Proceedings of the Annual Meeting of the Cognitive Science Society

This workshop is aimed at giving human interaction re-searchers the conceptual and practical apparatus to balancetheir representations of data (mixes of drawings and pho-tographs in the most part), so as to “maximally incite, but alsoconstrain” their representations, just as artists sometimes suc-ceed in doing (Streeck, Grothues, & Villanueva, 2009, p.28).Why—as Streeck points out—are the drawings and visuali-sations of interaction researchers so halting and timid, com-pared to the ways artists have responded to the same kinds ofrepresentational problems? Are these heavily segmented andsparsely constructed representations of interaction the resultof a prevailing positivistic outlook with regard to representingshared space, where interaction is presented as staggered anddiscrete physical events with apparently little to connect them.The workshop seeks to redress this situation by examining thesolutions that artists have arrived at when representing humaninteraction, and asking participants to engage in a series of ac-tivities and discussions which will re-frame their approachesto this issue.

Anthropology was a founding member of cognitive science(Bender et al., 2010; Gardner, 1985), sharing with othercognitive disciplines a deep interest in thinking and behav-ior. With its unique expertise in the cultural content, con-text, and constitution of cognition, it would still be essentialto any comprehensive endeavor to explore the human mind(Bloch, 2012), but rather has turned into cognitive science’s“missing discipline” (Boden, 2006), thus leaving importantquestions unanswered or even unasked. Given that substan-tial shares of knowledge are implicit and that cognition issituated, distributed, embodied, and grounded in variousother ways, anthropological approaches provide privilegedaccess to investigation: for arriving at reasonable hypothe-ses, ensuring ecological validity, and even for coming upwith new research questions and paradigms (Astuti &Bloch, 2012; Hutchins, 2010; Nersessian, 2006).In line with recent calls for rapprochement in Topics inCognitive Science (Bender et al., 2012; Beller & Bender,2015), our symposium brings together scholars that repre-sent different branches of contemporary anthropology withdistinct perspectives—including ‘traditional’ social anthro-pology, cognitive anthropology and ethno-linguistics, cogni-tive ecology, evolutionary anthropology, and archaeology—to present what they consider to be indispensable contribu-tions to cognitive science.With our selection of authors, we hope to demonstrate thevalue of anthropological approaches for cognitive science aswell as the potential benefits of cross-disciplinary collabora-tion. Cognitive archaeologist Overmann discusses a theo-retical perspective on how mind, behavior, and materialartifacts interact to shape human cognition. Combining theirexpertise in linguistics and evolutionary anthropology, Ráczand Jordan investigate the design principles of kinship sys-tems as near-universal conceptual tools. With his back-ground in (ethno-)linguistics and cognitive anthropology,Le Guen uses Yucatec Maya sign languages to illustrate theimportance of cultural practices for shaping cognitive be-havior. Based on Hutchins’ cognitive ecology approach,Solberg speaks to questions at the intersection of anthropol-ogy and philosophy of science by illuminating the culturalframework of science production in a biology lab. And so-cial anthropologist Astuti concludes by taking a bird’s eyeview on how efforts to understand the human mind cruciallybenefit from acknowledging its historical origins and fromtaking the specific sociocultural contexts into consideration.Based on work some of which is published in high-qualityjournals (such as Science, Nature, PNAS, BBS, TiCS, Cur-rent Anthropology, or Cognition), these participants willoffer invaluable contributions to a more diverse, more inclu-sive, and hence more comprehensive cognitive science.

