and priorprobability distribution. We also show under which conditionswe would expect the opposite pattern, an MT-MP asymmetry.

Young children often stretch terms to novel objects when theylack the proper adult words—a phenomenon known as overex-tension. Psychologists have proposed that overextension relieson the formation of a chain complex, such that new objectsmay be linked to existing referents of a word based on a diverseset of relations including taxonomic, analogical, and predicate-based knowledge. We build on these ideas by proposing a com-putational framework that creates chain complexes by multi-modal fusion of resources from linguistics, deep learning net-works, and psychological experiments. We test our models ina communicative scenario that simulates linguistic productionand comprehension between a child and a caretaker. Our re-sults show that the multimodal semantic space accounts forsubstantial variation in children’s overextension in the liter-ature, and our framework predicts overextension strategies.This work provides a formal approach to characterizing lin-guistic creativity of word sense extension in early childhood.

Investigating folk conceptions of choice and constraints havebeen problematic given that human actions are rarelyconsidered constrained. In this paper, we utilize humanoidrobots (more clearly influenced by determined programming)to empirically test children’s developing concepts of choiceand action. Using a series of agency attribution and choiceprediction tasks, we examined whether children differentiatefree will abilities between robots and humans. Resultsindicated that 5–7-year-old children similarly attributed theability to choose to both a robot and human child. However,for moral scenarios, participants considered the robot’s actionsto be more constrained than the human. These findingsdemonstrate that children appear to hold a nuancedunderstanding of choice across agents and across context.

Past research has shown that adults can access multiplemeanings for a word, but little work has examined howchildren process multiple meanings. We tested 48 4- to 7-year-old children and 48 adults in a touchscreen picturerecognition task. Two meanings of the same word weredisplayed on successive trials, which varied according towhether the 2 meanings were unrelated (homonyms), related(polysemes), or repeated (same-meaning). Adults identifiedthe second meaning more quickly than the first in allconditions and to the same extent. Children, however,identified the second meaning more quickly only onpolysemy and same-meaning trials. This difference suggeststhat children are less capable of co-activating unrelatedmeanings, which raises the possibility that children mustlearn to do so over development. Despite the ubiquity ofpolysemy in language, our work is the first to show thatchildren’s processing of word representations is organizedby similarity.

People frequently use metaphors to communicate and reasonabout complex topics. However, many studies of metaphoricalreasoning exclusively rely on researcher intuitions aboutdifferent metaphors and their associated entailments. Here wedescribe a more principled method for mapping the structureof metaphorical lay theories, focusing on metaphors forteaching. Across two studies, we identified four common, aptmetaphors for the teacher-student relationship and used factoranalysis to explore whether these metaphors reflectsystematically different intuitions about the qualities of collegeteachers. Our findings demonstrate that (1) people endorse avariety of different teaching metaphors, and (2) thesemetaphors bring to mind distinct, coherent clusters of teacherattributes. This work demonstrates a novel method forsystematically mapping the structure of metaphorical laytheories and sets the stage for future research on metaphoricalreasoning as well as innovative educational interventionscentered on shifting lay theories of teaching.

Word learning relies on the ability to master the sound con-trasts that are phonemic (i.e., signal meaning difference) ina given language. Though the timeline of phoneme develop-ment has been studied extensively over the past few decades,the mechanism of this development is poorly understood. Pre-vious work has shown that human learners rely on referentialinformation to differentiate similar sounds, but largely ignoredthe problem of taxonomic ambiguity at the semantic level (twodifferent objects may be described by one or two words de-pending on how abstract the meaning intended by the speakeris). In this study, we varied the taxonomic distance of pairs ofobjects and tested how adult learners judged the phonemic sta-tus of the sound contrast associated with each of these pairs.We found that judgments were sensitive to gradients in thetaxonomic structure, suggesting that learners use probabilisticinformation at the semantic level to optimize the accuracy oftheir judgements at the phonological level. The findings pro-vide evidence for an interaction between phonological learningand meaning generalization, raising important questions abouthow these two important processes of language acquisition arerelated.

How do you make sense of an unconventional graph? Buildingon research demonstrating that prior knowledge of graphicalconventions is difficult to overcome, we reconstrue graphreading as an insight problem. We hypothesize that imposing amental impasse during a particular type of graph reading taskwill improve comprehension by inducing a sense ofpuzzlement, prompting learners to reconsider theirinterpretation. We find support for this proposal in a between-subjects experiment in which participants presented with animpasse-formulated version of graph reading questions aresignificantly more likely to correctly interpret a graph featuringan unconventional coordinate system. We characterize thedifferential patterns of mouse movements for learners betweenconditions and discuss implications for the use of novelgraphical forms in science communication.

Recurrent neural network (RNN) models of sentence process-ing have recently displayed a remarkable ability to learn as-pects of structure comprehension, as evidenced by their abilityto account for reading times on sentences with local syntac-tic ambiguities (i.e., garden-path effects). Here, we investi-gate if these models can also simulate the effect of semanticappropriateness of the ambiguity’s readings. RNN-based esti-mates of surprisal of the disambiguating verb of sentences withan NP/S-coordination ambiguity (as in ‘The wizard guards theking and the princess protects ...’) show identical patters to hu-man reading times on the same sentences: Surprisal is higheron ambiguous structures than on their disambiguated counter-parts and this effect is weaker, but not absent, in cases of poorthematic fit between the verb and its potential object (‘Theteacher baked the cake and the baker made ...’). These resultsshow that an RNN is able to simultaneously learn about struc-tural and semantic relations between words and suggest thatgarden-path phenomena may be more closely related to wordpredictability than traditionally assumed.

Adjective ordering preferences (e.g., big brown bag vs. brownbig bag) are robustly attested in English and many unrelatedlanguages (Dixon, 1982). Scontras, Degen, and Goodman(2017) showed that adjective subjectivity is a robust predictorof ordering preferences in English: less subjective adjectivesare preferred closer to the modified noun. In a follow-up tothis empirical finding, Simoniˇc (2018) and Scontras, Degen,and Goodman (to appear) claim that pressures from success-ful reference resolution and the hierarchical structure of mod-ification explain subjectivity-based ordering preferences. Weprovide further support for this claim using large-scale sim-ulations of reference scenarios, together with an empirically-motivated adjective semantics. In the vast majority of cases,subjectivity-based adjective orderings yield a higher probabil-ity of successful reference resolution.

Understanding the cognitive structure of explanations— andthe cognitive processes that assemble them— is a milestonefor understanding how people learn and communicate. Re-cent research on explanatory coexistence suggests that peo-ple’s causal beliefs are less globally coherent than previouslythought: people use seemingly-competing supernatural and bi-ological causes to explain different aspects of the same phe-nomenon, or they assemble supernatural and biological causesinto single, coherent explanations (Legare & Gelman, 2008;Legare & Shtulman, 2018; Shtulman & Lombrozo, 2016).This coexistence— and unexpected coherence— of diversecausal mechanisms poses interesting questions about the roleof coherence and fragmentation in people’s mental models andexplanations. This paper presents a computational model ofexplanatory coherence in the well-characterized domain of dis-ease transmission, extending a previous cognitive model ofexplanation-based conceptual change (Friedman, Forbus, &Sherin, 2018). Our approach (1) retrieves diverse causal modelfragments based on the phenomenon to explain, (2) assem-bles coherent causal models using relevance-directed abduc-tive reasoning, and (3) selects explanatory paths that supportwithin-explanation and within-scenario coherence. Our modelsimulates the three different types of explanatory coexistencedetailed in the literature.

What is the descriptive content and guiding lay theory of transgender stereotypes? The recent rise in public visibilityand the numeric minority of this gender group make this an opportunity to understand not only the content of stereotypesapplied to transgender individuals today, but also the ontology of gender guiding the content of these stereotypes. Usingconvergent methods, we measure the descriptive content of transgender stereotypes and assess the role of essentialistbeliefs in guiding that content. We show that transgender categories are perceived less positively than cisgender categories,and that while perceptions of cisgender men and women differ sharply, those of transgender men and women show strikingsimilarity. Essentialist beliefs about gender exaggerate these patterns.

What is the mechanism by which linguistic knowledge is updated over time? In six experiments, we asked whether error-driven learning can explain how adults and children add new words to their vocabulary. Participants were exposed tonovel object labels that were more or less unexpected given participants linguistic knowledge. Two-to-four-year-olds werestrongly affected by expectations based on contextual constraint when choosing the referent of a new label. However,while adults formed stronger memory traces for novel words that violated a stronger prior expectation, childrens memorywas unaffected by the strength of their prior expectations. We conclude that the encoding of new words in memory followsthe principles of error-driven learning in adults, but not in preschoolers.

How sleep affects memory in older adults is a critical topic,since age significantly impacts both sleep and memory. Fordeclarative memory, previous research reports contradictoryresults, with some studies showing sleep-dependent memoryconsolidation and some other not. We hypothesize that thisdiscrepancy may be due to the use of recall as the memorymeasure, a demanding task for older adults. The present paperfocuses on the effect of sleep on relearning, a measure thatproved useful to reveal subtle, implicit memory effects.Previous research in young adults showed that sleeping afterlearning was more beneficial to relearning the same Swahili-French word pairs 12 hours later, compared with the sameinterval spent awake. In particular, those words that could notbe recalled were relearned faster when participants previouslyslept. The effect of sleep was also beneficial for retention aftera one-week and a 6-month delay. The present study used thesame experimental design in older adults aged 71 on averagebut showed no significant effect of sleep on consolidation, onrelearning, or on long-term retention. Thus, even when usingrelearning speed as the memory measure, the consolidatingeffect of sleep in older adults was not demonstrated, inalignment with some previous findings.

Drift diffusion (or evidence accumulation) models have found widespread use in the modelling of simple decision tasks. Extensions of these models, in which the model’s instantaneous drift rate is not fixed but instead allowed to vary over time as a function of a stream of perceptual inputs, have allowed these models to account for more complex sensorimotor decision tasks. However, many real-world tasks seemingly rely on a myriad of even more complex underlying processes. One interesting example is the task of deciding whether to cross a road with an approaching vehicle. This action decision seemingly depends on sensory information both about own affordances (whether one can make it across before the vehicle) and action intention of others (whether the vehicle is yielding to oneself). Here, we compared three extensions of a standard drift diffusion model, with regards to their ability to capture timing of pedestrian crossing decisions in a virtual reality environment. We find that a single variable-drift diffusion model (S-VDDM) in which the varying drift rate is determined by visual quantities describing vehicle approach and deceleration, saturated at an upper and lower bound, can explain multimodal distributions of crossing times well across a broad range vehicle approach scenarios. More complex models, which attempt to partition the final crossing decision into constituent perceptual decisions, improve the fit to the human data but further work is needed before firm conclusions can be drawn from this finding.

One powerful way children can learn word meanings is viacross-situational learning, the ability to discern consistentword-referent mappings from a series of ambiguous scenes andutterances. Various computational accounts of word learninghave been proposed, with mechanisms ranging from storingand testing a single hypothesized referent for each word, totracking multiple graded associations and selectively strength-ening some of them. Nearly all word learning models as-sume storage of some feasible word-referent mappings fromeach situation, resulting in a degree of learning proportionalto the number of co-occurrences. While these accumulativemodels would generally predict that incorrect co-occurrenceswould slow learning, recent empirical work suggests these ac-counts are incomplete: paradoxically, giving learners incorrectmappings early in training was found to boost performance(Fitneva & Christiansen, 2015). We test this finding’s general-ity in a new experiment with more items, consider system- anditem-level explanations, and find that a model with error-drivenlearning best accounts for this benefit of initially-inaccuratepairings.

Historically, research on preschool-aged children’sprobabilistic reasoning abilities has yielded mixed results.Although some findings have suggested that young childrencan successfully evaluate probabilities, others have suggestedthat they may use strategies that only approximate trueprobabilistic inference and therefore sometimes make errors(e.g., Girotto et al., 2016; Piaget & Inhelder, 1975). To explorethe factors that affect young children’s probabilistic reasoning,we developed a battery of problems that contained features thataffect the ease with which a problem is evaluated, and the typesof alternative strategies that can be applied to solve them. Thecurrent experiments (total N = 124) assessed 3- and 4-year-oldchildren’s probabilistic reasoning using an experimentalparadigm tailored to this age group. Results from bothexperiments suggest that young children are able to engage intrue probabilistic inference, as they performed well-abovechance on each problem. Nuances in children’s performanceare discussed, along with possibilities for future research.

Given the importance of developing analogical reasoning tobootstrapping children’s understanding of the world, why isthis ability so challenging for children? Two commonmechanisms have been implicated: 1) children’s inability toprioritize relational information during initial problem solving;2) children’s inability to disengage from salient distractors.Here, we use eye tracking to examine children and adults’looking patterns when solving scene analogies, allowing fordifferentiation between attention to relations versus tofeaturally salient distractors. In contrast to a recent study withpropositional analogies, our data suggest prioritization ofsource information does not differ between adults and children,nor is it predictive of performance; however, children andadults attend differently to distractors, and this attentionpredicts performance. These results suggest that feature-baseddistraction is a key way children and adults differ duringanalogical reasoning, and that the analogy problem formatshould be taken into account when considering children’sanalogical reasoning.

How do societies develop categories for continuous sets of novel phenomena, as in the domains of art and technology?Seminal work in the nativist tradition argues that given the same stimuli, people can independently produce the samecategories as a result of universal cognitive constraints. These constraints are said to account for cross-group coherence,where distinct communities and cultures have been shown to arrive at highly similar categories. Cross-group coherence iswidely seen as incompatible with functionalism, which holds that categories are defined through communication, leadingto divergent category systems. Here, we use an experiment to demonstrate that communication can generate either thedivergence or convergence of category systems, depending on the size of the social network (2, 6, 8, 24, and 50). We findthat large social networks amplify population biases, where a subset of slightly more frequent words become exponentiallymore likely to spread as network size increases.

We consider the problem of predicting how humans learn inter-actively in an adversarial Multi-Armed Bandit (MAB) setting.In a cybersecurity scenario, we designed defense algorithms toassign decoys to lure attackers. Humans play the role of cyberattackers in an experiment to try to learn the defense strategyafter repeated interactions. Participants played against one ofthree defense algorithms: a stationary strategy, a static game-theoretic solution, and an adaptive MAB strategy. Our resultsshow that humans have the most difficulty learning against theadaptive defense. We also evaluated five different models ofattack behavior and compared their predictions against humandata. We show that a modified version of Thompson Samplingand a cognitive model based on Instance-Based Learning The-ory are the best at replicating human learning against defensestrategies. We discuss how these models of human attacker caninform future cyberdefense tools.

Intuitively, human readers cope easily with errors in text; ty-pos, misspelling, word substitutions, etc. do not unduly disruptnatural reading. Previous work indicates that letter transposi-tions result in increased reading times, but it is unclear if thiseffect generalizes to more natural errors. In this paper, we re-port an eye-tracking study that compares two error types (let-ter transpositions and naturally occurring misspelling) and twoerror rates (10% or 50% of all words contain errors). We findthat human readers show unimpaired comprehension in spiteof these errors, but error words cause more reading difficultythan correct words. Also, transpositions are more difficult thanmisspellings, and a high error rate increases difficulty for allwords, including correct ones. We then present a computa-tional model that uses character-based (rather than traditionalword-based) surprisal to account for these results. The modelexplains that transpositions are harder than misspellings be-cause they contain unexpected letter combinations. It also ex-plains the error rate effect: expectations about upcoming wordsare harder to compute when the context is degraded, leading toincreased surprisal.

Semantic search and retrieval of information plays an im-portant role in creative idea generation. This study was de-signed to examine how semantic and temporal clustering varieswhen asking participants to generate ideas about uses for ob-jects compared with generating members of goal-derived cat-egories. Participants generated uses for three objects: brick,hammer, picture frame, and also generated members of thefollowing goal-derived categories: things to take in case of afire, things to sell at a garage sale, and ways to spend lotterywinnings. Using response-time analysis and semantic analysis,results illustrated that all six prompts generally led to exponen-tial cumulative response-time distributions. However, the pro-portion of temporally clustered responses, defined using theslope-difference algorithm, was higher for goal-derived cate-gory responses compared with object uses. Despite that, over-all pairwise semantic similarity was higher for object uses thanfor goal derived exemplars. The effect of prompt on pairwisesemantic similarity is likely the result of context-dependencyof exemplars from goal-derived categories. However, the cur-rent analysis contains a potential confound such that specialinstructions to give “common and uncommon” responses wereprovided only for the object-uses prompts. The confound islikely minimal, but future work is necessary to verify that theseresults would hold when the confound is removed.

Drawing is a versatile technique for visual communication,ranging from photorealistic renderings to schematic diagramsconsisting entirely of symbols. How does a medium spanningsuch a broad range of appearances reliably convey meaning? Anatural possibility is that drawings derive meaning from boththeir visual properties as well as shared knowledge betweenpeople who use them to communicate. Here we evaluate thispossibility in a drawing-based reference game in which twoparticipants repeatedly communicated about visual objects.Across a series of controlled experiments, we found that pairsof participants discover increasingly sparse yet effective waysof depicting objects. These gains were specific to thoseobjects that were repeatedly referenced, and went beyond whatcould be explained by task practice or the visual properties ofthe drawings alone. We employed modern techniques fromcomputer vision to characterize how the high-level visual fea-tures of drawings changed, finding that drawings of the sameobject became more consistent within a pair of participants anddivergent across participants from different interactions. Takentogether, these findings suggest that visual communicationpromotes the emergence of depictions whose meanings areincreasingly determined by shared knowledge rather than theirvisual properties alone.

Ambiguity has often been viewed as a hindrance to communi-cation. In contrast, Piantadosi et al. (2012) argued that ambi-guity may be useful in that it allows communication to be ef-ficient, and they found support for this argument in the spokenforms of modern English, Dutch, and German. The historicalorigins of this phenomenon cannot be probed in the case of spo-ken language, but they can for written language, as it leaves anenduring trace. Here, we explore ambiguity and efficiency inone of the earliest known written forms of language: Sumeriancuneiform. Sumerian cuneiform exhibits extensive ambiguity,and for that reason it has been considered to be poorly suited forcommunication. We find, however, that ambiguity in Sumeriancuneiform supports efficient communication, mirroring the ear-lier findings for spoken English, Dutch, and German. Thus, theearly stages of human writing exhibit evidence suggesting pres-sure for communicative efficiency.

When do children extend a construction (“rule”) productively?A recent Threshold proposal claims that a construction isproductive if and only if it has been witnessed applying to asufficient proportion of cases and sufficiently few exceptions.An alternative proposal, Communicate and Access (C&A),argues that children extend a construction productively becausethey wish to express an intended message and are unable toaccess a “better” (appropriate and more conventional) way todo so. Accessibility, in turn, is negatively affected byinterference from competing alternatives. In a preregisteredexperiment, 32 4-6-year-old children were provided withexposure to 2 mini-artificial languages for which the twoproposals make opposite predictions. Results support the C&Aproposal: children were more productive after witnessing 3rule-following cases than after 5, due to differences ininterference. We conclude that productivity is encouraged by adesire to communicate a message and is constrained byaccessibility and interference.

Language and music are structured at multiple temporal scalesand have been characterized as having meter: a hierarchical andperiodic alternation of the prominence of syllables/beats. Meter isthought to emerge from the entrainment of neural oscillators,affording temporal expectations and selective attention. Higher-levels of a metric hierarchy also tend to track syntactic phrasestructure, however, it is not clear within the framework oftemporal attending why this would be advantageous. Neuraloscillations have recently been shown to also track syntacticphrases. We propose that meter aligns to phrase structure so as tomake syntactic processing more efficient. In two experiments(both visual and auditory language), we show that certainalignments of meter to syntax influence sentence comprehensionand we suggest potential mechanisms for why certain alignmentstend to be preferred. Our results underline the rhythmicity of notonly low-level perception but also of higher-level cognitiveprocessing of syntactic sequences.

One design feature of human language is its combinatorialphonology, allowing it to form an unbounded set of mean-ingful utterances from a finite set of building blocks. Re-cent experiments suggest how this feature can evolve culturallywhen continuous signals are repeatedly transmitted betweengenerations. Because the building blocks of a combinatorialsystem lack independent meaning, combinatorial structure ap-pears to be in conflict with iconicity, another property salientin language evolution. To investigate the developmental tra-jectory of iconicity during the evolution of combinatoriality,we conducted an iterated learning experiment where partici-pants learned auditory signals produced using a virtual slidewhistle. We find that iconicity emerges rapidly but is gradu-ally lost over generations as combinatorial structure continuesto increase. This suggests that iconicity biases, whose pres-ence was revealed in a signal guessing experiment, manifest innuanced ways. We discuss implications of these findings fordifferent ideas about how biases for iconicity and combinato-riality interact in language evolution.

Everyday hearing requires inferring the causal factors that produce a sound, as when we separate the acoustic effectsof the environment (reverberation) from those of sound sources. Here we consider perceptual inferences from impactsounds, in which the resonance of a struck object provides cues to its material, but via acoustic effects that might benontrivial to disentangle from reverberation. We investigated whether and how humans separate the effects of objectresonance and reverberation in a material classification task. For comparison, we implemented a Bayesian observer thatinferred material from a generative model of object sounds without reverberation. Humans were robust to reverberation,whereas the model was not. However, human robustness was specific to reverberation consistent with the statistics ofnatural environments. The results suggest that humans use internal models of room and object acoustics to determine theirrespective contributions to sound, providing an example of causal inference in audition.

We report on initial findings from a pupillometry study thatinvestigated the influence of two extra-linguistic variables,namely Neuroticism and Disgust Sensitivity, on auditory lan-guage comprehension in adults. Results suggest that: (1) Lan-guage comprehension is influenced by extra-linguistic vari-ables and individual differences; (2) the processing of differ-ent kinds of linguistic errors, as opposed to clashes with anindividual’s value or belief system, are influenced by differ-ent extra-linguistic variables; and that (3) Disgust Sensitiv-ity at least partially predicts pupillary responses to utterancesclashing with an individual’s belief system. Results are dis-cussed with regards to linguistic anticipation, cognitive effortand arousal, and resource allocation.

How do people use human-made objects (artifacts) to learn about the people and actions that created them? We test the richness of people’s reasoning in this domain, focusing on the task of judging whether social transmission has occurred (i.e. whether one person copied another). We develop a formal model of this reasoning process as a form of rational inverse planning, which predicts that rather than solely focusing on artifacts’ similarity to judge whether copying occurred, people should also take into account availability constraints (the materials available), and functional constraints (which materials work). Using an artifact-building task where two characters build tools to solve a puzzle box, we find that this inverse planning model predicts trial-by-trial judgments, whereas simpler models that do not consider availability or functional constraints do not. This suggests people use a process like inverse planning to make flexible inferences from artifacts’ features about the source of design ideas.

The ability to recognize and make inductive inferences based onrelational similarity is fundamental to much of human highercognition. However, relational similarity is not easily defined ormeasured, which makes it difficult to determine whetherindividual differences in cognitive capacity or semanticknowledge impact relational processing. In two experiments, weused a multi-arrangement task (previously applied to individualwords or objects) to efficiently assess similarities between wordpairs instantiating various abstract relations. Experiment 1established that the method identifies word pairs expressing thesame relation as more similar to each other than to thoseexpressing different relations. Experiment 2 extended theseresults by showing that relational similarity measured by themulti-arrangement task is sensitive to more subtle distinctions.Word pairs instantiating the same specific subrelation werejudged as more similar to each other than to those instantiatingdifferent subrelations within the same general relation type. Inaddition, Experiment 2 found that individual differences in bothfluid intelligence and crystalized verbal intelligence correlatedwith differentiation of relation similarity judgments.

In prior research, arguments using both anecdotal andstatistical evidence are more persuasive than arguments usingeither alone (Allen, Bruflat, Fucilla, Kramer, McKellips,Ryan, & Spiegelhoff, 2000; Hornikx, 2005). However, it isless clear how people integrate information when the statisticsand the anecdotes present conflicting information. In threepreregistered experiments, we tested how people integrateconflicting information to judge the efficacy of a medicine ina clinical trial. Participants read either an anecdote fromsomeone in the trial, summary statistics about the trial, orboth types of information. We found that reading an anecdotefrom a member of the trial for whom treatment wasineffective reduced people’s beliefs in a medical treatmenteven when participants received strong evidence that thetreatment was effective. In Experiment 3, we found thatintroducing icon arrays increased the perceived efficacy of thetreatment but did not eliminate the effect of the anecdote.

We introduce a memory-augmented neural network, calledDifferentiable Working Memory (DWM), that captures somekey aspects of attention in working memory. We tested DWMon a suite of psychology inspired tasks, where the model had todevelop a strategy only by processing sequences of inputs anddesired outputs. Thanks to novel attention control mechanismscalled bookmarks, the model was able to rapidly learn a goodstrategy—generalizing to sequence lengths even two orders ofmagnitude larger than that used for training—allowing it to re-tain, ignore or forget information based on its relevance. Thebehavior of DWM is interpretable and allowed us to analyzeits performance on different tasks. Surprisingly, as the train-ing progressed, we observed that in some cases the model wasable to discover more than one successful strategy, possiblyinvolving sophisticated use of memory and attention.

Children are motivated to explore and learn about the world,but they vary in their degree of perseverance duringexploration. A growing body of literature suggests that ismalleable from an early age. Here, we ask whetherpedagogical questions empower children to persevere duringa difficult problem-solving task with a blicket detectormachine. Previous research has shown that when presentedwith a blicket detector, asking children “pedagogicalquestions” promotes more exploratory behaviors compared todirect instruction. A pedagogical question is a question askedby a knowledgeable person, whose intention is to teach ratherthan to seek an answer to that question. The current studyexamines whether pedagogical questions influence theamount of time children spend problem-solving beforeseeking help, compared to direct instruction, overheardpedagogical questions, and overheard questions asked by anaive other. We predicted that children who were asked apedagogical question prior to having the opportunity to playwith a machine would persevere longer in trying to make itwork, and would be less likely to ask for help. Results suggestthat pedagogical questioning encourages children to attemptmore hypothesis-test interventions in an effort to make themachine work. Results will be discussed in terms of the roleof pedagogical questioning in promoting perseverance duringproblem-solving.

Many students fail to develop adequate understanding ofmathematical equivalence in early grades, with detrimentalconsequences for later algebra learning. The changeresistance account (McNeil, 2014) proposes that studentsstruggle with equivalence because traditional arithmeticpractice overexposes students to mathematical expressionswhere all the operators are on the left of the equal sign.Students erroneously believe the equal sign means to “dosomething” or “give the answer” – and fail to see equations asrelations between two expressions. These operations-basedmisconceptions affect how they perceive, conceptualize, andapproach math problems and interfere with developingcorrect understandings of equivalence. The current paperexplores 1) are these misconceptions evident as encodingerrors in second graders? 2) do item properties make specificerror types more or less likely? 3) do misconceptions inencoding impact solving performance? and 4) can targetedtraining mitigate the effects of prior misconceptions on bothequation encoding and solving? We identify a category ofmisconception-based encoding errors that negatively impactsequation solving and replicate findings that a conceptuallyrich research-based intervention program is maximallyeffective in training students to overcome problematicmisconceptions.

Do we praise altruistic acts because they produce socialbenefits or because they require a personal sacrifice? Onthe one hand, utilitarianism demands that we maximize thesocial benefit of our actions, which could motivatealtruistic acts. On the other hand, altruistic acts signalreputation precisely because personal sacrifice is a strong,costly signal. Consistent with the reputational account,these studies find that in the absence of reputational cues,people mainly rely on personal cost rather than socialbenefit when evaluating prosocial actors (Study 1).However, when reputation is known, personal cost acts as amuch weaker signal and play a smaller role in moralevaluations (Study 2). We argue that these results have far-reaching implications for the psychology and philosophy ofaltruism, as well as practical import for charitable giving,particularly the effective altruism movement.

People assess others’ moral characters to predict what theywill do. Here, we study the computational mechanismsused to predict behavior from uncertain evidence aboutcharacter. Whereas previous work has found that peopleoften ignore hypotheses with low probabilities, we find thatpeople often account for the possibility of poor moralcharacter even when that possibility is relatively unlikely.There was no evidence that comparable inferences fromuncertain non-moralized traits integrate across multiplepossibilities. These results contribute to our understandingof moral judgment, probability reasoning, and theory ofmind.

Since we rarely view our own body movements in our dailylives, understanding the recognition of self-body movementcan shed light on the core of self-awareness and on therepresentation of actions. We first recorded nine simple andnine complex actions performed by individual participants,who also subsequently observed nine videos displayed on thescreen and imitated these actions. After a delay period of 35-40 days, participants were asked to identify their self- bodymovements presented as point-light displays amongst threeother actors who performed the same actions. Participants wereable to recognize themselves solely based on kinematics inpoint-light displays. However, self-recognition accuracyvaried according to the complexity of performed actions, withmore accurate self-recognition for complex than simpleactions. The ability of self-recognition with simple actionsshowed a significant relation with autistic traits (negativerelation: poorer self-recognition accuracy with more autistictraits), schizophrenic traits (quadratic non-linear relation,participants with the median degree of schizophrenia traitsperformed better than participants at the extremes), and withimitation actions and motor imagery traits (linear relation:increased self-recognition accuracy with greater motorimagery). We also found that participants did not recognizeactions that only required visual experience but could identifytheir self-generated actions that required motor experience,underscoring the importance of motor experience to therepresentation of self-body movements.

Learning new words does not only require infants to find words in continuous speech, but also be remember recentlysegmented words and link them to meaning. Prior research has shown that statistical learning supports word learning.However, as infant statistical learning was typically tested immediately after familiarization with a speech stream, weknow very little about whether infants experience with statistical regularities supports long-term memory and future wordlearning. The current study was designed to shed light on the relationship between statistical learning, word learning, andmemory. We found that while both co-occurrence statistics and syllable frequency information support word learning inthe moment, co-occurrence information alone supports long-term memory for recently segmented candidate object labels.

Infants are powerful learners. A large corpus of experimental paradigms demonstrate that infants readily learn distributional cues of name-object co-occurrences. But infants’ natural learning environment is cluttered: every heard word has multiple competing referents in view. Here we ask how infants start learning name-object co-occurrences in naturalistic learning environments that are cluttered and where there is much visual ambiguity. The framework presented in this paper integrates a naturalistic behavioral study and an application of a machine learning model. Our behavioral findings suggest that in order to start learning object names, infants and their parents consistently select a set of a few objects to play with during a set amount of time. What emerges is a frequency distribution of a few toys that approximates a Zipfian frequency distribution of objects for learning. We find that a machine learning model trained with a Zipf-like distribution of these object images outperformed the model trained with a uniform distribution. Overall, these findings suggest that to overcome referential ambiguity in clutter, infants may be selecting just a few toys allowing them to learn many distributional cues about a few name-object pairs.

Spoken language often includes ambiguity in meaning.Compounds such as “green teacup” can be interpreted with twodifferent meanings: “green colored teacup” and “cup for greentea.” We can assume there are two different underlyingsyntactic structures. Phonetic aspects have been studied in thedisambiguation process of such ambiguous phrases, but theroles of nonlinguistic information such as gestures have notbeen explored yet. We investigated whether people usegestures differently when they were asked to describe themeanings of Japanese compounds that can be interpreted astwo different meanings. We found that the timing of gesturesin relation to the target words of accompanying speech wasdifferent between right branching compounds and leftbranching compounds. Gestures seem to be used to suggestupcoming two words (adjective and noun) as a unit inbranching. Gestures can be a useful means to disambiguate themeanings of compounds.

Despite decision science have increased our understanding of human decision-making in different contexts, voters’ decision has been studied less from this point of view. Therefore, we investigated, how electorate- and candidate-related factors affect electorate’s (N=1334) valuation to the Prime Minister candidates (N=11) on the multiparty democracy. Electorates valuated candidates individually and through pairwise candidate comparison. We collected the data by using anonymous questionnaire and sent it via mass emailing and social media. We applied linear mixed-effects and Bayesian network models to analyze the data. Electorate-related variable Valence and candidate-related variables Trustworthiness and Righteousness was found as the strongest main effects. The pairwise analysis comparison highlighted voters’ personal characteristic. In particular, the interactions associated to valence, arousal and gender had high effect only in pairwise comparisons. Our results suggest that the pairwise comparisons - which is typical for elections, e.g., in USA - highlights the importance of emotional and gender-related factors.

Seasonal patterns vary dramatically around the world, andwe explore the extent to which systems of season categoriessupport efficient communication about the local environment.Our analyses build on a domain-general information-theoreticmodel of categorization across languages, and we identify sev-eral qualitative predictions that emerge when this model is ap-plied to season naming, including the prediction that systemswith even numbers of categories should be more common thansystems with odd sizes. We test the model quantitatively usinga collection of season systems drawn from the linguistic andanthropological literature and data specifying temperature andprecipitation in locations associated with these systems. Ourresults support the predicted even-odd asymmetry, and we alsofind that the model makes a number of successful predictionsabout the locations of boundaries between seasons.

A substantial body of research has demonstrated that childrenand adults (both native and non-native speakers) are sensitiveto the statistics of multiword sequences (MWS) and rely onknowledge of such statistics to facilitate their language pro-cessing and boost their acquisition. However, this researchwas primarily aimed at determining whether and to what ex-tent speakers can develop sensitivity to MWS statistics of asingle type of linguistic input: that of spoken language. Re-cently, there has been a growing awareness of the key role ofwritten input in the development of linguistic knowledge, as itprovides a source of substantial change in the statistics of anindividual’s language experience. The present study reports ona series of experiments designed to determine whether secondlanguage learners of English are able to develop sensitivity todistributional statistics of MWS inherent in different (register-specific) input types.

Methodological advances have made it possible to generatefMRI predictions for cognitive architectures, such as ACT-R, thus expanding the range of model predictions and mak-ing it possible to distinguish between alternative models thatproduce otherwise identical behavioral patterns. However, fortasks associated with relatively brief response times, fMRI pre-dictions are often not sufficient to compare alternative models.In this paper, we outline a method based on effective connec-tivity, which significantly augments the amount of informationthat can be extracted from fMRI data to distinguish betweenmodels. We show the application of this method in the caseof two competing ACT-R models of the Stroop task. Althoughthe models make, predictably, identical behavioral and BOLDtime-course predictions, patterns of functional connectivity fa-vor one model over the other. Finally, we show that the samedata suggests directions in which both models should be re-vised.

What mechanisms underlie people’s ability to use cross-situational statistics to learn the meanings of words? Here wepresent a large-scale evaluation of two major models of cross-situational learning: associative (Kachergis, Yu, & Shiffrin,2012a) and hypothesis testing (Trueswell, Medina, Hafri, &Gleitman, 2013). We fit each model individually to over 1500participants across seven experiments with a wide range ofconditions. We find that the associative model better capturesthe full range of individual differences and conditions whenlearning is cross-situational, although the hypothesis testingapproach outperforms it when there is no referential ambiguityduring training.

Capturing the effects of fatigue and, more generally, the effectsof physical and mental states on human performance has beena topic of research for many years. Recent models, especiallythose developed in a cognitive architecture, have shown greatpromise in capturing these effects by providing insight into thespecific cognitive and other components involved in taskperformance (like perception and motor movement). Inparticular, separate models have been developed to account forboth time-of-day and time-on-task effects related to fatigue. Inthis paper, we present a novel unified model, developed in theACT-R cognitive architecture, that captures both time-of-dayand time-on-task effects with a single set of mechanisms andparameters. We demonstrate how this unified model accountsfor quantitative and qualitative aspects of fatigued performancefrom two experiments, one focused on time-on-task effectsunder conditions of moderate fatigue, the other focusing ontime-of-day effects under conditions of severe fatigue in astudy of long-term (88-hour) sleep deprivation.

Locke argued that persons born blind do not possess true knowledge about color. While prior studies find some knowledgeof color among blind individuals, questions remain about the depth of this knowledge. Do blind individuals merely learninferentially shallow verbal associations (e.g., bananayellow)? We hypothesized instead that blind individuals are morelikely to acquire causally-relevant color information. Blind (n=20) and sighted adults (n=20) reported colors of naturalkinds (e.g. banana) and artifacts (e.g. car) and judged the likelihood that two instances of a type have the same color.Relative to the sighted, blind participants were less likely to know specific object colors (e.g. banana-yellow), but madeidentical inferences about color consistency (more consistent colors for natural kinds). Inferences were similar acrossgroups even for novel objects. Further, blind individuals gave detailed and coherent causal explanations of color origins.Inferentially rich knowledge of sensory categories can develop without first-person experience.

When several causes contributed to an outcome, we often single out one causal factor as being “more of a cause” than others. What explains this selection? Existing research suggests that people’s judgements of actual causation can be influenced by the degree to which they regard certain events as norm-deviant, or “abnormal” (Hart & Honoré, 1963; Kahneman & Miller, 1986; Hitchcock & Knobe, 2009; Halpern & Hitchcock 2015). In this paper, we argue that statistical abnormality influences causal judgements about human agents by changing the agents’ epistemic states (Epistemic Hypothesis). In Experiment 1, we replicate previous findings that people assign more causal strength to a statistically abnormally acting agent, but show that they also assign them more knowledge about the behaviour of their peers. In Experiment 2, we show that in case of equal epistemic uncertainty, people do not differentiate between statistically abnormal and normal causal agents. In Experiment 3, we explore the difference between type and token abnormality, and find that a token abnormal, but type normal behaviour still influences causal judgments, with people’s epistemic judgments mirroring these causal judgments. We discuss the implications of this research for current norm-frameworks in causal cognition.

We show that computational reinforcement learning can modelhuman decision making in the Iowa Gambling Task (IGT). TheIGT is a card game, which tests decision making under uncer-tainty. In our experiments, we found that modulating learningrate decay in Q-learning, enables the approximation of both thebehaviour of normal subjects and those who are emotionallyimpaired by ventromedial prefrontal lesions. Outcomes ob-served in impaired subjects are modeled by high learning ratedecay, while low learning rate decay replicates healthy sub-jects under otherwise identical conditions. The ventromedialprefrontal cortex has been associated with emotion based re-ward valuation, and, the value function in reinforcement learn-ing provides an analogous assessment mechanism. Thus rein-forcement learning can provide a good model for the role ofemotional reward as a modulator of the learning rate.

The adoption of conversational agents is growing at a rapidpace. Agents however, are not optimised to simulate key so-cial aspects of situated human conversational environments.Humans are intellectually biased towards social activity whenfacing more anthropomorphic agents or when presented withsubtle social cues. In this work, we explore the effects of simu-lating anthropomorphism and social eye-gaze in three conver-sational agents. We tested whether subjects’ visual attentionwould be similar to agents in different forms of embodimentand social eye-gaze. In a within-subject situated interactionstudy (N=30), we asked subjects to engage in task-orienteddialogue with a smart speaker and two variations of a socialrobot. We observed shifting of interactive behaviour by hu-man users, as shown in differences in behavioural and objec-tive measures. With a trade-off in task performance, socialfacilitation is higher with more anthropomorphic social agentswhen performing the same task.

The Rubber Hand Illusion (RHI) is an illusion of body ownership.This study investigates the RHI in furries: people who manifestinterest in anthropomorphic animals through various combinationsof costuming, roleplay, identification with a fursona, and unusualbodily experiences. Furry culture suggests two ways furries coulddiffer from non-furries in their RHI experience: (1) furries’malleable perception of bodily self and identity may result instronger feelings of illusory experience; alternatively, (2) furries’identification with non-human animals may result in weakerfeelings of self-ownership for a human prosthetic. Results supportthe latter hypothesis; furries felt less subjective embodimentcompared to non-furries. Moreover, proprioceptive drift waspredicted by the extent individual furries valued humanity and theirhuman bodies. The less esteem furries had for humanity and theirhuman form, the less drift toward the human rubber hand wasobserved. These findings suggest how embodiment is related tosubjectivity, identity, and practice.

Evaluations along a positivenegative dimension can be measured either explicitly (via self-report) or implicitly (via re-sponse interference tasks). Whether implicit evaluations encode relational information (e.g., A causes B) or only co-occurrence information (AB) has been debated extensively. 1,082 participants observed a machine being activated bycausally responsible stimuli and dispensing rewards in the presence of merely associated, but not causal, stimuli. Eval-uations of causally responsible vs. associated stimuli were measured implicitly and explicitly. Explicit and implicitevaluations of causally responsible stimuli were more positive than those of associated stimuli, both in the presence (Study1) and absence (Study 2) of verbal instructions about the operation of the machine. Study 3 eliminated temporal primacyand overshadowing as explanations of the effect. Supporting propositional theories, these findings suggest that implicitevaluations are sensitive not only to co-occurrence but also to relational information, whether conveyed verbally or learnedsolely from experience.

We report two artificial-language-learning experimentsinvestigating if the acquisition of sociolinguistic associations isfacilitated by two kinds of expectation violation: encounteringa variant (a) for the first time or (b) in an ungrammaticalcontext. Participants learned an artificial language with twodialects, each spoken by one of two alien species: Gulusand Norls. The two dialects differed with regard to a pluralsuffix: Gulus mostly used -dup, and Norls mostly used -nup.In the first learning phase, participants learned the languagewithout aliens; in the second learning phase, they wereexposed to it with alien interlocutors. In Experiment 1 wemanipulated whether -nup occurred in the first learning phase;in Experiment 2 we manipulated linguistic constraints on itsoccurrence. The acquisition of sociolinguistic association wasevaluated by asking participants to select suffixes given aliensand vice versa. We found that sociolinguistic acquisitionwas facilitated in Experiment 1, but not Experiment 2. InExperiment 2, however, a post hoc analysis revealed thatparticipants who had learned the grammatical context of thelinguistic conditioning did experience facilitation, while thosewho had not did not. Our results provide laboratory evidencethat unexpectedness facilitates the learning of sociolinguisticvariation.

People learn in fast and flexible ways that have not been emu-lated by machines. Once a person learns a new verb “dax,” heor she can effortlessly understand how to “dax twice,” “walkand dax,” or “dax vigorously.” There have been striking recentimprovements in machine learning for natural language pro-cessing, yet the best algorithms require vast amounts of experi-ence and struggle to generalize new concepts in compositionalways. To better understand these distinctively human abilities,we study the compositional skills of people through language-like instruction learning tasks. Our results show that peoplecan learn and use novel functional concepts from very fewexamples (few-shot learning), successfully applying familiarfunctions to novel inputs. People can also compose conceptsin complex ways that go beyond the provided demonstrations.Two additional experiments examined the assumptions and in-ductive biases that people make when solving these tasks, re-vealing three biases: mutual exclusivity, one-to-one mappings,and iconic concatenation. We discuss the implications for cog-nitive modeling and the potential for building machines withmore human-like language learning capabilities.

Subject of this article is the question whether the potential forautomatic defect analysis for symbolic timed ACT-R models asdemonstrated in earlier work can be developed into a scalable andcomprehensible technique. We present a formal, operational modelof an ACT-R architecture and a translation scheme of ACT-Rmodels into timed automata. We have applied this translationto ACT-R models and report on scalability experiments withautomatic defect analysis.