Why games? How could anyone consider action gamesas experimental paradigms for Cognitive Science? In 1973,as one of three strategies he proposed for advancing Cogni-tive Science, Allen Newell exhorted us to “accept a singlecomplex task and do all of it.” More specifically, he told usthat rather than taking an “experimental psychology as usualapproach” that, we should “focus on a series of experimentaland theoretical studies around a single complex task” so as todemonstrate that our theories of human cognition were pow-erful enough to explain, “a genuine slab of human behavior”with the studies fitting into a detailed theoretical picture. Ac-tion games represent the type of experimental paradigms thatNewell was advocating and the current state of programmingexpertise and laboratory equipment, along with the emer-gence of Big Data (Griffiths, 2015) and Naturally OccurringData Sets (NODS, Goldstone & Lupyan, 2016), provide thetechnologies and data needed to realize his vision. ActionGames enable us to escape from our field’s regrettable fo-cus on novice performance to develop theories that accountfor the full range of expertise through a twin focus on ex-pertise sampling (across individuals) and longitudinal studies(within individuals) of simple and complex tasks.This Symposium is inspired by the recent Action Gamesas Experimental Paradigms for Cognitive Science (Game-XP), issue of Topics in Cognitive Science (topiCS), April2017. It includes late-breaking work from some of the re-searchers represented in that topic as well as new work bynew researchers.

Recent studies of naturalistic face-to-face communication havedemonstrated temporal coordination patterns such as thesynchronization of verbal and non-verbal behavior, which providesevidence for the proposal that verbal and non-verbalcommunicative control derives from one system. In this study, weargue that the observed relationship between verbal and non-verbalbehaviors depends on the level of analysis. In a re-analysis of acorpus of naturalistic multimodal communication (Louwerse et al.,2012), we focus on measuring the temporal patterns of specificcommunicative behaviors in terms of their burstiness. Weexamined burstiness estimates across different roles of the speakerand different communicative channels. We observed moreburstiness for verbal versus non-verbal channels, and for moreversus less informative language sub-channels. These findingsdemonstrate a new method for analyzing temporal patterns incommunicative behaviors, and they suggest a more complexrelationship between verbal and non-verbal channels thansuggested by prior studies.

In this paper we examine an enactive approach to social cog-nition, a species of radical embodied cognition typically pro-posed as an alternative to traditional cognitive science. Ac-cording to enactivists, social cognition is best explained byreference to the social unit rather than the individuals that par-ticipate in it. We identify a methodological problem in thisapproach, namely a lack of clarity with respect to the modelof explanation it adopts. We review two complaints abouta mechanistic explanatory framework, popular in traditionalcognitive science, that prevent enactivists from embracing it.We argue that these complaints are unfounded and propose aconceptual model of enactive mechanistic explanation of so-cial cognition.

When people search for a target in a novel image they oftenmake use of eye movements to bring the relatively high acuityfovea to bear on areas of interest. The strategies that controlthese eye movements for visual search have been of substantialscientific interest. In the current article we report a new com-putational model that shows how strategies for visual searchare an emergent consequence of perceptual/motor constraintsand approximately optimal strategies. The model solves a Par-tially Observable Markov Decision Process (POMDP) usingdeep Q-learning to acquire strategies that optimise the trade-off between speed and accuracy. Results are reported for theDistractor-ratio task.

Recent studies find that school-age children learn better whenthey have active control during study. Yet little is knownabout how individual differences in strategy or cognitive con-trol skills may affect active learning for preschoolers, nor ifexperimental measures of active learning map onto real-worldlearning outcomes. The current study assesses 101 low-income5-year-olds on an active category learning task, and measuresof executive function, attention, and school readiness. We findthat preschoolers use an informative sampling strategy for cat-egories defined by stimuli features in 1D and when presentedwith a distractor dimension (2D). Children accurately classifyin 1D, but show mixed performance in 2D. Attention predictssampling accuracy, and working memory and inhibitory con-trol predict classification accuracy. Performance in the activelearning task predicts early math and pre-literacy skills. Thesefindings suggest that trial-by-trial learning decisions may re-veal insight into how cognitive control skills support the ac-quisition of knowledge.

Computational models of cognitive development have beenfrequently used to model impairments found in developmentaldisorders but relatively rarely to simulate behaviouralinterventions to remediate these impairments. One area ofcontroversy in practices of intervention is whether it is betterto attempt to remediate an area of weakness or to build on thechild’s strengths. We present an artificial neural networkmodel of productive vocabulary development simulatingchildren with word-finding difficulties. We contrast anintervention to remediate weakness (additional practice onnaming) with interventions to improve strengths (improvingphonological and semantic knowledge). Remediatingweakness served to propel the system more quickly along thesame atypical trajectory, while improving strengths producedlong-term increases in final vocabulary size. A combinationyielded the best outcome. The model represents the firstmechanistic demonstration of how interventions targetingstrengths may serve to improve behavioural outcomes indevelopmental disorders. The observed effects in the modelare in line with those observed empirically for children withword-finding difficulties.