In this study we investigated the role of conceptually rich explanations and anomalous evidence in children’s scientific belief revision. We also explored whether the order in which children experience these two learning opportunities influences their belief revision ability. Five-year-old children were assigned to one of two conditions, where they either first received conceptual explanations about buoyancy and then observed anomalous data in a guided activity (Explanation- First), or the reverse (Anomalies-First). Results showed that (1) conceptually rich explanations lead to more accurate predictions about which objects sink and which float than anomalous data presentation, and (2) when explanations and anomalous data were combined, children’s correct predictions increased significantly from pre-test to post-test when they received the conceptual information before the anomalous evidence (Explanation-First), but not in the opposite order condition (Anomalies-First). These results suggest that children are more likely to maintain their misconceptions when exposed to anomalies without prior instruction involving conceptually rich explanations.

During their first year, infants learn to name objects. To do so, they need to segment speech, extract the label and map it to the correct referent. While children successfully do so in the wild, previous results suggest they struggle to simultaneously learn segmentation and object-label pairings in the lab. Here, we ask if some of children’s difficulty is related to the uniform distribution they were exposed to, since it differs from that of natural language, and has high entropy (making it less predictable). Will a low entropy distribution facilitate children’s performance in these two tasks? We looked at children’s (mean age=10;4 years) simultaneous segmentation and object-label mapping of words in an artificial language task. Low entropy (created by making one word more frequent) facilitated children's performance in both tasks. We discuss the importance of using more ecologic stimuli in the lab, specifically- distributions with lower entropy.

The number-line task is a widely used task in diverse fields of study. In the task, a given number that varies every trial isestimated on a continuum flanked with 0 and an upper-bound number. An upper-bound of a number-line is often arbitrarilyselected by researchers, although this design variable has been shown to affect the non-linearity in estimates. Examiningestimates of varying given numbers (design variable 1) with varying upper-bound numbers (design variable 2) can be costlybecause adding a new design dimension into a number-line task could drastically increase the number of trials requiredfor examining the underlying representation of number. The present study aims to conduct a number-line task with thegiven number and the upper-bound being the design variables. A design optimization algorithm, Gaussian Process ActiveLearning (GPAL), made this new paradigm feasible without increasing the number of trials, by presenting only the mostinformative combinations of the design variables every trial. Our experimental data showed that the non-linearity of thenumber-line estimates increases with the upper-bound of the number line. The degree of non-linearity could predict a mathskill (i.e., addition proficiency), but only when the upper-bound was relatively large. The observed range-dependency of thenumber-line estimates would not be fully explored without systematically manipulating the upper-bound as an additionaldesign variable. As in the present number-line task, GPAL would be a useful tool for the research problems that requiremultidimensional design experiments to be solved.

The ability to reason about our own and others’ competenceinforms our everyday decisions. However, competence is anabstract concept which manifests in the objective properties ofthe task completed by an agent (i.e., task-based features, suchas quality of outcome or task difficulty) as well as the sub-jective properties of the agent (i.e., agent-based features, suchas dexterity, speed, focus). Thus, acquiring an integrated no-tion of competence may be a nontrivial challenge for youngchildren. Prior work on children’s understanding of compe-tence has often used explicit verbal cues to describe the rele-vant features, or experimental tasks that confounded these fea-tures. Here we examine how preschool-aged children evalu-ate the relative competence of two agents by systematicallymanipulating task-based and agent-based features without ex-plicit linguistic or gestural support. We find that 4- and 5-year-olds readily use perceptual cues to task-based (i.e., taskdifficulty) and agent-based (i.e., agent speed) features to in-fer competence (Exp.1-3) but not when when these perceptualcues are closely matched (Exp.4). These results suggest thata basic understanding of relative competence emerges earlierthan previously believed, but an abstract, adult-like concept ofcompetence may continue to develop throughout childhood.

Observers rapidly extract summary statistics from sets ofvisually presented items, like the mean size of a set of circles,or the mean expression of a set of faces. Their excellentability to report summary statistics stands in contrast to near-chance representation of any of the individuals. Here weasked to what extent this ‘ensemble perception’ signatureextends to a more abstract property: relations amongelements. Participants watched ten unique animations ofvisually patterned objects (hereafter, ‘shapes’) colliding witheach other and producing a new shape. Collisions conformedto ABA patterns, such that the result shape always matchedone of the collider shapes. Recognition tests showed thatparticipants accurately recalled the collisions they saw, butalso falsely accepted foils which conformed to the ABApattern but which were not in fact specifically seen (wererearrangements of the original shapes across collisions). Onthe other hand, they were much less likely to accept foilswhich did not conform to the pattern, but were equallydistinct rearrangements (e.g., AAB). This suggests thatparticipants represented the overall, common pattern betterthan the specifics of what they saw; the superior encoding ofthe summary relative to the individuals thus applies tosummaries of relations. However, in contrast to prior findingswith visual features, we did not find that recall of individualpatterns was entirely at chance. Our paradigm offers a way topursue future questions such as the pressures and motivationswhich might govern the trade-off between summarizingevidence vs. retaining individual experiences.

Young children learn language at an incredible rate. Whilechildren come prepared with powerful statistical learningmechanisms, the statistics they encounter are also prepared forthem: Children learn from caregivers motivated to communi-cate with them. Do caregivers modify their speech in orderto support children’s comprehension? We asked children andtheir parents to play a simple reference game in which the par-ent’s goal was to guide their child to select a target animal froma set of three. We show that parents calibrate their referringexpressions to their children’s language knowledge, produc-ing more informative references for animals that they thoughttheir children did not know. Further, parents learn about theirchildren’s knowledge over the course of the game, and cali-brate their referring expressions accordingly. These results un-derscore the importance of understanding the communicativecontext in which language learning happens.

The development of self-adaptive systems has attracted lots ofattention as they can adapt themselves autonomously to en-vironmental dynamics and maintain user satisfaction. How-ever, there are still tremendous challenges remained. One ma-jor challenge is to guarantee the reusability of the system andextend the adaptability with the changing deployment environ-ments. Another challenge is to ensure the adaptability copingwith the open and complex environments with the existence ofunknown. To solve these problems, we introduce a concep-tual self-adaptive model, decoupling the environment with thesystem. This model is a two-layer structure, based on internalcauses and external causes from attribution theory. The firstlayer, determining how the internal causes affect the adapta-tion behaviors, is independently designed and reusable; whilethe second layer, mapping the relationship between externalcauses with internal causes, is replaceable and dynamicallybound to different deployment environments.

Explanations not only increase understanding; they are oftendeeply satisfying. In the present research, we explore how thisphenomenological sense of “explanatory satisfaction” relatesto the functional role of explanation within the process ofinquiry. In two studies, we address the following questions: 1)Does explanatory satisfaction track the epistemic, learning-directed features of explanation? and 2) How doesexplanatory satisfaction relate to both antecedent andsubsequent curiosity? In answering these questions, weuncover novel determinants of explanatory satisfaction andcontribute to the broader literature on explanation and inquiry.

When predicting or explaining another person’s actions, weoften appeal to the physical effort they require; a person whoworks hard for something, for instance, must really like it (Liu,Ullman, Tenenbaum, & Spelke, 2017). But people are notonly motivated to avoid physical effort; they also seek to avoidmental effort (Shenhav et al., 2017; Kool & Botvinick, 2018).Here, we ask whether mental effort enters into preschoolers’understanding of other people’s actions. Across 4 experiments(N=112), we presented 4- and 5-year-old children with anagent (naive in Exp 1, 2 and 4, and knowledgeable in Exp 3)who can either move through a simple or complex maze envi-ronment with a specific goal (in Exp 1-3, to reach a play struc-ture beyond the mazes, and in Exp 4, to practice solving themazes). We found that children were sensitive to the physicaland mental effort associated with more complex mazes, and tothe trade-offs between effort and gain in skill. The intuitionthat choices impose costs on our bodies and minds appears toguide children’s understanding of other people.

Our everyday decisions are driven by costs, risk, and reward.How do people take these factors into account when they pre-dict and explain the decisions of others? In a two-part exper-iment, we assessed people’s perceptions of other people’s riskpreferences, relative to their own. In Part 1, participants re-ported their relative preference between a guaranteed payoutand lotteries with various probabilities and payouts, and madepredictions about other people’s preferences. In Part 2, partic-ipants estimated the lottery payout that generated a given rela-tive preference between a guaranteed payout and a lottery, bothfor themselves and others. We found considerable individualvariability in how people perceive the risk preferences of oth-ers relative to their own, and consistency in people’s percep-tions across our two measures. Future directions include for-mal computational models and developmental studies of howwe think about our own and each other’s decision-making.

The unacceptability of wh-extraction (e.g., question formation) out of certain syntactic structures, known as ‘island’ effects, has been a central topic in theoretical syntax for many years (Ross, 1967; Chomsky, 1973). A prominent example of islands is that extraction out of a sentential complement introduced by factive and manner-of-speaking verbs (‘What did John know/whisper that Mary bought?’) is less acceptable than extraction from a clause introduced by “bridge” verbs (‘What did John say that Mary bought?’). We aimed to replicate Ambridge and Goldberg (2008) who argued that extraction from a sentential complement is unacceptable in proportion to its discourse salience. We failed to replicate their results and found that there is no true island effect for such structures: instead there are separate, additive penalties based on two factors: (a) verb-frame frequency (cf. Dabrowska, 2008), and (b) the presence of extraction. These penalties give rise to apparent island effects as a result of the nonlinear relationship between true acceptability and acceptability ratings as measured in Likert scales and forced- choice tasks.

According to theories of anticipatory behavior control, actionplanning and control is realized by activating desired goalstates. From an event-predictive perspective, this activationshould focus sensorimotor processing on expected, upcomingevent boundaries. Previous studies have shown that periper-sonal hand space (PPHS) is remapped to the future hand lo-cation in a grasping task before the movement commences.Here, we investigated if the current hand posture interfereswith the anticipatory remapping of PPHS. Participants had tograsp virtual bottles from two differently oriented starting pos-tures. During the prehension, they received a vibrotactile stim-ulus on their right index finger or on their thumb, while a vi-sual stimulus appeared at the bottle, either matching the futurefinger position, or not. Participants had to name the stimu-lated finger. While the hand posture affected verbal responsetimes, the anticipatory remapping remained unchanged. Ap-parently, the predictive processes that realize the anticipatoryremapping, generalize over initial hand postures.

How do children’s visual concepts change across childhood,and how might these changes be reflected in their drawings?Here we investigate developmental changes in children’s abil-ity to emphasize the relevant visual distinctions between objectcategories in their drawings. We collected over 13K drawingsfrom children aged 2-10 years via a free-standing drawing sta-tion in a children’s museum. We hypothesized that older chil-dren would produce more recognizable drawings, and that thisgain in recognizability would not be entirely explained by con-current development in visuomotor control. To measure recog-nizability, we applied a pretrained deep convolutional neuralnetwork model to extract a high-level feature representation ofall drawings, and then trained a multi-way linear classifier onthese features. To measure visuomotor control, we developedan automated procedure to measure their ability to accuratelytrace complex shapes. We found consistent gains in the recog-nizability of drawings across ages that were not fully explainedby children’s ability to accurately trace complex shapes. Fur-thermore, these gains were accompanied by an increase in howdistinct different object categories were in feature space. Over-all, these results demonstrate that children’s drawings includemore distinctive visual features as they grow older.

How do humans compute approximate number? According to one influential theory, approximate number representationsarise in the intraparietal sulcus and are amodal (independent of any sensory modality). Alternatively, approximate numbermay be computed initially within sensory systems. We tested for approximate number representations in the visual systemusing steady state visual evoked potentials (SSVEPs). We recorded EEG from human subjects while they viewed dotcloudspresented at 30 Hz. Alternating the dotcloud numerosity at 15 Hz evoked a 15 Hz SSVEP detectable over the occipital lobe(Oz). The SSVEP amplitude increased as the numerical difference between dotclouds increased, indicating that subjectsvisual systems were differentiating dotclouds on the basis of their numerical ratios. Critically, subjects were unable toconsciously discriminate dotcloud numerosity, indicating the rapid presentation disrupted reentrant feedback to visualcortex. Approximate number appears to be computed within the visual system, independently of higher-order areas suchas the intraparietal sulcus.

In this paper, we describe three experiments involving simplephysical judgments and predictions, and argue their results aregenerally inconsistent with three core commitments of proba-bilistic mental simulation theory (PMST). The first experimentshows that people routinely fail to track the spatio-temporalidentity of objects. The second experiment shows that peopleoften incorrectly reverse the order of consequential physicalevents when making physical predictions. Finally, we demon-strate a physical version of the conjunction fallacy where par-ticipants rate the probability of two joint events as more likelyto occur than a constituent event of that set. These results high-light the limitations or boundary conditions of simulation the-ory.

This study examined age effects on real-life language use and within-person variations in language use across social contexts. We used the Electronically Activated Recorder (i.e., a portable audio recorder that periodically records sound snippets) to collect over 31,300 snippets (30 seconds long) from 61 young and 48 healthy older adults in Switzerland across four days. We examined vocabulary richness and grammatical complexity across the social contexts of (a) activities (i.e., socializing, working); and (b) conversation types (i.e., small talk, substantive conversation). Multilevel models showed that vocabulary richness and grammatical complexity increased during socializing and substantive conversations, but decreased in small talk. Moreover, young adults produced shorter clauses at work than not at work. Furthermore, compared with young adults, older adults used richer vocabulary and more complex grammatical structures at work; and used richer vocabulary in small talk. In contrast, young adults used richer vocabulary than older adults during non-socializing and non-working occasions, such as watching TV and exercising. Results are discussed in the context of cognitive aging research with a novel emphasis on context.

Extensive research on probability learning has reported on the ubiquity of the probability matching strategy—choosing options in proportion to their probability of being correct. The current paper explores why the optimal strategy in this task (always choosing the higher probability option) is not intuitive for participants, by examining their decisions in relation to their working memory capacities. We hypothesize that probability matching is a by-product of an automatic recency-based strategy produced by limits in working memory storage and that deliberate strategizing mediated by working memory processing can override recency in favor of optimal responding. A variant of the Expectancy-Valence Learning Model is fit to participant data from a two-choice probability learning task using hierarchical Bayesian modelling. Point estimates of the best-fitting parameter values are then correlated with working memory measures. Results indicate close relations between them, providing support for our hypothesis.

Sensorimotor information plays a fundamental role incognition. However, datasets of ratings of sensorimotorexperience have generally been restricted to several hundredwords, leading to limited linguistic coverage and reducedstatistical power for more complex analyses. Here, we presentmodality-specific and effector-specific norms for 39,954concepts across six sensory modalities (touch, hearing, smell,taste, vision, and interoception) and five action effectors(mouth/throat, hand/arm, foot/leg, head excluding mouth, andtorso), which were gathered from 4,557 participants whocompleted a total of 32,456 surveys using Amazon'sMechanical Turk platform. The dataset therefore representsone of the largest set of semantic norms currently available.We describe the data collection procedures, provide summarydescriptives of the data set, demonstrate the utility of thenorms in predicting lexical decision times and accuracy, aswell as offering new insights and outlining avenues for futureresearch. Our findings will be of interest to researchers inembodied cognition, cognitive semantics, sensorimotorprocessing, and the psychology of language generally. Thescale of this dataset will also facilitate computationalmodelling and big data approaches to the analysis of languageand conceptual representations.

Pervasive behavioral and neural evidence for predictiveprocessing has led to claims that language processing dependsupon predictive coding. In some cases, this may reflect aconflation of terms, but predictive coding formally is acomputational mechanism where only deviations from top-down expectations are passed between levels of representation.We evaluate three models’ ability to simulate predictiveprocessing and ask whether they exhibit the putative hallmarkof formal predictive coding (reduced signal when inputmatches expectations). Of crucial interest, TRACE, aninteractive activation model that does not explicitly implementprediction, exhibits both predictive processing and model-internal signal reduction. This may indicate that interactiveactivation is functionally equivalent or approximant topredictive coding, or that caution is warranted in interpretingneural signal reduction as diagnostic of predictive coding.

It is well established that readers form mental images when reading a narrative. The influence of mental imagery on theway people experience stories is however still unclear. In two experiments reported here, participants received instructionsaimed at encouraging or discouraging mental imagery before reading literary short stories. After reading, participantsanswered questions about their reading experiences. The results from the first experiment suggested an important roleof mental imagery in determining reading experiences. However, the results from the second experiment showed thatindividual trait differences in how imaginative participants are predicted reading experiences much better than guidedmental imagery. Moreover, the role of mental imagery did not extend to aspects of the reading experience other thanmental imagery. The implications of these results for the relationship between mental imagery and reading experiencesare discussed.

Embodied cognition theories predict that changing motor control would change cognitive control, as cognition is considered to emerge from action in this theoretical approach. We tested this prediction, by examining the attention and cognitive control capabilities of a group of school students (12-13-year-olds) trained to write using both hands (experimental group, N=28), compared to a group of age- matched children (control group, N=33) who did not receive such training. The key tasks used were the attentional network test (ANT) task and the hearts and flowers (HF) task. Results from the ANT task showed that there was no significant difference in the three attentional networks between the groups. However, results from the HF task showed that the experimental group had better inhibitory control. This second result provides support to the embodied cognition prediction that cognitive control and motor control are related, and the former can be changed to some extent by changing the latter.

Languages partition semantic space into linguistic cate-gories in systematic ways. In this study, we investigatea semantic space which has received sustained attentionin theoretical linguistics: person. Person systems con-vey the roles entities play in the conversational context(i.e., speaker(s), addressee(s), other(s)). Like other lin-guistic category systems (e.g. color and kinship terms),not all ways of partitioning the person space are equallylikely. We use an artificial language learning paradigm totest whether typological frequency correlates with learn-ability of person paradigms. We focus on first personsystems (e.g., ‘I’ and ‘we’ in English), and test the predic-tions of a set of theories which posit a universal set of fea-tures (±exclusive, and ±minimal) to capture this space.Our results provide the first experimental evidence forfeature-based theories of person systems.

How does cross-linguistic variation in grammatical structureaffect children’s acquisition of number words? In this study,we addressed this question by investigating the case study ofyoung speakers of French, a language in which the number oneand the indefinite article a are phonologically the same (i.e.,un). We tested how French-speaking children interpret un, andwhether it more closely resembles the English word a or one.We found that French-speaking children almost alwaysaccepted sets of 1 for un, but that their responses for sets of 2were more equivocal, with many children saying “Oui” (Yes)when asked whether there was un. Overall, French children’sinterpretation of un differed from how English-speakinginterpret both a and one. This suggests that French-speakingchildren’s interpretation of un reflects the ambiguity of theinput that they are exposed to. We conclude that Frenchmorphological structure may pose a challenge to French-speaking children in acquiring an exact numerical meaning forthe word un, potentially causing a delay in number wordlearning.

One of the pinnacles of human cognition is the creative insightof expert mathematics. While its concepts are abstract, theactual practice of mathematics is undeniably material andembodied. Mathematicians draw, sketch, write; having createdthese inscriptions, they interact with them. This iterated processof inscription is the engine of mathematical discovery. But howdoes this engine work? Here, using a new video corpus ofmathematical experts working on proofs, and deploying toolsfrom network and complexity science, we characterize thestructure and temporal dynamics of how mathematical expertscreate and interact with blackboard inscriptions. We findregularities in the structure of this activity (e.g., emergent‘communities’ of inscriptions) and its temporal dynamics (e.g.,‘bursty’ shifts in attention). By characterizing this activity, wegain a better understanding of the distributed ecosystem inwhich mathematical creativity occurs — including the ways thatmathematicians actively construct their own notational niches.

A growing body of research indicates that “active learning” im-proves episodic memory for material experienced during study.It is less clear how active learning impacts the integration ofthose experiences into flexible, generalizable knowledge. Thisstudy used a novel active transitive inference task to investi-gate how people learn a relational hierarchy through activeselection of premise pairs. Active control improved memoryfor studied premises as well as transitive inferences involv-ing items that were never experienced together during study.Active learners also exhibited a systematic search preference,generating sequences of overlapping premises that may fa-cilitate relational integration. Critically, however, advantagesfrom active control were not universal: Only participants withhigher working memory capacity benefited from the opportu-nity to select premise pairs during learning. These findingssuggest that active control enhances integrative encoding ofstudied material, but only among individuals with sufficientcognitive resources.

Users sometimes face anomalous behaviors of systems, such asmachine failures and autonomous agents. Predicting suchbehaviors of systems is difficult. We investigate the benefits ofthe memory-based strategy, which focuses on memorization ofinstances to predict anomalous and regular behaviors of thesystem, with ACT-R simulations with a cognitive model. Inthis study, we presumed the parameters defining the encodingprocesses on anomalous instances and regular instances in themodel of the memory-based strategy and performedsimulations to verify how these two parameters influenceprediction performance. The results of simulations showed that(1) regular instances are not encoded as default values in thememory-based strategy and that (2) such inactivity on regularinstances suppresses commission errors of regular instancesand does not suppress commission errors of anomalousinstances nor omission errors.

A central question for cognitive science is whether children’s linguistic productivity can be captured by item-based learning, or whether the learner must be guided by abstract, system-wide principles governed by innate constraints. Here, we present a computational model of early language acquisition which learns to discover and use lexically-based frames in a fully incremental, on-line fashion. The model is rooted in simple prediction- and recognition-based processes, subject to the same memory limitations as language learners. When exposed to English corpora of child-directed speech, the model is able learn developmentally plausible frames and use them to capture over 70% of the utterances produced by target children aged 2 to 5. Across a typologically diverse range of 29 languages, the model is able to capture over 68% of child utterances. Together, these findings suggest that much of children’s early linguistic productivity can be captured by item-based learning through computationally simple mechanisms.

Subject-auxiliary inversion in interrogatives has been a topic of great interest in language acquisition research, and has often been held up as evidence for the structure-dependence of grammar. Usage-based and nativist approaches posit different representations and processes underlying children’s question formation and therefore predict different causes for these errors. Here, we explore the question of whether input statistics predict children’s spontaneous non-inversion errors with wh- questions. In contrast to previous studies, we look at properties of the non-inverted, errorful forms of questions. Through a series of corpus analyses, we show that the frequency of uninverted subsequences (e.g., “she is going” in “what she is going to do?*”) is a good predictor of children’s errors, consistent with recent evidence for multiword units in children’s comprehension and production. This finding has implications for the types of mental representations and cognitive processes researchers ascribe to children acquiring a first language.

Human-level natural language understanding (NLU) of opentext is far beyond the current state of the art. In practice, ifdeep NLU is attempted at all, it is within narrow domains. Wereport a program of R&D on cognitively modeled NLU thatworks toward depth and breadth of processing simultaneous-ly. The current contribution describes lessons learned – scien-tifically and methodologically – from an exercise in applyingdeep NLU to open-domain texts. An overarching lesson wasthat although learning to compute sentence-level semanticsseems like a natural step toward computing full, context-sensitive, semantic and pragmatic meaning, corpus evidenceunderscores just how infrequently semantics can be cleanlyseparated from pragmatics. We conclude that a more compre-hensive methodology for automatic example selection and re-sult validation is needed as prerequisite for success in devel-oping NLU applications operating on open text.

A wealth of developmental evidence suggests that children es-sentialise natural kind but not artifact categories, and that bothadults and children use generic language less with artifacts aswell (Gelman, 2003). Here we further explore the latter resultusing a novel model for generic identification. We apply ourmodel to a much larger dataset than before, consisting of 26CHILDES corpora of naturalistic speech involving children ata variety of ages and in a variety of contexts. We found noconsistent preference for generic usage in animates over arti-facts. Follow-up analyses indicate that this result was probablydriven by our inclusion of a wider variety of nouns into ourdataset than previous work.

High-Variability Phonetic Training (HVPT) has been shown tobe effective in improving the perception of even the hardestsecond-language (L2) contrasts. However, little is known as towhether such training can improve phonological processing atthe lexical level. The present study tested whether this type oftraining also improves word recognition. Adult proficientFrench late learners of English completed eight online sessionsof HVPT on the perception of English word-initial /h/. Thissound does not exist in French and has been shown to bedifficult to process by French listeners both on the prelexical(Mah, Goad & Steinhauer, 2016) and the lexical level (Melnik& Peperkamp, 2019). In pretest and posttest participantscompleted an identification task as well as a lexical decisiontask. The results demonstrated that after training the learners’accuracy had improved in both tasks. The theoretical andapplied implications are discussed.

Evidence is typically consistent with more than one hypothesis.How do we decide which hypothesis to pursue (e.g., to subjectto further consideration and testing)? Research has shown thatexplanatory considerations play an important role in learningand inference: we tend to seek and favor hypotheses thatoffer good explanations for the evidence we invoke them toexplain. Here we report three studies testing the proposal thatexplanatory considerations similarly inform decisions concern-ing pursuit. We find that ratings of explanatory goodness predictpursuit (though to a lesser extent than they predict belief), andthat these effects hold after adjusting for subjective probability.These findings contribute to a growing body of work suggestingan important role for explanatory considerations in shapinginquiry.

Although most Americans support capital punishment, manypeople have misconceptions about its efficacy andadministration (e.g., that capital punishment deters crime). Cancorrecting people’s inaccurate attitudes change their support forthe death penalty? If not, are there other strategies that mightshift people’s attitudes about the death penalty? Some researchsuggests that statistical information can correct misconceptionsabout polarizing topics. Still, statistics might be irrelevant forsome people because they may support capital punishment forpurely retributive reasons, suggesting other argumentativestrategies may be more effective. In Studies 1 and 2, weexamined what attitudes shape endorsement of capitalpunishment and compared how two different interventionsshifted these attitudes. Altogether, our findings suggest thatattitudes about capital punishment are based on more than justretributive motives, and that correcting misconceptions relatedto its administration reduces support for capital punishment.

Repeated purchase decisions often violate assumptions of stan-dard economic or rational choice models, such as demonstrat-ing asymmetric or unstable responses to changes in underlyingpolicy, price, or tax variables. I propose a novel frameworkfor how such decisions can be interpreted through the lens of acognitive process model. This provides psychologically inter-pretable characterizations of individuals or population groups.It incorporates mental accounting, hedonic adaptation, confir-mation bias, and the influence of perceived trust and fairness.It shows how sequential experiences and contextual aspectssuch as political affiliation, are mediated by this cognitive pro-cess to produce evolving consumption patterns. This novel ap-proach can account for empirically observed violations of con-ventional choice models. The model is quantitatively fit to ex-perimental data for individual purchase decisions and demon-strates improved descriptive, predictive, and inference capabil-ities. A proof-of-concept analysis using this model to accountfor real world consumption trends is also demonstrated.

Developmental theory considers action prediction as one of several processes involved in determining how infants come to perceive and understand social events (Gredebäck & Daum, 2015). Action prediction is observed from early in life and is considered an important social-cognitive skill. However, knowledge about infant action prediction is limited to evidence from screen-based eye-tracking tasks. Little is known about action prediction in real-life action contexts. Our aim in the current study was to provide new evidence on whether and how infants anticipate actions in free-flowing parent-child interaction. Using dual head-mounted eye- tracking, we analyzed infants’ visual anticipations of their parents’ reaching actions while they played with objects together. Findings reveal that infants anticipate their parents’ actions at a rate higher than would be expected by chance.

This study investigated how beat gesture and contrastive pitchaccenting affect online contrastive reference resolution duringspoken discourse comprehension. Evidence from gazefixations indicated that beat gesture encouraged fixations totarget referents of contrastive referring expressions and thatcontrastive accenting encouraged fixations to competitorreferents of non-contrastive referring expressions. Notably,beat gesture and contrastive accenting acted independently,indicating that their effects are additive rather than interactive.Moreover, neither beat gesture nor contrastive accentingaffected an observed tendency to anticipate contrastivereferring expressions. Together, these results provide the firstevidence that beat gesture, like contrastive accenting, isinterpreted as a cue to contrast during online referenceresolution in spoken discourse comprehension.

One of the most fundamental and striking limitations of hu-man cognitive function is the constraint on the number ofcontrol-dependent processes that can be executed simultane-ously. However, the sources of this capacity constraint re-main largely unexplored. Previous work has attributed the con-straints on control-dependent processing to the sharing of rep-resentations between tasks in neural systems. Here, we exam-ine how shared representations interact with two other factorsin producing constraints on control-dependent processing. Wefirst demonstrate that the detrimental effects of shared repre-sentations on multitasking performance are contingent on theamount of conflict that is induced by the tasks that share rep-resentations. We then examine how the persistence of sharedrepresentations between tasks affects processing interferenceduring serial task execution. Finally, we discuss how this set ofmechanisms can account for various phenomena in neural ar-chitectures, including the psychological refractory period, taskswitch costs, as well as constraints on cognitive control.

There are two main types of bias in simple decision tasks,response bias and stimulus bias. Response bias is a startinglevel of evidence in favor of a biased response, whereas stim-ulus bias is the evaluation of stimuli in favor of a biased re-sponse. Previous research typically dissociates between thesetwo types of bias. Some studies suggest that it can be diffi-cult to induce response bias without stimulus bias (Ratcliff &McKoon, 2008; van Ravenzwaaij, Mulder, Tuerlinckx, & Wa-genmakers, 2012). We used a two-alternative forced-choicebrightness discrimination task in which we manipulated thepresentation length of the stimuli. We analyzed the data witha hierarchical diffusion model. The results show an overall re-sponse bias, as well as stimulus bias that increases as stimuluspresentation time decreases. We argue that the results suggesta need to revise how stimulus bias is conceptualized throughthe drift rate parameter of the diffusion model.

The concept of Nash equilibrium has played a profound rolein economics, and is widely accepted as a normative stance forhow people should choose their strategies in competitive envi-ronments. However, extensive empirical evidence shows thatpeople often systematically deviate from Nash equilibrium. Inthis work, we present the first resource-rational mechanisticapproach to one-shot, non-cooperative games (ONG), show-ing that a variant of normative expected-utility maximizationacknowledging cognitive limitations can account for impor-tant deviations from the prescriptions of Nash equilibrium inONGs. Concretely, we show that Nobandegani et al.’s (2018)metacognitively-rational model, sample-based expected util-ity, can account for purportedly irrational cooperation rates ob-served in one-shot, non-cooperative Prisoner’s Dilemma, andcan accurately explain how cooperation rate varies dependingon the parameterization of the game. Additionally, our workprovides a resource-rational explanation of why people withhigher general intelligence tend to cooperate less in OPDs, andserves as the first (Bayesian) rational, process-level explana-tion of a well-known violation of the law of total probability inOPDs, documented by Shafir and Tversky (1992), which hasresisted explanation by a model governed by classical proba-bility theory for nearly three decades. Surprisingly, our workdemonstrates that cooperation can arise from purely selfish,expected-utility maximization subject to cognitive limitations.

The St. Petersburg paradox is a centuries-old philosophicalpuzzle concerning a lottery with infinite expected payoff,on which people are, nevertheless, willing to place only asmall bid. Despite many attempts and several proposals, nogenerally-accepted resolution is yet at hand. In this work, wepresent the first resource-rational process-level explanation ofthis paradox, demonstrating that it can be accounted for by avariant of normative expected-utility-maximization which ac-knowledges cognitive limitations. Specifically, we show thatNobandegani et al.’s (2018) metacognitively-rational model,sample-based expected utility (SbEU), can account for majorexperimental findings on this paradox. Crucially, our resolu-tion is consistent with two empirically well-supported assump-tions: (1) people use only a few samples in probabilistic judg-ments and decision-making, and (2) people tend to overesti-mate the probability of extreme events in their judgment.

How does cooperation arise in an evolutionary context? We ap-proach this problem using a collective search paradigm whereinteractions are dynamic and there is competition for rewards.Using evolutionary simulations, we find that the unconditionalsharing of information can be an evolutionary advantageousstrategy without the need for conditional strategies or explicitreciprocation. Shared information acts as a recruitment sig-nal and facilitates the formation of a self-organized group.Thus, the improved search efficiency of the collective bestowsbyproduct benefits onto the original sharer. A key mecha-nism is a visibility radius, where individuals have uncondi-tional access to information about neighbors within a lim-ited distance. Our results show that for a variety of initialconditions—including populations initially devoid of prosocialindividuals—and across both static and dynamic fitness land-scapes, we find strong selection pressure to evolve uncondi-tional sharing.

We experimentally investigate, by means of a novel gesture-elicitation paradigm, the spontaneous spatial frames ofreference (FoRs) used by bilingual individuals who speakJapanese (which has been labeled as a “relative” language)and one of the endangered Ryukyuan languages (Miyako orShiraho) whose speakers have been reported to routinely useabsolute FoRs. How would these last elderly bilingualspeakers spontaneously resolve the clashing FoRs the twolanguages they speak bring forth? We find that despite thefact that Japanese and these Ryukyuan languages have fullcorresponding grammatical and lexical resources forexpressing both, relative and absolute FoR, Ryukyuanspeakers tend to markedly prefer the latter gesturally.Methodologically, the results, which are consistent with dataobtained with standard FoRs methods, corroborate thereliability of the novel gesture elicitation task, which adds tothe battery of techniques for studying FoRs a method thatassesses effortless spontaneous real-time cognition with highecologically validity.

The human ability to deceive others and detect deception haslong been tied to theory of mind. We make a stronger argu-ment: in order to be adept liars – to balance gain (i.e. maxi-mizing their own reward) and plausibility (i.e. maintaining arealistic lie) – humans calibrate their lies under the assumptionthat their partner is a rational, utility-maximizing agent. Wedevelop an adversarial recursive Bayesian model that aims toformalize the behaviors of liars and lie detectors. We comparethis model to (1) a model that does not perform theory of mindcomputations and (2) a model that has perfect knowledge ofthe opponent’s behavior. To test these models, we introduce anovel dyadic, stochastic game, allowing for quantitative mea-sures of lies and lie detection. In a second experiment, we varythe ground truth probability. We find that our rational modelsqualitatively predict human lying and lie detecting behaviorbetter than the non-rational model. Our findings suggest thathumans control for the extremeness of their lies in a mannerreflective of rational social inference. These findings provide anew paradigm and formal framework for nuanced quantitativeanalysis of the role of rationality and theory of mind in lyingand lie detecting behavior.

Even when we are not faced with any decision, we sometimesengage in offline cognition where we simulate various possi-ble actions we can take. In these instances, which options dowe tend to simulate? Computational models have suggestedthat it is better to focus our limited cognitive resources to-wards simulating and refining our representations of optionsthat appear, at first blush, to have higher values. Two exper-imental studies explore whether we use this strategy. Partic-ipants went through an ‘offline’ thinking phase, and an ‘on-line’ decision-making phase. Participants first freely viewedvarious options, which they had to simulate to determine theiractual values. They were later asked to decide between goodor bad options. Offline simulation produced faster online re-sponse times for the options that appeared to have highervalues, indicating a pre-computation benefit for these items.These results suggest that people focus their offline cognitionon the apparently good.

Risk and uncertainty are inherent in life, and how people perceive, respond to, and manage both are topics of great aca-demic interest. One critical insight is that people distinguish between types of uncertainty (see, e.g., Fox & lkmen, 2011)and, consequently, may respond to objectively equally probabilistic events differently (e.g., with more polarized predic-tions of those events outcomes). The current work identifies another way in which risk (a specific form of uncertainty)is differentiated: on the basis of causal depth (Sloman, Love, & Ahn, 1998). Specifically, in contexts where an uncertainoutcome (e.g., win/lose) is determined by a causal chain, people tend to prefer for the uncertainty to arise at lower causaldepth within the chain (i.e., at later causal stages). This occurs even though the causal depth at which the uncertainty arisesmakes no difference in the overall probability that the causal chain will generate one outcome or another.

Despite their diversity, languages around the world share aconsistent set of properties and distributional regularities. Forexample, the distribution of word frequencies, the distributionof syntactic dependency lengths, and the presence of ambigu-ity are all remarkably consistent across languages. We dis-cuss a framework for studying how these system-level proper-ties emerge from local, in-the-moment interactions of rational,pragmatic speakers and listeners. To do so, we derive a novelobjective function for measuring the communicative efficiencyof linguistic systems in terms of the interactions of speakersand listeners. We examine the behavior of this objective ina series of simulations focusing on the communicative func-tion of ambiguity in language. These simulations suggest thatrational pragmatic agents will produce communicatively effi-cient systems and that interactions between such agents pro-vide a framework for examining efficient properties of lan-guage structure and use more broadly.

Many real-world situations involve learning entirely or mostlybased on the information provided by other people, which cre-ates a thorny epistemological problem: how does one deter-mine which of those people to trust? Previous work has shownthat even populations of rational Bayesian agents, faced withthis problem, polarise into “echo chambers” characterised bydifferent beliefs and low levels of between-group trust. In thisstudy we show that this general result holds even when thereasoners have a more complex meaning space and can com-municate about their beliefs in a more nuanced way. However,even a tiny amount of exposure to a mutually trusted “groundtruth” is sufficient to eliminate polarisation. Societal and psy-chological implications are discussed.

Previous research with typically developing (TD) childrenand adults show an advantage of active control for episodicmemory as compared to conditions lacking this control. Thepresent study attempts to replicate this effect in autisticchildren. Six- to 12-year-old autistic children (n = 30) wereinstructed to remember as many of 64 presented objects aspossible. For half of the materials presented, participantscould decide the order and pacing of study (Active condition).For the other half, they passively observed the study decisionsof a previous participant (Yoked condition). We found thatrecognition memory was more accurate for objects studied inthe active as compared to the yoked condition, even after aweek-long delay. The magnitude of the effect was comparableto that obtained in previous studies with TD children andadults, suggesting a strong robustness for the benefits ofactive learning. We discuss how pedagogical approaches maybe encouraged to utilize self-directed learning strategies topromote inclusive learning.

How does one deal with the possibility of deception? Extant literature has mostly focused on identifying deception via cue detection. However, how we reason about the possibility of deception remains under-explored. We use a novel formalism to expose the complexity of this reasoning problem (e.g. separating the uncertainty of an honest mistake, from willful deception), in the process highlighting several reasoning errors regarding deception. Notably, we show reasoners to make substantial errors when reasoning about a (possibly) deceptive source in isolation (including base rate neglect errors), but find that reasoning improves when further (independently sourced) corroborative or contradicting reports are introduced.

Recent research (Pilditch, Fenton, & Lagnado, 2019) shows that people are susceptible to zero-sum thinking in evidence evaluation, where they dismiss or underweight the probative value of evidence that is equally predicted by multiple independent hypotheses. But such an assumption is only valid when explanations are mutually exclusive and exhaustive. The present work extends these findings by looking at the context of information selection, and the decisional consequences of the zero-sum fallacy. It uses an information metric to quantify the cost of the error in terms of overlooked information.

We provide new evidence concerning two views of episodic associations: The independent associations hypothesis (IAH)posits that associations are unidirectional and separately modifiable links (AB; AB); the associative symmetry hypothesis(ASH) considers the association to be a holistic conjunction of A and B representations. While existing literature focuseson tests that compare the correlation of forward and backward associations and favors ASH over IAH, we provide thefirst direct evidence of IAH by showing that forward and backward associations are separately modifiable for semanti-cally related pairs. In two experiments, participants studied 30 semantically unrelated and 30 semantically related pairsintermixed in a single list, and then performed a series of up to eight cued-recall test cycles. All pairs were tested in eachcycle, and the testing direction (A-? or B-?) alternated between cycles. Consistent with prior research, unrelated pairsexhibited associative symmetry accuracy and response times improved gradually on each test, suggesting that testing inboth directions strengthened the same association. In contrast, semantically related pairs exhibited a stair-like pattern,where performance did not change from odd to even tests when the test direction changed; it only improved between testsof the same direction. We conclude that episodic associations can have either a holistic representation (ASH) or separatedirectional representations (IAH), depending on the semantic relatedness of their constituent items.

Normative word frequency has played a key role in the study of human memory, but there is little agreement as to themechanism responsible for its effects. To determine whether word frequency affects binding probability or memoryprecision, we examined working memory for spatial positions of words. Each of three experiments included 300 trialsin which five words were presented sequentially around an invisible circle followed by one of those words shown in themiddle of the circle as a probe to test its location. Participants had to click on the associated location and the degree oferror around the circle was the dependent measure. Across experiments we varied word frequency, presentation rate andthe proportion of low frequency words on each trial. A mixture model dissociated memory precision, binding failure andguessing rates from the continuous distribution of errors. On trials that contained only low- or high-frequency words,low-frequency words lead to a greater degree of error in recalling the associated location. This was due to a higher word-location binding failure and not due to differences in memory precision or guessing rates. Slowing down the presentationrate eliminated the word frequency effect by reducing binding failures for low-frequency words. Mixing frequencies in asingle trial hurt high-frequency and helped low-frequency words, but frequency composition and presentation rate did notinteract. These findings support the idea that low-frequency words require more resources for binding and that the bindingfails when these resources are insufficient.

A growing body of literature suggests that making differentsampling assumptions about how data are generated can leadto qualitatively different patterns of inference based on thatdata. However, relatively little is known about how samplingassumptions are represented or when they are incorporated.We report the results of a single category generalisation exper-iment aimed at exploring these issues. By systematically vary-ing both the sampling cover story and whether it is given beforeor after the training stimuli we are able to determine whetherencoding or retrieval issues drive the impact of sampling as-sumptions. We find that the sampling cover story affects gen-eralisation when it is presented before the training stimuli, butnot after, which we interpret in favour of an encoding account.

“No Valid Conclusion” (NVC) is one of the most frequently se-lected responses in syllogistic reasoning experiments and cor-responds to the logically correct conclusion for 58% of thesyllogistic problem domain. Still, NVC is often neglected incomputational models or just treated as a byproduct of theunderlying inferential mechanisms such as a last resort whenthe search for alternatives is exhausted. We illustrate thatNVC represents a major shortcoming of current models for hu-man syllogistic reasoning. By introducing heuristic rules, wedemonstrate that slight extensions of the existing models resultin substantial improvements of their predictive performances.Our results illustrate the need for better NVC handling in cog-nitive modeling and provide directions for modelers on how tointegrate it into their approaches.

Verbs are widely analyzed as functions taking a discrete number of arguments (e.g., drink has two arguments but give has three). Recent studies, however, suggest that English verbs encode Instruments as more or less salient (e.g., the Instrument is more salient for slice, less salient for eat). We conducted a judgment task with adult speakers of Spanish and Mandarin and found that verbs in these languages also encode Instruments as having a relative degree of salience, inconsistent with the discrete model of participant encoding.

Traditional models of cognitive control account for a host ofclassic findings, but these classic tasks have limited our abil-ity to test a broader range of model predictions. In particu-lar, such models predict that control should vary parametricallyin response to cognitive demands and that control adjustmentsshould be targeted towards task-relevant stimulus features. Wedeveloped a task to probe these predictions across two exper-iments. Participants responded to one dimension of a stim-ulus while ignoring the other, and we parametrically variedthe conflict between those dimensions and the predictability ofthis conflict across trials. We found that control adjustments(1) varied parametrically in response to cognitive demands,(2) were sensitive to the predictability of those demands, and(3) were primarily targeted towards task-irrelevant dimensions.These results raise interesting questions about the structure ofcognitive control and demonstrate the utility of rich tasks forconstraining model predictions.

We provide a definition of ‘memory’ that is broad enough to apply to both natural and artificial systems. Inspired by computation and information theory, we define memory as a process that preserves information through time while maintaining its usefulness as an object to be computed. We defend the extensiveness of our definition by explaining how it applies to both brains and modern computers. We then consider potential objections to our definition. Our primary goal is to provide a definition of ‘memory’ that is broadly applicable across various cognitive sciences subfields.

The study of question asking in humans and machines hasgained attention in recent years. A key aspect of question ask-ing is the ability to select good (informative) questions froma provided set. Machines—in particular neural networks—generally struggle with two important aspects of question ask-ing, namely to learn from the answer to their selected ques-tion and to flexibly adjust their questioning to new goals. Inthe present paper, we show that people are sensitive to both ofthese aspects and describe a unified Bayesian account of ques-tion asking that is capable of similar ingenuity. In the first ex-periment, we predict people’s judgments when adjusting theirquestion-asking towards a particular goal. In the second ex-periment, we predict people’s judgments when deciding whatfollow-up question to ask. An alternative model based on su-perficial features, such as the existence of certain key wordsin the questions, was not able to capture these judgments to areasonable degree.

Music training is commonly thought to have a positive impacton children’s cognitive skills and academic achievement. Thisbelief relies on the idea that engaging in an intellectuallydemanding activity helps to foster overall cognitive function.We here present a meta-analysis of music-intervention studiesin children (N = 3,780, k = 204, m = 43). Consistent with thesubstantial findings in the field of cognitive training, the overalleffect size was small (

A foundational goal of linguistics is to investigate whethershared features of the human cognitive system can explainhow linguistic patterns are distributed across languages. Inthis study we report a series of artificial language learning ex-periments to test a hypothesised link between cognition and apersistent regularity of morpheme order: number morphemes(e.g., plural markers) tend to be ordered closer to noun stemsthan case morphemes (e.g., accusative markers) (Greenberg,1963). We argue that this typological tendency may be drivenby a bias favouring orders that reflect scopal relationships inmorphosyntactic composition (Bybee, 1985; Rice, 2000; Cul-bertson & Adger, 2014). We taught participants an artificiallanguage with noun stems, and case and number morphemes.Crucially, the input language indicated only that each mor-pheme preceded or followed the noun stem. Examples inwhich two (overt) morphemes co-occurred were held out—i.e.,no instances of plural accusatives. At test, participants wereasked to produce utterances, including the held-out examples.As predicted, learners consistently produced number closer tothe noun stem than case. We replicate this effect with freeand bound morphemes, pre- or post-nominal placement, andwith English and Japanese speakers. However, we also findthat this tendency can be reversed when the form of the casemarker is conditioned on the noun, suggesting an influence ofdependency length. Our results provide evidence that univer-sal features of cognition may play a causal role in shaping therelative order of morphemes.

The unique cumulative nature of human culture has often beenexplained by high-fidelity copying mechanisms found only inhuman social learning. However, transmission chain exper-iments in human and non-human primates suggest that cu-mulative cultural evolution (CCE) might not be dependent onhigh-fidelity copying after all. In this study we test whetherCCE is possible even with a non-copying task. We performedtransmission chain experiments in Guinea baboons and chil-dren where individuals observed and reproduced visual pat-terns on touch screen devices. In order to be rewarded, par-ticipants had to avoid touching squares that were touched bya previous participant. In other words, they were regardedfor innovation rather than copying. Results nevertheless ex-hibited two fundamental properties of CCE: an increase overgenerations in task performance and the emergence of sys-tematic structure. However, CCE arose from different mecha-nisms across species: children, unlike baboons, converged inbehaviour over generations by copying specific patterns in adifferent location, thus introducing alternative copying mech-anisms into the non-copying task. We conclude that CCE canresult from non-copying tasks and that there is a broad spec-trum of possible mechanisms that will lead to CCE aside fromhigh-fidelity transmission.

Dotov, Nie and Chemero (2010) conducted a set of exper-iments to demonstrate how phenomenology, particularly thework of Martin Heidegger, interfaces with experimental re-search in embodied cognitive science. Specifically, they drew aparallel between Heidegger’s notion of readiness-to-hand andthe concept of an extended cognitive system (Clark 2008) bylooking for the presence or absence of interaction-dominantdynamics (Holden, van Orden, and Turvey 2009; Ihlen andVereijken 2010) in a hand/mouse system. We share Dotov,Nie and Chemero’s optimism about the potential for cross-pollination between phenomenology and cognitive science, butwe think that it can be better advanced through a shift in fo-cus. First, we argue in favor of using Maurice Merleau-Ponty’sphenomenological theory as the philosophical foundation forexperimental research in embodied cognitive science. Sec-ond, we describe an audio-visual tracking task in virtual realitythat we designed and used to empirically investigate human-environment coupling and interactivity. In addition to provid-ing further support for phenomenologically-inspired empiricalcognitive science, our research also offers a more generaliz-able scientific treatment of the interaction between humans andtheir environments.

The successor function a recursive function S which states that for every natural number n, S(n) = n+1 underlies ourunderstanding of the natural numbers as an infinite class. Recent work has found that acquisition of this logical propertyis surprisingly protracted, completed several years after children master the counting procedure. While such work linkssuccessor knowledge with counting mastery, the exact processes underlying this developmental transition remain unclear.Here, we examined two possible mechanisms: (1) recursive counting knowledge, and (2) formal training with the +1 rulein arithmetic. We find that while both recursive counting and arithmetic mastery predict successor knowledge, arithmeticperformance is significantly lower than measures of recursive counting for all children. This dissociation suggests childrendo not generalize the successor function from trained mathematics; rather, we find evidence consistent with the hypothesisthat successor knowledge is supported by the extraction of recursive counting rules.

Parental input in the form of visual joint attention is hypothesized to serve a critical role in the development of infant attention, acting as a training ground by scaffolding an infant’s ability to sustain visual attention in real-time. We extended this hypothesis by studying the effects of parent speech on infant visual and manual attention. Thirty-four toddlers and their parents participated in a free-play study while wearing head-mounted eye trackers. Infant multimodal behaviors were measured in four ways: visual attention, manual action, hand-eye coordination, and joint visual attention with their parent. Overall, we found that longer durations of attention were accompanied by parent speech. Moreover, sustained attention, defined as behaviors lasting 3s or more, almost always occurred with parent speech. Individual differences in parent-infant coordination were also explored. These results suggest that parent-infant interactions create multimodal opportunities for infants to practice sustaining attention.

Memorizing and retrieving information about the spatial layoutof one’s surrounding is of crucial importance for humans. Wepropose a new theory of spatial memory of immediate envi-ronments and develop a corresponding computational realiza-tion. We detail how the theory explains key findings on humanspatial memory (use) and show that the computational real-ization accounts well for human behavior from three pertinentexperiments. One implication of the theory’s success is thatenduring spatial memory representations may best be concep-tualized as flexible combinations of representation structuresand reference frames.

Memory researchers have studied learning behavior andextracted regularities describing learning and forgetting overtime. Early work revealed forgetting curves and the benefitsof temporal spacing and testing for learning. Computationalmodels formally implemented these regularities to capturerelevant trends over time. As these models improved, theywere applied to adaptive learning contexts, where learningprofiles could be identified from responses to past learningevents to predict and improve future performance. Often times,past performance is expressed as accuracy alone. Here weexplore whether a model’s predictions can be improved ifpast performance is expressed by an integrated measure thatcombines accuracy and response times (RT). We present asimple, data-driven method to combine accuracy and RT on atrial-by-trial basis. This research demonstrates that predictionsmade using the Predictive Performance Equation improvewhen past performance is expressed as an integrated measurerather than accuracy alone.

In Joint Action (JA) tasks, individuals must coordinate theiractions so as to achieve some desirable outcome at the group-level. Group function is an emergent outcome of ongoing,mutually constraining interactions between agents. Here weinvestigate JA in dyads of improvising jazz pianists. Partic-ipants’ musical output is recorded in one of two conditions:a real condition, in which two pianists improvise together asthey typically would, and a virtual condition, in which a singlepianist improvises along with a “ghost partner” – a recordingof another pianist taken from a previous real trial. The con-ditions are identical except for that in real trials subjects aremutually coupled to one another, whereas there is only unidi-rectional influence in virtual trials (i.e. recording to musician).We quantify ways in which the rhythmic structures sponta-neously produced in these improvisations is shaped by mutualcoupling of co-performers. Musical signatures of underlyingcoordination patterns are also shown to parallel the subjectiveexperience of improvisers, who preferred playing in trials withbidirectional influence despite not explicitly knowing whichcondition they had played in. These results illuminate howmutual coupling shapes emergent, group-level structure in thecreative, open-ended and fundamentally collaborative domainof expert musical improvisation.

This study develops a cognitive model to explain the processof artistic creation in a dance domain. Many researchers in thefield of psychology and cognitive science have investigatedthe process of creativity and developed various theories thatexplain this process. Their efforts have mostly focused onhigher cognitive functions of artists and scientists. However,in recent years, several studies that have highlighted theimportance of the interaction between idea generation andidea externalization processes suggest that people can findand develop new aspects of images and ideas by perceivingand reflecting on the images and ideas they externalize. Thisstudy develops a cognitive model that explains this interactionprocess in dance creation by referring to a famous theory ofmotor learning, the closed-loop model. We also investigatedance creation of an expert breakdancer and check thevalidity of our proposed model.

Humans demonstrate remarkable abilities to perceive physi-cal and social events based on very limited information (e.g.,movements of a few simple geometric shapes). However, thecomputational mechanisms underlying intuitive physics andsocial perception remain unclear. In an effort to identify thekey computational components, we propose a unified psycho-logical space that reveals the partition between the perceptionof physical events involving inanimate objects and the percep-tion of social events involving human interactions with otheragents. This unified space consists of two prominent dimen-sions: an intuitive sense of whether physical laws are obeyedor violated; and an impression of whether an agent possessesintentions, as inferred from movements. We adopt a physicsengine and a deep reinforcement learning model to synthe-size a rich set of motion patterns. In two experiments, humanjudgments were used to demonstrate that the constructed psy-chological space successfully partitions human perception ofphysical versus social events.

Do some cultures think more abstractly than others? According to tests of formal logic and rule-based reasoning, Western-ers tend to think more abstractly than East Asians. Yet, rule-based reasoning is only one type of abstract thinking. Moregenerally, thinking abstractly involves discerning relationships and seeing the big picture. Here we argue that previous testsof attention, perception, and memory can be interpreted as showing that East Asians tend to think more abstractly thanWesterners. To test this hypothesis directly we gave a validated measure of abstract thinking (Vallacher & Wegner, 1989)to Chinese and US individuals. Participants chose either abstract or concrete definitions of events. Across six indepen-dent national samples (total N=1,798), Chinese participants tended to construe events more abstractly, and US participantsmore concretely. Within China, more independent (Western-like) groups chose more concrete definitions. Together, theseresults challenge the generalization that Westerners have a greater propensity for abstract thought.

Whereas first language (L1) research has demonstrated thatperceptions of creative ability are influenced by the complexityand diversity of language used to answer verbal tests ofcreativity, relatively little is known about the linguisticcomponents of bilingual creative task performance. In thisstudy, we analyze written transcripts of speech produced by466 Japanese learners of English produced during a creativenarrative task for features related to linguistic and cognitivedimensions of creativity. Then, we extract various linguisticfeatures and test whether these features can predict humanperceptions of creativity for the transcripts. Unlike L1 data,results suggest text length and L2 proficiency comprise themost parsimonious explanation of creativity scores in this L2data. At the same time, linguistic features related to positivesentiment explained a significant yet small amount ofadditional variance in perceptions of creativity, suggestingtexts with more positive language were perceived to be morecreative.

Interruptions are an inevitable part of every day life. Previousresearch suggests that interruptions can decrease performanceand increase errors and response time. Additionally, there isevidence that providing a lag time prior to an interruption canmitigate some of the interruption costs. The goal of this pa-per is to investigate the effects of interruptions and interrup-tion lags and explore possible strategies to attenuate interrup-tion costs. A novel sequential decision-making paradigm wasused, where the difficulty of the task and type of interruptionwere the two experimental manipulations. The results indicatethat there is a potential benefit to including a lag time whenpresented with interruptions.

Backup plans represent a safety net that can help ensure goal attainment. However, managing backup plans during goalpursuit can also deflect attention away from the initial plan. We examined how individuals sense of power, which is saidto facilitate goal pursuit, affects the extent to which one gets distracted by backup plans. Results from four studies showedthat when a backup plan was activated, greater sense of power was associated with lower self-reported distraction and betterperformance. Studies 2 and 3 further revealed mediating effects of distraction between sense of power and performance.Greater sense of power was associated with less distraction, which in turn was related to better performance. Our findingssuggest that when pursuing goals, individuals experiencing high power may be better at allocating their limited cognitiveresources to the initial plan.

The current study explored the impact that “academic” shame had on learning of the human circulatory system. Participants were randomly assigned to one of two conditions: a shame induction condition or a control condition (no shame induction). Results revealed that the shame induction manipulation was related to higher levels of state shame. Additionally, it was discovered that by and large “in the moment” shame and having a proneness to experiencing shame dampened down any subsequent learning. Implications to education and future research are discussed.

Notably, while English tends to prefer shorter before longercomplements (explained to us a very clear effect), Japanesedisplays the opposite tendency. Far less cross-linguistic workhas investigated possible differences in the ordering of nounswithin conjunctions (“binomials’), although a corpus studysuggests that the same factors predict binomial ordering inJapanese and English. To investigate the issue experimentally,we report Japanese and English speakers’ productions of namesof the members of couples that they knew personally. Resultsconfirm that conceptual accessibility is the most importantfactor in the ordering of familiar name binomials in bothlanguages. That is, both groups tended to name the memberthey felt closer to first. Length (syllables/mora) was not asignificant predictor in either language. Differences in thepreferred order of verbs’ complements are then attributable toother factors, possibly a very general preference to minimizethe average distance between semantically related elements.

How may intentionality, the capacity of mental states to beabout the world, emerge from neural processes? We proposea set of theoretical concepts that enable a simulated agent tohave intentional states as it perceives, acts, memorizes, plans,and builds beliefs about a simulated environment. The con-cepts are framed within Dynamic Field Theory (Sch ̈oner et al.,2015), a mathematical language for neural processes modelsat the level of networks of neural populations. Inspired bySearle’s analysis of the two directions of fit of intentional states(Searle, 1980), we recognize that process models of intentionalstates must detect the match of the world to the mind (for “ac-tion” intentions) or the match of the mind to the world (for“perceptual” intentions). Neural representations of Searle’scondition of satisfaction implement these detection decisionsthrough dynamic instabilities that are instrumental in enablingautonomous switches among intentional states.

It is well known that people tend to anthropomorphize in inter-pretations and explanations of the behavior of robots and otherinteractive artifacts. Scientific discussions of this phenomenontend to confuse the overlapping notions of folk psychology,theory of mind, and the intentional stance. We provide a clarifi-cation of the terminology, outline different research questions,and propose a methodology for making progress in studyingthe intentional stance toward robots empirically.

Iconic words are known to exhibit an imitative relationshipbetween a word and its referent. Many studies have workedto pinpoint sound-to-meaning correspondences for ideophonesfrom different languages. The correspondence patterns showsimilarities across languages, but what makes such language-specific correspondences universal, as iconicity claims to be,remains unclear. This could be due to a lack of consensus onhow to describe and test the perceptuo-motor affordances thatmake an iconic word feel imitative to speakers. We created andanalyzed a database of 1,888 ideophones across 13 languages,and found that 5 articulatory properties, physiologically acces-sible to all spoken language users, pattern according to seman-tic features of ideophones. Our findings pave the way for futureresearch to utilize articulatory properties as a means to test andexplain how iconicity is encoded in spoken language.

Cumulative cultural evolution is a distinctively human formof information-processing that endows our societies with im-probable and efficient technologies. But how objective is thisprocess? A widely held conjecture is that human cognitivebiases can constrain cumulative cultural evolution, and there-fore shape our discoveries. We present a Bayesian analysis ofa simple form of cumulative cultural evolution. This modelallows us to formulate and test the theoretical conjecture inan experimental setting. Across a series of behavioural ex-periments, we show that people’s inductive biases constrain apopulation’s ability to discover counter-intuitive virtual tech-nologies in a simple search problem. Our analysis highlightsformal relationships between cumulative cultural evolution,Bayesian inference, and stochastic optimization.

Constrained by results from classic behavioral experiments weprovide a neural-level cognitive architecture for modeling be-havior in working memory tasks. We propose a canonicalmicrocircuit that can be used as a building block for work-ing memory, decision making and cognitive control. The con-troller controls gates to route the flow of information betweenthe working memory and the evidence accumulator and setsparameters of the circuits. We show that this type of cognitivearchitecture can account for results in behavioral experimentssuch as judgment of recency, probe recognition and delayed-match-to-sample. In addition, the neural dynamics generatedby the cognitive architecture provides a good match with neu-rophysiological data from rodents and monkeys. For instance,it generates cells tuned to a particular amount of elapsed time(time cells), to a particular position in space (place cells) andto a particular amount of accumulated evidence.

Semantic knowledge can facilitate or distort new memories,depending on their alignment. We aimed to quantifydistortions in memory by examining how categorymembership biases new encoding. Across two experiments,participants encoded and retrieved image-locationassociations on a 2D grid. The locations of images weremanipulated so that most members of a category (e.g. birds)were clustered near each other, but some were in randomlocations. Memory for an item’s location was more precisewhen it was near members of the same category.Furthermore, typical category members’ retrieved locationswere more biased towards their semantic neighbors, relativeto atypical members. This demonstrates that the organizationof semantic knowledge can explain bias in new memories.

It is well-known that in free recall participants tend to recallwords presented close together in time in sequence, reflectinga form of temporal binding in memory. This contiguity effectis robust, having been observed across many different experi-mental manipulations. In order to explore a potential boundaryon the contiguity effect, participants performed a free recalltask in which items were presented at rates ranging from 2 Hzto 8 Hz. Participants were still able to recall items even atthe fastest presentation rate, though accuracy decreased. Im-portantly, the contiguity effect flattened as presentation ratesincreased. These findings illuminate possible constraints onthe temporal encoding of episodic memories.

When introducing a novel mathematical idea, should wepresent learners with abstract or concrete examples of thisidea? Considerable efforts have been made over the last decadeto settle this question in favor of either abstract or concreterepresentations. We contribute to this discussion through acritical replication and extension of a well-known study in thisarea. Whereas the target article argues for the generalsuperiority of abstract representations, we demonstrate thatseemingly minor modifications of the study design indicateotherwise. Our results suggest that the previously reported“advantage of abstract examples” manifested not becauseabstract examples are advantageous in general, but because theearlier studies utilized concrete examples that arepedagogically suboptimal.

Ambiguity pervades language. One prevalent kind ofambiguity is indirect requests. For example, “My office isreally hot” could be intended not only as a complaint aboutthe temperature, but as a request to turn on the AC. How docomprehenders determine whether a speaker is making arequest? We ask whether the prosody of an utterance providesinformation about a speaker’s intentions. In a behavioralexperiment, we find that human listeners can identify whichof two utterances a speaker intended as a request, suggestingthat speakers can produce discriminable cues. We then showthat the acoustic features associated with an utterance allow aclassifier to detect the original intent of an utterance (74%accuracy). Finally, we ask which of these features predictlistener accuracy on the behavioral experiment.

People vary in what they know, yet models of languageprocessing do not take this variability into account. Weharnessed the temporal sensitivity of event-related brainpotentials alongside individual differences in Harry Potter (HP)knowledge to investigate the extent to which the availabilityand timing of information relevant for real-time wordcomprehension are influenced by variation in degree of domainknowledge. We manipulated meaningful (category, event)relationships between sentence contexts about HP stories andcritical words (endings), assessed via behavioral ratings and bymeasuring similarity of word embeddings derived from a high-dimensional semantic model trained on HP texts. Individuals’ratings were sensitive to these relationships according to thedegree of their domain knowledge. During reading, N400amplitudes (neural measures of semantic retrieval) alsoreflected this variability, suggesting the degree to whichinformation relevant for word understanding is availableduring real-time sentence processing varies as a function ofindividuals’ domain knowledge.

Our knowledge of the world is an organized lexico-semantic network in which concepts can be linked by relations, such as “taxonomic” relations between members of the same stable category (e.g., cat and sheep), or association between entities that occur together or in the same context (e.g., sock and foot). Prior research has focused on the emergence of knowledge about taxonomic relations, whereas association has received little attention. The goal of the present research was to investigate how semantic organization development is shaped by both taxonomic relatedness and associations based on co-occurrence between labels for concepts in language. Using a Cued Recall paradigm, we found a substantial influence of co-occurrence in both 4-5-year-olds and adults, whereas taxonomic relatedness only influenced adults. These results demonstrate a critical and persistent influence of co- occurrence associations on semantic organization. We discuss these findings in relation to theories of semantic development.

When people behave impatiently, prioritizing sooner outcomes at the expense of latter ones, is it because they valueachieve their goal sooner, or because they value receiving the benefits sooner? Prior research has often confounded goalgradient (the stronger motivational effect of more proximal goals) and time discounting effects on decision-making. Wefirst establish a preference to invest in the earlier of two equally difficult goals (e.g, a first-goal preference) that could beexplained either by relative goal gradients or by differences in time discounted value. We then experimentally separatethe timing of goal completion and reward receipt. We find separate and disassociated large goal gradient and somewhatsmaller time discounting effects. Our results suggest that goal gradient effects may provide a partial, but substantial,explanation of time discounting and, consequently, can inflate estimated discount rates when not accounted for.

Most theories of kind representation suggest that people positinternal, essence-like factors believed to underlie kindmembership and the observable properties of members.Across two studies (N = 234), we show that adults canconstrue properties of social kinds as products of both internaland structural (stable external) factors. Internalist andstructural construals are similar in that both support formalexplanations (i.e., “category member has property P due tocategory membership C”), generic claims (“Cs have P”), anda particular pattern of generalization to individuals when theindividuals’ category membership and structural position arepreserved. Our findings thus challenge these phenomena assignatures of essentialist thinking. However, once categorymembership and structural position are unconfounded,different patterns of generalization emerge across internalistand structural construals, as do different judgmentsconcerning category definitions and property mutability.These findings have important implications for reasoningabout social kinds.

Most research on how children learn the mapping betweenwords and world has assumed that language is arbitrary, andhas investigated language learning in contexts in which objectsreferred to are present in the environment. Here, we reportanalyses of a semi-naturalistic corpus of caregivers talking totheir 2-3 year-old. We focus on caregivers’ use of non-arbitrarycues across different expressive channels: both iconic(onomatopoeia and representational gestures) and indexical(points and actions with objects). We ask if these cues are useddifferently when talking about objects known or unknown tothe child, and when the referred objects are present or absent.We hypothesize that caregivers would use these cues moreoften with objects novel to the child. Moreover, they would usethe iconic cues especially when objects are absent becauseiconic cues bring to the mind’s eye properties of referents. Wefind that cue distribution differs: all cues except points are morecommon for unknown objects indicating their potential role inlearning; onomatopoeia and representational gestures are morecommon for displaced contexts whereas indexical cues aremore common when objects are present. Thus, caregiversprovide multimodal non-arbitrary cues to support children’svocabulary learning and iconicity – specifically – can supportlinking mental representations for objects and labels.

Formal theories of grammar and traditional parsing models,insofar as they presuppose a categorical notion of grammar,face the challenge of accounting for gradient judgments ofacceptability. This challenge is traditionally met by explaininggradient effects in terms of extra-grammatical factors, positinga purely categorical core for the language system. We presenta new way of accounting for gradience in a self-organizedsentence processing (SOSP) model, which generates structureswith a continuous range of grammaticality values. We focuson islands, a family of syntactic domains out of whichmovement is generally prohibited. Islands are interestingbecause, although most linguistic theories treat them asfully ungrammatical and uninterpretable, experimental studieshave revealed gradient patterns of acceptability and evidencefor their interpretability. We report simulations in whichSOSP largely respects island constraints, but in certain cases,consistent with empirical data, coerces elements that blockdependencies into elements that allow them.

What are the categorical distinctions shared between conceptual and visual representations? One distinction may be between bounded and unbounded entities. Previous research in sign language has shown that even non-signers associate signs with repetitive motion with atelic verbs, such as “run”, and signs with sudden motion with telic verbs, such as “arrive”. In our first study, we show this distinction holds even when the visual stimuli depicted bear no intrinsic linguistic reference: we used non-linguistic random dot motions. In our second study, we demonstrate this association occurs spontaneously, even when subjects are not making explicit semantic judgments about verbs. We use a cross- modal lexical decision task in which verbs and non-words appear superimposed on bounded or unbounded dot stimuli. We find congruency when the motion boundedness matches the conceptual boundedness of the verb. Together, these studies provide evidence for an automatic link between visual and conceptual boundedness in the mind.

The ability to anticipate, attend and respond appropriately to specific stimuli is involved in the execution of everyday tasks. The current investigation examined the relations between cognitive skills measured by the NIH Toolbox and changes in the power of mu oscillations during anticipation of and in response to a tactile stimulus. Electroencephalographic (EEG) activity was measured after a visuospatial cue directed adults (n=40) to monitor their right or left hand for upcoming tactile stimulation. In the 500 ms prior to the onset of the tactile stimulus, a desynchronization was apparent 8 – 14 Hz at contralateral central sites, consistent with prior investigations of mu rhythm; a widespread synchronization was apparent in the 250 ms proceeding delivery of the tactile stimulus. The extent of contralateral reduction in mu power was associated with speed processing ability, while ipsilateral mu power was associated with flanker performance and marginally correlated with card sort performance. Regression further probe the significance and specificity of these effects. Increases in mu power following onset of the tactile stimulus were not associated with any behavioral measures. Mu modulation during attention to a specific bodily location appears related to variability in the broader ability to regulate behavior in a goal- directed manner, and perhaps to speed of stimulus processing.

Recurrent Neural Networks (RNNs) trained on a languagemodeling task have been shown to acquire a number of non-local grammatical dependencies with some success (Linzen,Dupoux, & Goldberg, 2016). Here, we provide new evidencethat RNN language models are sensitive to hierarchical syntac-tic structure by investigating the filler–gap dependency andconstraints on it, known as syntactic islands. Previous workis inconclusive about whether RNNs learn to attenuate theirexpectations for gaps in island constructions in particular orin any sufficiently complex syntactic environment. This papergives new evidence for the former by providing control studiesthat have been lacking so far. We demonstrate that two state-of-the-art RNN models are are able to maintain the filler–gapdependency through unbounded sentential embeddings and arealso sensitive to the hierarchical relationship between the fillerand the gap. Next, we demonstrate that the models are ableto maintain possessive pronoun gender expectations throughisland constructions—this control case rules out the possibil-ity that island constructions block all information flow in thesenetworks. We also evaluate three untested islands constraints:coordination islands, left branch islands, and sentential subjectislands. Models are able to learn left branch islands and learncoordination islands gradiently, but fail to learn sentential sub-ject islands. Through these controls and new tests, we provideevidence that model behavior is due to finer-grained expecta-tions than gross syntactic complexity, but also that the modelsare conspicuously un-humanlike in some of their performancecharacteristics.

Learning constitutes an essential part of human experienceover the life course. Independent of the domain, it ischaracterized by changes in performance. But what cognitivemechanisms are responsible for these changes and how dosituational features affect the dynamics? To inspect that inmore detail, this paper introduces a cognitive modelingapproach that investigates performance-related changes inlearning situations. It leverages the cognitive architectureACT-R to model learner behavior in an interrupted learningtask in two conditions of task complexity. Comparisons withthe original human dataset indicate a good fit in terms of bothaccuracy and reaction times. Although interruption effects aremore obvious in the human data, they are prevalent as well inthe model. Furthermore, the model can map the learningeffects, particularly in the easy task condition. Based on theexisting mapping of ACT-R module activity with fMRI data,simulated neural activity is computed to investigate underlyingcognitive mechanisms in more detail. The resulting evidenceconnects learning and interruption effects in both taskconditions with activation-related patterns to explain changesin performance.

It is unknown whether modality affects the efficiency with which humans learn novel word forms and their meanings, with previous studies reporting both written and auditory advantages. The current study implements controls whose absence in previous work likely offers explanation for such contradictory findings. In two novel word learning experiments, participants were trained and tested on pseudoword - novel object pairs, with controls on: modality of test, modality of meaning, duration of exposure and transparency of word form. In both experiments word forms were presented in either their written or spoken form, each paired with a pictorial meaning (novel object). Following a 20- minute filler task, participants were tested on their ability to identify the picture-word form pairs on which they were trained. A between subjects design generated four participant groups per experiment 1) written training, written test; 2) written training, spoken test; 3) spoken training, written test; 4) spoken training, spoken test. In Experiment 1 the written stimulus was presented for a time period equal to the duration of the spoken form. Results showed that when the duration of exposure was equal, participants displayed a written training benefit. Given words can be read faster than the time taken for the spoken form to unfold, in Experiment 2 the written form was presented for 300 ms, sufficient time to read the word yet 65% shorter than the duration of the spoken form. No modality effect was observed under these conditions, when exposure to the word form was equivalent. These results demonstrate, at least for proficient readers, that when exposure to the word form is controlled across modalities the efficiency with which word form-meaning associations are learnt does not differ. Our results therefore suggest that, although we typically begin as aural-only word learners, we ultimately converge on developing learning mechanisms that learn equally efficiently from both written and spoken materials.

How does time pressure influence attitudes towards uncer-tainty? When time is limited, do people engage in differentexploration strategies? We study human exploration in a rangeof four-armed bandit tasks with different reward distributionsand manipulate the available time for each decision (limitedvs. unlimited). Through multiple behavioral and model-basedanalyses, we show that reactions towards uncertainty are influ-enced by time pressure. Specifically, participants seek out un-certain options when time is unlimited, but avoid uncertaintyunder time pressure. Moreover, larger relative differences inuncertainty between options slowed down reaction times anddampened the drift rate of a linear ballistic accumulator model.These results shed new light on the differential effect of uncer-tainty and time pressure on human exploration.

Prior work on early social learning suggests that children are sensitive to adults pedagogical demonstrations and ver-bal instructions. Yet, people also display various emotional expressions when interacting with children. Here we showthat young children draw rich causal inferences and guide their own exploration based on others expressions of surprise.Preschoolers (age:3.0-4.9) saw an experimenter discover a function of a novel causal toy. Then, either the same experi-menter or a nave confederate expressed surprise while playing with the toy behind an occluder. Children explored the toymore broadly to search for a hidden function following the experimenters surprise than following the confederates surprise,suggesting that children integrated others expressions of surprise and others epistemic states to infer the presence of hiddenfunctions and explore accordingly. This study synthesizes perspectives from literature on social learning, exploration, andaffective cognition towards a more comprehensive science of learning. Preprint:https://psyarxiv.com/ckh6j

It has been proposed that language evolves under the joint con-straints of communicative expressivity and cognitive ease. Weexplore this idea in the historical emergence of word forms.We hypothesize that new word forms that enter the lexiconshould reflect a trade-off between predictability and distinc-tiveness. An emergent word form can be highly predictable ifit efficiently reuses elements from the existing word forms, re-sulting in low cognitive load. An emergent word form shouldalso be sufficiently distinctive from the existing lexicon, facil-itating communicative expressivity. We test our hypothesis byexamining the properties of 34,478 emergent word forms overthe past 200 years of Modern English. We show how wordforms at future time t + 1 are bounded statistically betweenn-gram generated word forms (highly predictable) and slangwords that are outside the standard lexicon (highly distinctive)at time t. Our work supports the view of cognitive economy inlexical emergence.

The ability to estimate task difficulty is critical for many real-world decisions such as setting appropriate goals for ourselvesor appreciating others’ accomplishments. Here we give a computational account of how humans judge the difficultyof a range of physical construction tasks (e.g., moving 10 loose blocks from their initial configuration to their targetconfiguration, such as a vertical tower) by quantifying two key factors that influence construction difficulty: physical effortand physical risk. Physical effort captures the minimal work needed to transport all objects to their final positions, and iscomputed using a hybrid task-and-motion planner. Physical risk corresponds to stability of the structure, and is computedusing noisy physics simulations to capture the costs for precision (e.g., attention, coordination, fine motor movements)required for success. We show that the full effort-risk model captures human estimates of difficulty and construction timebetter than either component alone. Preprint link https://arxiv.org/abs/1905.04445.

Adults with extensive science education exhibit cognitiveconflict when reasoning about counterintuitive scientificideas, such as whether clouds have weight or whether bacterianeed nutrients. Here, we investigated whether elementary-school-aged children show the same conflict and whether thatconflict can be reduced by targeted instruction. Seventy-eight5- to 12-year-olds verified, as quickly as possible, statementsabout life and matter before and after a tutorial on thescientific properties of life or matter. Half the statements wereconsistent with intuitive theories of the domain (e.g., “frogsreproduce”) and half were inconsistent (e.g., “cactusesreproduce”). Participants verified the latter less accurately andmore slowly than the former, both before instruction andafter. Instruction increased the accuracy of participants’verifications for counterintuitive statements within thedomain of instruction but not their speed. These resultsindicate that children experience conflict between scientificand intuitive conceptions of a domain in the earliest stages ofacquiring scientific knowledge but can learn to resolve thatconflict in favor of scientific conceptions.

Many studies have investigated the roles that area and number play in visual quantity estimation. Yet, recent work has shown that perceived area is not equal to true, mathematical area. This simple fact calls into question many findings in numerical cognition and suggests a new theoretical perspective: that area estimation plays a dominant role in visual quantity estimation. We examine two ‘case studies’: (1) a ‘general magnitude’ account of visual quantity estimation, which posits bi-directional influences between area and number. In contrast with prior work, controlling for perceived area reveals a unidirectional relation between area and number (Experiments 1 and 2), and (2) acuity of area and number estimation (Experiment 3). We show how an understanding of the perception of area forces a reevaluation of several findings concerning the relative acuity of number and area estimation. Combined, and in contrast to many prior studies, our findings suggest a dominant role of area in visual quantity estimation.

The cognitive system readily detects statistical relationships where the presence of an object predicts a specific outcome. What is less known is how the mind generates predictions when multiple objects predicting different outcomes are present simultaneously. Here we examine the rules with which predictions are made in the presence of two objects that are associated with two distinct outcomes. In three experiments, participants first implicitly learned that an object predicted a specific target location in a visual search task. When two objects predicting two different target locations were present simultaneously, participants were reliably faster to find the target when it appeared in the conjunctive location than in disjunctive locations. This was true even if participants were not consciously aware of the association between the objects and target locations. The results suggest that in the presence of multiple predictors, statistical learning generates implicit expectations about the outcomes in a conjunctive fashion.

It has been argued that semantic categories across languagesreflect pressure for efficient communication. Recently, thisidea has been cast in terms of a general information-theoreticprinciple of efficiency, the Information Bottleneck (IB) prin-ciple, and it has been shown that this principle accounts forthe emergence and evolution of named color categories acrosslanguages, including soft structure and patterns of inconsistentnaming. However, it is not yet clear to what extent this ac-count generalizes to semantic domains other than color. Herewe show that it generalizes to two qualitatively different se-mantic domains: names for containers, and for animals. First,we show that container naming in Dutch and French is near-optimal in the IB sense, and that IB broadly accounts for softcategories and inconsistent naming patterns in both languages.Second, we show that a hierarchy of animal categories derivedfrom IB captures cross-linguistic tendencies in the growth ofanimal taxonomies. Taken together, these findings suggest thatfundamental information-theoretic principles of efficient cod-ing may shape semantic categories across languages and acrossdomains.

How do learners gather new information during wordlearning? We present evidence that adult learners will chooseto receive additional training on object-label associations thatreduce ambiguity about reference during cross-situationalword learning. This ambiguity-reduction strategy is related toimproved test performance. We find mixed evidence thatchildren (4-8 years of age) show a similar preference to seekinformation about words experienced in ambiguous wordlearning situations. In an initial experiment, children did notpreferentially select object-label associations that remainedambiguous during cross-situational word learning. However,this may be explained by some children having relatively highcertainty about object-label associations for which they didnot see evidence disconfirming their initial hypothesis. In asecond experiment that increased the relative ambiguity oftwo sets of novel object-label associations, we found evidencethat children preferentially make selections that reduceambiguity about novel word meanings.

Speakers often face choices as to how to structure their in-tended message into an utterance. Here we investigate the in-fluence of contextual predictability on the encoding of linguis-tic content manifested by speaker choice in a classifier lan-guage, Mandarin Chinese. In Mandarin, modifying a nounwith a numeral obligatorily requires the use of a classifier.While different nouns are compatible with different SPECIFICclassifiers, there is a GENERAL classifier that can be used withmost nouns. When the upcoming noun is less predictable,using a more specific classifier would reduce the noun’s sur-prisal, potentially facilitating comprehension (predicted to bepreferred under Uniform Information Density, Levy & Jaeger,2007), but the specific classifier may be dispreferred from aproduction standpoint if the general classifier is more easilyavailable (predicted by Availability-Based Production; Bock,1987; Ferreira & Dell, 2000). Here we report a picture-namingexperiment confirming two distinctive predictions made byAvailability-Based Production.

Decision makers often reject mixed gambles offering equalprobabilities of a larger gain and a smaller loss. This importantbehavioral pattern is generally seen as evidence for lossaversion, a psychological mechanism according to whichlosses are given higher utility weights than gains. In this paperwe consider an alternate mechanism capable of generatinghigh rejection rates: A predecisional bias towards rejectionwithout the calculation of utility. We use a drift diffusionmodel of decision making to simultaneously specify and testfor the effects of these two psychological mechanisms in agambling task. Our results indicate that high rejection rates formixed gambles result from multiple different psychologicalmechanisms, and that a predecisional bias applied prior to thecomputation of utility (rather than loss aversion) is the primarydeterminant of this important behavioral tendency.

Choice behavior can be influenced by many different types of incidental contextual effects, including those pertaining topresentation format, emotion, social belief, and cognitive capacity. Many of these contextual effects form the basis ofnudges, used by academics and practitioners to shape choice. In this paper, we use data from a very large-scale choiceexperiment to uncover a space of contextual effects. We construct this space by analyzing fifteen contextual effects usingthe parameters of the drift diffusion model (DDM). DDM is a quantitative theory of decision making whose parametersoffer a theoretically compelling characterization of the cognitive underpinnings of choice behavior. By representing a largenumber of contextual effects in terms of how they influence the parameters of the DDM, our space is able to preciselymeasure, quantify, and compare the contextual effects, and interpret these effects in terms of their behavioral, mechanistic,and statistical implications.

It remains unclear whether multimedia facilitates or impairsknowledge acquisition. Here we examined whether subtitlesand video content facilitate comprehension of documentariesconsisting of statements of facts and whether thecomprehension depends on participants’ cognitive abilitiesand eye movement strategies during video watching. Wefound that subtitles facilitated comprehension regardless ofparticipants’ cognitive abilities or eye movement strategiesfor video watching. In contrast, with video content but notsubtitles, comprehension depended on participants’ auditoryworking memory, task switching ability, and eye movementstrategy. Through the Eye Movement analysis with HiddenMarkov Models (EMHMM) method, we found that acentralized (looking mainly at the screen center) eyemovement strategy predicted better comprehension asopposed to a distributed strategy (with distributed regions ofinterest) after contributions from cognitive abilities werecontrolled. Thus, whether video content facilitatescomprehension of documentaries depends on the viewers’ eyemovement strategy in addition to cognitive abilities.

Manufacturing systems of the future need to have flexible re-sources and flexible routing to produce extremely personalizedproducts, even of lot size equal to one. In this paper we haveproposed a framework, which is designed to achieve this goal.Towards this we have integrated an established cultural evolu-tion model to achieve desirable flexibility of resources and ac-ceptable routing time. Promising results are evidenced througha simple proof-of-concept agent-based simulation. The simu-lation results reveal that the products need to move less in morediversified cultural groups when looking for suitable resources.It was also observed that the more time we provide for culturaldissemination, the cultural groups become increasingly coher-ent due to homophily. For scenarios, which require diversifica-tion of resources, we need to find a balance between coherenceand diversification. This paper provides first insights into theseaspects for a production shop floor.

When deciding whether to explore, people must consider both their need for information, and the cost of obtaining it.Thus, to judge why someone explores (or decides not to), we must consider not only their actions, but also the cost ofinformation. Do children attend to the cost of agents exploratory choices when inferring what others know or desireto know? In Experiment 1, four- and five-year-olds judged that an agent who rejected an opportunity to gain low-costinformation must have already known it. In Experiment 2, four- and five-year-olds judged that an agent who incurreda greater cost to gain information had a greater epistemic desire. In two control experiments, we show that these resultscannot be explained by a low-level heuristic linking competence with knowledge. Our results suggest that childrens Theoryof Mind includes expectations about how costs interact with epistemic desires to produce action.

When deciding how to act in new situations, we expect agents todraw on relevant prior experiences. This expectation underliesmany of our mental-state inferences, allowing us to infer agents’prior knowledge from their current actions. Do children sharethis expectation, and use it to infer others’ epistemic states? InExperiment 1, we find that five- and six-year-olds (but not four-year-olds) attribute additional knowledge to agents whose priorexperiences cannot explain their success. In Experiment 2, wefind that six-year-olds (but not younger children) also attributegreater knowledge to agents whose prior experience cannotexplain their failure. We show that by age five or six, childrenexpect ignorant agents’ beliefs (and therefore their actions) to beguided by their prior knowledge. This work adds to a growingbody of research suggesting that, while infants can representmental states, the ability to infer mental states continues todevelop throughout early childhood.

Early mathematics skills are an important predictor of lateracademic, economic and personal success. Children bornpreterm, about 10% of the US population, have an increasedrisk of deficits in mathematics. These deficits may be relatedto lower levels of executive functions and processing speed.We investigated the development of mathematics skills,working memory, inhibitory control and processing speed ofhealthy children born very preterm (between 25 and 32 weeksgestational age, n=51) and full-term (n=29). Children weretested annually from ages 5 to 7 years. We found persistentlower overall mathematics skills in the preterm group, drivenby differences in more informal skills (e.g. counting) at earliertime points, and by differences in more formal skills (e.g.calculation) at later time points. We did not find significantdifferences between preterm and full-term born children inspatial working memory capacity or processing speed.However, these cognitive measures were significantpredictors of mathematics skills in the preterm but not thefull-term group, hinting towards the use of different strategieswhen solving problems.

Understanding, predicting, and learning from other people’sactions are fundamental human social-cognitive skills. Littleis known about how and when we consider other’s actionsand outcomes when making our own decisions. We developeda novel task to study social influence in decision-making: thesocial multi-armed bandit task. This task assesses how peoplelearn policies for optimal choices based on their ownoutcomes and another player's (observed) outcomes. Themajority of participants integrated information gained throughobservation of their partner similarly as information gainedthrough their own actions. This lead to a suboptimal decision-making strategy. Interestingly, event-related potentials time-locked to stimulus onset qualitatively similar but theamplitudes are attenuated in the solo compared to the dyadicversion. This might indicate that arousal and attention afterreceiving a reward are sustained when a second agent ispresent but not when playing alone.