The study of memory for texts has had an long tradition of re-search in psychology. According to most general accounts oftext memory, the recognition or recall of items in a text is basedon querying a memory representation that is built up on the ba-sis of background knowledge. The objective of this paper is todescribe and thoroughly test a Bayesian model of this generalaccount. In particular, we develop a model that describes howwe use our background knowledge to form memories as a pro-cess of Bayesian inference of the statistical patterns that areinherent in a text, followed by posterior predictive inference ofthe words that are typical of those inferred patterns. This pro-vides us with precise predictions about what words will be re-membered, whether veridically or erroneously, from any giventext. We then test these predictions using data from a memoryexperiment using a relatively large sample of randomly chosentexts from a representative corpus of British English.

Over the past two-and-a-half decades, numerous empiricalstudies have demonstrated a relationship between numbersand space. A classic interpretation is that these spatial-numerical associations (SNAs) are a product of a stablemental number line (MNL) in the mind, yet others haveargued that SNAs are a product of transient mappings thatoccur in working memory. Importantly, although the latterinterpretation has no implications for the representation ofnumber, the former suggests that the representation ofnumber is inherently spatial. Here, we tease apart questionsof spatial representation (à la an MNL perspective) andspatial strategy (à la alternative accounts). In a novel place-the-number task, we demonstrate that numbersautomatically bias spatial attention whereas other ordinalsequences (i.e., letters) do not. We argue that this isevidence of an inherently spatial representation of numberand explore how this work may help answer futurequestions about the relationship between space andnumber.

The use of subject pronouns by bilingual speakers using both a pro-drop and a non-pro-drop language (e.g. Spanish heritage speakers in the USA) is a well-studied topic in research on cross-linguistic influence in language contact situations. Previous studies looking at bilinguals with different proficiency levels have yielded conflicting results on whether there is transfer from the non-pro-drop patterns to the pro-drop language. Additionally, previous research has focused on speech patterns only. In this paper, we study the two modalities of language, speech and gesture, and ask whether and how they reveal cross-linguistic influence on the use of subject pronouns in discourse. We focus on elicited narratives from heritage speakers of Turkish in the Netherlands, in both Turkish (pro- drop) and Dutch (non-pro-drop), as well as from monolingual control groups. The use of pronouns was not very common in monolingual Turkish narratives and was constrained by the pragmatic contexts, unlike in Dutch. Furthermore, Turkish pronouns were more likely to be accompanied by localized gestures than Dutch pronouns, presumably because pronouns in Turkish are pragmatically marked forms. We did not find any cross-linguistic influence in bilingual speech or gesture patterns, in line with studies (speech only) of highly proficient bilinguals. We therefore suggest that speech and gesture parallel each other not only in monolingual but also in bilingual production. Highly proficient heritage speakers who have been exposed to diverse linguistic and gestural patterns of each language from early on maintain monolingual patterns of pragmatic constraints on the use of pronouns multimodally.

Recent dual process models proposed that the strength of competing intuitions determines reasoning performance. A key challenge at this point is to search for boundary conditions; identify cases in which the strength of different intuitions will be weaker/stronger. Therefore, we ran two studies with the two-response paradigm in which people are asked to give two answers to a given reasoning problem. We adopted base-rate problems in which base rate and stereotypic information can cue conflicting intuitions. By manipulating the information presentation order, we aimed to manipulate their saliency; and by that, indirectly the activation strength of the intuitions. Contrary to our expectation, we observed that the order manipulation had opposite effects in the initial and final response stages. We explain these results by taking into account that the strength of intuitions is not constant but changes over time; they have a peak, a growth, and a decay rate.