Large-scale behavioral datasets enable researchers to use com-plex machine learning algorithms to better predict human be-havior, yet this increased predictive power does not always leadto a better understanding of the behavior in question. In thispaper, we outline a data-driven, iterative procedure that allowscognitive scientists to use machine learning to generate mod-els that are both interpretable and accurate. We demonstratethis method in the domain of moral decision-making, wherestandard experimental approaches often identify relevant prin-ciples that influence human judgments, but fail to generalizethese findings to “real world” situations that place these prin-ciples in conflict. The recently released Moral Machine datasetallows us to build a powerful model that can predict the out-comes of these conflicts while remaining simple enough to ex-plain the basis behind human decisions.

How do people understand concepts such as dog, aggressive dog, dog house or house dog? The meaning of a concept depends crucially on the concepts around it. While this hypothesis has existed for a long time, only recently it has become possible to test it based on neuroimaging and quantify it using computational modeling. In this paper, a neural network is trained with backpropagation to map attribute- based semantic representations to fMRI images of subjects reading everyday sentences. Backpropagation is then extended to the attributes, demonstrating how word meanings change in different contexts. Across a large corpus of sentences, the new attributes are more similar to the attributes of other words in the sentence than they are to the original attributes, demonstrating that the meaning of the context is transferred to a degree to each word in the sentence. Such dynamic conceptual combination effects could be included in natural language processing systems to encode rich contextual embeddings to mirror human performance more accurately.

Numerosity is informative for living organisms. It cantransmit, among many things, amount of food available,heading direction of the troop, which group could win aterritorial dispute, the decision of were to build a beehive.Given its ecological importance, we test the hypothesisthat numerosity captures visual selection. In five exper-iments we confirmed that an irrelevant visual stimulusthat was numerically large slowed down participants indetecting a task-relevant visual target (Exp. 1 and 2). Thiscapture was not driven by sensory variables that could cor-relate with numerosity: cumulative area (Exp. 3) and ele-ment size (Exp. 4). We also confirmed that the underlyingnumerosity representations were analogue, not set-based(Exp. 5). In a crowded visual scene numerosity is a rele-vant cue for visual selection, but represented only in ap-proximate/coarse fashion.

Probabilities can be described by a numerator and a de-nominator and students and decision-makers are not in-different to numerical values of the components. For in-stance, when people compare two equal ratios their choicesgravitate to the option with larger number, even if theyknow both ratios are equal. To the date, however, it is un-clear if whole number biases are present in other cultures.We tested a farming-foraging group living in the Bolivianrain forest in a simple 2AFC ratio comparison task. Af-ter appropriate training, the Tsimane were highly accu-rate in this task, confirming that visual proportional rea-soning is present across cultures. Importantly, they hada strong tendency to favor large numbers in equal ratiocomparisons, similar to what is found in educated popu-lations. Even though our sample size is moderate (n=76),the whole number bias we found occurred under good pro-portional reasoning. The bias may be a general feature ofcognition, rather than a cultural or education artifact, thatmay help humans solve ambiguous situations.

A novel experimental method is motivated and applied in aneffort to test for effects of category learning on perceptualdiscrimination so as to clearly distinguish category boundaryeffects of expansion and compression from changes insensitivity to stimulus dimensions. The method includes acontrol group performing a task that, like category learning,requires attention to one systematically varying stimulusdimension rather than another. Discrimination accuracy istracked over time and measured using a psychophysicalstaircase procedure tailored to individual participants thatdoesn’t rely on memory. Initial results suggest improvementin discrimination accuracy over time, particularly on thedimension relevant to the categorization or control task, butno evidence of category boundary effects or effects ofcategory learning on dimension perception stronger than thoseof the control task. Possible reasons for this and directions forfurther research are briefly discussed .

It is presently unclear how localized, word association networkrepresentations compare to distributed, spatial representationsin representing distant concepts and accounting for primingeffects. We compared and contrasted 4 models of representingsemantic knowledge (5018-word directed and undirected stepdistance networks, an association-correlation network andword2vec spatial representations) to predict semantic primingperformance for distant concepts. In Experiment 1, responselatencies for relatedness judgments for word-pairs followed aquadratic relationship with network path lengths and spatialcosines, replicating and extending a pattern recently reportedby Kenett, Levi, Anaki, and Faust (2017) for an 800-wordHebrew network. In Experiment 2, response latencies toidentify a word through progressive demasking showed a lineartrend for path lengths and cosines, suggesting that simpleassociation networks can capture distant semanticrelationships. Further analyses indicated that spatial modelsand correlation networks are less sensitive to directassociations and likely represent more higher-levelrelationships between words.

Non-numerical magnitudes such as cumulative area, element size, and density influence the perception of number. How-ever, it is unclear whether interactions between number and non-numerical magnitudes reflect independent representationsthat interface vis--vis other systems (e.g., language) or, conversely, reflect holistic perception of number and other mag-nitudes. In the present work, we found converging evidence that number and cumulative area are perceptually integraldimensions. Whether assessed explicitly (Experiment 1) or implicitly (Experiment 2), perceived similarity for dot arraysthat varied parametrically in number and area was best modeled by Euclidean, as opposed to city-block, distance. Criti-cally, we also found that the integrality of number and area is comparable to other integral dimensions (Exp. 1: bright-ness/saturation; Exp. 2: radial frequency components), but different from separable dimensions (Exp. 1: shape/color; Exp.2: thickness/curvature). In summary, these findings suggest that non-symbolic number perception is holistic, such that theprocessing of non-numerical magnitudes is obligatory.

In this paper we present a model of three central aspects ofprobabilistic cognition: event prediction, feature formation,and attention allocation. While most models of probabilisticreasoning take a parameter estimation and error minimisationapproach (sometimes referred to as ‘predictive coding’, and of-ten described in terms of Bayesian updating), our model takesa contrasting frequentist hypothesis-testing approach. Thischoice is motivated by a series of recent results suggesting thatpeople’s probabilistic reasoning follows frequentist probabilitytheory. In simulation tests we demonstrate that this frequentistmodel, in which predictive features are formed by a process ofnull hypothesis significance testing, can give a successful ac-count of event prediction and attentional switching behaviour.

Many industry professionals are poorly calibrated, overestimatingtheir ability to make accurate forecasts. Previous research hasdemonstrated that an individual’s calibration in a specific domaincan be improved through calibration training in that domain;however devising a training program for each specific domainwithin a field is laborious. A more efficient method would be ifindividuals from different disciplines could undertake the samegeneral training and transfer the skills learnt to their respective,specific domains. This study investigated whether calibrationtraining in a general domain was transferable to the specificdomain of petroleum engineering. The results showed that, whilstthe feedback training was effective within the general domain,there was only limited transfer to the specific domain. This isargued to be due to recognition failure, where the participantsfailed to recognise that the skill learnt through training in thegeneral domain could be transferred to the specific domain.

Efficient data compression is essential for capacity-limited sys-tems, such as biological memory. We hypothesize that the needfor efficient data compression shapes biological perception andperceptual memory in many of the same ways that it shapesengineered systems. If true, then the tools that engineers useto analyze and design systems, namely rate-distortion theory(RDT), can profitably be used to understand perception andmemory. To date, researchers have used deep neural networksto approximately implement RDT in high-dimensional spaces,but these implementations have been limited to tasks in whichthe sole goal is compression with respect to reconstruction er-ror. Here, we introduce a new deep neural network architecturethat approximately implements RDT in a task-general manner.An important property of our architecture is that it is trained“end-to-end”, operating on raw perceptual input (e.g., pixels)rather than an intermediate level of abstraction, as is the casewith most psychological models. We demonstrate that ourframework can mimick categorical biases in perception andperceptual memory in several ways, and thus generates spe-cific hypotheses that can be tested empirically in future work.

We show, contrary to some recent claims in the literature, thatprototype distributional semantic models (DSMs) are capa-ble of representing multiple senses of ambiguous words, in-cluding infrequent meanings. We propose that word2vec con-tains a natural, model-internal way of operationalizing the dis-ambiguation process by leveraging the two sets of represen-tations word2vec learns, instead of just one as most workon this model does. We evaluate our approach on artifi-cial language simulations where other prototype DSMs havebeen shown to fail. We furthermore assess whether these re-sults scale to the disambiguation of naturalistic corpus exam-ples. We do so by replacing all instances of sampled pairsof words in a corpus with pseudo-homonym tokens, and test-ing whether models, after being trained on one half of the cor-pus, were able to disambiguate pseudo-homonyms on the ba-sis of their linguistic contexts in the second half of the cor-pus. We observe that word2vec well surpasses the baselineof always guessing the most frequent meaning to be the rightone. Moreover, it degrades gracefully: As words are moreunbalanced, the baseline is higher, and it is harder to surpassit; nonetheless, Word2vec succeeds at surpassing the baseline,even for pseudo-homonyms whose most frequent meaning ismuch more frequent than the other.

The Remote Associates Test (RAT, CRA) is a classical creativ-ity test used to measure creativity as a function of associativeability. The RAT has been administered in different languages.Nonetheless, because of how embedded in the language thetest is, only a few items are directly translatable, and most ofthe time the RAT is created anew in each language. This pro-cess of manual (and in two cases computational) creation ofRAT items is guided by the researchers’ understanding of thetask. However, are the RAT items in different languages com-parable? In this paper, different RAT stimuli datasets are an-alyzed qualitatively and quantitatively. Significant differencesare observed between certain datasets in terms of solver per-formance. The potential sources of these differences are dis-cussed, together with what this means for creativity psycho-metrics and computational vs. manual creation of stimuli.

Narrative processing is an important skill to model bothfrom a cognitive science perspective and a computa-tional modeling perspective which applies to intelligentagents. Communication between humans often involvesstorytelling patterns that make the mundane exchange ofinformation more interesting and with proper emphasison important communicative goals. Current narrativegeneration models evaluate their generations basedon either a priori domain semantics (e.g. game statefor an in-game conversation with player agents) orgeneric text quality measures (e.g. coherence). However,in utilizing storytelling as a communicative tool forreal-world interactions, domain-specific approaches failto generalize and text quality measures fail to ensurethat the narrative is perceived as interesting. Hence, suchgeneration needs to consider the cognitive processesinvolved in the perception of narrative. Using theories ofcognitive interest, we present results of an investigationof whether word embeddings (e.g. GloVe (Pennington,Socher, & Manning, 2014)) could be used to model andestimate cognitive interestingness in stories.

People predict incoming words during online sentencecomprehension based on their knowledge of real-world eventsthat is cued by preceding linguistic contexts. We used thevisual world paradigm to investigate how event knowledgeactivated by an agent-verb pair is integrated with perceptualinformation about the referent that fits the patient role. Duringthe verb time window participants looked significantly more atthe referents that are expected given the agent-verb pair.Results are consistent with the assumption that event-basedknowledge involves perceptual properties of typicalparticipants. The knowledge activated by the agent iscompositionally integrated with knowledge cued by the verbto drive anticipatory eye movements during sentencecomprehension based on the expectations associated not onlywith the incoming word, but also with the visual features of itsreferent.

Human decision-making deviates from the optimal solution,i.e. the one maximizing cumulative rewards, in many sit-uations. Here we approach this discrepancy from the per-spective of computational rationality and our goal is to pro-vide justification for such seemingly sub-optimal strategies.More specifically we investigate the hypothesis, that humansdo not know optimal decision-making algorithms in advance,but instead employ a learned, resource-constrained approxima-tion. The idea is formalized through combining a recently pro-posed meta-learning model based on Recurrent Neural Net-works with a resource-rational objective. The resulting ap-proach is closely connected to variational inference and theMinimum Description Length principle. Empirical evidenceis obtained from a two-armed bandit task. Here we observepatterns in our family of models that resemble differences be-tween individual human participants.

Although selective attention is useful in many situations, it alsohas costs. In addition to ignoring information that may becomeuseful later, it can have long term costs, such as learnedinattention – difficulty in learning from formerly irrelevantsources of information in novel situations. In the current study wetracked participants’ gaze while they completed a categorylearning task designed to elicit learned inattention. Duringlearning an unannounced shift occurred such that information thatwas most relevant became irrelevant, whereas formerly irrelevantinformation became relevant. We assessed looking patternsduring initial learning to understand how different aspects ofattention allocation contribute to learned inattention. Our resultsindicate that learned inattention depends on both the overall levelof selectivity (measured as entropy of proportion of looking toeach feature) and the extent to which participants optimizedattention (becoming more selective over time).

A fundamental challenge in object recognition is torecognize an image when it is projected across differentretinal locations, an ability known as translation tolerance.Although the human visual system can overcome thischallenge, the mechanisms responsible remain largelyunexplained. The ‘trained-tolerance’ approach holds thatan object must be experienced across different retinallocations to achieve translation tolerance. Previous studieshave supported this approach by showing that the visualsystem struggles to generalize recognition of novelobjects to translations as small as 2° of visual angle. Thepresent paper outlines a series of eyetracking studies thatshow novel objects can be recognized at translations as faras 18° from the trained retinal location, challenging thestandard account of translation tolerance in neuroscienceand psychology.

Using the prepositions in and on, Jamrozik and Gentner (2015; 2014; 2011) explored a particular factor of meaning that was hypothesized to serve as a metaphorical link between spatial and abstract concepts. Across several studies, these researchers have provided evidence for the idea that there is a “continuum of control” that exists for both spatial and abstract uses of in and on. Our research explores other potential meaning factors that might play a role in non-spatial uses of in and on. Our results replicate and extend Jamrozik and Gentner’s (2011) findings. We advocate using a multi-componential approach as research involving indirect metaphors continues moving forward.

How do children’s beliefs about a causal system influence theirexploration of that system? Children watched an experimentertry to make a machine play music by placing blocks on top; oneblock always activated the machine and the other block neverdid (Deterministic condition), or one block activated the ma-chine a higher proportion of times than the other (Probabilisticcondition). Subsequently, we measured children’s exploratorybehaviors without feedback (the machine never activated). Wepredicted that children in the two conditions would differ intheir beliefs about how the system should work, leading to dif-ferent hypotheses about why the machine was no longer work-ing, and to differential exploration. Compared to the Proba-bilistic condition, children in the Deterministic condition in-tervened more often with the previously more effective block,experimented more with how to activate the machine, and ex-plored for less time. Children’s exploration provides a rich,nuanced view of their causal reasoning.

Quantity can be expressed in a variety of ways and at dif-ferent levels of precision. One factor that influences numer-ical description of elements in a visual scene is how long thescene is observed. We extend a previous incremental modelof numerical perception to model quantified description undertime constraints. Our extended model predicts that as presen-tation duration decreases and as the quantity of items to beenumerated increases, the frequency of inexact quantifiers willincrease. We conducted two human subject elicitation stud-ies to test these predictions. Our findings were consistent withour model’s predictions. Additionally, we demonstrate that ournovel model of incremental numerical perception and quanti-fied description closely predicts the precise proportion of exactnumerical responses generated by in these experiments.

Modern society frequently requires that we express our sub-jective senses in objective, shared formal systems; this en-tails mapping multiple internal variables onto a common scale.Here we ask whether we accomplish this feat in the case ofestimating number by learning a single mapping between ex-plicit numbers and one integrated subjective estimate of nu-merosity, or if we separately map different perceptual featuresonto numbers. We present people with arrays of dots and askthem to report how many dots there are; we rely on the sys-tematic under/overestimation of number at higher quantities toestimate error in the mapping function. By comparing how thiserror changes over time, as the mapping fluctuates for differentvisual cues to numerosity, we can evaluate whether these cuesshare a single mapping, or are mapped onto number individu-ally. We find that area, size, and density all share a commonmapping, indicating that people obtain a unified subjective es-timate of numerosity before mapping it onto the formal numberline.

Accounts of human physical reasoning based on simulationfrom a noisy physics engine have enjoyed considerable suc-cess in recent years. However, simulating complex physicaldynamics can be a computationally expensive process, and itis possible that people use faster, cheaper shortcuts to makepredictions and inferences in complicated physical scenarios.Here we asked people to predict the eventual destination of aball on a 2D bumper table (in the style of Smith, de Peres, Vul,and Tenenbaum (2017)). We designed scenarios that we ex-pected would modulate the use of heuristics and simulation:the bumper table provided varying degrees of containment toconstrain future outcomes and to make a containment heuris-tic more useful, and could have more or less internal struc-ture to vary the reliability of noisy simulation. As the con-tainment heuristic becomes more useful, and as simulation be-comes more expensive, we expected that people would switchfrom using simulation to rely more on rapid heuristic-basedpredictions and therefore respond faster. Instead, we foundthat even when containment was very predictive, people wereprogressively slower and less accurate as simulation complex-ity increased, indicating that they persisted in using simulationrather than containment heuristics.

Children and adults prefer simpler to more complex explanations, a penchant they share with scientists and philoso-phers. While the preference has been widely remarked, its mechanisms and justification remain contested (Kitcher1987,Lombrozo 2007, Lombrozo2015). Explanations for the simplicity preference have included over-hypotheses, resourcerationality, pragmatic justifications, and quirks of the hypothesis generation process. We present a model of key resultsfrom Pacer and Lombrozo (Pacer2017) and show that one form of the simplicity bias can be explained on probabilisticgrounds alone. This modeling work provides an explanation for one manifestation of the simplicity bias, and allows usto formalize questions within the ’Explanation for Best Inference’ Framework (Lombrozo2015), asking explicitly whatmakes the best explanation ’best.’

Language understanding requires listeners to quickly compresslarge amounts of perceptual information into abstract linguis-tic categories. Critical cues to those categories are distributedacross the speech signal, with some cues appearing substan-tially later. Speech perception would thus be facilitated if gra-dient sub-categorical representations of the input are main-tained in memory, allowing optimal cue integration. How-ever, indiscriminate maintenance of the high-dimensional sig-nal would tax memory systems. We hypothesize that speechperception balances these pressures by maintaining gradientrepresentations that are expected to facilitate category recog-nition. Two perception experiments test this hypothesis. Be-tween participants, an initial exposure phase manipulated theutility of information maintenance: in the High-Informativitygroup, following context always was informative; in the Low-Informativity group, following context always was uninforma-tive. A subsequent test phase measured the extent to whichparticipants maintained gradient representations. The Low-Informativity group showed less maintenance, compared to theHigh-Informativity group (Experiment 1). We then increasedthe task demands and made the targets of the manipulation lessobvious to participants (Experiment 2). We found a qualita-tively similar pattern. Together, these results suggest that lis-teners are capable of allocating memory to gradient representa-tions of the speech input based on the expected utility of thoserepresentations.

In this paper we explore the effect of normal aging on executive function and present a computational account of the effectof aging in a standard executive task. We tested 25 younger adults and 25 older adults (both with no known neurologicalcondition) on the Wisconsin Card Sorting Task (WCST), a classic test of executive function. The test produces multiplemeasures related to the types of error made by participants, the rate of learning, and so on. As hypothesised, results showno difference between the groups in the number of perseverative errors (i.e., in continuing with a previously successful rulein the presence of negative feedback), but a significantly increased tendency for older adults relative to younger adults tocommit set loss errors (i.e., to switch away from a rule despite positive feedback). We fit an existing neurocomputationalmodel of the task to the experimental data by searching through the models parameter space in order to find the best set ofparameter values for the two different age groups. This leads to a proposition regarding the effect of aging on the value ofthe epsilon ctx parameter, which we argue elsewhere reflects cortical dopamine concentration. We further reanalyse thedata by clustering participants by performance (rather than by age) and show that there are multiple points in parameterspace that fit each cluster of participants. We argue on the basis of this and the behavioural data, that different parametervalues reflect different solutions to optimizing task performance, and that older participants may compensate for changesin epsilon ctx (reflecting dopamine concentration) by effortful changes in other parameters (specifically, by increasingattentional focus).

We propose a neural network model that accounts for the emer-gence of the taxonomic constraint and for the whole objectconstraint in early word learning. Our proposal is based onMayor and Plunkett (2010)’s neurocomputational model of thetaxonomic constraint and extends it in two directions. Firstly,we deal with realistic visual and acoustic stimuli. Secondly,we model the well-known whole object constraint in the visualcomponent. We show that, despite the augmented input com-plexity, the proposed model compares favorably with respectto previous systems.

Children learning language in multilingual settings have tolearn that objects take different labels within each differentlanguage to which they are exposed. Previous research hasshown that adults can learn one-to-one and two-to-one word-object mappings via cross-situational statistical learning(CSSL), and that socio-pragmatic cues may differentiallyinfluence monolingual and bilingual adults’ learning of suchmappings. However, the extent to which monolingual andbilingual learners can keep track of multiple labels frommultiple speakers has not yet been investigated. Wemanipulated the number of speakers in a CSSL task thatinvolved learning both mapping types. We successfullyreplicated previous studies that found that both monolingualsand bilinguals could learn both types of mappings via CSSL.In addition, we found that bilinguals showed a steeper learningrate for two-to-one mappings than monolinguals, andbilinguals were more likely to accept two words for the sameobject than monolinguals. These results show that the effect ofspeaker identity on tracking word-object mappings variesaccording to language experience.

We explore a nonparametric approach to cognitive modeling.Traditionally, models in cognitive science have been paramet-ric. As such, the model relies on the assumption that the datadistribution can be defined by a finite set of parameters. How-ever, there is no guarantee that such an assumption will hold,and it may introduce undesirable biases. For these reasons, anonparametric approach to model building is appealing. Wepropose a novel framework that combines Gaussian Processeswith active learning (GPAL), and evaluate it in the context ofdelay discounting (DD), a well-studied task in decision mak-ing. We evaluate GPAL in a simulation and a behavioral exper-iment, and compare it against a traditional parametric model.The results show that GPAL is a suitable modeling frameworkthat is robust, reliable, and efficient, exhibiting high sensitivityto individual differences.

There has been much discussion around the Linguistic-Savings Hypothesis (LSH), which postulates that language can affect intertemporal choices of its speakers; the validity of this claim has remained controversial. To test the LSH independent from the possible influencing factors, such as cultural differences, we focused on the Japanese language, which features asymmetric tense marking, in that past tense is grammatically marked but future tense is not. Adopting a within-participant design, we compared the discounting behavior between past and future gains in native Japanese participants. Our results revealed that Japanese speakers tended to discount the values placed on rewards in an asymmetry way: to discount the value of past gains more heavily than that of future gains. We believed our results corroborated the LSH and linguistic relativity.

Despite the fact that priming is one of the most studiedphenomena in cognitive psychology, many questions remainabout exactly when, why and under what task conditions weought to observe priming in the lab, and what types ofrelationships between words or concepts reliably lead topriming. This project contrasted two priming experimentswhere the primary manipulation was the decision the subjectswere making about words (as well as manipulating otherfactors, like relatedness proportion, known to affect priming).We found evidence that: 1) automatic priming forsemantically related words does happen under someconditions, but 2) semantic priming, and whether it happensindependent of association, is dependent on the task in whichparticipants are engaged. These results provide evidence forthe context sensitive nature of the activation of semanticmemory.

With two experiments, we begin an inquiry into theperceived explanatory value of mathematical entities ineveryday explanations. This work is motivated by aphilosophical debate about the role mathematical entitiesplay in explanation. Simply put, are the mathematicalentities themselves explanatory, or is mathematical talkelliptical or shorthand for talk about the physical entities weare concerned with? Across the two experiments, we foundclear evidence that situational factors affected how themathematical entities were considered. However, whenthose situational factors are accounted for, participantstended to see more explanatory value for mathematicalentities that point to other objects involved in the explanationas opposed to mathematical entities that assume theexplanatory role themselves.

In mathematical cognition, problem difficulty is a central vari-able. In the present study, problem difficulty was operational-ized through five arithmetic operators — addition, subtrac-tion, multiplication, division, and modulo — and through thenumber of carries required to correctly solve a problem. Thepresent study collected data from human participants solvingarithmetic problems, and from multilayer perceptrons (MLPs)that learn arithmetic problems. Binary numeral problems werechosen in order to minimize other criteria that may affect prob-lem difficulty, such as problem familiarity and the problemsize effect. In both humans and MLPs, problem difficulty washighest for multiplication, followed by modulo and then sub-traction. The human study found that problem difficulty wasmonotonically increasing with respect to the number of car-ries, across all five operators. Furthermore, a strict increasewas also observed for addition in the MLP study

Do children’s flexible causal inferences promote more cre-ative causal discovery for observing adults? Inspired by a taskin which children are more likely to consider unconventionalcausal forms (Lucas, Bridgers, Griffiths, & Gopnik, 2014;Wente et al., 2019), we designed a new method in which child-adult pairs work together to solve a causal task and assessedthe relative influence of each member of the pair on the other’scausal inference. Consistent with previous research, childrenwere better than parents at learning the unusual conjunctive re-lationship, suggesting that children make more flexible causalinferences than adults. Our research also revealed a surpris-ing and new result – that observing a child explore broadlyhelped parents to be more flexible and open-minded in theircausal learning. In contrast, a child observing an adult’s ex-ploratory interventions had no negative consequence on thechild’s ability to infer the correct relation. Follow-up exper-iments explored the degree to which this child-led bootstrap-ping for adults was due to the particular exploratory evidencegenerated by the child during play, or merely the presence ofa child. Results suggest that both factors may play a role inshaping adult’s causal inferences.

How do we seek information from our environment to find solutions to the questions facing us? We pose an open-endedvisual search problem to adult participants, asking them to identify targets of questions in scenes guided by only an in-complete question prefix (e.g. Why is..., Where will...). Participants converged on visual targets and question completionsgiven just these function words, but the preferred targets and completions for a given scene varied dramatically dependingon the query. We account for this systematic query-guided behavior with a model linking conventions of linguistic refer-ence to abstract representations of scene events. The ability to predict and find probable targets of incomplete queries maybe just one example of a more general ability to pay attention to what problems require of their solutions, and to use thoserequirements as a helpful guide in searching for solutions.

Both behavioral studies and the neurophysiological data modelling suggested female advantage in memory for objects, however, most research pertained to long-term memory, whereas data from visual working memory (VWM) are scanty. In a large sample of 2044 people, the number of objects supposedly encoded in VWM was measured during the change detection task. The stimuli were either relatively familiar geometric shapes or less familiar Greek symbols. Controlling for the general ability level, a small but significant advantage for memorizing shapes in VWM was found in females over males, but no effect was observed for memorizing abstract symbols. The present results support neuroimaging models of human cognitive architecture, suggesting that female VWM relies on a more complex network of domain-specific brain modules, as compared to males. Consequently, formal models of VWM and related cognitive processes should account for sex and material type.

In the present study, we combined tDCS and EEG to examine theelectrophysiological responses to the tDCS-induced effects onthe face inversion effect showed in recent studies. A double-blindprocedure with a between-subjects design (n=48) was used withthe subjects, recruited from the student population, beingrandomly assigned to either tDCS anodal or sham condition. ThetDCS stimulation was delivered over the DLPFC at Fp3 site for10 min at an intensity of 1.5mA while subjects engaged in anold/new recognition task traditionally used to obtain theinversion effect. The behavioural results generally confirmedprevious findings. Critically, the results from the N170 show aneffect of tDCS. Specifically, the tDCS procedure was able tomodulate the N170 peak component by reducing the inversioneffect on the latencies (i.e. less delay between upright andinverted faces) and by increasing the inversion effect on theamplitudes (i.e. larger N170 for inverted vs upright faces). Weinterpret the results based on the previous literature in regard tothe inversion effect on the N170 component.

The phenomenon of insight (also called “Aha!” or “Eureka!”moments) is considered a core component of creative cogni-tion. It is also a puzzle and a challenge for statistics-basedapproaches to behavior such as associative learning and rein-forcement learning. We simulate a classic experiment on in-sight in pigeons using deep Reinforcement Learning. We showthat prior experience may produce large and rapid performanceimprovements reminiscent of insights, and we suggest theo-retical connections between concepts from machine learning(such as the value function or overfitting) and concepts frompsychology (such as feelings-of-warmth and the einstellung ef-fect). However, the simulated pigeons were slower than thereal pigeons at solving the test problem, requiring a greateramount of trial and error: their “insightful” behavior was sud-den by comparison with learning from scratch, but slow bycomparison with real pigeons. This leaves open the questionof whether incremental improvements to reinforcement learn-ing algorithms will be sufficient to produce insightful behavior.

People often navigate new environments and must learn abouthow actions map to outcomes to achieve their goals. In this pa-per, we are concerned with how people direct their search andtrade off between selecting informative actions and actions thatwill be most immediately rewarding when they are faced withnew tasks. We find that some people selected globally infor-mative actions and were able to generalize from few observa-tions in order learn new reward structures efficiently. Theseparticipants also displayed the ability to transfer knowledgeacross similar tasks. However, a consistent proportion of par-ticipants behaved sub-optimally, caring more about observingnovel information instead of maximizing reward. Across fourexperiments, we present evidence that participants’ motivationto explore was influenced by 1) how much they already knewabout the underlying task structure and 2) whether their obser-vations remained available. We discuss possible explanationsbehind people’s exploratory drive.

Kinematic specification of dynamics (KSD) states that full- body kinematic patterns of daily activities are reflective of a person’s plans, goals, and intentions. Furthermore, it has been shown that observers of those activities are well attuned to differences between those kinematic patterns. However, despite a substantial body of research on the identification of intentional motion, it is not yet clear what the essential kinematic information is required to perceive the intention from the kinematic pattern. Therefore, we analyzed four different intentional full body motions (sit-to-stand transitions: stand, press-stand, press-sit, and reach-up), to determine the essential kinematic information that differentiates them. We utilized principal component analysis (PCA), linear mixed models, and hierarchical multinomial logistic regression to create two predictive regression models that allow us to successfully identify and distinguish the four intentional motions.

The importance of verbal and visuospatial working memory(WM) for co-speech gesture comprehension was tested in twoexperiments using the dual task paradigm. Healthy, college-aged participants encoded either a dot locations in a grid(Experiment 1), or a series of digits (Experiment 2), andrehearsed them as they performed a discourse comprehensiontask. The discourse comprehension task involved watching avideo of a man describing household objects, and judgingwhich of two words probes was most related to the video.Following the discourse comprehension task, participantsrecalled either the verbally or visuo-spatially encodedinformation. In both experiments, performance on thediscourse comprehension task was faster when gesturalinformation was congruent with the speech than when it wasincongruent. Moreover, performance on the discoursecomprehension task was impacted both by increasing the loadon the visuospatial WM system (Experiment 1) and the verbalWM system (Experiment 2). However, in both studies effectsof WM load and gesture congruency were additive,suggesting they were independent.

Cognitive tests used to measure individual differences are gen-erally designed with equality in mind: the same “broadly ac-ceptable” items are used for all participants. This has unknownconsequences for equity, particularly when a single set of lin-guistic stimuli are used for a diverse population of languageusers. We hypothesized that differences in language varietywould result in disparities in psycholinguistically meaningfulproperties of test items in two widely-used cognitive tasks, re-sulting in large differences in performance. As a proxy for in-dividuals’ language use, we administered a self-report surveyof media consumption. We identified two substantial clustersfrom the survey data, roughly orthogonal to a priori groups re-cruited into the study (university students and members of thesurrounding community). We found effects of both populationand cluster membership. Comparing item-wise differences be-tween the clusters’ language models did not identify specificitems driving performance differences.

Learning to read English requires learning the complex statis-tical dependencies between orthography and phonology. Pre-vious research has focused on how these statistics are learnedin neural network models provided with as much training asneeded. Children, however, are expected to acquire this knowl-edge in a few years of school with only limited instruction. Weexamined how these mappings can be learned efficiently, de-fined by tradeoffs between the number of words that are explic-itly trained and the number that are correct by generalization.A million models were trained, varying the sizes of randomly-selected training sets. For a target corpus of about 3000 words,training sets of 200–300 words were most efficient, producinggeneralization to as many as 1800 untrained words. Composi-tion of the 300 word training sets also greatly affected general-ization. The results suggest directions for designing curriculathat promote efficient learning of complex material.

The iconic cross-modal correspondence between fundamentalfrequency and location in vertical space (“high is up”) has longbeen described in the literature. However, an explanation forthis relationship has not been proposed. We conducted an ex-periment in which participants shot at cans projected on thewall in different vertical positions. We found that mean funda-mental frequency was significantly influenced by vertical headposition. Moving the head upwards changes the position of thelarynx, which pulls on the cricothyroid muscle and changes thefundamental frequency. We thus propose that the iconic rela-tionship between fundamental frequency and vertical space isgrounded in the body.

While previous models of economic decision-making offer de-scriptive accounts of behavior, they often overlook the com-putational complexity of estimating expected utility. Here,we seek to understand how both environmental and individualconstraints on cognition shape our daily decision. Informedby the predictions of a recently-proposed resource-rationalprocess model of risky choice, sample-based expected utility(SbEU; Nobandegani, da Silva Castanheira, Otto, & Shultz,2018), we reveal that both time pressure and individual dif-ferences in processing speed have a convergent effect on riskpreferences during a risky decision-making task. Under severetime constraints, participants’ risk preferences manifested astrong framing effect compared to little time pressure in whichchoice adhered to the classic fourfold pattern of risk prefer-ences. Similarly, individual differences in processing speed,measured using an established task, predicted similar effectsupon risk attitudes as extrinsic time pressure. These findingsreveal a converging contribution of environmental and individ-ual limitations on risky choice, and provide empirical supportfor SbEU as a resource-rational process model of risky deci-sion making. Notably, SbEU serves as a single-process modelof two well-established biases, and the transition between thetwo, in risky choice.

Previous studies have shown that readers often overestimatethe similarity between their perspective and the perspective ofprotagonists in a story. This egocentric projection is argued tooriginate from readers’ tendency to use their own knowledgeas a frame of reference from which they (insufficiently) adjustaway to account for protagonists’ less informed perspective.This experimental study demonstrated that readers usefeedback about protagonists’ knowledge status to drawinferences that are more accurate on future perspective-takingtrials. Readers who were given the opportunity to learn throughfeedback not only adjusted their perspective-judgment morethan those who did not receive feedback, these readers alsoshowed less egocentric projection on future assessments.

Context is important for abstract concept processing, but amechanism by which it is encoded and re-instantiated withconcepts is unclear. We used a source-memory paradigm todetermine whether episodic context is attended more whenprocessing abstract concepts. Experiment 1 presentedabstract and concrete words in colored boxes at encoding. Attest, memory for the frame color was worse for abstractconcepts, counter to our predictions. Experiment 2 showedthe same pattern when colored boxes were replaced withmale and female voices. Experiment 3 presented words fromencoding in the same or different box color to determinewhether a greater advantage is conferred by context retentionin memory for abstract concepts. There was instead adisadvantage: abstract concepts were less likely to beidentified when the encoding color was retained at test.Concrete concepts are more sensitive to simple episodicdetail, and in abstract concepts, arbitrary context may besuppressed.

What does it take to learn a new word? Many of the words welearn, we have learned from language itself – by encounteringthem in various informative contexts. Here, we investigate thelimits of learning from context by studying how people learnnew words from very sparse contexts, at the extreme, a contextin which all content words are replaced by nonsense words. Wefind that participants exposed to even such extremely sparsecontexts nevertheless learn something about the meaning ofwords embedded in those contexts. Performance tended to bebetter when knowledge was assessed by first directing people’sattention to the part of speech of the target words.

In the present study, the correlations of eye blink rate (EBR) with the effective execution of early and late creative idea generation were explored. Participants engaged in a real- world idea generation task. Resting state EBR (before the task) and task-evoked EBR (during the task) were measured using eye-tracking. The results showed that resting state EBR negatively correlated with the amount of generated ideas during early stage, but not late stage idea generation. Task- evoked EBR did not correlate with the amount of generated ideas during early nor late stage idea generation. However, the change in EBR (from resting state to during early or late stage idea generation) positively correlated with the amount of ideas generated during early, but not during late stage idea generation. The contribution of this study is that it shows that EBR predicts and dissociates the effective execution of early and late stage creative idea generation.

Prior research showed that young children prefer to seek helpfrom actors who have demonstrated active learning compe-tence. What inferences do people make based on the abil-ity to search effectively, for example by asking informativequestions? This project explores across two experiments towhat extent adults and children (3- to 9-year-olds) general-ize the ability to ask informative questions to other abili-ties/characteristics. We presented participants with one mon-ster who always asked informative questions and one whoalways asked uninformative questions. Participants had tochoose which monster they thought was more likely to pos-sess/was better at 12 different characteristics/abilities. Our re-sults show a clear developmental trend. Three- and 4-year-olds draw unsystematic inferences from the monsters question-asking expertise. Five- and 6-year-olds identified the betterquestion asker as better at everything. Seven- to 9-year-oldsshowed adult-like response patterns, selectively associating theability to ask good questions to related characteristics/abilities.

Prior knowledge has long been recognized as an important predictor of learning, yet the term prior knowledge is often applied to related but distinct constructs. We define a specific form of prior knowledge, ancillary knowledge, as knowledge of concepts and skills that enable learners to gain the most from a target lesson. Ancillary knowledge is not prior knowledge of the lesson’s target concepts and skills, and may even fall outside the domain of the lesson. Nevertheless, ancillary knowledge affects learning of the lesson, e.g., lower ancillary knowledge can hinder performance on lesson-related tasks. We measured ancillary knowledge, prior knowledge of the domain, and controlled for general ability, and found that (a) stronger ancillary knowledge and general ability predicted better performance on transfer tasks, but (b) prior knowledge of the domain did not. This research suggests that enhancing instruction by remediating gaps in ancillary knowledge may improve learning in introductory-level courses.

Children quickly gain enormous linguistic knowledge duringearly development, in part due to low-level features of theirparents’ speech. Some posit that parents contribute to theirchild’s language development by tuning their own languageaccording to their child’s developmental abilities and needs(Bruner, 1985; Snow, 1972). Here, we investigate this hypoth-esis by examining ‘alignment’ at the level of syntax and func-tion words in a large-scale corpus of parent-child conversationsand measuring its association with language development out-comes. To do so, we employ a statistical model of alignment toestimate its presence in our dataset and its predictive impact ona measure of vocabulary development. Our results corroborateprevious findings, showing strong alignment for both parentsand children; in addition, we demonstrate that parental align-ment is a significant predictor of language maturity indepen-dent of demographic features, suggesting that parental tuninghas strong ties to a child’s language development.

Is the nested sets approach to improving accuracy on Bayesian word problems simply a way of prompting a natural frequencies solution, as its critics claim? Conversely, is it in fact, as its advocates claim, a more fundamental explanation of why the natural frequency approach itself works? Following recent calls, we use a process-focused approach to contribute to answering these long-debated questions. We also argue for a third, pragmatic way of looking at these two approaches and argue that they reveal different truths about human Bayesian reasoning. Using a think aloud methodology we show that while the nested sets approach does appear in part to work via the mechanisms theorised by advocates (by encouraging a nested sets representation), it also encourages conversion of the problem to frequencies, as its critics claim. The ramifications of these findings, as well as ways to further enhance the nested sets approach and train individuals to deal with standard probability problems are discussed.

While many solutions to the apparent civic online reasoningdeficit have been put forth, few consider how reasoning is of-ten moderated by the dynamic relationship between the user’svalues and the values latent in the online content they are con-suming. The current experiment leverages Moral FoundationsTheory and Distributed Dictionary Representations to developa method for measuring the alignment between an individual’svalues and the values latent in text content. This new measureof alignment was predictive of bias in an argument evaluationtask, such that higher alignment was associated with higherratings of argument strength. Finally, we discuss how theseresults support the development of adaptive interventions thatcould provide real-time feedback when an individual may bemost susceptible to bias.

Computational thinking (CT) refers to a range of problem-solving skills applicable to computer science and everyday life.Although recent research in developmental cognitive sciencesuggests mental capacities relevant to CT may emerge quiteearly in life, research on CT, and computer science educa-tion more generally, has made little contact with this litera-ture. As a way to better bridge these fields, we explore thedevelopment of problem decomposition, a critical feature ofCT, in the spatial domain. We ask whether young childrencan break a complex spatial problem down into subcompo-nents that can be reassembled to solve the overarching prob-lem. Across two experiments (Exp.1: 4- to 7-year-olds; Exp.2:3- to 5-year-olds) that involve constructing block structures,we demonstrate that some of the key capacities underlyingproblem decomposition begin to emerge in preschool years anddevelop throughout early childhood. Although preschool-agedchildren struggle to solve an open-ended decomposition prob-lem that requires generation and execution of decompositionplans, even 4-year-olds can successfully evaluate the viabilityof these plans. These results suggest that experimental meth-ods in developmental cognitive science can inform CS edu-cation research that focuses on promoting CT; by identifyingwhen and how CT concepts emerge in early childhood, we canbetter create age-appropriate educational tools.

In their current functional form, ACT-R’s retrieval equations do not account for the left side of the RT-distance relation, that is, that as memory activation decreases, so does response time for retrieval failures. To accommodate this effect, I propose that the memory system uses the familiarity of the encoded object to gauge how much effort it should devote to retrieval. I quantify the degree of familiarity through the match score, which is the output of a global matching process. Familiarity, in turn, directly determines what the retrieval threshold should be. Adding a familiarity process orthogonal to recollection is in line with neuroimaging results, which uncover parallel familiarity and retrieval processes. The developments in this paper extend ACT-R’s memory theory into a dual process theory.

Recent work shows that distributional semantic models can be used to decode patterns of brain activity associated withindividual words and sentence meanings. However, it is yet unclear to what extent such models can be used to study anddecode brain activity patterns associated with specific aspects of semantic composition such as the negation function. Inthis paper, we investigate the extent to which distributional semantic models of action-verbs correlate with brain activityassociated with negated and affirmative sentences containing hand-action verbs. Our results show reduced correlations forsentences where the verb is in the negated context, as compared to the affirmative one, within brain regions implicated inaction-semantic processing. The results lend support to the idea that negation involves reduced access to aspects of theaffirmative representation and pave the way for further testing alternate distributional-based semantic models of negationagainst human semantic processing in the brain.

Feedback is essential for many kinds of learning, but the cognitive processes involved in learning from feedback areunclear. In models of category learning, feedback is typically treated as an error signal without a temporal component. Weconducted two simple category learning experiments that manipulated the duration of feedback (1s vs. 9s) and investigatedthe effect on learning and gaze. In two different category structures, participants in the longer feedback condition learnedfaster. The analysis of gaze data showed several findings. Participants in the 9s condition had longer fixations, and in bothconditions and experiments, participants spent far more time looking at stimulus features than the feedback. Overall, ourfindings provide empirical support for the idea that feedback processes, and temporal factors more generally, have muchto tell us about how people learn categories.

Metrics seeking to predict financial risk-taking behaviors typically exhibit limited validity. This is due to the fluid nature of an individual’s risk taking, and the influence of the mode and medium, which presents a decision. This paper presents two experiments that investigate how an existing risk elicitation task’s predictive capacity may be enhanced through the application of an interactive model of visual reasoning in a digitized version. In the first experiment, 60 participants demonstrated their reasoning process. In the second experiment, 225 participants were randomly assigned into three groups, with the validated risk elicitation task compared as a control to interactive digital and non-interactive digital stimuli with pie charts. The experiments yielded significant results, highlighting that when participants interact with a graph to reason their choices, it leads to consistent choices. The findings have implications for improvement of the risk task's validity and the deployment of digital interactive assessments beyond laboratory settings.

We propose a unified explanation of contrastive and assimila-tive adaptation aftereffects from the perspective of higher-levelcognitive processes: rational category learning and categoricalperception. We replicate (twice) previously reported assimila-tive and contrastive effects (Uznadze illusion in visual modal-ity), propose a rational computational model of the process,and evaluate our model performance against the Bayesian lo-gistic regression baseline. We conclude by discussing theo-retical implications of our study and directions for further re-search.

We used a population-level connectionist model ofcognitive development to unify a range of empiricalfindings on the influence of socioeconomic status (SES) onbehavior and brain development. The model capturedqualitative patterns of development in behavior and brainstructure, including reductions in connectivity acrossdevelopment (gray matter, cortical thickness) as behavioralaccuracy increases. Individual differences in SES wereimplemented by altering the level of stimulation available inthe environment. At the brain level, the model simulatednon-linear effects of SES on cortical surface area (Noble etal., 2015), and faster cortical thinning across development inchildren from lower SES backgrounds (Piccolo et al., 2016).At the behavioral level, the model simulated the effect ofSES on IQ, whereby gaps are observed to widen acrossdevelopment (von Stumm & Plomin, 2015). The model’smain shortcoming was insufficient growth in connectionmagnitude across development in lower SES groups,implying that some aspects of the growth of connectionstrengths may be maturational (e.g., myelination) rather thanexperience dependent.

The linguistic-simulation approach to cognition predicts that language can enable more efficient conceptual processing than sensorimotor-affective simulations of concepts. We proposed that this has implications for working memory, whereby use of linguistic labels enables more efficient representation of concepts in a limited-capacity store than representation via full sensorimotor simulation. In two pre-registered experiments, we asked participants to remember sequences of real-world objects, and used articulatory suppression to selectively block access to linguistic information, which we predicted would impair accuracy and latency of performance in an object memory recognition task. We found that blocking access to language at encoding impaired memory performance, but blocking access at retrieval unexpectedly facilitated speed of responding. These results suggest that working memory for object concepts normally relies on language but people can flexibly adapt their memory strategies when language is unavailable. Moreover, our data suggest that a sequence of up to 10 object concepts can be held in working memory when relying on sensorimotor information alone, but this capacity increases when linguistic labels are available.

Young children are selective in deciding whom to help (i.e.,they preferentially assist and share resources with prosocialversus antisocial others; Hamlin, Wynn, Bloom, & Mahajan,2011; Vaish, Carpenter, & Tomasello, 2010) but are they alsoselective in deciding how to offer help? Here we show two tofive-year-olds (N = 32; mean: 42.41 months; range 27-68months) characters who are distressed for different reasons:they are hurt, bored, or sad. Children of all ages tried to helpthe agent but the selectivity of children’s responses variedwith age and condition; in particular, children’s responses toboredom and sadness became increasingly differentiated withage.

Transfer learning is fundamental for intelligence; agents ex-pected to operate in novel and unfamiliar environments mustbe able to transfer previously learned knowledge to new do-mains or problems. However, knowledge transfer manifestsat different levels of representation. The underlying compu-tational mechanisms in support of different types of transferlearning remain unclear. In this paper, we approach the transferlearning challenge by decomposing the underlying computa-tional mechanisms involved in bottom-up associative learningand top-down causal schema induction. We adopt a Bayesianframework to model causal theory induction and use the in-ferred causal theory to transfer abstract knowledge betweensimilar environments. Specifically, we train a simulated agentto discover and transfer useful relational and abstract knowl-edge by interactively exploring the problem space and extract-ing relations from observed low-level attributes. A set of hier-archical causal schema is constructed to determine task struc-ture. Our agent combines causal theories and associative learn-ing to select a sequence of actions most likely to accomplishthe task. To evaluate the proposed framework, we compareperformances of the simulated agent with human performancein the OpenLock environment, a virtual “escape room” with acomplex hierarchy that requires agents to reason about causalstructures governing the system. While the simulated agent re-quires more attempts than human participants, the qualitativetrends of transfer in the learning situations are similar betweenhumans and our trained agent. These findings suggest humancausal learning in complex, unfamiliar situations may rely onthe synergy between bottom-up associative learning and top-down schema reasoning.

Many actions have both an intended primary effect and unin-tended, but foreseen side effects. In two experiments we inves-tigated how people morally evaluate such situations. While anegative side effect was held constant across conditions in Ex-periment 1, we varied features of the positive primary effect.We found that judgments of moral justification of actions weresensitive to the numerical ratios of helped versus harmed enti-ties as well as to the kind of state change that was induced byan agent’s action (saving entities from harm versus improvingtheir status quo). Judgments of moral responsibility for sideeffects were only sensitive to the latter manipulation. In Ex-periment 2, we found initial support for a subjective utilitarianexplanation of the moral justification judgments.

Modern deep neural networks require a tremendous amountof data to train, often needing hundreds or thousands of la-beled examples to learn an effective representation. For thesenetworks to work with less data, more structure must be builtinto their architectures or learned from previous experience.The learned weights of convolutional neural networks (CNNs)trained on large datasets for object recognition contain a sub-stantial amount of structure. These representations have par-allels to simple cells in the primary visual cortex, where re-ceptive fields are smooth and contain many regularities. In-corporating smoothness constraints over the kernel weightsof modern CNN architectures is a promising way to improvetheir sample complexity. We propose a smooth kernel regu-larizer that encourages spatial correlations in convolution ker-nel weights. The correlation parameters of this regularizer arelearned from previous experience, yielding a method with ahierarchical Bayesian interpretation. We show that our corre-lated regularizer can help constrain models for visual recogni-tion, improving over an L2 regularization baseline.

The semantics of spatial terms has attracted substantialattention in the cognitive sciences, revealing both compellingsimilarities and striking differences across languages.However, much of the evidence regarding cross-linguisticvariation pertains to fine-grained comparisons betweenindividual lexical items, while cross-linguistic similarities arefound in more coarse-grained studies of the conceptual spaceunderlying semantic systems. We seek to bridge this gap,moving beyond the semantics of individual terms to ask whatthe comparison of spatial semantic systems may reveal aboutthe conceptualization of locations in English and MandarinChinese and about the nature of potential universals in thisdomain. We subjected descriptions of 116 spatial scenes tomultidimensional scaling analyses in order to reveal thestructures of the underlying conceptual spaces in eachlanguage. In addition to revealing overlaps and divergences inthe conceptualization of space in English and Mandarin, ourresults suggest a difference in complexity, whereby Mandarinterms are accommodated by a lower-dimensional similarityspace than are English terms.

This study utilises an experiment famous in quantum physics,the Stern-Gerlach experiment, to inform the structure of an ex-perimental protocol from which a quantum cognitive decisionmodel can be developed. The ’quantumness’ of this modelis tested by computing a discrete quasi-probabilistic Wignerfunction. Based on theory from quantum physics, our hypothe-sis is that the Stern-Gerlach protocol will admit negative valuesin the Wigner function, thus signalling that the cognitive de-cision model is quantum. A crowdsourced experiment of twoimages was used to collect decisions around three questions re-lated to image trustworthiness. The resultant data was used toinstantiate the quantum model and compute the Wigner func-tion. Negative values in the Wigner functions of both imageswere encountered, thus substantiating our hypothesis. Find-ings also revealed that the quantum cognitive model was amore accurate predictor of decisions when compared to pre-dictions computed using Bayes’ rule.

The use of abstract higher-level knowledge (overhypotheses) allows humans to learn quickly from sparse data, and make predictions in new situations. Previous research has suggested that humans may be the only species capable of abstract knowledge formation, but this remains controversial, and there is also mixed evidence for when this ability emerges over human development. Kemp et al. (2007) proposed a computational model of overhypothesis formation from sparse data. We provide the first direct test of this model: an ecologically valid paradigm for testing two species, capuchin monkeys (Sapajus spp.) and 4-5-year-old human children. We compared performance to predictions made by models with and without the capacity to learn overhypotheses. Children’s choices were consistent with the overhypothesis model predictions, whereas monkeys performed at chance level.

The ability to imagine alternative possibilities plays a crucialrole in everyday cognitive functioning beginning in earlychildhood. Across two studies, we ask whether individualdifferences in young children’s (Mean Age = 5.01; SD = 0.78Range = 2) fluency in generating alternative possibilitiesrelates to a particular type of social-cognitive counterfactualjudgment, namely children’s belief in the possibility to “actotherwise” when actions go against stated strong desires (i.e.“free will”). We found that the fluency of generating ideaswas a consistent individual difference that held regardless ofdomain. We also found that individual children’s fluencypredicted judgments of free will for themselves (Study 2) butnot for others (Study 1). Our findings raise new questionsabout how counterfactual thinking enables children toovercome psychological barriers to self-control, and howstimulating the imagination facilitates developing cognitionsthat rely on it.

Children learn the meaning of number words by going througha systematic set of stages of knowledge that culminates in theirmastery of counting. Theoretical work has long suggested thatchildren’s acquisition of counting is not procedural, butsemantic: all counters understand that counting computescardinality. Yet, recent research has cast doubt on whetherearly counters truly understand the meaning of these words.Here we show that early counters also have an immatureunderstanding of how one-to-one correspondence between anordered list and a set of objects can be used to compute exactcardinality. Nonetheless, this understanding is improved whencues to quantity, such as size, are highlighted. Our results addto a growing body of work suggesting that counting is not afinal stage in children’s path to number, but a powerful toolthat they can use to build and strengthen their intuitions aboutcardinalities.

Up to 80% of words have multiple, related meanings (polysemy), yet work on early word learning has almost uniformlyassumed one-to-one mappings between form and meaning. Using a looking-while-listening procedure, we present thefirst evidence that toddlers (n=40) can recognize multiple meanings for common nouns, e.g., dog collar, shirt collar. In anEnglish-meaning condition, toddlers were tested on their ability to recognize multiple English meanings for polysemouswords such as cap (e.g., a baseball cap and a bottle cap). Another condition prompted toddlers with the same Englishwords (e.g., cap), but target referents instead corresponded to the words polysemous extension in an unfamiliar language,(e.g., lid is a meaning for Spanishs cap, tapa). Toddlers looked to the correct targets above chance on both trial types,but with greater accuracy on English-meaning trials, demonstrating a recognition of familiar word-meaning pairs and anability to infer potential new meanings.

Humans understand language based on the rich backgroundknowledge about how the physical world works, which in turn,allows us to reason about the physical world through language.In addition to the properties of objects (e.g., boats require fuel)and their affordances, i.e., the actions that are applicable tothem (e.g., boats can be driven), we can also reason about if–then inferences between what properties of objects imply thekind of actions that are applicable to them (e.g., that if we candrive something then it likely requires fuel).In this paper, we investigate the extent to which state-of-the-art neural language representations, trained on a vast amount ofnatural language text, demonstrate physical commonsense rea-soning. While recent advancements of neural language mod-els have demonstrated strong performance on various types ofnatural language inference tasks, our study based on a datasetof over 200k newly collected annotations suggests that neurallanguage representations still only learn associations that areexplicitly written down.1

Cognitive development is often characterized in term of dis-continuities, but these discontinuities can sometimes be appar-ent rather than actual and can arise from continuous develop-mental change. To explore this idea, we use as a case study thefinding by Stager and Werker (1997) that children’s early abil-ity to distinguish similar sounds does not automatically trans-late into word learning skills. Early explanations proposedthat children may not be able to encode subtle phonetic con-trasts when learning novel word meanings, thus suggestinga discontinuous/stage-like pattern of development. However,later work has revealed (e.g., through using simpler testingmethods) that children do encode such contrasts, thus favoringa continuous pattern of development. Here we propose a prob-abilistic model describing how development may proceed ina continuous fashion across the lifespan. The model accountsfor previously documented facts and provides new predictions.We collected data from preschool children and adults, and weshowed that the model can explain various patterns of learningboth within the same age and across development. The find-ings suggest that major aspects of cognitive development thatare typically thought of as discontinuities, may emerge fromsimpler, continuous mechanisms.

Human analogical ability involves the re-use of abstract, struc-tured representations within and across domains. Here, wepresent a generative neural network that completes analogiesin a 1D metric space, without explicit training on analogy.Our model integrates two key ideas. First, it operates overrepresentations inspired by properties of the mammalian En-torhinal Cortex (EC), believed to extract low-dimensional rep-resentations of the environment from the transition probabil-ities between states. Second, we show that a neural networkequipped with a simple predictive objective and highly generalinductive bias can learn to utilize these EC-like codes to com-pute explicit, abstract relations between pairs of objects. Theproposed inductive bias favors a latent code that consists ofanti-correlated representations. The relational representationslearned by the model can then be used to complete analogiesinvolving the signed distance between novel input pairs (1:3:: 5:? (7)), and extrapolate outside of the network’s trainingdomain. As a proof of principle, we extend the same architec-ture to more richly structured tree representations. We suggestthat this combination of predictive, error-driven learning andsimple inductive biases offers promise for deriving and utiliz-ing the representations necessary for high-level cognitive func-tions, such as analogy.

In this paper, we examine the success of developmentaldistributional analysis in English, German and Dutch. Weembed the mechanism for distributional analysis within anexisting model of language acquisition (MOSAIC) thatencodes increasingly long utterances, and compare resultsagainst a measure of ‘noun richness’ in child speech. We showthat, cross-linguistically, the mechanism’s success in buildingan early noun class is inversely related to the complexity of thedeterminer and noun gender system, and that merging ofdeterminers gives very similar results across languages. Theseresults suggest that children may represent grammaticalcategories at multiple levels of abstraction that reflect both thelarger category as well as its finer structure.

Given identical informational content, the order in which youreceive spatial information may heavily influence the correct-ness of your mental representation. This can reveal importantinsights into the specifics of human spatial cognition and theway we integrate information. Despite its importance in ev-eryday life, its causes and the mental processes involved stillremain an open question. Most cognitive models so far havefocused on modeling only answer distributions or just the mostfrequent answer given by all participants.In this paper we take a rather radical approach: We turn tothe individual spatial reasoner and focus our analyses on thestream of spatial information and related reaction times, i.e.,how the spatial information is represented and cognitively pro-cessed. By spanning a space of 243 cognitive spatial models,some of which outperform the current state-of-the art models,it is possible to test the goodness of general principles under-lying such models.

Achieving linguistic proficiency requires identifying words from speech, and discovering the constraints that govern the way those words are used. In a recent study of non-adjacent dependency learning, Frost and Monaghan (2016) demonstrated that learners may perform these tasks together, using similar statistical processes — contrary to prior suggestions. However, in their study, non-adjacent dependencies were marked by phonological cues (plosive- continuant-plosive structure), which may have influenced learning. Here, we test the necessity of these cues by comparing learning across three conditions; fixed phonology, which contains these cues, varied phonology, which omits them, and shapes, which uses visual shape sequences to assess the generality of statistical processing for these tasks. Participants segmented the sequences and generalized the structure in both auditory conditions, but learning was best when phonological cues were present. Learning was around chance on both tasks for the visual shapes group, indicating statistical processing may critically differ across domains

Convergent and divergent thought are promoted as keyconstructs of creativity. Convergent thought is defined andmeasured in terms of the ability to perform on tasks wherethere is one correct solution, and divergent thought is definedand measured in terms of the ability to generate multiplesolutions. However, these characterizations of convergent anddivergent thought presents inconsistencies, and do not capturethe reiterative processing, or ‘honing’ of an idea thatcharacterizes creative cognition. Research on formal modelsof concepts and their interactions suggests that differentcreative outputs may be projections of the same underlyingidea at different phases of a honing process. This leads us toredefine convergent thought as thought in which the relevantconcepts are considered from conventional contexts, anddivergent thought as thought in which they are consideredfrom unconventional contexts. Implications for the assessmentof creativity are discussed.

Although Darwinian models are rampant in the socialsciences, social scientists do not face the problem thatmotivated Darwin’s theory of natural selection: the problemof explaining how lineages evolve despite that any traits theyacquire are regularly discarded at the end of the lifetime of theindividuals that acquired them. While the rationale forframing culture as an evolutionary process is correct, it doesnot follow that culture is a Darwinian or selectionist process,or that population genetics provides viable starting points formodeling cultural change. This paper lays out step-by-steparguments as to why a selectionist approach to culturalevolution is inappropriate, focusing on the lack ofrandomness, and lack of a self-assembly code. It summarizesan alternative evolutionary approach to culture: self-otherreorganization via context-driven actualization of potential.

To understand the representations learned by neural networks(NNs), various methods of measuring unit selectivity havebeen developed. Here we undertake a comparison of four suchmeasures on AlexNet: localist selectivity (Bowers et al., 2014);precision (Zhou et al., 2015); class-conditional mean activityselectivity CCMAS (Morcos et al., 2018); and top-class se-lectivity. In contrast with previous work on recurrent neuralnetworks (RNNs), we fail to find any 100% selective ‘local-ist units’ in AlexNet, and demonstrate that the precision andCCMAS measures are misleading and suggest a much higherlevel of selectivity than is warranted. We also generated ac-tivation maximization (AM) images that maximally activatedindividual units and found that under (5%) of units in fc6 andconv5 produced interpretable images of objects, whereas fc8produced over 50% interpretable images. Furthermore, theinterpretable images in the hidden layers were not associatedwith highly selective units. We also consider why localist rep-resentations are learned in RNNs and not AlexNet.

Children exploit regular links between the meanings of wordsand the syntactic structures in which they appear to learn aboutnovel words. This phenomenon, known as syntactic bootstrap-ping, is thought to play a critical role in word learning, espe-cially for words with more opaque meanings such as verbs.We present a computational word learning model which re-produces such syntactic bootstrapping phenomena after expo-sure to a naturalistic word learning dataset, even when undersubstantial memory constraints. The model demonstrates howexperimental syntactic bootstrapping effects constitute rationalbehavior given the nature of natural language input. The modelunifies computational accounts of word learning and syntacticbootstrapping effects observed in the laboratory, and offers apath forward for demonstrating the broad power of the syntax–semantics link in language acquisition.

In natural languages, closed-class items predict open-classitems but not the other way around. For example, in English, ifthere is a determiner there will be a noun, but nouns can occurwith or without determiners. Here, we asked whether languagelearners’ computations are also asymmetrical. In threeexperiments we exposed adults to a miniature language withthe one-way dependency “if X then Y”: if X was present, Ywas also present, but X could occur without Y. We createddifferent versions of the language in order to ask whetherlearning depended on which of these categories was an open orclosed class. In one condition, X was a closed class and Y wasan open class; in a contrasting condition, X was an open classand Y was a closed class. Learning was significantly betterwith closed-class X, even though learners’ exposure wasotherwise identical. Additional experiments demonstrated thatthe perceptual distinctiveness of closed-class items driveslearners to analyze them differently; and, crucially, that theprimary determinant of learning is the mathematicalrelationship between closed- and open-class items and not theirlinear order. These results suggest that learners privilegecomputations in which closed-class items are predictive of,rather than predicted by, open-class items. We suggest that thedistributional asymmetries of closed-class items in naturallanguages may arise in part from this learning bias.

In this paper, we study cross-cultural differences in strategicreasoning in turn-taking games, as related to game-theoreticnorms as well as affective aspects such as trust, degrees of risk-taking and cooperation. We performed a game experiment toinvestigate how these aspects play a role in reasoning in simpleturn-based games, known as centipede-like games, across threecultures, that of The Netherlands, Israel and India. While thereis no significant main effect of nationalities on the behaviourof players across games, certain unexpected interactive effectsare found in their behaviour in particular games.

When people understand language about their own actions they activate premotor regions they use to perform these actions.Do people understand language about other peoples actions by imagining how they perform these actions themselves, orhow they perceive others performing them? Here, we recorded BOLD fMRI while left- and right-handers read about andthen imagined their own unimanual actions (e.g. you write) or others actions (e.g. she writes). When imagining theirown manual actions, participants preferentially activated PMC circuits controlling their dominant hand. By contrast, whenimagining others actions, participants PMC activity reflected both how they perform actions themselves and how theytypically see actions performed by right-handers (about 90% of people they see). Language-induced motor imagery forour own actions reflects how we use our own bodies, whereas imagery for others actions also reflects how others use theirbodies, even if their bodies differ from our own.

Scientific norms value skepticism; many religious traditionsvalue faith. We test the hypothesis that these differentattitudes towards inquiry and belief result in differentinferences from epistemic behavior: Whereas the pursuit ofevidence or explanations is taken as a signal of commitmentto science, forgoing further evidence and explanation is takenas a signal of commitment to religion. Two studies (N = 401)support these predictions. We also find that deciding topursue inquiry is judged more moral and trustworthy, withmoderating effects of participant religiosity and scientism.These findings suggest that epistemic behavior can be a socialsignal and shed light on the epistemic and social functions ofscientific vs. religious belief.

Event cognition is a rapidly developing and promising research area. Meanwhile, some domains are not considered in detail in this scope. In particular, event cognition is not precisely explored from the perspective of cognitive development. In this paper, we compare the capacity to cut a visual narrative into events for kindergarten students, primary school students, high school students and adults. “The pear film” by W. Chafe (1975) is used as the material for our experiment. We also examine a correlation between event comprehension and other cognitive skills for primary school students. Our work provides clear evidence that, in contrast with high school students and adults, kindergarten students and primary school students perceive visual narrative on the surface level.

Becoming a proficient reader is a critical skill that supportsfuture learning. Toward the end of the primary grades,reading becomes increasingly automatized, and children beginto transition from learning-to-read to reading-to-learn. Yet,the design of beginning reader books may be suboptimal fornovice readers. Colorful illustrations that contain irrelevantinformation (i.e., seductive details) presented in closeproximity to the text may increase attentional competitionbetween these sources of information; thus, hamperingdecoding and reading comprehension. Study 1 examines thishypothesis by experimentally manipulating components of thebook design (e.g., presence/absence of seductive details) andinvestigating its effect on attention and reading performancein first grade students. In Study 2, we conduct an analysis inwhich we identify common design features in books forbeginning readers and examine the prevalence of designfeatures that were found to tax attention in Study 1 and inprior research. Collectively this work identifies an importantopportunity in which instructional materials can be optimizedto better support children as they learn-to-read.

We investigated whether induced affective states can affect the process and outcomes of decisions under risk. A mood induction task was used to elicit a positive or negative mood in a sample of adult participants (N=48). The participants then responded to 28 decision problems, each offering a choice between two mixed-domain risky alternatives. The dependent variables of interest were decision-making choices, as well as an eye-tracking based attentional measure: the total fixation durations for certain critical aspects of the two presented risky decision options. Mood condition did not have a significant main effect on participants’ choices, or on mean total fixation time for problems. However, fixation times showed a three- way interaction between mood condition, domain (gain versus loss), and time (block). The fixation time data also provided some general insights into participants’ patterns of attention allocation during decision-making. They generally spent more time looking at values compared to probabilities, and more time looking at potential gains compared to losses (although this difference declined over time, especially for positive-mood participants).

Human concept learning is surprisingly robust, allowing forprecise generalizations given only a few positive examples.Bayesian formulations that account for this behavior requireelaborate, pre-specified priors, leaving much of the learningprocess unexplained. More recent models of concept learningbootstrap from deep representations, but the deep neural net-works are themselves trained using millions of positive and neg-ative examples. In machine learning, recent progress in meta-learning has provided large-scale learning algorithms that canlearn new concepts from a few examples, but these approachesstill assume access to implicit negative evidence. In this paper,we formulate a training paradigm that allows a meta-learningalgorithm to solve the problem of concept learning from fewpositive examples. The algorithm discovers a taxonomic prioruseful for learning novel concepts even from held-out supercat-egories and mimics human generalization behavior—the firstto do so without hand-specified domain knowledge or negativeexamples of a novel concept.

Recent work on adversarial examples demonstrates a brittleness of many state-of-the-art machine learning systems. Weinvestigate one human analog, asking: What fraction of natural speech can be turned into illusions which alter humans per-ception or result in different people having significantly different perceptions? Using generated videos, we first empiricallyestimate that 17% of words occurring in natural speech have some susceptibility to the McGurk effect–the phenomenonby which adding a carefully chosen video clip to the audio channel affects the viewers perception of the message. We de-velop a bag-of-phonemes prediction model for word-level illusionability that we extend with natural language modeling tobuild a sentence-level framework. We train an instantiation using Amazon Mechanical Turk evaluations on sentence-levelillusions. Finally we generate several new instances of the Yanny/Laurel illusion, demonstrating that it is not an isolatedoccurrence. The surprising density of illusionable instances warrants further investigation from cognitive and securityperspectives.

An experiment aiming to assess the use of stopping rules in information acquisition was performed. An exploratoryexperimental paradigm was used. Participants (47 healthy individuals) were requested to make a decision in 24 financialscenarios with the possibility of buying information pieces. Participants were able to accept, reject or choose not todecide. Behavioral, EEG, ECG and Eyetracker data were recorded and integrated offline for analysis. Results showedthat participants followed primarily Bayesian calculations in order to determine when to cease information acquisition anddecide. Participants would tend to rely more on the valences (BAL) of the information acquired (positive or negative)than on sheer quantity. Acceptance tended to be made with mean positive BAL, rejection with mean negative BAL andprocrastination with mean zero BAL. Uncertainty was seen to affect the information acquisition and decision process;EEG data suggest Slow Cortical Potentials at fronto-central electrodes for risk with low consequences and uncertaintywith high consequences. Eyetracker data shows greater mean fixation time for decisions and information areas of interest(AOI). Heart rate data shows no difference in scenarios and/or information acquisition behavior, meaning that the decisionscenarios did not elicit significant emotional engagement. Integrated psychophysiological measures were of importantassistance to the conclusions given that they provided information as to what happened or not both behaviorally andphysiologically.

Outside university, students encounter disciplinary practices mediated by technological resources. In this sense, the real world is decidedly resource-rich. In contrast, most educational assessments remain decidedly resource-poor. Situated versus mindbased perspectives of cognition fundamentally differ in the role they ascribe to such resources in cognition and learning. To mindbased perspectives, they are a source of input, to situated perspectives they are constitutive to cognition itself. We assessed the validity of resource-rich versus resource-poor assessments of learning outcomes from resource-rich versus resource-poor learning activities. The study implemented an in-class 2x2 between-subjects experimental design in an introductory programming course with 192 first semester BSc engineering students. Both types of assessment were sensitive to differences in learning outcomes, indicating validity for both. Results indicate resource-rich assessments may be more ecologically valid, while – intriguingly – the resource-poor assessments were more sensitive to transfer of learning. Furthermore, the resource-rich learning activities better facilitated learning for transfer.

Listeners must integrate multiple sources of information toconstruct an interpretation of a sentence. We concentrate hereon the alignment of prosodic and syntactic grouping duringonline sentence comprehension. We present the results from apupillometry study on the use of prosodic boundaries in resolv-ing well-known attachment ambiguities. Using growth curveanalyses to capture the non-linear dynamics of pupil dilation,we found increased pupil excursions for sentences that weredisambiguated towards the dispreferred, non-local relation, es-pecially when accompanied by supporting prosodic informa-tion. However, when prosodic and structural information didnot align, pupillary response was muted, potentially indicatinga failure to commit to a specific interpretation. More generally,the study shows how the currently under-utilized pupillometrymethod offers insights into spoken language comprehension.

This study investigated how participant’s specificity in shar-ing of information in collaborative problem solving was criti-cal to them reaching a successful shared perspective. We ana-lyzed participants’ communication strategies in a collaborativetask designed to make finding common ground challenging.We set out to better understand the difference between suc-cessful and unsuccessful collaborations by conducting a cog-nitive task analysis. From participants’ utterances, we inferredcognitive processes associated with repeating communicationmoves and coded those processes as if-then production rules.We thereby specified the communication strategies used duringinteractions and developed a production-rule model to explainwhether and how shared perspective developed or not. Ourcognitive task analysis indicated that although all collaboratingpairs described the objects they were seeing with a variety offeatures, the successful pairs were more specific in using com-binations of features. Quantitatively, we found significant cor-relations between frequency of combined feature statementsand success in sharing perspectives.

Decision models are formal algorithms that are used to represent decision processes and predict choice across a wide rangeof disciplines. These models are often highly complex, which makes it difficult to understand the relationships betweendifferent models, the unique features of individual models and, in turn, the fundamental properties of choice behaviorcaptured by these models. We address this issue in a large-scale computational analysis that uses parameter bootstrappingcross-fitting techniques to derive pairwise measures of decision model distances. Our analysis includes over 80 prominentmodels of risky and intertemporal choice, and results in an ontology of decision models, with data-driven model clustersand hierarchies that synthesize over seven decades of quantitative research on human choice behavior.

Visual thinking plays a central role in human cognition, yet weknow little about the algorithmic operations that make itpossible. Starting with outputs of a JIM-like model of shapeperception, we present a model that generates object file-likerepresentations that can be stored in memory for futurerecognition, and can be used by a LISA-like inference engineto reason about those objects. The model encodes structuralrepresentations of objects on the fly, stores them in long termmemory, and simultaneously compares them to previouslystored representations in order to identify candidate sourceanalogs for inference. Preliminary simulation results suggestthat the representations afford the flexibility necessary forvisual thinking. The model provides a starting point forsimulating not only object recognition, but also reasoningabout the form and function of objects.

People often reason about omissions. One line of researchshows that people can distinguish between the semantics ofomissive causes and omissive enabling conditions: forinstance, not flunking out of college enabled you (but didn’tcause you) to graduate. Another line of work shows that peoplerely on the normative status of omissive events in inferringtheir causal role: if the outcome came about because theomission violated some norm, reasoners are more likely toselect that omission as a cause. We designed a novel paradigmthat tests how norms interact with the semantics of omissiveenabling conditions. The paradigm concerns the circuitry of amechanical device that plays music. Two experiments used theparadigm to stipulate norms and present a distinct set ofpossibilities to participants. Participants chose which causalverb best described the operations of the machine. The studiesrevealed that participants’ responses are best predicted by theirtendency to consider the semantics of omissive relations. Incontrast, norms had little to no effect in participants’ responses.We conclude by marshaling the evidence and considering whatrole norms may play in people’s understanding of omissions.

The learning and generalisation of grammatical regularities is fundamental to successful language acquisition and use. Research into statistical learning has started to consider how this process occurs through the implicit detection and assimilation of grammatical regularities. This study focuses on how adults and children generalise regularities and explores the role of explicit knowledge in this process. Across three experiments, adults and children learnt an artificial language containing two semantic categories denoted by a co-occurring determiner and suffix. Explicit knowledge of the regularities was associated with generalisation performance in adults but not children, even when adult word level knowledge was similar to children’s. The implications of these results for developmental theories of grammatical generalisation are discussed.

Speech-acts can have literal meaning as well as pragmaticmeaning, but these both involve consequences typically in-tended by a speaker. Speech-acts can also have unintentionalmeaning, in which what is conveyed goes above and beyondwhat was intended. Here, we present a Bayesian analysis ofhow, to a listener, the meaning of an utterance can significantlydiffer from a speaker’s intended meaning. Our model em-phasizes how comprehending the intentional and unintentionalmeaning of speech-acts requires listeners to engage in sophisti-cated model-based perspective-taking and reasoning about thehistory of the state of the world, each other’s actions, and eachother’s observations. To test our model, we have human partic-ipants make judgments about vignettes where speakers makeutterances that could be interpreted as intentional insults or un-intentional faux pas. In elucidating the mechanics of speech-acts with unintentional meanings, our account provides insightinto how communication both functions and malfunctions.

Previous studies have shown that people can make adaptive in-ferences based on memory-based simple heuristics such asrecognition, fluency, or familiarity heuristic. In the presentstudy, we discussed the adaptive nature of memory-based sim-ple heuristics in a group decision making setting. In particular,we examined how the diversity of memory affected group de-cision making when group members were assumed to make in-ferences based on the familiarity heuristic. We predicted that,when the group members’ memories were diverse, group deci-sion making would become more accurate. To examine thisprediction, we conducted a behavioral experiment and com-puter simulations, and our results generally supported the pre-diction. We discuss the role of diverse memories in generatingadaptive group decision making.

Humans readily form social impressions of faces at a glance,whether assessing trustworthiness, attractiveness, or domi-nance. However, little is understood about how such compu-tations are carried out neurally. Here, we leverage a computa-tional model of human face perception to quantify and charac-terize the extent to which macaque monkey face patch neuronsencode information relevant for social trait perception. Specif-ically, we use a social trait prediction model to estimate thesocial trait ratings for face stimuli viewed by monkeys duringa neural recording experiment. We find that, while the monkeyface patch neurons are linearly tuned to facial features differ-ent from those used by humans to make social judgments, thesubspace spanned by the face patch neurons and the subspacespanned by the facial features supporting human social percep-tion are highly overlapping. This result implies that the infor-mation present in the monkey face patch neurons are largelysufficient, after linear decoding, to support human social per-ception, thus shedding light on the biological origin of humansocial processing of faces.

We examined if using meta-memory judgments to controlrestudy choices has a positive impact on undergraduatestudents’ memory performance, or whether simply makingmeta-memory judgments improved memory performance. 72undergraduates at the University of Exeter were randomlydivided into three groups. Participants in group A, had a chanceto make meta-memory judgments and restudied the words theychose (self- selection). Participants in group B, also mademeta-memory judgments, but restudy for this group wasmatched to that of Group A (control 1). Group C did not havea chance to make meta-memory judgments and were alsomatched to Group A for restudy opportunities (control 2). Theresults indicated that making meta-memory judgments had apositive overall impact on memory performance ifundergraduates were allowed to control their restudyopportunities. Groups B and C showed no differences inmemory performance, which means that making meta-memoryjudgments did not automatically improve memoryperformance. Group A restudied more of the words that theyhad rated as least well learned, and there were no significantdifferences between groups on test for the restudied words.

The present study explores the impact of including social media messages on learning from television programs that broadcast pseudoscientific claims. Seventy-seven university students were allocated to one of three experimental conditions: viewing television content with messages supporting the claim, with opposing messages, or without any messages presented. Memory retention did not differ among the conditions. However, social media messages influenced validation of the arguments claimed in the video. The participants who watched the video with opposing messages showed significant decrease in positive attitude toward the pseudoscientific technology that claimed to be effective in the video. Additionally, the participants who watched the video with supporting messages made fewer critical comments and showed willingness to donate more to the activity using the pseudoscientific technology. The impact of including social media messages and the process of attitude change are discussed.

Speech input is often noisy and ambiguous. Yet listenersusually do not have difficulties understanding it. A keyhypothesis is that in speech processing acoustic-phoneticbottom-up processing is complemented by top-downcontextual information. This context effect is larger when theambiguous word is only separated from a disambiguating wordby a few syllables compared to many syllables, suggesting thatthere is a limited time window for processing acoustic-phoneticinformation with the help of context. Here, we argue that therelative weight of bottom-up and top-down processes may bedifferent for languages that have different phonologicalproperties. We report an experiment comparing two closelyrelated languages, Danish and Norwegian. We show thatDanish speakers do indeed rely on context more thanNorwegian speakers do. These results highlight the importanceof investigating cross-linguistic differences in speechprocessing, suggesting that speakers of different languagesmay develop different language processing strategies.

How can people learn to make better decisions and be-come more far-sighted? To make the underlying learningmechanisms more accessible to scientific inquiry, we developa computational method for measuring the time course ofexperience-dependent changes in people’s planning strategies.We validated our method on simulated and empirical data: onsimulated data its inferences were significantly more accuratethan simpler approaches, and when evaluated on human datait correctly detected the plasticity-enhancing effect of perfor-mance feedback. Having validated our method, we illustratehow it can be used to gain new insights into the time courseand nature of cognitive plasticity. Future work will leverageour method to i) reverse-engineer the learning mechanismsenabling people to acquire complex cognitive skills such asplanning and problem-solving and ii) measure individual dif-ferences in cognitive plasticity.

So-called insight problems are believed to tap into sudden, creative thinking that is crucial for real problems. In contrast, recent findings suggest that solving insight problems depends on the same cognitive mechanisms that underpin systematic, analytical thinking. However, existing studies may have low ecological validity, because insight problems were usually presented in static formats (on paper, computer screen) which allowed no physical interaction with the problem elements. This study administered 8 established insight problems either in the static or interactive variants. It also probed two markers of analytical thinking: working memory capacity and reasoning ability. Virtually no difference in performance was observed between the static and interactive variants of insight problems with regard to (1) solution rate, (2) subjective experience of suddenness, pleasure, and relief accompanying the solutions, as well as (3) correlations with the working memory capacity and analytical reasoning tests. These results suggest that externalized/embodied/situated factors play no substantial role in insight problem solving and the crucial parts of this process seem to occur in the mind of a solver.

Scientists seek to discover truths that are interesting andimportant. We characterized these notions by askinglaypeople to assess the importance, interestingness,surprisingness, practical value, scientific impact, andcomprehensibility of research reported in the journalsScience and Psychological Science. These judgments wereinterrelated in both samples, with interest predicted bypractical value, surprisingness, and comprehensibility, andimportance predicted mainly by practical value. However,these judgments poorly tracked the academic impact of theresearch, measured by citation counts three and seven yearslater. These results suggest that although people haveinternally reliable notions of what makes scienceinteresting and important, these notions do not trackscientific findings’ actual impact.

Research using short-term chess training programs has indicated an enhancement of cognitive functioning among children.The aim of the study was to investigate the effect of 1-year systematic chess training on the creativity and intelligence ofchildren. A pretestposttest with control group design was used. Children who were studying in two government schoolsand two private schools (grades 39) were selected randomly. They were then randomly assigned to experimental andcontrol groups, with 88 (50 boys, 38 girls) children in the experimental group and 90 (57 boys, 33 girls) children inthe control group. The experimental group underwent weekly 1-hour chess training for 1 year, while the control groupwas actively involved in extracurricular activities offered by the school during the same period. Creativity was measuredby WallachKogan Creativity Test (Indian adaptation) and intelligence was measured by subtests of Wechsler IntelligenceScale for Children: Fourth edition (WISC-IV), India. Analysis of covariance (ANCOVA) revealed significant improvementin total creativity and Full Scale Intelligence Quotient (FSIQ) for experimental group compared to the control group. Chesstraining as part of school activities appears to have a wide spectrum of outcomes.

Categories carve up the world in a structured way, allowingpeople to inductively reason about the properties of novel ex-emplars. Children are still in the process of learning categorystructure, and often fail to leverage the inductive power of theserepresentations to their advantage. For example, young chil-dren generally fail to recognize the value of sampling diverseexemplars to support category-wide generalization. This studyinvestigates whether teaching children the structure within anatural category increases diversity-based inductive reasoning.In an informal science learning environment, we presented 259children aged 5 to 8 years with exemplars of the three maintypes of birds: raptors, songbirds, and waterbirds. After a shortdialogue pointing out the various within-type similarities andbetween-type differences, children’s diversity-based inductivereasoning did not significantly improve, despite them evidenc-ing a better understanding of the category’s structure. Instead,children tended to avoid sampling waterbirds, the least typicalcluster of birds. These patterns suggest that children’s neglectof sample diversity is unlikely to be solely due to their relativeignorance of category structure.

The ability to maintain attentive state over a period of time (i.e., Selective Sustained Attention) is important for higher- order cognition but challenging to assess in preschool-age children. The TrackIt task was developed to address this challenge and has been argued to be sensitive to age-related differences in selective sustained attention in 3- to 5-year-old children. However, it remains unclear whether this improvement with age also (or predominantly) reflects improvement in children’s knowledge of different shapes used as stimuli in this task in prior studies. The current study addressed this possibility. Consistent with prior studies, we found clear age-related improvement in performance on TrackIt. However, we did not find evidence that shape knowledge played a role in TrackIt performance for children aged 2 to 5, suggesting that increased knowledge of geometric shapes is not sufficient to explain age-related improvement in performance and helping to validate TrackIt as an assessment of Selective Sustained Attention.

This paper investigates whether a curiosity-based strategy could be beneficial to word learning. Children are active conversation partners and exert considerable influence over the topics that are discussed in conversation with their parents. As the choice of topics is likely to be intrinsically motivated, a formalization of curiosity is implemented in a word learning model. The model receives annotated Flickr30k Entities images as input, and is trained in two conditions. In the curious condition, the model chooses objects to talk about from the scene according to the curiosity mechanism, whereas in the random condition, the model receives randomly chosen objects as input. The goal of this study is to show how a curious, active choice of topics by a language learner improves word learning compared to random selection. Curiosity is found to make word learning faster, increase robustness, and lead to better accuracy.

Few experiments have examined how people reason aboutdurative relations, e.g., "during". Such relations posechallenges to present theories of reasoning, but manyresearchers argue that people simulate a mental timeline whenthey think about sequences of events. A recent theory positsthat to mentally simulate durative relations, reasoners do notrepresent all of the time points across which an event mightendure. Instead, they construct discrete tokens that stand inplace of the beginnings and endings of those events. The theorypredicts that when reasoners need to build multiple simulationsto solve a reasoning problem, they should be more prone toerror. To test the theory, an experiment provided participantswith sets of premises describing durative relations; theyassessed whether the sets were consistent or inconsistent. Theresults of the experiment validated the theory's prediction. Weconclude by situating the study in recent work on temporalthinking.

One method for uncovering the subprocesses of mental processes is the “Additive Factors Method” (AFM). The AFM uses reaction time data from factorial experiments to infer the presence of separate processing stages. This paper investigates the conceptual status of the AFM. It argues that one of the AFM’s underlying assumptions is problematic in light of recent developments in cognitive neuroscience. Discussion begins by laying out the basic logic of the AFM, followed by an analysis of the challenge presented by neural reuse. Following this, implications are analysed and avenues of response considered. Keywords: additive factors method; seriality assumption; anatomical modularity; neural reuse.

Previous research has found that verbs are more likely to adapttheir meaning to the semantic context provided by a noun thanthe reverse (verb mutability). One possible explanation for thiseffect is that verbs are more polysemous than nouns, allowingfor more sense-selection. We investigated this possibility bytesting polysemy as a predictor of semantic adjustment. Ourresults replicated the verb mutability effect. However, wefound no evidence that polysemy predicts meaning adjustmentin verbs. Instead, polysemy was found to predict meaningadjustment in nouns, while semantic strain was found to predictmeaning adjustment in verbs (but not nouns). This suggeststhat processes of meaning adjustment may be different fornouns vs verbs.

The purpose of the present study was to investigate the effects of addressee of verbalization, self or other, on insight problem solving. Thirty-five participants were assigned to one of the three conditions: toward-self verbalization, toward-other verbalization, or irrelevant verbalization (control). A 3-minute verbalization phase was inserted after 5 minutes of solving the T-puzzle. The participants were asked to write down their thoughts during the first 5 minutes as a record in the toward- self verbalization condition, and as an instruction for other participants in the toward-other verbalization condition. The participants in the control condition were required to write down their concerns. After that, they were asked to engage in the puzzle again for 10 minutes. The results showed a detrimental verbalization effect while allowed a wide range of effects for the self vs other distinction going in either direction. We are using this study as a basis for a pre-registered repo

In a noisy but structured world, memory can be improvedby enhancing limited stimulus-specific memory with statisti-cal information about the context. To do this, people have tolearn the statistical structure of their current environment. Wepresent a Sequential Monte Carlo (particle filter) model of howpeople track the statistical properties of the environment acrossmultiple contexts. This model approximates non-parametricBayesian clustering of percepts over time, capturing how peo-ple impute structure in their perceptual experience in order tomore efficiently encode that experience in memory. Each trialis treated as a draw from a context-specific distribution, wherethe number of contexts is unknown (and potentially infinite).The model maintains a finite set of hypotheses about how thepercepts encountered thus far are assigned to contexts, updat-ing these in parallel as each new percept comes in. We applythis model to a recall task where subjects had to recall the posi-tion of dots (Robbins, Hemmer, & Tang, 2014). Unbeknownstto subjects, each dot appeared in one of a few pre-defined re-gions on the screen. Our model captures subjects’ ability tolearn the inventory of contexts, the statistics of dot positionswithin each context, and the statistics of transitions betweencontexts—as reflected in both recall and prediction.