The focus of the present work concerns the nature ofdeontological decisions. We test the hypothesis that it ispossible to specify deontological moral choices based on anunemotional rule, norm or principle and that such moralchoices can be distinguished from emotion-driven ones.Using a novel paradigm for moral choice that we call TheRefugees’ Dilemma, we provide evidence for such a rule-based route to moral choice. We show that participants withhigh scores in a Cognitive Reflection Test (CRT) were morelikely to adopt utilitarian or rule-based responses, as opposedto emotional ones. We also found that rule-based respondentsreported the highest average psychological distance, more sothat even utilitarian respondents. These findings show howemotional and rule-based influences can be separated with theappropriate scenario and challenges the approach of assumingboth influences can be combined into a single deontologicalroute in dual-process models.

Natural pedagogy emerges early in development (Knudsen &Liszkowski, 2012), but good teaching requires presenting ev-idence specific to learners’ knowledge (Shafto, Goodman, &Griffiths, 2014). How might the development of Theory ofMind (ToM) relate to the ability to select pedagogical evi-dence? We present a training study in which we investigatedthe link between preschool-aged children’s false-belief under-standing and their ability to select evidence for teaching. Ourresults suggest that children with more advanced ToM abili-ties were better evidence selectors, even when controlling foreffects of age and numerical conservation abilities. We alsofound that children who improved more in false-belief under-standing from pre- to post-test performed better on the peda-gogical tasks over the course of the training. Finally, we reporttentative evidence for a link between the pedagogical train-ing and improvements in ToM. Our findings suggest importantconnections between ToM and evidential reasoning in naturalpedagogy in early childhood.

The ability to recognize and evaluate reliable informants is acritical skill for effective social learning. Building on priorwork showing children’s sensitivity to informants who omitrelevant information, here we asked whether children’s teacherevaluations incorporate information about 1) the epistemicstate of the teacher, and 2) the amount and value of informationtaught. Preschool-aged children rated informants who taughtlearners about a novel toy with four functions; we systemat-ically varied the number and value of functions the teachersknew and taught. Our results indicate that children exoner-ated unintentional omissions of teachers who had incompleteknowledge, and provided graded ratings based on the degree ofomission. These findings are consistent with the predictions ofprior computational work, and suggest that the ability to reasonabout others’ knowledge plays an important role in children’sinferences about others’ efficacy as informants.

Over the last few decades, cognitive scientists have developed sophisticated formal models of human categorization,and computer vision researchers have achieved increasingly impressive performance in natural image classification. In thispaper, we combine the strengths of these approaches, using representations from a convolutional neural network to evaluatecognitive models of categorization against >300,000 human judgments of natural images. We find that a prototype modelperforms best overall, and that an exemplar model performs best when the network’s most abstract features are used. Altogether,our results demonstrate that the optimal categorization strategy over a set of stimuli is deeply linked to how they are represented,suggesting that any satisfying characterization of categorization behavior over naturalistic stimuli must consider it the resultof a dual process of feature learning and strategy selection. The paradigm we present herein offers one avenue to begin thisundertaking.

Patterns of crosslinguistic variation in the expression of wordmeaning are informative about semantic organization, but mostmethods to study this are labor intensive and obscure the gra-dient nature of concepts. We propose an automatic method forextracting crosslinguistic co-categorization patterns from par-allel texts, and explore the properties of the data as a potentialsource for automatically creating semantic representations forcognitive modeling. We focus on indefinite pronouns, com-paring our findings against a study based on secondary sources(Haspelmath 1997). We show that using automatic methods onparallel texts contributes to more cognitively-plausible seman-tic representations for a domain.

Across languages, people differ in which of the three basicframes of reference (FoRs) they prefer when describingspatial relations: absolute, intrinsic, or relative. But how muchvariation is there with regard to the relative FoR, which isanchored in the observer and occurs as one of three variants?Is the reflection variant canonical, as assumed by manyscholars? And how are objects in a person’s back referred to:by turning towards the objects? Results from two studies, onewith speakers of Norwegian and Farsi, the other with speakersof German and Japanese, reveal that reflection is notcanonical, but that translation and even rotation are used aswell. In addition, turning towards objects arranged in aperson’s back is very rare; what people use instead is abackward projection strategy that goes without rotation.