Recently, cognitive science researchers have revealed that human cognition involves the body and is a kind of self- organization phenomenon emerging from dynamic interaction across body-brain-environment. Some of the data obtained from such cognitive, behavioral, or physiological activities are often complicated in terms of non-stationarity and nonlinearity. Researchers have proposed several analytical tools and frameworks. Recurrence analysis is one of the nonlinear data analyses developed in nonlinear dynamics. It has been applied to various research fields, including cognitive science, for language (categorical) data or motion (continuous) data. However, most previous studies have applied recurrence methods individually to categorical or continuous data. We aimed to integrate these methods to investigate the relationship between speech (categorical) and motion (continuous) directly. To do so, we added temporal information (a time stamp) to categorical data and applied the joint recurrence analysis methods to visualize and quantify speech-motion coordination during a rap performance. Our pilot study suggested the possibility of visualizing and quantifying it.

We present a novel method for constructing neurally imple-mented spatial representations that we show to be useful forbuilding models of spatial cognition. This method representscontinuous (i.e., real-valued) spaces using neurons, and iden-tifies a set of operations for manipulating these representa-tions. Specifically, we use “fractional binding” to construct“spatial semantic pointers” (SSPs) that we use to generate andmanipulate representations of spatial maps encoding the posi-tions of objects. We show how these representations can betransformed to answer queries about the location and identitiesof objects, move the relative or global position of items, andanswer queries about regions of space, among other things.We demonstrate that the neural implementation in spiking net-works of SSPs have similar accuracy and capacity as the math-ematical ideal.

Previous work has argued that young children do not answercounterfactual questions (e.g. “what would have happened?”)by constructing simulations of alternative possibilities in theway adults do. Here, we propose that children can engage insimulation when answering these questions, but considerdifferent counterfactual possibilities than adults. While mostprevious research has relied on narrative stimuli, we use causalperception events, which are understood even in infancy. InExperiment 1, we replicate earlier findings that childrenstruggle with counterfactual reasoning, but show that they arecapable of conducting the required simulations in a predictiontask. In Experiment 2, we use a novel multiple-choice methodthat allows us to study not only when children get it right, butalso how they get it wrong. We find evidence that 4-year-oldsengage in simulation, but preserve only some features of whatactually happened and not others.

Transmission of information by means of language is a po-tentially lossy process. Especially adjunct information, suchas the graded degree of evidence, is a piece of informationthat seems prima facie likely to be distorted by reproductionnoise. To investigate this issue, we present the results of a two-step iterated narration study: first, we collected a corpus of250 crime story reproductions that were produced in parallelreproduction chains of 5 generations in depth, for 5 differentseed stories; a second separate large-scale experiment then tar-geted readers’ interpretation of these reproductions. Crucially,strength of evidence for the guilt of each story’s suspect(s)was manipulated in the initial seed stories. Across genera-tions, readers’ guilt perceptions decreased when the evidencewas originally strong, but remained stable when evidence wasoriginally weak. Analysis of linguistic measures revealed thatdissimilarity between a seed story and its reproduction, storylength, and amount of hedging language affected the readers’own guilt perception and the readers’ attribution of guilt per-ception to the author differently. The results provide evidencethat evidential information indeed influences guilt perceptionin complex ways.

Animal behavior is not driven simply by its current observa-tions, but is strongly influenced by internal states. Estimatingthe structure of these internal states is crucial for understand-ing the neural basis of behavior. In principle, internal statescan be estimated by inverting behavior models, as in inversemodel-based Reinforcement Learning. However, this requirescareful parameterization and risks model-mismatch to the ani-mal. Here we take a data-driven approach to infer latent statesdirectly from observations of behavior, using a partially ob-servable switching semi-Markov process. This process has twoelements critical for capturing animal behavior: it captures non-exponential distribution of times between observations, andtransitions between latent states depend on the animal’s actions,features that require more complex non-markovian models torepresent. To demonstrate the utility of our approach, we applyit to the observations of a simulated optimal agent performinga foraging task, and find that latent dynamics extracted by themodel has correspondences with the belief dynamics of theagent. Finally, we apply our model to identify latent states inthe behaviors of monkey performing a foraging task, and findclusters of latent states that identify periods of time consistentwith expectant waiting. This data-driven behavioral model willbe valuable for inferring latent cognitive states, and thereby formeasuring neural representations of those states.

Understanding how a person thinks, i.e., measuring a singleindividual’s cognitive characteristics, is challenging becausecognition is not directly observable. Practically speaking, stan-dardized cognitive tests (tests of IQ, memory, attention, etc.),with results interpreted by expert clinicians, represent the stateof the art in measuring a person’s cognition. Three areas ofAI show particular promise for improving the effectiveness ofthis kind of cognitive testing: 1) behavioral sensing, to morerobustly quantify individual test-taker behaviors, 2) data min-ing, to identify and extract meaningful patterns from behav-ioral datasets; and 3) cognitive modeling, to help map ob-served behaviors onto hypothesized cognitive strategies. Webring these three areas of AI research together in a unified con-ceptual framework and provide a sampling of recent work ineach area. Continued research at the nexus of AI and cogni-tive testing has potentially far-reaching implications for soci-ety in virtually every context in which measuring cognition isimportant, including research across many disciplines of cog-nitive science as well as applications in clinical, educational,and workforce settings.

The interrelatedness of lexical items, typically defined in termsof semantic or phonological overlap, has been shown toinfluence language learning. Given that language also containssequential structure, we investigate here whether temporaloverlap among words, formalized in graph theoretical terms asdisplaying the property of community structure, might alsohave consequences for learning. We create a graph organizedinto clusters of densely interconnected nodes with relativelysparse external connections. After assigning a novelpseudoword to each node in the graph, we generate acontinuous sequence of visually-presented items by walkingalong its edges. Word-by-word reading times suggest thatlearners are indeed sensitive to temporal overlap.Compellingly, we also demonstrate that prior exposure tosequences organized into temporal communities influencesperformance on a subsequent word recognition task.

The internal representations of three dimensional objectswithin visual memory are only partially understood. Previousresearch suggests that 3D object perception is viewpoint de-pendent, and that the visual system stores viewpoint perspec-tives in a biased manner. The aim of this project was to ob-tain detailed estimates of the distributions of 3D object viewsin shared human memory. We devised a novel experimentalparadigm based on transmission chains to investigate memorybiases for the 3D orientation of objects. We found that memorytends to be biased towards orthogonal diagrammatic perspec-tives aligned with the ends of the standard basis for a set ofcommon 3D objects, and that these biases are strongest for sideviews as well as top or bottom views for a small set of bilater-ally symmetric objects. Finally, we found that views sampledfrom the modes were easier to categorize in a recognition task.

Words influence cognition well before infants know their specific meanings. For example, three-month-olds are more likelyto form visually-based categories when exemplars are paired with spoken words than with sine-wave tones. We testedwhether structure in infants environment can foster this effect. Caregivers often use exaggerated showing gestures whenlabeling objects, presenting words in synchrony with object motion, and creating amodal temporal structure in auditoryand visual modalities. Because attention to amodal structure attenuates encoding information specific to just one modality,we hypothesized that it can lead auditory signals to impact visually-based categorization. Indeed, when 3-month-olds arefamiliarized to videos in which tones occur in synchrony with object motion, tones subsequently facilitate categorization,just like words. Moreover, familiarizing infants to word-object synchrony enhances their subsequent categorization in thepresence of words. These results suggest that structure in infants environment may contribute to the special effects thatwords have on categorization.

We investigate whether children preferentially selectinformative actions and make accurate inferences from theoutcome of their own interventions in a causal learning task.Four- to six-year-olds were presented with a novel systemcomposed of two gears that could operate according to twopossible causal structures (single or multiple cause). Giventhe choice between interventions (i.e., removing one of thegears to observe the remaining gear in isolation), childrendemonstrated a clear preference for the action that revealedthe true causal structure, and made subsequent causaljudgments that were consistent with the outcome observed.Experiment 2 addressed the possibility that performance wasdriven by children’s tendency to select an intervention thatwould produce a desirable effect (i.e., spinning gears), ratherthan to disambiguate the causal structure. The results replicateour initial findings in a context in which the informativeaction was less likely to produce a positive outcome than theuninformative one. We discuss these results in terms of theirsignificance for understanding both the development ofscientific reasoning and the role of self-directed actions inearly learning.

Do language learners benefit from exposure to input that is more predictable and has lower entropy? Frequency is known to facilitate learning (more frequent words acquired earlier). However, frequency is only one measure of the distributional structure of the linguistic input. Here, we show that entropy also impacts language learning: adults show better word segmentation in an artificial language when the sequence has lower entropy (created by making one word more frequent). Segmentation improved both for the language as a whole, and for the less frequent words, despite appearing half the number of times. These results illustrate the facilitative effect of entropy reduction on language learning. Theoretically, they show that the effect of frequency is relative, not absolute, and that language learners are sensitive to more complex measures of the environment. Methodologically, they suggest that the prevalent use of uniform distributions in word segmentation studies may underestimate learners’ abilities.

Inductive reasoning is constrained by several principles that govern how we choose to generalize evidence to new cases. Here we focus on diversity principle of induction, which describes the tendency to favor inductive arguments that include a diverse sample of evidence over those that include a homogenous sample of evidence. Several studies reveal that adherence to the diversity principle is influenced by a range of conceptual processes, such as an individuals’ prior knowledge or expectations about the categories and properties represented in the evidence. In the two experiments reported here we examined a contextual factor of the available evidence – the spatial separation of evidence exemplars – that we expected would impact how people reason about diverse samples. We found that when the pictures (Experiment 1) or labels (Experiment 2) used to represent evidence exemplars were presented far apart (approximately 10 cm), participants showed a greater willingness to endorse arguments with diverse exemplars than those with homogenous sample, relative to when these exemplars were placed in close proximity (approximately 1 cm apart). We discuss these results as they relate to existing models of induction.

Function learning research has highlighted the importance ofhuman inductive biases that facilitate long-range extrapola-tions. However, most previous research is focused on aggre-gate errors or single-criterion extrapolations. Thus, little isknown about the underlying psychological space in which con-tinuous relationships are represented. We ask whether peoplecan learn the distributional properties of new classes of rela-tionships, using Markov Chain Monte Carlo with People, andfind that (1) people are able to track not just the expected pa-rameters of a linear function, but information about the vari-ability of functions in a specific context and (2) in many casesthese spaces over parameters exhibit multiple modes.

We propose that when humans learn sets of relationships theyare able to learn the abstract structure or type of a family of re-lationships, and exploit that knowledge to improve their abilityto learn and generalize in the future, especially in the face ofsparse or ambiguous data. In two experiments we found thatparticipants choose patterns and extrapolate in ways consistentwith sets of previously learned relations, as measured by ex-trapolation judgments and forced-choice tasks. We take theseresults to suggest that humans can detect shared abstract re-lations and apply this learned regularity to perform rapid andflexible generalization.

Contemporary evidence suggests that sleep contributes to theextraction of gist from previously encoded experiences, aprocess that relies on compressed memory replay. While thefunctional significance of the time compression is not fullyunderstood, a recent ‘temporal scaffolding’ model suggestedthat compression allows associating encoded events thathappened in disparate times, a critical feature when extractinggist of a temporal nature. We examined this hypothesis usinga novel behavioral paradigm. Subjects were first presentedwith word pairs that could form a new composite word ifcombined (e.g., car, pet --> carpet), and then tested onwhether they falsely recognize seeing the composite word.When subjects napped in between exposure and testing, falsememories of composite words increased, with reaction timesfor false recognition correlating to time spent in slow wavesleep. These results confirm the functional role of timecompression in memory replay, supporting the temporalscaffolding model.

We describe a cognitive mechanism of moral judgment, universalization, that has received little attention up to now. Underuniversalization, an action’s moral permissibility is determined by calculating what the outcome would be if all people whoare similarly situated to the actor also acted in that way. This mechanism is particularly well-suited to capture our moraljudgments of free-rider cases, where one person doing the action increases utility but many people doing it decreasesutility. Universalization fits into an agreement-based (contractualist) theory of moral cognition, and explains properties ofour moral judgments that an outcome-based or rule-based approach cannot. We show patterns of universalization reasoningin young children as well as adults.

When encountering unfamiliar physical objects, children andadults often perform structured interrogatory actions such asgrasping and prodding, so revealing latent physical propertiessuch as masses and textures. However, the processes drivingand supporting these curious behaviors are still largely mys-terious. In this paper, we develop and train an agent able toactively uncover latent physical properties such as the massand force of objects in a simulated physical “micro-world’.Concretely, we used a simulation-based-inference frameworkto quantify the physical information produced by observationand interaction with the evolving dynamic environment. Weused model-free reinforcement learning algorithm to train anagent to implement general strategies for revealing latent phys-ical properties. We compare the behaviors of this agent to thehuman behaviors observed in a similar task.

Suspense is an affective state that contributes to our enjoy-ment of experiences such as movies and sports. Ely, Frankel,and Kamenica (2015) proposed a formal definition of suspensewhich depends on the variance of subjective future beliefsabout an outcome of interest (e.g., winning a game). In orderto evaluate this theory, we designed a task based on the cardgame Blackjack where a variety of suspense dynamics can beexperimentally induced. By presenting participants with iden-tical sequences of information (i.e., card draws), but manip-ulating contextual knowledge (i.e., their understanding of therules of the game) we were able to show that self-reported sus-pense follows the predictions of the model. Follow-up modelcomparison further showed an advantage for the “suspense asvariance of future beliefs” account over a number of alterna-tive definitions of suspense, including some that depend onlyon current uncertainty (not the future). This paper is an initialattempt to link aspects of formal models of information anduncertainty with affective cognitive states.

Outcome bias describes the tendency of people to alter theirrating of a decision’s quality according to whether theoutcome is good or bad – despite equivalencies in availableinformation and decision processes – which has the potentialto undermine learning about causal structures and diagnosticinformation in many fields, including medicine. Herein, asample of 181 doctors and medical students is shown todisplay outcome bias in medical and non-medical scenarios –with their susceptibility correlating across the domains, r =0.38. Analyses showed that rational and intuitive decisionstyles and a medical risk tolerance measure offered littlepredictive power. Instead, the strongest drivers of biassusceptibility were the Age and professional Level ofparticipants, with more senior personnel showing lessoutcome bias. We argue that this could reflect improvedlearning across a doctor’s career or result from increasingconfidence making them less likely to change their initialjudgement of decision quality.

The categorization literature often considers two types of cat-egories as equivalent: (a) standard categories and (b) negationcategories. For example, category learning studies typicallyconflate learning categories A and B with learning categoriesA and NOT A. This study represents the first attempt at de-lineating these two separate types of generated categories. Wespecifically test for differences in the distributional structure ofgenerated categories, demonstrating that categories identifiedas not what was known are larger and wider-spread comparedto categories that were identified with a specific label. We alsoobserve consistency in distributional structure across multiplegenerated categories, replicating and extending previous find-ings. These results are discussed in the context of providing afoundation for future modeling work.

Mounting evidence suggests that human category learning is achieved by multiple qualitatively distinct biological and psychological systems. In an information-integration (II) categorization task, optimal performance requires switching away from rule and adopting a procedural response strategy. However, many participants perseverate with rules. This article attempts at understanding the difference between optimal and suboptimal participants in II categorization. To this end, we collected data in the Iowa Gambling Task (IGT) and an II categorization task. Performance in the IGT was used to estimate each participant’s sensitivity to reward, punishment, and propensity to explore. The results show that optimal participants in the II task explored more in the IGT than suboptimal participants. However, optimal participants in the II task did not show higher sensitivity to punishment or lower sensitivity to reward. We conclude by discussing the implications of these findings on system-switching and theoretical work on multiple-systems model of perceptual category learning.

Curiosity plays a critical role in our daily behaviors and interactions. Yet, very little is known about its psychological and neural underpinnings. By reframing curiosity as the motivation to obtain reward – where the reward is information –, and using frequency-based metrics of frontal brain lateralization, we aimed to investigate the neural correlates of curiosity in the frontal cortex and its effects on subsequent learning. Twenty-one undergraduate students participated in this two-day study by answering 35 general interest trivia questions, while EEG data was being recorded, also indicating their curiosity towards the question. One week later, participants were asked to write down the correct answers to each one of the questions. The results of this study suggested that frontal brain asymmetry (FBA) predicts memory recall, but is not directly correlated with self-reported curiosity. Study limitations and future directions are discussed.

One previously unexplained observation about numeral sys-tems is the shared tendency in numeral expressions: Numer-als greater than 20 often have the larger constituent numberexpressed before the smaller constituent number (e.g., twenty-four as opposed to four-twenty in English), and systems thatoriginally adopt the reverse order of expression (e.g., four-and-twenty in Old English) tend to switch order over time. Toexplore these phenomena, we propose the view of Rapid In-formation Gain and contrast it with the established theory ofUniform Information Density. We compare the two theoriesin their ability to explain the shared tendency in the orderingof numeral expressions around 20. We find that Rapid Infor-mation Gain accounts for empirical patterns better than the al-ternative theory, suggesting that there is an emphasis on infor-mation front-loading as opposed to information smoothing inthe design of large compound numerals. Our work shows thatfine-grained generalizations about numeral systems can be un-derstood in information-theoretic terms and offers an opportu-nity to characterize the design principles of lexical compoundsthrough the lens of informative communication.

Verb learning is important for young children. While most previous research has focused on linguistic and conceptual challenges in early verb learning (e.g. Gentner, 1982, 2006), the present paper examined early verb learning at the attentional level and quantified the input for early verb learning by measuring verb-action co-occurrence statistics in parent- child interaction from the learner’s perspective. To do so, we used head-mounted eye tracking to record fine-grained multimodal behaviors during parent-infant joint play, and analyzed parent speech, parent and infant action, and infant attention at the moments when parents produced verb labels. Our results show great variability across different action verbs, in terms of frequency of verb utterances, frequency of corresponding actions related to verb meanings, and infants’ attention to verbs and actions, which provide new insights on why some verbs are harder to learn than others.

The nature of morphological processing has remained acontroversial topic in psycholinguistic research. Some studies(e.g., Rastle, Davis, & New, 2004) have argued that whenwe read words like corner and talker, we automaticallydecompose them into existing morphemes like talk, corn, and-er, regardless of whether it is semantically plausible (e.g.,talker) or not (e.g., corner). Recent studies, however, havechallenged this view, by showing early semantic effects ofthe whole complex word (J ̈arvikivi & Pyykk ̈onen, 2011; L ̃oo& J ̈arvikivi, 2019; Milin, Feldman, Ramscar, Hendrix, &Baayen, 2017). Using a masked priming paradigm, the presentstudy only found effects of morphological decomposition fortrue morphological relations (e.g., talker) as well as effectsof frequency and affective properties of whole words, furtherchallenging automatic decomposition accounts. Finally, wealso report that individual differences such as participants’self-reported scholarly reading and openness to new experi-ence, affect processing.

Infants learn language by exposure to streams of speech pro-duced by their caregivers. Early on, they manage to segmentword forms out of this continuous input, which is either di-rectly addressed to them, or directed to other adults, thus over-heard. It has been suggested that infant-directed speech is sim-plified and could facilitate language learning. This study aimedto investigate whether features such as utterance length, seg-mentation entropy and lexical diversity could account for anadvantage in segmentability of infant-directed speech. A largeset of word segmentation algorithms was used on an ecolog-ically valid corpus, consisting of 18 sets of recordings gath-ered from French-learning infants aged 3-48 months. A se-ries of textual analyses confirmed several simplicity featuresof infant-, compared to adult-directed speech. A small seg-mentation advantage was also documented, which could notbe attributed to any of those corpus features. Some particular-ities of the data invite further research on more corpora.

Humans make plans with remarkable flexibility by leveraging symbolic representations. How are these representationslearned? We present a model that starts out with a language of low-level physical constraints and, by observing expertdemonstrations, builds up a library of high-level concepts that afford planning and action understanding. We demonstrateits versatility through experiments inspired by developmental psychology literature.

We present a computational model of attentional capture inhumans. The model distinguishes between automatic mecha-nisms that directly determine the focus of visual attention, anddeliberate mental actions an individual can perform to influ-ence these mechanisms. The automatic mechanisms select anobject as the focus of attention and enhance its location andfeatures, so that nearby or similar objects are likely to be se-lected in the future. The deliberate actions include engagingwith a selected object to further enhance its features, and re-trieving a previously selected object from memory. By per-forming these actions, the model is able to exert limited top-down control over capture, increasing the probability that task-relevant objects will be attended and irrelevant objects will beignored. To evaluate the model, we conduct a simulation of arecent visual search study, demonstrating that the model canaccount for three established factors that are known to influ-ence capture.

We report a first effort to model the solution of meaningful four-termvisual analogies, by combining a machine-vision model (ResNet50-A) that can classify pixel-level images into object categories, with acognitive model (BART) that takes semantic representations of wordsas input and identifies semantic relations instantiated by a word pair.Each model achieves above-chance performance in selecting the bestanalogical option from a set of four. However, combining the visualand the semantic models increases analogical performance above thelevel achieved by either model alone. The contribution of vision toreasoning thus may extend beyond simply generating verbalrepresentations from images. These findings provide a proof ofconcept that a comprehensive model can solve semantically-richanalogies from pixel-level inputs.

Research has consistently demonstrated that people treat digital technology-based environments such as VR as if they were real. This is consistent with neural reuse and predictive processing theories. Neural circuits that have developed to perform real world actions are reused when performing tasks in computer mediated environments. The current research investigates some of the factors that could support users in leveraging their existing real world representations. A reasonable hypothesis is that users are more likely to emulate existing real world processing if technological artifacts are congruent with their experiential basis. This work investigates the perceived cues of task risks, movement realism and effector realism in performing actions. Effector design is manipulated (gesturing, wand, vs. knife), and participants cut a vegetable in a simulated environment. Participants evoked real world sensory motor contingency when technological artifacts are congruent with their experiential basis.

We propose a cognitively plausible method for representingand querying spatial relationships in a neural architecture. Thistechnique employs a fractional binding operator that capturescontinuous spatial information in spatial semantic pointers(SSPs). We propose a model that takes an image with severalobjects, parses the image into an SSP memory representation,and answers queries about the objects. We demonstrate thatour model allows us to not only store and extract objects andtheir spatial information, but also perform queries based on lo-cation and in relation to other objects. We show that we canquery images with 2, 3, and 4 objects with relative spatial lo-cations. We also show that the model qualitatively reproducesKosslyn’s famous map experiment.

The cognitive system readily learns when multiple cues jointly predict a specific outcome. What is less known is how the mind generates predictions when only a single cue is present. In four experiments, participants were first exposed to two objects followed by a circle with a specific size or a specific numeric value. Afterwards, participants viewed a single object and estimated the associated size or value. Finally, participants recalled the size or value that followed the initial two objects. We found that the estimated size associated with the single object was significantly smaller than 100% but significantly larger than 50% of the recalled size associated with the two objects. No participants were consciously aware of the associations. The results reveal a new consequence of statistical learning on automatic inferences: When multiple objects were previously associated with an outcome, the single object is implicitly expected to predict a subadditive outcome.

Sequential testimonies where more or less reliable sourcesargue about an issue are central to public debates. Often, themajority of sources may argue that a hypothesis is true whilea minority dissenter may claim the opposite (e.g. scientistsand lobbyists in the climate change debate).In this paper, we show that people are sensitive to sourcereliability as well as the structural relationship between thesources. Participants follow Bayesian predictions for revisingbelief in the hypothesis and the reliability of the competingsources given majority consent, minority dissent, and sharedreliability between sources. Shared reliability and dissent is akey issue for public debate and belief revision. The paperprovides novel insight into the workings of these aspects.

Misinformation, and its impact on society, has become anincreasingly topical field of study of late. A body of literatureexists that suggests misinformation can retain an influenceover beliefs despite subsequent retraction, known as theContinued Influence Effect (CIE). Researchers have arguedthis to be irrational. However, we show using a Bayesianformalism why this argument is overly assumptive, pointingto (previously overlooked) considerations of reliability of, anddependence between, misinforming and retracting sources.We demonstrate that lay reasoners intuitively endorseassumptions that demarcate CIE as a rational process, basedon the fact misinformation precedes its retraction. Moreover,despite using established CIE materials, we further upturn theapplecart by finding participants show CIE, and appropriatelypenalize the reliabilities of contradicting sources.

This study experimentally investigated the effect of sugges-tions from a physically present robot on human creative gener-ation. In the experiments, we used a creative task in which theparticipants were required to draw creatures living on a planetother than the Earth, and a physically present robot, which pro-vided suggestions for creative drawing to the participants withspeech sounds and physical movements. First, the results ofthe pilot experiment confirmed that drawing creativity was en-hanced for the participants supported by a robot; however, theywere unlikely to refer to the suggestions. Based on the re-sults, two hypotheses were developed: the suggestions from arobot offered a variety of different perspectives and facilitatedmetacognition (Hypothesis 1), and the suggestions worked asdistractions and suppressed fixated perspectives (Hypothesis2). The experiment was conducted to investigate these hy-potheses. As a result, Hypothesis 1 was supported. The resultswere discussed based on previous studies.

Despite the lack of invariance problem (the many-to-manymapping between acoustics and percepts), we experiencephonetic constancy and typically perceive what a speakerintends. Models of human speech recognition have side-stepped this problem, working with abstract, idealized inputsand deferring the challenge of working with real speech. Incontrast, automatic speech recognition powered by deeplearning networks have allowed robust, real-world speechrecognition. However, the complexities of deep learningarchitectures and training regimens make it difficult to usethem to provide direct insights into mechanisms that maysupport human speech recognition. We developed a simplenetwork that borrows one element from automatic speechrecognition (long short-term memory nodes, which providedynamic memory for short and long spans). This allows thenetwork to learn to map real speech from multiple talkers tosemantic targets with high accuracy. Internal representationsemerge that resemble phonetically-organized responses inhuman superior temporal gyrus, suggesting that the modeldevelops a distributed phonological code despite no explicittraining on phonetic or phonemic targets. The ability to workwith real speech is a major advance for cognitive models ofhuman speech recognition.

Emergence of collective cooperation in an inherently selfishsociety is a paradox that has preoccupied biologists, sociol-ogists, and cognitive scientists alike for centuries. We pro-pose a computational model and demonstrate through simula-tions how collective cooperation can emerge from selfish inter-ests: the goal of improving each individual’s own rewards. Wealso demonstrate how the same selfish interests lead to the dy-namic emergence of a network of interconnected agents. Ourmodel includes two simple mechanisms: Selfish-Trust (ST)and Selfish-Connection (SC). ST involves the possibility of re-lying on others in a society of agents when it is beneficial tothe individual, and SC involves the possibility of connecting toother agents when those agents help improve the individual’sown benefit. Our simulation results suggest that collective co-operation can emerge from ST and a complex dynamic net-work can emerge from ST and SC. The simulated data demon-strate an important property of many living organisms: pat-terns of temporal complexity, which are essential to transferinformation among agents of any society of living beings.

Convolutional neural networks (CNNs) were inspired by hu-man vision and, in some settings, achieve a performance com-parable to human object recognition. This has lead to the spec-ulation that both systems use similar mechanisms to performrecognition. In this study, we conducted a series of simulationsthat indicate that there is a fundamental difference between hu-man vision and vanilla CNNs: while object recognition in hu-mans relies on analysing shape, these CNNs do not have sucha shape-bias. We teased apart the type of features selectedby the model by modifying the CIFAR-10 dataset so that, inaddition to containing objects with shape, the images concur-rently contained non-shape features, such as a noise-like mask.When trained on these modified set of images, the model didnot show any bias towards selecting shapes as features. In-stead it relied on whichever feature allowed it to perform thebest prediction – even when this feature was a noise-like maskor a single predictive pixel amongst 50176 pixels.

Previous research suggests that people’s folk conception of themind is organized along a few fundamental dimensions; butstudies disagree on the exact number of those dimensions. Withan expanded item pool of mental capacities, variations ofquestion probes, and numerous judged agents, four studies pro-vide consistent evidence for three dimensions of perceivedmind: Affect (A), Moral and Mental Regulation (M), and Real-ity Interaction (R). The dimensions are not simply bundles ofsemantically related features but capture psychological func-tions of the mind—to engage with its own processes, with otherminds, and with the social and physical world. Under someconditions, two of the three dimensions further divide: Adivides into negative and positive (social) affect, and M dividesinto moral cognition and social cognition. We offer a 20-iteminstrument to measure people’s 3- and 5-dimensionalrepresentations of human and other minds.

Studies have claimed that blind people’s spatial representations are different from sighted people, and blind people display superior auditory processing. Due to the nature of auditory and haptic information, it has been proposed that blind people have spatial representations that are more sequential than sighted people. Even the temporary loss of sight—such as through blindfolding—can affect spatial representations, but not much research has been done on this topic. We compared blindfolded and sighted people’s linguistic spatial expressions and non- linguistic localization accuracy to test how blindfolding affects the representation of path in auditory motion events. We found that blindfolded people were as good as sighted people when localizing simple sounds, but they outperformed sighted people when localizing auditory motion events. Blindfolded people’s path related speech also included more sequential, and less holistic elements. Our results indicate that even temporary loss of sight influences spatial representations of auditory motion events.

Predictive neural networks, such as word2vec, have seenimpressive recent popularity as an architecture to learndistributional semantics in the fields of machine learning andcognitive science. They are particularly popular because theylearn continuously, making them more space efficient andcognitively plausible than classic models of semantic memory.However, a major weakness of this architecture is catastrophicinterference (CI): The sudden and complete loss of previouslylearned associations when encoding new ones. CI is an issuewith backpropagation; when learning sequential data, the errorsignal dramatically modifies the connection weights betweennodes—causing rapid forgetting of previously learnedinformation. CI is a huge problem for predictive semanticmodels of word meaning, because multiple word sensesinterfere with each other. Here, we evaluate a recentlyproposed solution to CI from neuroscience, elastic weightconsolidation, as well as a Hebbian learning architecture fromthe memory literature that does not produce an error signal.Both solutions are evaluated on an artificial and naturallanguage task in their ability to insulate a previously learnedsense of a word when learning a new one.

What do people learn from experience with repeated decisions?Is it merely implicit behavioral tendencies? If so, wouldarticulating or summarizing what is learned change behavior?Online participants (N=126) experienced 100 trials of adecisions-from-experience problem with outcome feedback.Some participants then verbally summarized what they hadlearned and estimated the probability of the risky gain eitherfor themselves (Self condition) or for another hypotheticalplayer (Other condition); others did not summarize (Controlcondition). Finally, they faced 20 more decision trials.Verbalizing a social message to another person significantlyincreased sure choices (that is, decreased risk-taking) insubsequent decision making. In general, participantsunderestimated the probabilities of both certain and riskyprospects, and articulating a summary message (Self or Other)seemed to increase this conservatism.

A long-standing assumption in cognitive science has been thatconcepts are shared among individuals for common words.However, given that concepts are formed by the data we ob-serve, and observations vary wildly across individual experi-ences, our concepts are not likely identical. Here, we presentdata in which 104 participants answer questions regarding theirbeliefs about the definitions of common everyday words, andthe degree to which they think others agree. Our results sug-gest that even for common words, there exist many distinctextensions of ordinary and political concepts across individu-als. There is also a pervasive bias which leads individuals tooverestimate the degree to which others agree, which may ex-plain why “talking past each other” is an anecdotally commonexperience when discussing important topics.

Previous research has argued that learners infer word order pat-terns when learning a new language based on knowledge aboutunderlying structure, rather than linear order (Culbertson &Adger, 2014). Specifically, learners prefer typologically com-mon noun phrase word order patterns that transparently reflecthow elements like nouns, adjectives, numerals, and demon-stratives combine hierarchically. We test whether this resultstill holds after removing a potentially confounding strategypresent in the original study design. We find that when learn-ers are taught a naturalistic “foreign” language, a clear prefer-ence for noun phrase word order is replicated but for a subsetof modifier types originally tested. Specifically, participantspreferred noun phrases with the order N-Adj-Dem (as in “mugred this”) over the order N-Dem-Adj (as in “mug this red”).However, they showed no preference between orders N-Adj-Num (as in “mugs red two”) and N-Num-Adj (as in “mugstwo red”). We interpret this sensitivity as potentially reflectingan asymmetry among modifier types in the underlying hierar-chical structure.

The acquisition of the grammar of a second language requires a variety of cognitive mechanisms, including inductive reasoning. In the current study, we examine the cognitive underpinnings of grammar learning with an explicit-inductive (rule search) learning task, designed to capture more of the complexity associated with grammar learning than purely deductive tasks. Research in language aptitude has shown that working memory capacity (WMC) is a key predictor of grammar learning outcomes. Inductive reasoning and grammatical sensitivity are other established aptitude factors. The goal of the present study was to determine the degree to which relevant variables predict learning on an explicit- inductive grammar learning task. Our results indicate that both WMC and inductive reasoning ability predict learning over three days of grammar training.

This study examined the roles recognition of the creative process behind artworks plays in cognitive processes of art-viewing. To this end, we conducted an experiment (N = 45) in which prior experience of participants was manipulatedand investigated whether and how creative experience influences subsequent cognitive processes while viewing artworks.We revealed that having creative experience before art viewing changes viewers recognition of the creative process behindartworks and causes them to have a more positive impression of the artworks. It was also revealed that these two changesare correlated. In particular, the emotion of admiration, which is considered a kind of social emotion, was found to behighly correlated with the recognition of assessed difficulty of the creative process. These results suggest the importance ofrecognition of the creative process behind artworks and contribute to understanding the cognitive process of art-viewing.

The mental model of one’s body plays an important role in de-termining subsequent actions. We changed the mental modelusing visual information and observed the effects of suchchange on pain perception. These effects were compared tothe effects of changes in the sense of body ownership, which isthe sensation that something is a part of one’s own body. Someresearchers have shown that the sense of ownership is a factormodulating pain perception. In our experiments, we manipu-lated the visibility of participants’ limbs using Mixed Reality(MR) techniques and measured their perceived pain and feel-ings while observing their limbs. Results showed the sensationthat nothing can touch one’s limbs decreased the strength ofperceived pain.

Recent studies demonstrate strong, concurrent relationships between language and EF, particularly during early childhood.However, the literature remains controversial with respect to this relationship. Whereas some studies cite a bidirectionalrelationship, others suggest that EF is predictive of language gains, while others suggest that it is language which affectsEF through conversational practice. Further controversy remains in the literature regarding which components of EF areengaged in the processes. The bidirectionality of current research in this area suggests that perhaps EF and languageare best fitted by a curvilinear relationship. This is compounded by the fact that a large number of these studies haveemployed linear statistical analyses to examine the relationship of the two constructs. Thus, in order to further specifythe relationship between EF and language development, we examined monolingual and bilingual infants and toddlers todetermine the utility of a curvilinear model to assess the EF and language relationship, what aspect of language inhibitorycontrol most correlates to EF, and whether there is a monolingual/bilingual difference. Results indicate that the EF andlanguage early childhood relationship is best fitted by a curvilinear model.

The initial phase of linguistic production by children is char-acterized by rote-learned, lexically restricted forms and con-structions. Only during later phases of language acquisitiondo they develop flexibility across a paradigm and mix lexicaland grammatical material more freely. In the development ofverb morphology, a correlation between the use of tense andaspect has been observed in many languages. It has been sug-gested that this leads to an intermediary state of paradigm cat-egorization based on temporal categories. So far the flexibilityof individual verbs occurring in different tense-aspect combi-nations has not been examined in detail. Here we evaluate theflexibility of verb use in a large longitudinal corpus of 4 Rus-sian children. We compute the Shannon entropy of verb stemsdistributed over individual grammatical forms. Results showthat children do not pass through a stage of paradigm cate-gorization based on aspecto-temporal categories. After a briefitem-specific phase of rote learned forms, they quickly becomeflexible users of verbs in both aspects.

Previous research by Kaniel & Lubow in 1986 found that youngchildren (aged 4-5 years) exhibited poorer learning (latentinhibition) to pre-exposed stimuli than older children (aged 7-10years). The aim of our research was to develop a computer-based,child-friendly study that would replicate the work of Kaniel &Lubow. Sixty-three children took part in our experiment. Thisconsisted of a pre-exposure/study phase in which participants wereasked to press computer keys in response to clipart pictures ofanimals and dinosaurs. Each animal or dinosaur picture waspreceded by one of two “warning signals” which acted as the pre-exposed stimuli (to which no response was required). In the testphase that followed, the participants had to either press thespacebar or withhold their response to each pre-exposed stimulusand two novel stimuli. They learnt which response was correct bytrial and error using the feedback provided. The accuracy andreaction time of the responses during the test phase were analysedand indicated that the youngest children showed significantlylower mean accuracy and longer mean response times to the pre-exposed stimuli than to stimuli they had not been pre-exposed to.In contrast, the older children showed no significant differences intheir responses to pre-exposed and novel stimuli. These results areconsistent with those found by Kaniel & Lubow and could be takenas evidence for latent inhibition in young children. Further studiesare proposed in which variations in pre-exposure procedure areused to rule out explanations based on response inhibition ornegative priming.

Intervention selection is at once crucial in causal learning andchallenging for causal learners. While the optimal strategy ismaximizing the expected information gain (EIG), both chil-dren and adults often combine it with suboptimal ones suchas the positive test strategy (PTS). In the current study, wesought to facilitate causal learning from intervention by asking5- to 7-year-olds to explain why they chose a certain interven-tion to identify the true structure of a three-node causal sys-tem that might work in one of two ways. Our findings suggestthat while engaging in self-explaining did not help children se-lect more informative interventions, asking them to think abouttheir intervention choices (explaining or reporting) might helpthem better utilize interventional data to infer causal structures.

An agent decides against her preferences, if she considersan option x better than another option y but neverthe-less decides to do y. A central tenet of rational choi-ce theory states that individuals do not decide againsttheir preferences, whereby we find two kinds of potentialcounterexamples in the literature: akrasia, also known asweak-willed decisions, and decisions based on so-calleddeontic constraints such as obligations or commitments.While there is some empirical evidence that weak-willedchoices are a real phenomenon, leading scholars in phi-losophy of economics debate whether choices based oncommitments can be counter-preferential. As far as weknow, however, nobody so far has tried to settle this de-bate empirically. This paper contributes to both debatessince we present some empirical evidence that (i) akrasiacan also be strong-willed and (ii) choices made on the ba-sis of commitments can indeed be counter-preferential.We will conclude that people can decide against theirpreferences without being unreasonable.

Adaptive learning systems that generate spacing intervalsbased on learner performance enhance learning efficiency andretention (Mettler, Massey & Kellman, 2016). Recentresearch in factual learning suggests that initial blocks ofpassive trials, where learners observe correct answers withoutovertly responding, produce greater learning than passive oractive trials alone (Mettler, Massey, Burke, Garrigan &Kellman, 2018). Here we tested whether this passive + activeadvantage generalizes beyond factual learning to perceptuallearning. Participants studied and classified images ofbutterfly genera using either: 1) Passive Only presentations,2) Passive Initial Blocks followed by active, adaptivescheduling, 3) Passive Initial Category Exemplar followed byactive, adaptive scheduling, or 4) Active Only learning. Wefound an advantage for combinations of active and passivepresentations over Passive Only or Active Only presentations.Passive trials presented in initial blocks showed the bestperformance, paralleling earlier findings in factual learning.Combining active and passive learning produces greaterlearning gains than either alone, and these effects occur fordiverse forms of learning, including perceptual learning.

We compare the performance of humans (English and Frenchlisteners) versus an unsupervised speech model in a perceptionexperiment (ABX discrimination task). Although the ABXtask has been used for acoustic model evaluation in previousresearch, the results have not, until now, been compared di-rectly with human behaviour in an experiment. We show that astandard, well-performing model (DPGMM) has better accu-racy at predicting human responses than the acoustic baseline.The model also shows a native language effect, better resem-bling native listeners of the language on which it was trained.However, the native language effect shown by the models isdifferent than the one shown by the human listeners, and, no-tably, the models do not show the same overall patterns ofvowel confusions.

Conspiracy theories are “alternative” explanations ofwell-understood events or phenomena. What makes themattractive explanations to so many people? We investigatewhether people ascribe characteristics typical of goodexplanations to conspiracy theories and whether they areperceived as more appealing explanations when they arearticulated as a refutation of the official version of events. Intwo experiments, participants read explanations of fourconspiracy theories and rated them along six dimensions ofexplanatory quality. We find that some explanatory virtues areascribed to conspiracy theories even by people who do notbelieve the conspiracy. Contrary to our predictions, we alsofind that framing a conspiracy as a refutation did not generallyelicit higher ascriptions of explanatory virtues. These resultssuggest that explanatory considerations may play a morecentral role in conspiracist beliefs than was previously thought.

Most statistical learning studies focus on the learning oftransitional probabilities between adjacent elements in asequence, however, other statistical regularities may un-derpin different aspects of processing language and regu-larities in other domains. Here, we investigate how con-junctive statistical regularities (of the form A and B to-gether predict C) can be learned, and how this learningis impacted by similarity in representations analogousto that in unambiguous words, homonyms with mul-tiple unrelated meanings, and polysemes with multiplerelated meanings. We observed that provided the stimu-lus structure is relatively simple, participants are readilyable to learn conjunctive probabilities and display sen-sitivity to relatedness among representations. These re-sults open new theoretical possibilities for exploring thedomain-generality of how the learning and processingsystems merge conjunctive information in simple labo-ratory tasks and in natural language.

Although process data indicate that people often rely on sim-plifying processes when choosing between risky options, cur-rent models of heuristics cannot predict people’s choices veryaccurately. To address this apparent paradox, it has been pro-posed that people might adaptively choose from a toolboxof simple strategies. But which strategies are contained inthis toolbox? And how do people decide when to use whichdecision strategy? Here, we develop a model according towhich the decision maker selects a decision strategy for a givenchoice problem rationally from a toolbox of strategies; the con-tent of the toolbox is estimated for each individual decisionmaker. Using cross-validation on an empirical data set, we findthat this model of strategy selection from a personal adaptivetoolbox predicts people’s choices better than any single strat-egy (even when it is allowed to vary across participants) andbetter than previously proposed toolbox models. Our modelcomparisons show that both inferring the content of the tool-box and rational strategy selection are critical for accuratelypredicting people’s risky choices. Furthermore, our analysisreveals considerable individual differences in the set of strate-gies people are equipped with and how they choose amongthem; these individual differences could partly explain whysome people make better choices than others. These findingsrepresent an important step towards a complete formalizationof the notion that people select their cognitive strategies froma personal adaptive toolbox.