Theories of language acquisition and perceptual learning increasingly rely on statistical learning mechanisms. The current meta-analysis aims to clarify the robustness of this capacity in infancy within the word segmentation literature. Our analysis reveals a significant, small effect size for conceptual replications of Saffran, Aslin, & Newport (1996), and a nonsignificant effect across all studies that incorporate transitional probabilities to segment words. In both conceptual replications and the broader literature, however, statistical learning is moderated by whether stimuli are naturally produced or synthesized. These findings invite deeper questions about the complex factors that influence statistical learning, and the role of statistical learning in language acquisition.

The Cone of Direct Gaze (CoD) is described as the range ofeye gaze deviations over which an observer reports gaze asbeing directed towards them. The CoD has been found tonarrow with age across childhood (Mareschal et al. 2016). Weinvestigated whether reinforcement learning, so critical inshaping eye gaze responses in infancy, was able to accountfor the emergence of a CoD and its narrowing in childhood.To this end, we adapted Triesch et al.'s (2006) reinforcementlearning model by (1) defining a topology over objectlocations, and (2) introducing opponent non-linear rewardprofiles for looking at objects and caregivers. In Simulation 1we show that these modifications give rise to a functionalCoD in which there is reduced eye gaze following andincreased fixation on the caregiver for locations with a smallcaregiver eye gaze eccentricity. In Simulation 2 we show thatthe width of this effect reduces with learning, suggesting thatdevelopmental decreases in the CoD may be driven byreinforcement learning. In Simulation 3 we explore howchanges in model parameters can explain the CoD in highanxiety populations. Finally, the model provides one way ofunifying the developmental gaze-following and CoDliteratures, until now considered largely independent.

Preferential activation to faces in the brain’s fusiform gyrus hasled to the proposed existence of a face module termed theFusiform Face Area (FFA) (Kanwisher et. al, 1997). However,arguments for distributed, topographical object-formrepresentations in FFA and across visual cortex have beenproposed to explain data showing that FFA activation patternscontain decodable information about non-face categories(Haxby et. al, 2001; Hanson & Schmidt, 2011). Using two deepconvolutional neural network models able to perform human-level object and facial recognition, respectively, wedemonstrate that both localized category representations(LCRs) and high-level face-specific representations allow forsimilar decoding accuracy between non-preferred visualcategories as between a preferred and non-preferred category.Our results suggest that neuroimaging of a cortical “module”optimized for face processing should yield significantdecodable information for non-face categories so long asrepresentations within the module are activated by non-facestimuli.

Cognitive models have long been used to study linguistic phe-nomena spanning the domains of phonology, syntax, and se-mantics. Of these domains, semantics is somewhat unique inthat there is little clarity concerning what a model needs to beable to do in order to provide an account of how the mean-ings of complex linguistic expressions, such as sentences, areunderstood. To help address this problem, we introduce a tree-structured neural model that is trained to generate further sen-tences that follow from an input sentence. These further sen-tences chart out the “inferential role” of the input sentence,which we argue constitutes an important part of its meaning.The model is trained using the Stanford Natural Language In-ference (SNLI) dataset, and to evaluate its performance, we re-port entailment prediction accuracies on a set of test sentencesnot present in the training data. We also report the results of asimple study that compares human plausibility ratings for bothground-truth and model-generated entailments for a randomselection of sentences in this test set. Finally, we examine anumber of qualitative features of the model’s ability to gener-alize. Taken together, these analyses indicate that our modelis able to accurately account for important inferential relation-ships amongst linguistic expressions.

Economic interactions often imply to gauge the trustworthiness of others. Recent studies showed that when makingtrust decisions in economic games, people have some accuracy in detecting trustworthiness from the facial features of unknownpartners. Here we provide evidence that this face-based trustworthiness detection is a fast and intuitive process by testing itsperformance at split-second levels of exposure. Participants played a Trust game, in which they made decisions whether totrust another player based on their picture. In two studies, we manipulated the exposure time of the picture. We observed thattrustworthiness detection remained better than chance for exposure times as short as 100 ms, although it disappeared with anexposure time of 33ms. We discuss implications for ongoing debates on the use of facial inferences for social and economicdecisions.

Four experiments explored the effect of diversity of contrasting negative evidence on inductive inferences drawnfrom a single-item target. In Experiments 1 and 2, we found that increasing the diversity of a contrast set led people to inferthat a target exemplar corresponded to a higher level category and led to greater generalization of a novel property associatedwith the target. Further, we demonstrated two boundary conditions in which the effect only occurred when the contrast setwas consistent with a higher level category that both united the contrast exemplars and distinguished them from the target(Experiment 4) and when contrast and target shared an obvious parent category (Experiment 5). Taken together, these findingsdemonstrate that increasing the diversity of a contrast increases generalization from a target, but only if the contrast set is drawnfrom a single category that excludes, but shares a common parent with, the target.

Event timing and interventions are important and intertwinedcues to causal structure, yet they have typically been studiedseparately. We bring them together for the first time in an ex-periment where participants learn causal structure by perform-ing interventions in continuous time. We contrast learning inacyclic and cyclic devices, with reliable and unreliable cause–effect delays. We show that successful learners use interven-tions to structure and simplify their interactions with the de-vices and that we can capture judgment patterns with heuristicsbased on online construction and testing of a single structural

Recent studies using artificial language learning have arguedthat the cross-linguistic frequency of harmonic word orderpatterns–in which heads are ordered consistently before or af-ter dependents across syntactic categories–reflects a cognitivebias (Culbertson, Smolensky, & Legendre, 2012; Culbertson& Newport, 2015a). These studies suggest that English speak-ing adults and children favor harmonic orders of nouns anddifferent nominal modifiers (adjectives, numerals). However,because they target English learners, whose native languageis harmonic in the nominal domain (Num-Adj-N), this pref-erence may be based on transfer rather than a universal biasfor harmony. We present new evidence from French-speakingchildren, whose native language is non-harmonic in this do-main (Num-N-Adj). Our results reveal clear effects of nativelanguage transfer, but also evidence that a harmonic pattern isfavored even in this population of learners.

When young children see others fail to achieve a goal, theyspontaneously help. But there are many reasons why someonemight fail, and consequently, many ways to help. In order tohelp effectively, we need to understand why someone is fail-ing, so we can address the cause. One important distinction iswhether the failure is due to the agent’s own actions or some-thing external to her in the world. Here we show that 24- to48-month-olds can use their past experience to reason aboutthe probable cause of another person’s failure and provide helpappropriate for that cause. Children’s help targeted the worldwhen their prior knowledge suggested that the source of fail-ure was external to the agent, and targeted the person’s actionswhen this source appeared to be internal to the agent.

This study investigates whether making and observing(human) gestures facilitates learning about non-humanbiological movements and whether correspondence betweengesture and to-be-learned movement is superior to non-correspondence. Functional near-infrared spectroscopy wasused to address whether gestures activate the human mirror-neuron system (hMNS) and whether this activation mediatesthe facilitation of learning. During learning, participantsviewed the animations of the to-be-learned movements twice.Depending on the condition, the second viewing wassupplemented with either a self-gesturing instruction (Y/N)and/or a gesture video (corresponding/non-corresponding/no).Results showed that high-visuospatial-ability learners showedbetter learning outcomes with non-corresponding gestures,whereas those gestures were detrimental for low-visuospatial-ability learners. Furthermore, the activation of the inferior-parietal cortex (part of the hMNS) tended to predict betterlearning outcomes. Unexpectedly, making gestures did notinfluence learning, but cortical activation differed for learnerswho self-gestured depending on which gesture they observed.Results and implications are discussed.