People are often faced with choices where there is a conflictbetween seeking reward and gathering information. In manyof these cases there exists a functional relationship betweenthe features associated with actions and their correspondingrewards. Accounts of how people make decisions in thesecircumstances have not considered how peoples’ strategiesdepend on the complexity of this function, as well as theperson’s goal. In a sequential decision making task we foundthat people chose between a number of different explorationstrategies, but that strategy selection did not necessarily alignwith goal or account for function complexity.

Ingroup favoritism, the tendency to favor ingroup over out-group, is often explained as a product of intergroup conflict, orcorrelations between group tags and behavior. Such accountsassume that group membership is meaningful, whereas humandata show that ingroup favoritism occurs even when it confersno advantage and groups are transparently arbitrary. Anotherpossibility is that ingroup favoritism arises due to perceptualbiases like outgroup homogeneity, the tendency for humans tohave greater difficulty distinguishing outgroup members thaningroup ones. We present a prisoner’s dilemma model, whereindividuals use Bayesian inference to learn how likely oth-ers are to cooperate, and then act rationally to maximize ex-pected utility. We show that, when such individuals exhibitoutgroup homogeneity bias, ingroup favoritism between arbi-trary groups arises through direct reciprocity. However, thisoutcome may be mitigated by: (1) raising the benefits of coop-eration, (2) increasing population diversity, and (3) imposing amore restrictive social structure.

Children do not learn language from passive observation ofthe world, but from interaction with caregivers who want tocommunicate with them. These communicative exchanges arestructured at multiple levels in ways that support support lan-guage learning. We argue this pedagogically supportive struc-ture can result from pressure to communicate successfully witha linguistically immature partner. We first characterize onekind of pedagogically supportive structure in a corpus analy-sis: caregivers provide more information-rich referential com-munication, using both gesture and speech to refer to a singleobject, when that object is rare and when their child is young.Then, in an iterated reference game experiment on MechanicalTurk (n = 480), we show how this behavior can arise from pres-sure to communicate successfully with a less knowledgeablepartner. Lastly, we show that speaker behavior in our experi-ment can be explained by a rational planning model, withoutany explicit teaching goal. We suggest that caregivers’ desireto communicate successfully may play a powerful role in struc-turing children’s input in order to support language learning.

Both examples and verbal explanations play an important rolein learning new concepts and categories. At the same time,learning from verbal explanations is not accounted for in mostcategory learning models, and is not studied in the traditionalcategory learning paradigm. We propose a rational categorycommunication model that formally describes the process ofcommunicating a category structure using both verbal expla-nations and visual examples in a pedagogical setting. We buildour model based on the assumption that verbal instructions arebest suited for communication of crude constraints on a cat-egory structure, while exemplars complement it by providingmeans for finer adjustments. Our empirical study demonstratesthat verbal communication is indeed more robust to changesin stimuli dimensionality, but that its efficiency is adverselyaffected when distinguishing between categories requires per-ceptual precision. Communicating through examples has a re-versed pattern. We hope that both the proposed experimentalparadigm and the computational model would facilitate furtherresearch into the relative roles of verbal and exemplar commu-nication in category learning.

e effortlessly grasp the correspondence between a drawingof an object and that physical object in the world, even whenthe drawing is far from realistic. How are visual objectconcepts organized such that we can both recognize theseabstract correspondences and also flexibly exploit them whencommunicating them to others in a drawing? Here we considerthe notion that the compositional nature of object conceptsenables us to readily decompose both objects and drawings ofobjects into a common set of semantically meaningful parts.To investigate this, we collected data on the part informationexpressed in drawings by having participants densely annotatedrawings of real-world objects. Our dataset contained bothdetailed and sparser drawings produced in different commu-nicative contexts. We found that: (1) people are consistentin what they interpret individual strokes to represent; (2)single strokes tend to correspond to single parts, with strokesrepresenting the same part often being clustered in time; and(3) both sparse and detailed drawings of the same object em-phasize similar part information, although detailed drawingsof different objects are more distinct from one another thansparse drawings. Taken together, our results support the notionthat people flexibly deploy their abstract understanding ofthe compositional part structure of objects to communicaterelevant information about them in context. More broadly,they highlight the importance of structured knowledge forunderstanding how pictorial representations convey meaning.

Constraints on control-dependent processing have become afundamental concept in general theories of cognition that ex-plain human behavior in terms of rational adaptations to theseconstraints. However, theories miss a rationale for why suchconstraints would exist in the first place. Recent work suggeststhat constraints on the allocation of control facilitate flexibletask switching at the expense of the stability needed to supportgoal-directed behavior in face of distraction. Here, we formu-late this problem in a dynamical system, in which control sig-nals are represented as attractors and in which constraints oncontrol allocation limit the depth of these attractors. We deriveformal expressions of the stability-flexibility tradeoff, showingthat constraints on control allocation improve cognitive flexi-bility but impair cognitive stability. Finally, we provide evi-dence that human participants adapt higher constraints on theallocation of control as the demand for flexibility increases butthat participants deviate from optimal constraints.

Researchers have long been interested in using laboratory mea-sures of cognitive control to predict a person’s cognitive con-trol/self control success outside the lab. We used a computa-tional approach to identify which lab-based performance mea-sures provide the most valid individual difference measuresof one’s ability and/or motivation to exert cognitive control.We simulated performance across an array of cognitive controltasks, and estimated the degree to which different performancemetrics (e.g., congruency effects, conflict adaptation, and de-mand avoidance) could theoretically provide valid estimatesof processes underlying control allocation. By performing di-mension reduction on these performance metrics, we furtherrevealed latent dimensions that can index separate mechanismsof control-demanding behavior. Our results suggest that indi-vidual differences in measures of cognitive control can orig-inate from multiple factors, several of which are unrelated tocapacity for cognitive control. We conclude by discussing im-plications of these analyses for assessing individual differencesin cognitive control phenomena.

The extent to which the mind is modular is a foundational concern in cognitive science. Much of this debate has centeredon the question of the degree to which input systems, i.e., sensory systems such as vision, are modular (see, e.g., Fodor1983; Pylyshyn 1999; MacPherson 2012; Firestone & Scholl 201; Burnston 2017; Mandelbaum 2017). By contrast,researchers have paid far less attention to the question of the extent to which our main output system, i.e., the motorsystem, qualifies as such. I will argue that the motor system should be construed as quasi-modular, at best, in that it isinformationally encapsulated only to a certain degree, and in a way that can be strategically modulated by the agent. I willexplore the implications of this result for nearby philosophical puzzles relating to different aspects of action control.

The round number effect refers to discontinuity around roundnumbers (“0.300”, “4 hours”) in frequency distribution,indicating that people consider the round numbers as goals orreference points for their performances. This study aimed toexamine the round number effect by exploring the followingtwo issues: (1) examination of Japanese baseball data, and (2)comparison between batters who exceeded the regulationnumber of at-bat of season and those who did not. Resultsindicate the following three points; (1) the round numbereffect was found in Japanese baseball data, (2) but it wasfound only for the batters who exceeded provision bat numberof season, and (3) magnitude of the effect was stronger inJapanese than Major League Baseball data. Generaldiscussion argued these results in terms of players’ motivationand disposition.

Affordances offer AI research an alternative from representations for linking perception to action in autonomous systems. Affordances are based in the informational structure of the environment and the somatic capacities of the agent and arise in their interaction. AI implementations of affordance perception typically utilize relatively basic, natural affordances such as the graspability of a handle. Culturally- scaffolded affordances, such as the letter-mailing capacity of a postbox, pose a more intractable problem for affordance- based robotics. This class of affordances requires acculturation and is highly culture-specific. AI implementations of affordance perception typically bypass this difficulty by making recourse to representations. I begin by reviewing affordance perception and the difference between natural and cultural affordances. I then critically discuss implementations of cultural affordance perception in autonomous agents. Finally, I argue that AI affordance perception does not require a robust representationalism in order to implement cultural affordances.

A mainstay of models that account for the access of lexicalknowledge is that auditory words compete for selection basedon form similarity, commonly seen in an inhibitory effect togreater phonological neighborhood density (PND). PND is ametric that states that two words are neighbors if they differ bythe addition, deletion or substitution of a single phoneme. Adrawback to this account is that there is competing evidenceeven among the European languages investigated thus far. Wesought to verify whether the inhibitory effect of greater PNDwould hold for Mandarin Chinese in two auditory wordrepetition tasks with monosyllabic and disyllabic Mandarinwords. Results of Experiment 1 showed a facilitative effect togreater PND. Experiment 2 added a non-verbal distractor taskto lessen the putative effect of working memory load during thetask. The facilitative effect to greater PND was confirmedalong with a significant post-hoc interaction with memorydecay, operationalized as the duration spent on the distractortasks. The facilitative effects extend previous reports ofdifferential behavior due to linguistic typology.

The purpose of this paper is to determine whether a par-ticular context factor among the variables that a researcheris interested in causally affects the route-choice behavior ofdrivers. To our knowledge, there is limited literature that con-sider the effects of various factors on route choice based oncausal inference.Yet, collecting data sets that are sensitive tothe aforementioned factors are challenging and the existingapproaches usually take into account only the general factorsmotivating drivers route choice behavior. To fill these gaps,we carried out a study using Immersive Virtual Environment(IVE) tools to elicit drivers route choice behavioral data, cov-ering drivers’ network familiarity, education level, financial-concern, etc, apart from conventional measurement variables.Having context-aware, high-fidelity properties, IVE data af-fords the opportunity to incorporate the impacts of human-related factors into the route choice causal analysis and ad-vance a more customizable research tool for investigatingcausal factors on path selection in network routing. This causalanalysis provides quantitative evidence to support drivers di-version decision. The study also provides academic sugges-tion and reference for investing in public infrastructure anddeveloping efficient strategies and policies to mitigate trafficcongestion.

The intrinsic integration hypothesis proposes that using coregame mechanisms to teach learning material makeseducational games more fun to play and better for learning.Our study tests the intrinsic integration hypothesis with twoeducational versions of Battleship that were designed for thisexperiment, in the domain of complex numbers. We examinethe learning gains and motivation of 58 participants whointeracted with either the intrinsically-integrated orextrinsically-integrated version of the game. Our resultscontradict previous findings supporting the intrinsicintegration hypothesis: participants reported similar levels ofmotivation from both versions of the game and participantswho interacted with the extrinsically-integrated versionlearned significantly more as measured by pretest to posttestgains. This work contributes empirical data to the debateconcerning intrinsic integration, and it highlights the need foradditional studies exploring the integration of learningmaterial into educational games.

Negation is a complex, abstract concept, despite the ubiquityof words like “no” and “not” in even young children’s speech.One challenging aspect to words like “no” and “not” is thatthese words can serve many functions in speech, giving ustools to express an array of concepts such as denial, refusal,and nonexistence. Is there a single concept of “negation” thatunites these separate negative functions – and if so, doesunderstanding this concept require the structure of humanlanguage? In this paper we present a study demonstrating thatadults spontaneously identify a concept of negation in theabsence of explicit verbal instructions, even when theexemplars of negation are perceptually varied and representmany different functions of negation. Furthermore, tying upparticipants’ language ability using verbal shadowing impairsparticipants’ ability to identify a concept of negation, but doesnot impair participants’ ability to identify an equally complexcontrol concept (natural kinds). We discuss our findings inlight of theories regarding the representation of negation andthe relationship between language and thought.

We assessed the neural substrates mediating a recentlydemonstrated preference for environments with high levels ofinstrumental divergence – a formal index of flexible operantcontrol. Across choice scenarios, participants chose betweengambling environments that differed in terms of bothinstrumental divergence and expected monetary pay-offs.Using model-based fMRI, we found that activity in theventromedial prefrontal cortex scaled with a divergence-based measure of expected utility that reflected the value ofboth divergence and monetary reward. Implications for aneural common currency for information theoretic andeconomic variables are discussed.

Intervening on causal systems can illuminate their underlyingstructures. Past work has shown that, relative to adults, youngchildren often make intervention decisions that confirm sin-gle hypotheses rather than those that discriminate alternativehypotheses. Here, we investigated how the ability to make in-formative intervention decisions changes across development.Ninety participants between the ages of 7 and 25 completed40 different puzzles in which they had to intervene on vari-ous causal systems to determine their underlying structures.We found that the use of discriminatory strategies increasedthrough adolescence and plateaued into adulthood. Our resultsidentify a clear developmental trend in causal reasoning, andhighlight the need to expand research on causal learning mech-anisms in adolescence.

Children often fail to control variables when conducting testsof hypotheses, yielding confounded evidence. We propose thatgetting children to think of alternative possibilities throughcounterfactual prompts may scaffold their ability to controlvariables, by engaging them in an imagined intervention that isstructurally similar to controlled actions in scientificexperiments. Findings provide preliminary support for thishypothesis. Seven- to 10-year-olds who were prompted to thinkcounterfactually showed better performance on post-testcontrol of variables tasks than children who were given controlprompts. These results inform debates about the contributionof counterfactual reasoning to scientific reasoning, and suggestthat counterfactual prompts may be useful in science learningcontexts.

People make decisions based on probabilistic information everyday and often use innacurate, heuristic decision rules. Although a great deal of research has investigated the developmental trajectory of accurate probability ojudgements, very little research has investigated how the learning process unfolds. In the current study a microgenetic experimental design was deployed to investigate the influence of feedback on children's probailistic decision making strategies. Seven- to ten-year-old children (N = 50) first performed a computer-based task to asses the type of strategy they use in a probabilitstic judgement task. Next, children recieve feedback on a series of 24 trials and then perform a post-test consisting of the same computer based strategy assessment. Findings revealed that some strategies may benefit from feedback more than others. These results suggest that children can learn about the proportional nature of probability from feedback alone and that the amount and type of feedback influence the learning process.

This study investigated direct and indirect effects of executivefunctions (EF) on reading comprehension in 87 adolescents(mean age = 14.0 years, SD = 1.5). The operation span taskwas used to measure the updating aspect of working memory,the plus-minus task to measure task-switching, and thenumerical Stroop task to measure inhibitory control. Literacyskills tasks assessed nonword decoding, text recall/inference,and passage comprehension. Regression models indicated thatEF measures accounted for significant variance in literacyskills after controlling for age and fluid intelligence. Workingmemory was associated with passage comprehension, task-switching with nonword decoding, and inhibitory control withnonword decoding as well as text recall/inference. Parallelmediation models tested for indirect effects of EF constructsvia decoding and text recall/inference. Working memoryshowed direct and indirect effects on passage comprehension,the latter mediated by text recall/inference. Task-switchingwas associated with decoding, but its relation to passagecomprehension was not significant. Inhibitory control showedindirect effects on passage comprehension via decoding andtext recall/inference. Results indicate overlapping but distinctcontributions of EF to literacy skills.

We examined whether probability weighting in decisions madeunder risk changed depending on the difference in evaluationmethods. In particular, we focused on two methods, joint eval-uation (JE) and separate evaluation (SE). We conducted a be-havioral experiment and found that participants put more prob-ability weight on small probability when using the SE methodthan when using JE, and that for large probabilities, the inversewas observed (i.e., participants put more weight in JE). We an-alyzed these results using a cognitive model and found that par-ticipants’ subjective value of money does not change owing todifferences in evaluation methods. However, beliefs concern-ing uncertain events shifted depending on evaluation methods,which led to the differences in probability weight. In this paper,we also discuss psychological mechanisms that produce differ-ent judgments or evaluations between SE and JE.

We describe a system that can associate images with Englishproverbs. We start from a corpus of proverbs, harvest relatedimages from the web and use this data to train two variants ofa convolutional neural network. We then collect a small set ofannotations, and use these to combine the outputs of the twonetworks into a single prediction for each input image. Wecarry out feature selection experiments on a set of features de-rived from the images and from the predicted proverbs, anddemonstrate that the metaphoricity of the proverbs plays a sig-nificant role in classification accuracy. An empirical evalua-tion with human raters confirms the system’s ability to abstractfrom the raw bits in the images and to learn meaningful, non-trivial associations.

Exploration and Exploitation represent two mutually exclusive goals associated with choices within an environment:search too little and the lack of information will make it difficult to distinguish good options penalizing the agent inthe long run (exploiting) or search too much and suffer sub-optimal performance in the short term (exploring). Striking abalance between exploiting and exploring requires the learner to behave optimally in different environments. Managingthis trade-off is an important process of our lives but isnt completely understood from a cognitive science perspective. Tothis end we present the findings from an experiment where the main objective was to examine how much the presence ofcompetition and threats affects both behaviors: the presence of competition directs greater exploration and the presenceof threats reduces this behavior, suggesting that learners prioritize their learning behavior in response to the presence ofdifferent types of agents in the environment.

Many studies on memory retrieval in language processing haveidentified similarity-based interference as a key determinant ofcomprehension. The broad consensus is that similarity-basedinterference reflects erroneous retrieval of a non-target itemthat matches some of the retrieval cues. However, themechanisms responsible for such effects remain debated.Activation-based models of retrieval (e.g., Lewis & Vasishth,2005) claim that any differences in processing difficulty due tointerference in standard RT measures and judgments reflectdifferences in the speed of retrieval (i.e., the amount of time ittakes to retrieve a memory item). But this claim is inconsistentwith empirical data showing that retrieval time is constant dueto the use of a direct-access procedure (e.g., McElree, 2000,2006). According to direct-access accounts, differences injudgments or RTs due to interference arise from differences inthe quality or availability of the candidate memoryrepresentations, rather than differences in retrieval speed. Toadjudicate between these accounts, we employed a novelmethodology that combined a high-powered (N = 200) two-alternative forced-choice study on interference effects withdrift diffusion modeling to disassociate the effects of retrievalspeed and representation quality. Results showed that thepresence of a distractor that matched some of the retrieval cueslowered asymptotic accuracy, reflecting an effect ofrepresentation quality, but did not affect retrieval speed,consistent with a direct-access procedure. These results suggestthat the differences observed in RTs and judgment studiesreflect differences in the ease of integrating the retrieved itemback into the current processing stream, rather than differencesin retrieval speed.

Natural language is replete with figurative expressions like my lawyer is a shark, and listeners are expected to intuitively understand the intended, rather than the literal, meaning of such expressions. But what cognitive resources are involved in attaining meaning for such sentences? Most research into metaphor comprehension has employed offline reading tasks that provide no insight into the time-course of metaphor processing. In order to investigate the moment-by-moment on-line processes involved in metaphor comprehension, the present study used a naturalistic cross-modal lexical decision paradigm (Swinney, 1979) with novel brief masked target presentations during and after the vehicle word (shark). Results obtained from a preliminary sample demonstrated priming of related target words across conditions, but no significant differences between conditions. These results may best be interpreted as supporting an exhaustive-access account of metaphor interpretation, which suggests that literal and metaphorical interpretations are simultaneously accessed during the early stages of metaphor/simile interpretation.

A plurality of the categories we hold exhibit family resemblance (FR; i.e., many characteristic but few defining features), suggesting FR may occupy a central role in human category formation. However, research in unsupervised learning has shown that when people are asked to sort an array of novel items into categories, they ubiquitously use a unidimensional (UNI) rule – despite the availability of a FR solution. This work suggests that, perhaps, FR similarity is not a core tendency in category formation. Here, we question whether the UNI bias is a result of the sorting paradigm. Specifically, we speculate the paradigm conflates two components vital for category formation: production and evaluation. Across three experiments we show that when evaluation is separated from generation – by using a novel forced-choice task that pits different category organizational schemes against one another – people exhibit a FR over UNI preference. The implications of these results are discussed.

Rock, Paper, Scissors (RPS) is a competitive game. There arethree actions: rock, paper, and scissors. The game’s rules aresimple: scissors beats paper, rock beats scissors and paper beatsrock (all signs stalemate against themselves). Over multiplegames with the same opponent, optimal play according to aNash Equilibrium requires subjects to play with genuinerandomness. To examine randomness judgments in the contextof competition, we tested subjects with identical sequences intwo conditions: one produced from a dice roll, one fromsomeone playing rock, paper, scissors. We compared thesefindings to models of subjective randomness from Falk andKonold (1997) and from Griffiths and Tenenbaum (2001),which explain assessments of randomness as a function ofalgorithmic complexity and statistical inference, respectively.In both conditions the models fail to adequately describesubjective randomness judgements of ternary outcomes. Wealso observe that context influences perceptions of randomnesssuch that some isomorphic sequences produced fromintentional play are perceived as less random than dice rolls.We discuss this finding in terms of the relation betweenpatterns and opponent modeling.

I present a novel approach to modelling spatial mental im-agery within the ACT-R cognitive architecture. The proposedmethod augments ACT-R’s representation of visual objects toenable the processing of spatial extent and incorporates a set oflinear and affine transformation functions to allow the manip-ulation of internal spatial representations. The assumptions ofthe modified architecture are then tested by using it to developmodels of two classic mental imagery phenomena: the mentalscanning study of Kosslyn, Ball, and Reiser (1978) and mentalrotation (Shepard & Metzler, 1971). Both models provide veryclose fits to human response time data.

We see the world as continuous with smooth movements of objects and people, even though visual inputs can consist of stationary frames. The perceptual construction of smooth movements depends not only on low-level spatiotemporal features but also high-level knowledge. Here, we examined the role of causality in guiding perceptual interpolation of motion in the observation of human actions. We recorded videos of natural human-object interactions. Frame rate was manipulated to yield short and long stimulus-onset-asynchrony (SOA) displays for a short clip in which a catcher prepared to receive a ball. The facing direction of the catcher was either maintained intact to generate a meaningful interaction consistent with causality, or was transformed by a mirror reflection to create a non-causal situation lacking a meaningful interaction. Across three experiments, participants were asked to judge whether the catcher’s action showed smooth movements or sudden changes. Participants were more likely to judge the catcher’s actions to be continuous in the causal condition than in the non-causal condition, even with long SOA displays. This causal interpolation effect was robust to manipulations of body orientation (i.e. upright versus inverted). These findings indicate that causality in human actions guides interpolation of body movements, thereby completing the history of an observed action despite gaps in the sensory information. Hence, causal knowledge not only makes us see the future, but also fills in information about recent history.

Previous research evaluating the influence of category knowl-edge on memory found that children, like adults, rely on cat-egory information to facilitate recall (Duffy, Huttenlocher, &Crawford, 2006). A model that combines category and targetinformation (Integrative) provides a superior fit to preschoolersrecall data compared to a category only (Prototype) and targetonly (Target) model (Macias, Persaud, Hemmer, & Bonawitz,in revision). Utilizing data and computational approaches fromMacias et al., (in revision), we explore whether individual andage-related differences persist in the model fits. Results re-vealed that a greater proportion of preschoolers recall was bestfit by the Prototype model and trials where children displayedindividuating behaviors, such as spontaneously labeling, werealso best fit by the Prototype model. Furthermore, the best fit-ting model varied by age. This work demonstrates a rich com-plexity and variation in recall between developmental groupsthat can be illuminated by computationally evaluating individ-ual differences.

When reasoning about evidence, we must carefully consider the impact of different structures. For instance, if in the process of evaluating multiple reports, we find they rely on the same, shared evidence, then the support proffered by those reports is dependent on that evidence. Critically, normative accounts suggest that such a dependency results in redundant information across reports (reducing evidential support), relative to reports based on distinct items of evidence. In the present work we disentangle the structural and observation-based indicators of this form of dependency. In so doing, we present novel findings that lay reasoners are not only insensitive to shared evidence structures when updating their beliefs, but also that reasoners do not necessarily prefer more diverse sources of evidence. Finally, we replicate prior effects in reasoning under uncertainty, including conservative sequential updating, and difficulty in integrating contradictory reports.

Humans develop an ability for Theory of Mind (ToM) by theage of six, which enables them to infer another agent’s men-tal state and to differentiate it from one’s own. Much evi-dence suggests that humans can do this in a presumably op-timal way and, correspondingly, a Bayesian Theory of Mind(BToM) framework has been shown to match human infer-ences and attributions. Mostly, this has been investigated withspecific, explicit mentalizing tasks. However, other researchhas shown that humans often deviate from optimal reasoningin various ways. We investigate whether typical BToM modelsreally capture human ToM reasoning in tasks that solicit moreintuitive reasoning. We present results of an empirical studywhere humans deviate from Bayesian optimal reasoning in aToM task but instead exhibit egocentric tendencies. We alsodiscuss how computational models can better account for suchsub-optimal processing.

Mind wandering involves internally focused attention, and isoften conceptualized as the opposite of external attention thatis oriented towards the task at hand. Individuals vary ac-cording to the amount they mind wander as well as with re-gards to the pattern of oscillations between mind wanderingthoughts and externally directed, focused thought. Assumingthat mind wandering is influenced by episodic contents, we ex-plore the proposition that mind wandering frequency is relatedto the manner in which individuals deal with the contents ofepisodic memory, as reflected by a maximizing decision mak-ing style. Based on previous studies measuring cognitive pro-cesses, we assume that mouse trajectories towards a particu-lar response on the screen are continuously updated by time-dependent and temporally-dynamic cognitive processes. Asa behavioral methodology, mouse tracking provides potentialcues to help predict mind wandering. In our experiment, a to-tal of 274 students completed a decision making questionnaire,episodic and associative memory tests (during which mousemovements were recorded) and a working memory task, dur-ing which mind wandering thoughts were assessed. We foundcertain mouse movement characteristics to be significantly pre-dictive of mind wandering. Also, a maximizing decision mak-ing style appeared to be related to a particular type of mindwandering, namely, task-related interference.

Creating effective educational interventions that correctpeople’s misconceptions is difficult. This has led manyresearchers to conclude that people do not properly attend tonew information in a way that they should. However, even if ascientifically-grounded intervention fails, it is still possiblethat other interventions would be effective. Yet, it is notpractically feasible to systematically explore and test theentire hypothesis space of possible interventions. Here, weexamined whether researchers could use online arguments todevelop effective educational interventions, in effect,narrowing the intervention hypothesis space. Across twoexperiments (N = 1, 816), we found that argumentscrowdsourced from Reddit’s Change My View were aseffective or more effective at changing beliefs thaninterventions developed by academics and published intop-tier scientific journals. These results suggest thatresearchers can build on successful crowdsourced argumentsto develop effective educational interventions likely to correctpeople’s misconceptions in more naturalistic settings.

Curiously, people assign less punishment to a person who attempts and fails to harm somebody if their intended victim happens to suffer the harm for coincidental reasons. This “blame blocking” effect provides an important evidence in support of the two-process model of moral judgment (Cushman, 2008). Yet, recent proposals suggest that it might be due to an unintended interpretation of the dependent measure in cases of coincidental harm (Prochownik, 2017; also Malle, Guglielmo, & Monroe, 2014). If so, this would deprive the two-process model of an important source of empirical support. We report and discuss results that speak against this alternative account.

Over the last 20 years, causal-model theory has produced muchknowledge about causal-based categorization. However, per-sistent violations to the normative causal-model theory areprevalent. In particular, violations to the Markov conditionhave been repeatedly found. These violations have receiveddifferent explanations. Here, we develop a model that startsfrom generally accepted cognitive phenomena (e.g., process-ing limitations, the relevance of inference in cognitive process-ing) and assumes that people are not fully causal nor fully asso-ciative when performing causal-based categorization, offeringa new explanation for Markov violations.

Humans can learn and reason about abstract concepts quickly,flexibly, and often from very little data. Here, we study howpeople learn novel concepts within a binary grid domain, andfind that even this minimal task nonetheless necessitates theinference of highly structured parts as well as their compo-sitional relationships. Furthermore, by changing the presen-tation condition of the learning examples, we reveal differentapproaches involved in learning such visual concepts: giventhe same images, human generalizations differ between rapidand static presentation conditions. We investigate this differ-ence by developing several computational models that vary intheir use of structured primitives and composition. We find thatlearning in the rapid presentation condition is best described asinference in simple models, while learning in the static presen-tation condition is best described as inference in a more struc-tured space of graphics programs.

The answers people give when asked to “think of theunexpected” for everyday event scenarios appear to be moreexpected than unexpected. There are expected unexpectedoutcomes that closely adhere to the given information in ascenario, based on familiar disruptions and common plan-failures. There are also unexpected unexpected outcomes thatare more inventive, that depart from given information, addingnew concepts/actions. However, people seem to tend toconceive of the unexpected as the former more than the latter.Study 1 tests these proposals by analysing the object-conceptspeople mention in their reports of the unexpected and theagreement between their answers. Study 2 shows that object-choices are weakly influenced by recency, that is, the order ofsentences in the scenario. The implications of these results forideas in philosophy, psychology and computing are discussed.

Converging evidence suggests that children’s linguistic and theory of mind (ToM) development are linked. Specifically, learning the sentential complement grammatical structure has been shown to play a causal role in the development of some false belief reasoning skills. Here, we extend this line of work to examine this relationship in the wild by means of a corpus analysis of children’s speech during the typical period of ToM development. We show that children’s use of the sentential complement grammatical structure increases immediately preceding the ToM development period and plateaus shortly thereafter. Furthermore, we find that parents’ child-directed speech follows a similar pattern.

When does a reasoner respond that ”no valid conclusion” (NVC)follows in a syllogistic reasoning task? Cognitive theories aim totrace it back to theory specific inference processes. In contrast,systemic theories explain it by depleted cognitive resources amongothers. This paper investigates possible theories to explain NVCresponses in an experiment with 139 participants. Using mixedmodels we analyze the association of NVC responses with reactiontimes, the validity as well as the entropy of a syllogism, and howNVC responses change over time. As expected, the number ofNVC responses is lower than logically expected, participantsrespond NVC more often for invalid syllogisms, and the likelihoodto respond NVC increases over the time-course of the experiment.Surprisingly, however, only for valid syllogisms, are the entropyand the RTs associated with NVC responses. Consequently,for invalid syllogisms, NVC responses seem to be generateddifferently as compared to valid ones.

In the present study, we examine whether the design of thelearning environment can impact causal inference in veryyoung children. Specifically, we assess whether the physicalfeatures of a novel toy can facilitate children’s recognition ofan abstract, relational hypothesis (same-different) that theytypically fail to discover. Three-year-olds were presented withan identical pattern of evidence that was consistent with arelational hypothesis (i.e., pairs of same or different blockscause a toy to activate) using one of two causal toys. In thestandard condition, blocks were placed in pairs on top of thetoy, while in the relational condition, each block was placedinside one of two transparent openings on either side of thetoy. The physical design of the latter toy was intended tohighlight the relationship between pairs of blocks. Resultssuggest that even 3-year-olds’ causal inferences are sensitiveto design, with children in the relational condition more likelyto infer the abstract relation than those in the standard case.These results provide strong evidence that design serves as aconstraint on causal inference in early childhood. Findings arediscussed in terms of their implications for creating intuitivelearning environments for young children.

The complex judgments we make about the innumerable ob-jects in the world are made on the basis of our representa-tion of those objects. Thus a model of judgment should spec-ify (a) our representation of the many objects in the world,and (b) how we use this knowledge for making judgments.Here we show that word embeddings, vector representationsfor words derived from statistics of word use in corpora, proxythis knowledge, and that accurate models of judgment can betrained by regressing human judgment ratings (e.g., femininityof traits) directly on word embeddings. This method achieveshigher out-of-sample accuracy than a vector similarity-basedbaseline and compares favorably to human inter-rater relia-bility. Word embeddings can also identify the concepts mostassociated with observed judgments, and can thus shed lighton the psychological substrates of judgment. Overall, we pro-vide new methods and insights for predicting and understand-ing high-level human judgment.

Thematic roles such as Agent and Instrument have a long- standing place in theories of event representation. Nonetheless, the structure of these categories has been difficult to determine. We investigated how instrumental events, such as someone slicing bread with a knife, are categorized in English. Speakers described a variety of typical and atypical instrumental events, and we determined the similarity structure of their descriptions using correspondence analysis. We found that events where the instrument is an extension of an intentional agent were most likely to elicit similar language, highlighting the importance of agency in structuring instrumental categories.

The current study used an eye tracker to examine how auditoryinput affects the latency of visual fixations and speeded responseson a Serial Response Time Task (SRTT). In Experiment 1,participants viewed a sequence of visual stimuli that appeared indifferent locations on a computer monitor and the same sequencerepeated throughout the experiment. The visual sequence waseither presented in silence or paired with uncorrelated sounds(i.e., sounds did not predict visual target location). Participantsmade more fixations and were more likely to fixate on the visualstimuli when visual sequences were presented in silence thanwhen paired with sounds. Participants in Experiment 2 werepresented with the same sequences, but they also had to determineif each visual stimulus was red or blue. The presence of auditorystimuli had no effect on accuracy (red vs. blue), however, therewas some evidence that auditory stimuli delayed the latency offirst fixations to the visual stimuli and discriminating the imagesas red or blue was also slower relative to the unimodal visualbaseline. While visual stimuli often dominate auditory processingon spatial tasks, the current findings show that auditory stimulican also slow down visual detection on a task that is better suitedfor the visual modality. These findings are consistent with apotential mechanism underlying auditory dominance effects,which posits that auditory stimuli may attenuate and/or delay theencoding of visual information.

Creative ideas emerge from a meshwork of dynamic elements.Resources internal and external to the agent configure a cognitiveecosystem that scaffolds performance. In addition, capitalizing onfortuitous external cues may trigger new ideas. We examined theseelements to determine how they come into play during a simpleword production task. Participants were video recorded as theygenerated new words from 7 letter tiles in three differentenvironments (i) high interactivity where the titles could be movedat will (ii) low interactivity where they could not, and (iii) lowinteractivity where the order of the tiles could be shuffled but onceshuffled no additional actions were allowed. Overall, interactivityhad a marginally positive impact on performance, whileindependent measures of participants’ verbal fluency were strongpredictors of performance in all environments. Based on a detailedcoding of the video data, a finer-grained analysis of behaviour in thehigh interactivity condition revealed that the time participants spentmanipulating the tiles was a significant predictor of performance.The video data also allowed us to measure the average latency to theproduction of a new word after shuffling the letters in the lowinteractivity condition as an index of how ‘lucky’ the reset was:Shorter average latencies were a significant predictor of overallword production. These data indicate that interactivity, serendipity,and internal cognitive resources determine problem-solvingperformance in this task.

A number of recent models of semantics combine linguistic information, derived from text corpora, and visual information, derived from image collections, demonstrating that the resulting multimodal models are better than either of their unimodal counterparts, in accounting for behavioural data. However, first, while linguistic models have been extensively tested for their fit to behavioural semantic ratings, this is not the case for visual models which are also far more limited in their coverage. More broadly, empirical work on semantic processing has shown that emotion also plays an important role especially for abstract concepts, however, models integrating emotion along with linguistic and visual information are lacking. Here, we first improve on visual representations by choosing a visual model that best fit semantic data and extending its coverage. Crucially then, we assess whether adding affective representations (obtained from a neural network model designed to predict emojis from co-occurring text) improves model’s ability to fit semantic similarity/relatedness judgements from a purely linguistic and linguistic-visual model. We find that adding both visual and affective representations improve performance, with visual representations providing an improvement especially for more concrete words and affective representations improving especially fit for more abstract words.

Models of visual saliency normally belong to one of twocamps: models such as Experience Guided Search (E-GS),which emphasize top-down guidance based on task features,and models such as Attention as Information Maximisation(AIM), which emphasize the role of bottom-up saliency. Inthis paper, we show that E-GS and AIM are structurally simi-lar and can be unified to create a general model of visual searchwhich includes a generic prior over potential non-task relatedobjects. We demonstrate that this model displays inattentionalblindness, and that blindness can be modulated by adjustingthe relative precisions of several terms within the model. Atthe same time, our model correctly accounts for a series ofclassical visual search results.

In software development, deliverables in an upstream process are reviewed to ensure their quality and to reduce error propagation to the downstream process. Methods are available for evaluating the review quality. In this study, we considered the defect detection process in a review of Requirement Definition Documents and tested a potential relationship between the gaze patterns and review quality. Specifically, we analyzed the relationship between the gaze patterns, with a primary focus on the blink rate, in a review of RDDs and detection accuracy. A significant nonlinear correlation between the blink rate and the detection accuracy was observed; moreover, the subsequent regression analysis also verified the blink rate as the best predictor of the review quality, notwithstanding the use of other gaze patterns. This result indicates that the blink rate is a major predictor of a type of review performance.

This article used an empirical experiment and a computationalmodel to test the hypothesis that humans rely on the visualsystem to solve the traveling salesperson problem (TSP). Wetested two consequences of this hypothesis: (1) humans shouldperform better on Euclidean TSP than not–Euclidean TSP; (2)a model of the visual system should account for performance inEuclidean TSP. Participants were asked to solve Euclidean ornot–Euclidean TSP, and a pyramid model of the visual systemwas used to solve the same tours as the humans. The resultsshow that deviations from the optimal tour were smaller in Eu-clidean problems than in not–Euclidean problems, and the fitof the pyramid model to human performance was worse onnot–Euclidean problems then on Euclidean problems. Theseresults suggest that participants solve Euclidean problems withthe visual system, but that other mechanisms are needed to suc-cesfully solve non–visual problems.

While there is substantial evidence showing that assistanceprovided to students during problem-solving activitiesinfluences learning outcomes, it is not yet clear how to bestdesign educational technologies to maximize learning throughvarious types of assistance. One common type of assistancecorresponds to hints delivered by an educational technology.To date, however, there is little research on the impact ofdifferent types of hints, including high-level hints vs. specificbottom-out hints. Our research takes a step in filling this gap,through an experimental study with an intelligent tutoringsystem we implemented in the domain of algebra (N = 50).We did not find evidence that the type of hint, high level vs.bottom out, influenced learning, with both types of hintsproducing similar outcomes. We did, however, find supportfor the conclusion that the number of hints accessed interactedwith the type of hint to influence learning, and specifically,that accessing more hints was correlated with learning butonly in the high-level hint condition.

It is often assumed that cross-linguistically more prevalent distinctions are easier to learn (Typological Prevalence Hypothesis - TPH). Prior work supports this hypothesis in phonology, morphology and syntax but has not addressed semantics. Using an Artificial Language Learning paradigm, we explore the learnability of semantic distinctions within the domain of evidentiality (i.e. the linguistic encoding of information sources). Our results support the TPH, since the most prevalent evidential system was learned best while the most rare evidentiality system yielded the worst learnability results. Furthermore, our results indicate that, cross- linguistically, indirect information sources seem to be marked preferentially (and acquired more easily) compared to direct sources. We explain this pattern in terms of the pragmatic need to mark indirect, potentially more unreliable sources over direct sources of information.

There is an influential body of research arguing that category exceptions have a special status in memory compared to regular category members. However, the memory advantage for category exceptions has typically been demonstrated using one very specific category structure (Differentiation). Here we present a study examining whether the reported memory advantage is specific to this particular structure or whether it can be generalized to other kinds of exceptions (Isolation and Odd-ball). We compare three different types of category exceptions that have varying memory demands due to different levels of feature binding required for accurate categorization. The results suggest that only those exceptions that require binding together multiple features are remembered better than regular, rule-following items. The present work clarifies that the memory advantage for exceptions characterizes certain kinds of exceptions rather than exceptions in general.

Our knowledge of words consists of a lexico-semantic network in which different words and their meanings are connected by relations, such as similarity in meaning. This research investigated the integration of new words into lexico-semantic networks. Specifically, we investigated whether new words can rapidly become linked with familiar words given exposure to distributional regularities that are ubiquitous in real-world language input, in which familiar and new words either: (1) directly co-occur in sentences, or (2) never co-occur, but instead share each other’s patterns of co-occurrence with another word. We observed that, immediately after sentence reading, familiar words came to be primed not only by new words with which they co-occurred in sentences, but also by new words with which they shared co- occurrence. This finding represents a novel demonstration that new words can be rapidly integrated into lexico-semantic networks from exposure to distributional regularities.

The aim of this study is to refine a cognitive model forthe takeover in highly automated driving. The focus lieson the impact of objective complexity on the takeoverand resulting outcomes. Complexity consists of variousaspects. In this study, objective complexities are di-vided into the complexity of the non-driving-related task(no-task, listening, playing, reading, searching) and thetraffic complexity (relevant vehicles in the driving envi-ronment). The impact of a non-driving related tasks’complexity on the takeover is evaluated in empiricaldata. Following, the cognitive model is run through sit-uations of different traffic complexities and compared toempirical results. The model can account for empiricaldata in most of the objective complexities. Additionally,model predictions are tested on significant variations indifferent complexities until the action decision is made.In more complex traffic conditions, the model predictslonger times on different processing steps. Altogether,the model can be used to explain cognitive mechanismsin differently complex traffic situations.

Cognitive enrichment for captive animals is the idea that cog-nitive stimulation can improve animal welfare. In zoos, cog-nitive enrichment not only helps the animals themselves butalso contributes to zoo missions of educating the public, sup-porting research, and more. Technology-based cognitive en-richment tools are increasingly popular for a variety of rea-sons, though they also present unique challenges for designand deployment. In this paper, we present a short review oftechnology-based cognitive enrichment programs in zoo set-tings, and then describe the design and development processwe used to create a new, touchscreen-based enrichment appfor a group of orangutans at Zoo Atlanta. We discuss initialobservations about the orangutans’ use of this app, as well aslessons learned by our research team.

We investigated the application of non-linear analysis techniques for capturing stability of neural oscillatory activity within and across brains. Recurrence Quantification Analysis (RQA), a technique that has been applied to detect stability and flexibility of motor performance, was extended to observe and quantify changes in patterns of non-linear neural activity. Participants synchronized their finger-tapping with a confederate partner who tapped at two different rhythms while neural activity was recorded from both partners using electroencephalography (EEG). Auto-recurrence (intra-brain) and cross-recurrence (inter-brain) of EEG activity were able to distinguish differences across tapping rhythms in stability of neural oscillatory activity. We also demonstrated the efficacy of RQA to capture how both period and phase changes in neural dynamics evolve over time.

Despite constantly using energy and having extensive interactions with household appliances, people consistently mis-estimate the amount of energy that is used by home appliances. This poses major problems for conservation efforts, while also presenting an interesting case study in human perception. Since many forms of energy used are not directly perceptible, and since the amount of energy that is being used by an appliance is often difficult to infer from appearances alone, people often rely on cues. Some of these cues are more reliable than others and previous literature has investigated which of these cues people rely on. However, past literature has always studied these proximal cues in isolation— despite the fact that, during real-world perception, people are always integrating a variety of cues. Here, we investigate how people rely on a variety of cues, and how individual differences in the reliance on those cues predicts the ability to estimate home energy use.

What drives people’s exploration in complex scenarios wherethey have to actively acquire information? How do peopleadapt their selection of queries to the environment? We explorethese questions using Entropy Mastermind, a novel variantof the Mastermind code-breaking game, in which participantshave to guess a secret code by making useful queries. Partici-pants solved games more efficiently if the entropy of the gameenvironment was low; moreover, people adapted their initialqueries to the scenario they were in. We also investigatedwhether it would be possible to predict participants’ querieswithin the generalized Sharma-Mittal information-theoreticframework. Although predicting individual queries was dif-ficult, the modeling framework offered important insights onhuman behavior. Entropy Mastermind opens up rich possibili-ties for modeling and behavioral research.

Speakers exhibit variability in their choice between uncertaintyexpressions such as might and probably. Recent work hasfound that listeners cope with such variability by updating theirexpectations about how a specific speaker uses uncertainty ex-pressions when interacting with a single speaker. However, itis still unclear to what extent listeners form speaker-specificexpectations for multiple speakers and to what extent listenersare adapting to a situation independent of the speakers. Here,we take a first step towards answering these questions. In Ex-periment 1, listeners formed speaker-specific expectations af-ter being exposed to two speakers whose use of uncertaintyexpressions differed. In Experiment 2, listeners who were ex-posed to two speakers with identical use of uncertainty expres-sions formed considerably stronger expectations than in Exper-iment 1. This suggests that listeners form both speaker-specificand situation-specific expectations. We discuss the implica-tions of these results for theories of adaptation.

Mentalization is an important ability to acquire for children,as it allows humans to understand the mental state of others oroneself, that underlies overt behavior (Fonagy & Target,1996). In the current study we examined the relationshipbetween development of mentalization ability in children andtheir experience of playing with a doll by observing child-mother interaction and by using functional near-infraredspectroscopy (fNIRS). 44 dyads of children aged 2 to 3 andtheir mothers were divided into two groups (high and low)depending on the frequency of doll-play experience. Weexamined mother-speech interaction during the doll play. Wealso used fNIRS system to measure cerebral hemodynamicactivation in the frontal and temporal regions during theobservation of video clips showing hindering and helpingbehaviors. The results showed that a mother’s proxy talk wasrelated to a child’s doll directed speech in the high group, butnot in the low group. fNRIS data showed that cerebralactivation in the helping condition was more increased in thelow group than the high group. This suggests that doll-playexperience facilitates the development of mentalization,which enables children to be aware of and understand other'spsychological states.

The rise of ubiquitous computing has cemented ubiquitous reproduction (UR) as a defining feature of contemporary human environments. UR is most obvious on our televisions and smartphones but has homogenised most material aspects of our lives. Emerging technologies such as 3D printing and robotics will ensure that this trend intensifies. UR is an issue of global scale that is relatively intractable to qualitative treatment. This paper introduces a novel quantitative approach to cognitive science and to analysis of UR. The approach uses the finiteness of cognition to establish a minimal ontology with which to model cognitive diversity under UR. It demonstrates that, despite widespread valorisation of diversity, cognitive diversity must be declining at a global level. The implications of this for creativity are that the arc for creative impact is growing shorter as the need to be immediately intelligible promotes the formulaic at the expense of the interpretable.

We traced the development of sensitivity to symmetric relational patterns by creating a symmetry match-to-sample task. Children saw a symmetric standard made up of two shapes and choose between two novel alternatives: a symmetric pair and an asymmetric pair. We found that young children chose randomly between the two alternatives. Children were not reliably above chance until 8-to 9 years of age. In a second study, we found that young children could succeed in making symmetric relational matches if the triads were designed to invite informative comparisons. These findings show that relational insight of symmetry develops relatively late. However, as with other relations, comparison processes can promote sensitivity to the symmetry relation.

Honorifics in Korean appear as verbal inflections and havebeen considered as markers of politeness. This study inves-tigates the pragmatic effects of honorifics, and suggests thathonorifics can contribute to the semantic interpretation of verbphrases in complex ways. Native Korean speakers reporteddifferent inferred meanings of “did very well” and “did verypoorly” based on the normative or non-normative honorificforms. We found significant effects of non-normative hon-orifics in positive assessments: over-polite honorifics broughtnegative interpretations. This suggests that pragmatic listen-ers interpret utterances based on the interaction between lit-eral meanings, honorifics, and the normativity of the hon-orifics within a relationship context, to obtain an estimate ofthe speaker’s intended meaning. This is inconsistent with theprevious explanations of honorific usage as discernment or vo-litional politeness. We suggest that non-literal meaning infer-ences reflect listeners treating the honorifics as signals to po-tential communicative goals.

Unscaffolded problem-solving before receiving instruction cangive students opportunities to entertain their exploratory hy-potheses at the expense of experiencing initial failures. Priorliterature has argued for the efficacy of such Productive Fail-ure (PF) activities in preparing students to “see” like an expert.Despite growing understanding of the socio-cognitive mecha-nisms that affect learning from PF, the necessity of success orfailure in initial problem-solving attempts is still unclear. Con-sequently, we do not know yet whether some ways of succeed-ing or failing are more efficacious than others. Here, we reportempirical evidence from a recently concluded classroom PF in-tervention (N=221), where we designed scaffolds to explicitlypush student problem-solving towards success via structuring,but also radically, towards failure via problematizing. Our ra-tionale for explicit failure scaffolding was rooted in facilitatingproblem-space exploration. We subsequently compared thedifferential preparatory effects of success-driven and failure-driven problem-solving on learning from subsequent instruc-tion. Results suggested explicit failure scaffolding during ini-tial problem-solving to have a higher impact on conceptual un-derstanding, compared to explicit success scaffolding. Thistrend was more salient for the task topic with greater difficulty.

Productive Failure (PF) is a learning design that intentionallydesigns for and uses failure in preparatory problem-solving forlearning. Over the past decade, there has been growing ev-idence supporting the effectiveness of learning from PF. Thepurpose of this paper, however, is to critically examine evi-dence for when PF fails. We analyze 95 experimental compar-isons from 57 studies reported in 44 articles into the extent towhich they conform to PF design criteria. These criteria, asoutlined in the original PF work, span the problem-solving ac-tivity, the participation structures, and the social surround. Re-sults suggest lack of design fidelity as a critical factor for whenPF fails to outperform alternative instructional approaches onconceptual knowledge and/or transfer.

Effective problem solving requires both exploration and ex-ploitation. We analyze data from a group problem-solving taskto gain insight into how people use information from past expe-riences and from others to achieve explore-exploit trade-offs incomplex environments. The behavior we observe is consistentwith the use of simple, reinforcement-based heuristics. Partic-ipants increase exploration immediately after experiencing alow payoff, and decrease exploration immediately after expe-riencing a high or improved payoff. We suggest that whetheran outcome is perceived as “high” or “low” is a dynamic func-tion of the outcome information available to participants. Thedegree to which the distribution of observed information re-flects the true range of possible outcomes plays an importantrole in determining whether or not this heuristic is adaptive ina given environment.

Previous research on adjective ordering in linguistics andpsychology has focused primarily on the unmarked or defaultorder of adjectives, as in large blue car. Inverted word order,as in blue large car, which violates the proposed semanticconstraints on adjective placement, received relatively littleattention. In two studies we show that the inverted order is notas limited in scope as previous researchers have argued. Wepropose that the inverted word order reflects the subjectivedistance principle: the attribute that is psychologically closerto the speaker is mentioned first. Our explanation draws onresearch on word order in binomials, thus connecting twopreviously unrelated research traditions on word order inlinguistics and cognitive psychology.

The mnemonic benefits of animate (e.g., Tiger) over inanimate (e.g., Table) stimuli have been demonstrated across several different memory paradigms. Given the ubiquity of inanimate, computer-generated voices we investigated if the animacy of a presentation source confers mnemonic benefits. We asked: is information delivered by a human voice better remembered than information presented by a computer-generated voice? Word-lists were presented auditorily by either a human or a computer-generated voice and memory was measured using a free recall assessment. In Experiment 1, words presented in a human voice were better remembered than words presented in a computer voice. Experiment 2 demonstrated that beliefs about the animacy of a computer-generated voice were not sufficient for any benefits to accrue, suggesting a possible boundary condition for the effect. Both experiments replicated the mnemonic benefits of animate words and demonstrated further extensions of the effect to spoken word presentation.

The goal of the present study was to demonstrate the potentialapplication of Item Response Theory (IRT) outside itstraditional use in assessing questionnaires by applying it todata from behavioural task. We did this by validating aperspective taking task called the Director Task used to assessTheory of Mind (ToM) abilities in young adults. IRT andconvergent validity analyses indicated that, contrary to ourhypotheses, the Director Task had an unduly narrow range ofresponding for measuring ToM. Furthermore, the DirectorTask did not correlate with other established measures ofToM. Our results suggest that the task should be used withcaution when assessing a young adult population.Furthermore, since convergent validity was not established, itis uncertain what specifically the task measures. Overall, weshow how IRT may serve as a useful tool in evaluatingbehavioural measures.

We employ a scenario-sentence-verification paradigm to inves-tigate the role of scenario-given alternatives for the process-ing of affirmative and negative sentences. We show that forboth affirmative and negative sentences the N400 amplitude islarger if the context model provides multiple alternatives fora true sentence continuation relative to the case when it pro-vides only a unique referent. Additionally, we observe a latepositivity effect for negative relative to affirmative sentences,independent of the context model.

The classic drift diffusion model of the 2AFC choice processassumes that observers select evidence accumulation thresh-olds to optimize some desired level of accuracy across the ex-periment. We argue that it is more ecologically natural to as-sume that decision-makers set this threshold adaptively, usinginformation from recent trials to adjust it for upcoming ones.To test this hypothesis, we designed and conducted a pair ofrandom dot motion discrimination experiment where the co-herence parameter that controls task difficulty varies across tri-als in a predictable manner. To analyze data from these exper-iments, we also designed a hierarchical drift diffusion modelthat allows decision-makers to adapt their evidence thresholdbased on the trend of difficulty of previous trials. Our resultssuggest that observers rationally integrate cross-trial informa-tion about trial difficulty into perceptual decision-making byadjusting their internal evidence thresholds. We briefly discussthe implications of the existence of such trial-level effort infer-ence on contemporary models of the choice process.

This paper provides empirical evidence that human decision- makers use prospective regret minimization as their dominant decision strategy when regret calculations are cognitively easier to perform, and use expected utility maximization when they aren't. We designed a simple decision problem wherein utility maximization and expected regret minimization yield distinctly difference choices, and manipulated the cognitive effort involved in making regret calculations across respondent samples to arrive at our results. While previous research has associated ecological considerations like sense of responsibility and familiarity with this difference, we show that, at least in experimental settings, cognitive calculability in regret space appears to predominantly drive this difference. We also show that this preference for regret minimization can be countermanded by changing the distribution of options presented to the respondent, posing a challenge to simple sequential accounts of strategy selection learning which sequence strategy selection and application in order.

There has been little cross-fertilization between research on ac-tive learning and teaching, despite extensive conceptual simi-larities. The current study aims to bridge the gap by show-ing that engaging in active learning can influence subsequentteaching performance. In a one-dimensional boundary teach-ing task, participants who first took the role of an active learnerwent on to become better teachers than participants who didnot. In order to disentangle the effect of active selection ofsamples from their information content, the performance ofactive learners was compared to that of yoked passive learn-ers. While prior passive learning also significantly boostedteaching performance, it did so to a lesser extent. However, inpaired comparisons, teachers with active learning experiencedid not differ significantly from their yoked-passive learningcounterparts. Based on the current results we cannot arguefor a teaching benefit specific to active learning as opposed toa more general improvement caused by experiencing the taskfrom the learner’s perspective. However, we suggest that thisis a promising line of inquiry using more complex learning andteaching tasks.

A common result is that comparison settings (i.e., several stimuli introduced simultaneously) favor conceptualization and generalization. In a comparison setting, we manipulated the semantic distance between the two training items (e.g., two bracelets versus a bracelet and a watch), and the semantic distance between the training items and the test items (e.g., a pendant versus a bow tie). We tested 5- and 8-year-old children’s generalization of novel names for objects. This study is the first one to study the temporal dynamics of comparison in a generalization task with eye-tracking data. The eye movement data revealed clear patterns of exploration in which participants first focused on the training items and compared them with each of the choice options. We also compared the search profiles for correct answers and errors. The results show that participants first found commonalities in the learning items, which they compared with each items in the solution set. This pattern is consistent with an alignment view of generalization.

Computational models of visual working memory (VWM) generally fall into two categories: slots-based models and resources-based models. Slots-based models theorise that the capacity of memory is defined by a finite number of items. Each slot can only contain one item and once an item is in memory it is remembered accurately. If an item is not in memory, however, there is no memory of the item at all. By contrast, resources-based models claim that all items, rather than just a few enter memory. However, unlike the slots model they are not necessarily remembered accurately. On the surface, these models appear to make distinct predictions. However, as these models have been developed and expanded to capture empirical data, they have begun to mimic each other. Further complicating matters, Donkin, Kary, Tahir and Taylor (2016) proposed that observers were capable of using either slot- or resource-based encoding strategies. In the current experiment, we aimed to test the claim that observers adapt their encoding strategies depending on the task environment by observing how participants move their eyes in a VWM experiment. We ran participants on a standard colour recall task (Zhang and Luck, 2008) while tracking their eye movements. All participants were asked to remember either 3 or 6 items in a given trial, and we manipulated whether the number of items was held constant for a block of trials, or varied randomly. We expected to see participants use more resource-like encoding when the number of items to remember was predictable. Contrary to these expectations, we observed no difference between blocked and unblocked conditions. Further, the eye gaze data was only very weakly related to behaviour in the task. We conclude that caution should be taken in interpreting eye gaze data in VWM experiments.

Choice blindness is a cognitive phenomenon describing that when people receive false feedback about a choice they just made, they often accept the outcome as their own. Little is known about what predisposes people to correct manipulations they are subjected to in choice blindness studies. In this study, 118 participants answered a political attitude survey and were then asked to explain some of their responses out of which three had been manipulated to indicate an opposite position. Just over half (58.4%) of the manipulations were corrected. We measured extremity, centrality and commitment for each attitude, and one week prior to the experiment we assessed participants’ preference for consistency, need for cognition and political awareness. Only extremity was able to predict correction. The results highlight the elusiveness of choice blindness and speak against dissonance and lack of motivation to engage in cognitively demanding tasks as explanations why the effect occurs.

The current study investigates how children act on a standardexploreexploit bandit task relative to adults. In Experiment 1,we used childfriendly versions of the bandit task and foundthat children did not play in a way that maximized payout.However, children were able to identify the machines thathad the highest level of payout and overwhelmingly preferredit. We also show that children’s exploration is not random. Forexample, children selected the bandits from left to rightmultiple times. In Experiment 2, we had adults complete thetask in Experiment 1 with different sets of instructions. Whentold to maximize learning, adults explored the task in muchthe same way that children did. Together, these results suggestthat children are more interested in exploring than exploiting,and a potential explanation for this is that children are tryingto learn as much about the environment as they can.

Slang is a common device for expressivity in natural lan-guage. While slang has been studied extensively as a socialphenomenon, its cognitive bases are not well understood. Weformulate the processes of slang generation as a categoriza-tion problem. We explore a set of cognitive models of catego-rization that recommend slang words based on intended refer-ents of the speaker beyond the existing senses of words. Wetest these models against a large repertoire of slang sense def-initions from the Online Slang Dictionary and show that thecategorization models predict slang word choices substantiallybetter than chance, without explicit consideration of externalsocial factors. We also show that words similar in existingsenses tend to extend to similar novel slang senses, reflecting aprocess of parallel semantic change. Our work helps to groundtheories of slang in cognitive models of categorization and pro-vides the potential for machine processing of informal naturallanguage.

There has been a great deal of interest in howgeneralizations and exceptions are learned, withparticular interest in how speakers learn to avoidovergeneralizations. Do overgeneralizations disappearonly because exceptions become more stronglyrepresented or does the generalization itself becomesuppressed? Novel labels were constructed bycombining 56 syllables with one of two prefixes, andeach label was assigned a unique image. Most labels withthe first prefix were paired with images from ageneralization category, whereas exceptional labels werepaired with images from a different semantic category.All labels with the second prefix appeared with a thirdcategory (“baseline”). Participants used a computermouse to choose one of two images for each label.Mouse-tracking results show that the generalizationitself became suppressed over time in the context ofexceptional labels. A post-test demonstrated thatexceptions were learned with item-specific precision.

People act upon their desires, but often, also act in adherenceto implicit social norms. How do people infer these unstatedsocial norms from others’ behavior, especially in novel so-cial contexts? We propose that laypeople have intuitive the-ories of social norms as behavioral constraints shared acrossdifferent agents in the same social context. We formalize in-ference of norms using a Bayesian Theory of Mind approach,and show that this computational approach provides excellentpredictions of how people infer norms in two scenarios. Ourresults suggest that people separate the influence of norms andindividual desires on others’ actions, and have implications formodelling generalizations of hidden causes of behavior.

Exposure to Inconsistent message has been demonstrated as auseful method to alleviate belief in false rumor. However, thedata from previous research included unexplained variation inresponse to inconsistent message. Existing research alsoincluded the possibility that participants skipped out onreading and therefore they were not exposed to a message.We used an eye tracker to eliminate the possibility. Eyetracking data revealed that participants not only did not skipbut they paid more visual attention to inconsistent messagescomparing with consistent messages. Despite the overalleffectiveness of inconsistent message, some responsesshowed continued belief in rumors even after the exposure.Eye-tracking analyses demonstrated that when participantshad positive pre-belief for a rumor, more visual attention toinconsistent message predicted strengthened the belief. Wediscuss when exposure to inconsistent message does not workwell as a way for harnessing belief in false rumor.

Creativity is an essential factor in successful advertising wherecatchy and memorable media is produced to persuade the au-dience. The creative elements in the visual design and in theslogan of an advertisement elevate the overall appeal providinga perceptually grounded attractive message. In this study, wepropose the exploitation of creativity cues in textual and visualinformation for the appreciation prediction of multimodal ad-vertising prints. Moreover, as a novel dimension space of mul-timodality, we propose using the human sense (i.e., sight, hear-ing, taste, and smell) information embedded in the language.Our findings show that sensorial information is an invaluableindication of whether the advertisement is appreciated or not.Furthermore, combining linguistic and visual models signif-icantly improves the unimodal appreciation detection perfor-mances.

In everyday life, people see, describe and remember motion events. We tested whether the type of motion event information (path or manner) encoded in speech and gesture predicts which information is remembered and if this varies across speakers of typologically different languages. We focus on intransitive motion events (e.g., a woman running to a tree) that are described differently in speech and co-speech gesture across languages, based on how these languages typologically encode manner and path information (Kita & Özyürek, 2003; Talmy, 1985). Speakers of Dutch (n = 19) and Turkish (n = 22) watched and described motion events. With a surprise (i.e. unexpected) recognition memory task, memory for manner and path components of these events was measured. Neither Dutch nor Turkish speakers’ memory for manner went above chance levels. However, we found a positive relation between path speech and path change detection: participants who described the path during encoding were more accurate at detecting changes to the path of an event during the memory task. In addition, the relation between path speech and path memory changed with native language: for Dutch speakers encoding path in speech was related to improved path memory, but for Turkish speakers no such relation existed. For both languages, co-speech gesture did not predict memory speakers. We discuss the implications of these findings for our understanding of the relations between speech, gesture, type of encoding in language and memory.

In real world contexts of reasoning about evidence, that evi-dence frequently arrives sequentially. Moreover, we often can-not anticipate in advance what kinds of evidence we will even-tually encounter. This raises the question of what we do to ourexisting models when we encounter new variables to consider.The standard normative framework for probabilistic reasoningyields the same ultimate outcome whether multiple pieces ofevidence are acquired in sequence or all at once, and it is in-sensitive to the order in which that evidence is acquired. Thisequivalence, however, holds only if all potential evidence isincorporated in a single model from the outset. Hence little isknown about what happens when evidence sets are expandedincrementally. Here, we examine this contrast formally and re-port the results of the first study, to date, that examines howpeople navigate such expansions.

Generic statements convey generalizations about categories,but how generic predications combine is unclear. “Elephantslive in Africa and Asia” does not mean that individual ele-phants live on both continents. In addition, such conjunc-tive generics pose interesting questions for theories of incre-mental processing because the meaning of the sentence canchange part-way through: “Elephants live in Africa” would im-ply most or all do, but “Africa and Asia” implies some live ineach. We extend a recently proposed computational model ofgeneric language understanding with an incremental process-ing mechanism that can begin to interpret an utterance beforea speaker has finished their sentence. This model makes novelpredictions about partial interpretations of conjunctive genericsentences, which we test in two behavioral experiments. Theresults support a strong view of incrementality, wherein lis-teners continuously update their beliefs based on expectationsabout where a speaker will go next with their utterance.

Imagining the future and remembering the past both involve mental time travel. This commonality could indicate shared mental processes, as held by the Constructive Episodic Simulation Hypothesis (Schacter & Addis, 2008), or else interactive processes that complement one another, a possibility we call the Complementarity Hypothesis. According to the Complementarity Hypothesis, future thoughts are constructed from schemas making them episodically poor, whereas past thoughts are constructed from schemas and direct retrieval of memory traces, making them relatively episodically rich. We tested these hypotheses using machine learning to data mine mental operations in language, much as a geologist can recover physical processes from the geological record. People’s natural, unprompted talk on web blogs was automatically analyzed for past, present, and future references using a temporal orientation classifier. In Study 1, we found that perceptual details were mentioned more often in past than future talk, implying greater use of episodic processing in past than future thinking. In Study 2, a neural network using schemas generated from Latent Dirichlet Allocation better predicted the content of references to the future than the past, implying that constructive processes are more common in future than past thinking. In Study 3, we used the results from the two prior studies to construct an episodic-by-constructive process space. We adapted techniques from fMRI analysis to analyze this space for clusters of activity, as if the frequency of past and future thinking were BOLD responses in cortical space. We found that past and future thinking occupy highly separable regions of processing space, supporting the Complementarity Hypothesis.

Causal and temporal reasoning are fundamentally linked, butfew studies have directly examined how the ability to makecausal inferences about the past vs. the future develops. We useda counterfactual reasoning task to explore 4- to 6-year-oldchildren’s understanding of the causal relationships among past,present, and future events. Like adults, even 4-year-olds judgedthat future, but not past, events could be altered by interventionsin the present. This early sensitivity to the causal asymmetrybetween the past and future became more pronounced with age.We also found that children and adults selectively andappropriately use evidence about the present to make inferencesabout past events. Implications for theoretical accounts of thedevelopment of causal reasoning and abstract concepts of timeare discussed.

A large body of research in the cognitive sciences relies on examining statistical differences. While the approach of examining differences can aid in explaining behavior, it does not necessarily mean that these differences have predictive power. Yet, understanding behavior both involves explaining and predicting behavior. As a point in case, the current study used a naturalistic email dataset to examine statistical differences and predictive power in fraudulent activities. Differences between 1st and 3rd person pronoun use in liars and people telling the truth are widely reported in the literature. The current study aimed to test for the effect of fraudulent events on pronoun use in emails using the Enron corpus and additionally applied a machine learning approach to estimate whether pronoun use predicts fraud. While the ratio between 1st and 3rd person pronoun use was related to fraud, this construct did not have predictive power. The current study highlights an important conclusion for the cognitive sciences: The importance of not only testing for differences, but of also applying predictive models. In this way it can be determined whether effects of a construct on an outcome can also predict the outcome.

Using a novel adaptation of the visual world eye-trackingparadigm we investigated children’s and adults’ onlineprocessing of reference in a naturalistic language context.Participants listened to a 5-minute long storybook whilewearing eye-tracking glasses. The gaze data were analyzedrelative to the onset of referring expressions (i.e., full nounphrases (NPs) and pronouns) that were mentionedthroughout the story. We found that following the mentionof a referring expression there was an increase in theproportion of looks to the intended referent for both childrenand adults. However, this effect was only found early on inthe story. As the story progressed, the likelihood thatparticipants directed their eye gaze towards the intendedreferent decreased. We also found differences in the eye gazepatterns between NPs and pronouns, as well as betweenchildren and adults. Overall these findings demonstrate thatthe mapping between linguistic input and corresponding eyemovements is heavily influenced by discourse context.

Language processing depends on the integration of bottom-upinformation with top-down cues from several differentsources—primarily our knowledge of the real world, ofdiscourse contexts, and of how language works. Previousstudies have shown that factors pertaining to both the senderand the receiver of the message affect the relative weighting ofsuch information. Here, we suggest another factor that maychange our processing strategies: perceptual noise. Wehypothesize that listeners weight different sources of top-downinformation more in situations of perceptual noise than innoise-free situations. Using a sentence-picture matchingexperiment with four forced-choice alternatives, we show thatdegrading the speech input with noise compels the listeners torely more on top-down information in processing. We discussour results in light of previous findings in the literature,highlighting the need for a unified model of spoken languagecomprehension in different ecologically valid situations,including under noisy conditions.

Expert face recognition has been marked by holistic processingand left-side bias/right hemisphere involvement. Hencerecognition for Chinese characters, sharing many visualperceptual properties with face perception, was thought toinduce stronger holistic processing and left-side bias effect.However, Hsiao & Cottrell (2009) showed that expertise inChinese character recognition involved reduced holisticprocessing, while Tso, Au & Hsiao (2014) suggested this effectmay be modulated by writing experiences; in contrast, left-sidebias was found to be a consistent expertise marker regardlessof writing experiences. Here we examine holistic processingand left-side bias effect of Chinese character recognitionbetween adolescents with and without dyslexia. Students withdyslexia were found to recognize Chinese characters with astronger holistic processing effect than the typical controls.However, compared with the controls, dyslexics showed amore reduced left-side bias in processing mirror-symmetricChinese characters. The theoretical and educationalimplications of these results were discussed.

Prior studies have reported that bilingualism enhancescognitive ability due to the regular conflict management oftwo language systems (Bialystok, 2015). Here, we explorewhether infant bilingualism improves cognitive ability at 9.5months. Twenty-four monolingual English and 23 bilingualFrench-English infants were first trained to predict a rewardon the right based on a set of tone-shape rule structure (AABpattern). Infants were later trained to predict a differentreward on the left based on another set of new rule structure(ABB pattern). Correct anticipation of reward locationsindicates successful learning. If bilingualism improvesinfants’ cognitive skills, bilingual infants would be better atlearning a new pattern-reward association. However, we didnot find evidence that bilinguals looked at the correct locationmore than monolinguals or learned the new pattern-rewardassociation faster. Thus, our results suggest bilingualism maynot enhance cognitive ability at 9.5 months, as least using thecurrent paradigm.

Humans have an intuitive understanding of physics. They canpredict how a physical scene will unfold, and reason about howit came to be. Adults may rely on such a physical representa-tion for visual reasoning and recognition, going beyond visualfeatures and capturing objects in terms of their physical prop-erties. Recently, the use of draped objects in recognition wasused to examine adult object representations in the absence ofmany common visual features. In this paper we examine youngchildren’s reasoning about draped objects in order to examinethe develop of physical object representation. In addition, weargue that a better understanding of the development of theconcept of cloth as a physical entity is worthwhile in and ofitself, as it may form a basic ontological category in intuitivephysical reasoning akin to liquids and solids. We use two ex-periments to investigate young children’s (ages 3–5) reasoningabout cloth-covered objects, and find that they perform signif-icantly above chance (though far from perfectly) indicating arepresentation of physical objects that can interact dynamicallywith the world. Children’s success and failure pattern is similaracross the two experiments, and we compare it to adult behav-ior. We find a small effect, which suggests the specific featuresthat make reasoning about certain objects more difficult maycarry into adulthood.

Despite wide variation among natural languages, there are lin-guistic properties universal to all (or nearly all) languages. Animportant challenge is to explain why these linguistic universalshold. One explanation employs a learnability argument: seman-tic universals hold because expressions that satisfy them areeasier to learn than those that do not. In an exploratory studywe investigate the relation between learnability and complexityand whether the presence of semantic universals for quantifierscan also be explained by differences in complexity. We developa novel application of (approximate) Kolmogorov complexityto measure fine-grained distinctions in complexity between dif-ferent quantifiers. Our results indicate that the monotonicityuniversal can be explained by complexity while the conserva-tivity universal cannot. For quantity we did not find a robustresult. We also found that learnability and complexity patterntogether in the monotonicity and conservativity cases that weconsider, while that pattern is less robust in the quantity cases.

Preschool children’s preference for knowledgeable agents over ignorant and inaccurate agents (Sabbagh & Baldwin, 2001; Koenig & Harris, 2005; Rakoczy et al., 2015), is generally interpreted as epistemic vigilance. However, Kushnir and Koenig (2017) recently found that without a contrasting accurate agent, preschoolers will learn new information from an agent who professed ignorance, but not from one who was inaccurate. Employing a two-speaker design contrasting an agent who professed ignorance about familiar object labels with a speaker whose knowledge state was not revealed, we found that preschoolers (N = 41; 3.50-4.89 years, M = 4.08 years) avoided requesting and endorsing novel information from the ignorant agent in the same domain as her previous ignorance (i.e., labels). In different domains, however, (i.e. novel function learning, resource sharing, etc.) they were at chance in choosing the ignorant agent. This suggests that preschoolers’ view of ignorance is situational, rather than uniformly negative.

An important question is whether speakers consider listeners’expectations when choosing whether to use a pronoun. It hasbeen suggested that certain thematic roles are more expectedto be mentioned again, and are therefore more likely to bepronominalized. In the present study, we aim to disentanglepredictability effects on pronoun use from thematic-role ef-fects. To this end, we conducted two web-based continuationexperiments in Dutch, in which the next-mention biases asso-ciated with Source-Goal and Agent-Patient verbs were ma-nipulated to create a shift in the bias. Experiment 1 confirmedthat the manipulations changed the biases. Experiment 2showed that while thematic role mainly influenced demon-strative and full pronoun use for non-subjects, next-mentionbiases played a role in the choice between reduced and fullpronouns and between pronouns and full NPs, irrespective ofthematic role or grammatical function. Thus, thematic roleand predictability seem to affect pronoun use in differentways.

This paper investigates which cognitive abilities predict theinterpretation of complex sentences by older adults.Participants performed a picture-selection task after hearingcomplex and simpler sentences, as well as a broad testbattery of cognitive tests. The results show that differentcognitive factors serve as predictors for the interpretation ofcomplex sentences compared to simpler sentences. Forcomplex sentences, verbal intelligence, cognitive flexibility,and working memory capacity are strong predictors. Ourstudy thus shows that older adults' interpretation of sentencesof varying complexity is influenced by different cognitiveabilities, and stresses the need to take such individualdifferences into account when studying language processing.

Social dilemmas conceptually suggest distinguishing direct individual and group-level effects (also involving indirect effects on others). Furthermore, the success of organizations appears to rely on identifying not only individual excellence but positive impact on others as well. In ‘Two-Level Personnel Evaluation Tasks’ (T-PETs) participants as human resource managers evaluate employees when individual and group contributions are dissociated. Von Sydow, Braus, & Hahn (2018) have suggested a potential ‘Tragedy of Personnel Evaluation’: A group-serving employee with the smallest individual contribution but by far the greatest po- sitive effect on the group’s overall earnings was often rated the most negatively. Here we investigate, in two experiments with conflicting information, whether emphasizing the group can avert the ‘tragic’ outcome. Our results suggest that the tragedy is not as complete as suggested, and that contextual information can mitigate the tragedy. Nonetheless, the results also corroborate the stability of underestimating the impact of team players.

Research on the acquisition of morphology commonly predicts that agglutinating systems should be easier to learn than fusional systems. This is argued to be due to compositional transparency: the mapping between morphemes and meanings is one-to-one in agglutinating systems, but not in fusional systems. This is supported by findings in first and second language learning (Goldschneider & DeKeyser 2001, Slobin 1973), typology (Dressler 2003, Haspelmath & Michaelis 2017), and language evolution (Brighton 2002). We present findings from a series of artificial language learning experiments which complicate this picture. First, we show that when only two features (e.g., NOUN CLASS and NUMBER) are morphologically encoded, the learnability of fusional and agglutinating systems does not differ significantly. This finding holds when learners are given an additional cue to morpheme segmentation–which in principle should make the agglutinating system easier. However, the error patterns of the two groups provide some evidence that learners might have a bias for transparent structures. Our results suggest that the advantages of agglutinating over fusional systems may be overstated, particularly when a small number of features are encoded. Since agglutinating systems likely bear additional costs (e.g., segmentation, longer word length, and the online cost of mapping between morphemes and meanings), such systems do not guarantee learning ease under all circumstances.

The purpose of these studies was to investigate the role of distributed cognition in defusing the impact of evaluative pressure caused by performance-approach goals on mental arithmetic performance. Performance-approach goals can generate worrying thoughts that can deplete working memory resources. However, some of these working memory limitations can be compensated by off-loading the internal cognitive process to the external environment. We tested this prediction in two experiments. Participants carried out modular arithmetic tasks in a performance-approach goal or mastery- approach goal condition crossed with interactivity or no interactivity. Performance-approach goal manipulation hampered cognitive performance (accuracies), (Experiment 1). However, these negative effects were defused with the help of interactivity (Experiment 2). Interestingly, the mastery-focused individuals had a performance drop in the interactive condition (Experiment 1 and Experiment 2). Finally, experiment 2 reported higher maths anxiety levels for the performance- focused individuals. Reasons for the findings and future implications will be discussed.

Human adults share the ability to approximate large quantitieswithout counting with newborn infants and non-human species.This ability is supported by the Approximate Number System(ANS) - a primitive and domain-specific cognitive system thatsupports noisy numerical decisions. How does the ANS supportactive exploratory decisions? Using a numerical comparisontask, we found that the amount of active information seekingdoes not simply increase as the decision becomes more difficult.Instead, there seems to be an inverted U-shaped relationshipbetween trial difficulty and how much one chooses to seekinformation. Additionally, this effect is not modulated byparticipants’ performance, suggesting that participants’exploratory decisions based on ANS representations are drivenby the utility of information seeking actions.

We present a novel statistical analysis of color categorizationusing a standard method from semantic typology. Our ap-proach shows that crosslinguistic color naming data exhibitslatent dimensions whose order of relative importance matchesthe evolutionary ordering of emergence of those distinctions.Moreover, we show that the importance ordering of these di-mensions holds even when controlling for frequency of the dis-tinctions by looking at languages within each stage of evolu-tion. Additionally, we find that the extreme points of the latentcolor dimensions correspond well to a small set of “univer-sal” focal colors. Thus we show that a simple mathematicalmethod simultaneously derives a consistent match both to theevolutionary stages and to the universal foci.

This study explores how the vocabulary growth trajectories of typically developing and late-talker children change in relation to their word learning biases. Forty late talkers and 44 typically developing toddlers visited the lab once a month for one year starting at about 18 months of age. Word-learning trajectories were tracked using a parent-reported vocabulary measure, and shape and material bias measures were collected using the novel noun generalization task each month. A two-level hierarchical linear model was utilized for the longitudinal analyses. Results indicate that, at the first visit, a stronger shape bias was significantly associated with a larger vocabulary in typically developing talkers. In late talkers, however, a stronger initial shape bias was associated with a smaller vocabulary. Over the course of the study, for every additional visit, stronger shape biases were associated with larger vocabularies in late talkers, but not in typically developing toddlers. Results for the material bias mirrored the shape bias results. These findings suggest different possible underlying mechanisms for the two groups of children, as well as avenues for the design of language interventions that might support young late talkers.

We contrast three views of how words contribute to a listener’s understanding of a sentence and compare corresponding quantitative models of how the listener’s probabilistic prediction on sentence completion is affected in online comprehension. The Semantic Similarity Model presupposes that the predictor of a word given a preceding discourse is their semantic similarity. The Relevance Model maintains that utterances are chosen to maximize relevance. The Bayesian Pragmatic Model assumes a relevance- guided modulation of a word’s lexical meaning that can be regarded as a Bayesian update of statistical regularities stored in memory. In addition to a Cloze test, we perform an EEG study, recording the event-related potential on the predicted word and take the N400 component to be inversely correlated with the word’s predictive probability. In a multiple regression analysis, we compare the three models with regard to Cloze values and N400 amplitudes. The Bayesian Pragmatic Model best explains the data.

Learning about novel objects not only involves noticinginformation that makes the object unique, but also what makesobjects the same. Yet, these two levels of learning involvedifferent pieces of information, meaning that learning one wellcould come at the cost of the other. Moreover, children maycategorize in a fundamentally different way, resulting in theselevels of learning interacting differently. To investigate this,we had adults and children perform a categorization taskfollowed by an item recognition test. We found that adultsshowed a trade-off, such that the ability to categorize itemscame at the cost of memory for those items. Using a subset ofmore unique lures, children’s memory trended towards a trade-off with category learning. However, this was only observedamong the older children. This suggests that adults’ efficientlearning comes at a cost, and this trade-off may start to appearin the elementary school years.

This paper examines the interaction between prior beliefs andpragmatic inferences, focusing on exhaustivity effects. Wepresent three experiments that tests how prior beliefs influenceboth interpretation and production of language, and comparethe results with the predictions of the Rational Speech Actmodel, a Bayesian model of linguistic interpretation. We findthat prior beliefs about conditional probabilities have no affecton language production, but do affect interpretation, producinganti-exhaustivity effects. We find that the RSA model achievesa relatively good fit both for the human production and inter-pretation data, but only for highly-implausible utterance costs.

The Prosodic Bootstrapping Theory (PBT) states that prosodic and phonetic cues assist infant language learners to segmentthe speech stream into words and assemble those words into phrase structures. However, many of the studies demonstratinga link between prosody and syntax were conducted on small data sets and on a narrow range of syntactic structures.This work uses a state-of-the-art parser to syntactically annotate the BU Radio News Corpus of around 16,000 diversesentences, which are prosodically tagged and annotated. A decision tree classifier was fit, using six prosodic features andachieving 87% accuracy at differentiating words internal to major syntactic phrases vs. words that mark phrase boundaries.However, the models tested are unable to differentiate between phrasal categories based on prosodic information alone.These results provide new evidence in support of the Prosodic Bootstrapping Theory, suggesting it is possible to identifyphrasal boundaries based on prosodic information alone.

Humans often rely on past experiences stored in long-termmemory to predict the outcome of an event. In traditional lab-based experiments (e.g., causal learning, probability learning,etc.), these observations are compressed into a successiveseries of learning trials. The rapid nature of this paradigmmeans that completing the task relies on working memory. Incontrast, real-world events are typically spread out over longerperiods of time, and therefore long-term memory must be used.We conducted a 24 day smartphone study to assess how wellpeople can learn causal relationships in extended timeframes.Surprisingly, we found few differences in causal learning whensubjects observed events in a traditional rapid series of 24 trialsas opposed to one trial per day for 24 days. Specifically,subjects were able to detect causality for generative andpreventive datasets and also exhibited illusory correlations inboth the short-term and long-term designs. We discusstheoretical implications of this work.

Studies of human expertise suggest that experts and novices “see“ problems differently. Experts not only acquire a bodyof domain-specific strategies and knowledge, but also learn to quickly identify when those concepts apply to problemswithin the domain. We propose modeling these elements as an iterative process of domain-specific language (DSL)learning, while jointly training a neural network to recognize when learned concepts apply to new problems. We showthat the algorithm solves problems more accurately and quickly than either a neural network alone, or a model that simplyacquires new concepts without learning when to use them. We also examine the implicit problem representations learnedby the neural network recognition model, and find that they increasingly come to reflect abstract relationships betweenproblems, rather than surface features, as the model acquires domain expertise. A full paper and additional details areavailable at: https://sites.google.com/view/neurally-guided-expertise-mit

Neuropsychological investigations with frontal patientshave revealed selective deficits in selecting the relationalanswer to pictorial analogy problems when the correctoption is embedded among foils that exhibit highsemantic or visual similarity. In contrast, normal age-matched controls solve the same problems with near-perfect accuracy regardless of whether high-similarityfoils are present (in the absence of speed pressure).Using more sensitive measures, the present study soughtto determine whether or not normal young adults aresubject to such interference. Experiment 1 used eye-tracking while participants answered multiple-choice 4-term pictorial analogies. Total looking time was longerfor semantically similar foils relative to an irrelevantfoil. Experiment 2 presented the same problems in atrue/false format with emphasis on rapid responding andfound that reaction time to correctly reject falseanalogies was greater (and errors rates higher) for thosebased on semantically or visually similar foils. Thesefindings demonstrate that healthy young adults aresensitive to both semantic and visual similarity whensolving pictorial analogy problems. Results areinterpreted in relation to neurocomputational models ofrelational processing.

Perseveration, or stickiness parameters have been added to reinforcement-learning (RL) models to capture autocorrelationin choices. Here, we systematically examined whether perseveration terms simply improve a models ability to fit noisein the data, thereby making them overly flexible. We simulated data with basic versions of a Delta and Prediction-ErrorDecay model with no perseveration terms added, and for half of the simulated data sets we added random noise to expectedRL values on each trial. We then performed cross-fitting analyses where the simulated data sets were fit by the basicdata-generating models as well as extended models with perseveration terms added. The addition of perseveration termsimproved model fit, particularly when noise was added to the simulation process. Parameter recovery was generally poorerfor the extended models. These results suggest simpler models may be more useful for prediction and generalization tonovel environments, as well as for theory development.

From social networks to public transportation, graph structuresare a ubiquitous feature of life. How do humans learn functionson graphs, where relationships are defined by the connectiv-ity structure? We adapt a Bayesian framework for functionlearning to graph structures, and propose that people performgeneralization by assuming that the observed function valuesdiffuse across the graph. We evaluate this model by askingparticipants to make predictions about passenger volume in avirtual subway network. The model captures both generaliza-tion and confidence judgments, and provides a quantitativelysuperior account relative to several heuristic models. Our worksuggests that people exploit graph structure to make general-izations about functions in complex discrete spaces.

Sentence comprehension in part involves introducing, stor-ing, and retrieving information about individuals. Natural lan-guages provide various means for performing this computa-tional work. One popular idea is that indefinite noun phrasesprovide instructions for updating the discourse model byadding a new discourse referent, while definite noun phrasespresuppose the existence of a discourse referent available inmemory, as well as instructions for retrieving it. When no an-tecedent is available, the definite’s presupposition fails to besatisfied, resulting in the so-called ‘presupposition failure’ andpragmatic infelicity. However, under certain conditions, def-inite noun phrases can felicitously be used even when no an-tecedent is available in memory. In such cases, a conversa-tional repair strategy called ‘presupposition accommodation’can rescue the discourse by adding the required referent. Itis natural to expect greater processing costs for adding a dis-course referent with a definite than with an indefinite: althoughboth involve the process of adding a referent, definites gothrough a stage of presupposition failure and a subsequent de-cision to accommodate. The experimental challenge has beento apply a method sensitive enough to detect expected costsin discourse, even when the participant is unaware of the pre-supposition failure and repairs it rapidly. The present studyaddresses this challenge by using EEG to capture temporallyfine-grained processing differences between definite and indef-inite noun phrases when both introduce new discourse refer-ents in plausible and implausible contexts. Our main findingis that definite noun phrases elicit the Left Anterior Negativ-ity (LAN) effect, compared to indefinite noun phrases, bothin implausible contexts where there is a sense of oddness andin perfectly coherent contexts. We take this as evidence of aspecific cognitive stage at which presupposition failure is de-tected and when an accommodation decision occurs. This alsosupports the idea that, when encountering a definite, the LANis tightly linked to working memory processes involving thesearch for discourse elements that are presupposed to exist inmemory. When none are found, definites are subsequently ac-commodated and bridged to other entities in the discourse.