Prior research indicated that information processing isinfluenced by the proximity of the hands to information:visuospatial processing is fostered near the hands, whereastextual processing might not be affected or even inhibitednear the hands. This study investigated how the proximity ofthe hands to digital information in pop-ups influenceslearning outcomes on multi-touch devices. Depending on thedistance between the information in the pop-ups and thehands of the users there were three conditions: (1) all pop-upsopened near the hands, (2) all pop-ups opened far from thehands, and (3) pop-ups with visuospatial information openednear the hands, whereas pop-ups with textual informationopened far from the hands (mixed condition). Results showedbetter learning outcomes when visuospatial pop-ups arepresented near the hands, whereas there was no difference inlearning outcomes between near and far presented textualpop-ups. Results and implications for multi-touch designs arediscussed.

While research on emotions has become one of the most pro-ductive areas at the intersection of cognitive science, artifi-cial intelligence and natural language processing, the diversityand incommensurability of emotion models seriously hampersprogress in the field. We here propose kNN regression as asimple, yet effective method for computationally mapping be-tween two major strands of emotion representations, namelydimensional and discrete emotion models. In a series of ma-chine learning experiments on data sets of textual stimuli wegather evidence that this approach reaches a human level ofreliability using a relatively small number of data points only.

Acoustic and contextual cues to linguistic categories (e.g.,phonemes or words) tend to be temporally distributed acrossthe speech signal. Optimal cue integration thus requires main-tenance of subcategorical information over time. At the sametime, previous work suggests that finite sensory memory orprocessing capacity strongly limits how much subcategoricalinformation can be maintained (or for how long). We ar-gue that previous work might have over-interpreted the roleof these limitations. In two perception experiments, we findno limit in the ability to maintain subcategorical information.We also find that maintenance seems to be the default, neitherlimited to perceptually particularly ambiguous signals, nor alearned strategy specific to our experiment. In contrast, listen-ers’ decision for how long to delay categorization, we find, is afunction of perceptual ambiguity. It is therefore crucial to dis-tinguish between in-principle abilities (even when they reflectdefault processing), and decisions made within the bounds ofthose abilities.

The estimation of duration has been shown to follow Bayesian inference, where people use their prior belief to calibrate the estimation. This explains timing biases such as the range bias where a duration is reproduced as longer when previously encountered durations were longer than shorter. However, it is unclear whether prior belief is based on previously perceived or reproduced durations. In 4 experiments, we show that the range bias occurs between short and long reproduction ranges but not between short and long perception ranges. Further analyses also show that the prior is updated by the most recent reproduced (but not perceived) duration. Together these results support a task-oriented Bayesian inference account of time reproduction, where people use the perceived duration and their past reproduction experience to make an inference about how much time to reproduce.

Previous research has shown that the sequence in whichconcepts are studied changes how well they are learned. In aseries of experiments featuring naturalistic concepts(psychology concepts) and naïve learners, we extend previousresearch by showing that the sequence of study changes therepresentation the learner creates of the study materials.Interleaved study leads to the creation of relatively interrelatedconcepts that are represented by contrast to each other andbased on discriminating properties. Blocked study, instead,leads to the creation of relatively isolated concepts that arerepresented in terms of their central and characteristicproperties. The relative benefits of these representationsdepend on whether the test of conceptual knowledge requirescontrastive or characteristic information. These results arguefor the integrated investigation of the benefits of differentsequences of study as depending on the characteristics of thestudy and testing situation as a whole.

Learning by doing refers to learning practices that involvecompleting activities as opposed to explicit learning (e.g.,reading). Although the benefits of learning by doing have beendescribed before, it is still relatively uncommon in instructionalpractice. We investigated how much students employ learningby doing in online courses, and whether it is associated withimproved learning outcomes. Spending more time completingactivities had a larger impact on learning outcomes thanspending more time reading, even in the case of mostlydeclarative content, such as in a Psychology course. Moreover,learning by doing is more efficient: grade improvements of 1standard-deviation require 10-20% less time in learning bydoing than reading. Finally, we contrast this evidence withstudents’ a priori intuitions on best study strategies for theironline course. Students overestimate the value of explicitlearning through reading, and underestimate the value of activelearning.

 We present a schema-based model of a classicneuropsychological task, the Wisconsin Card Sorting Task(WCST), where competition between motor and cognitiveschemas is resolved using a variation of a neuroanatomicallydetailed model of the basal ganglia (Gurney et al., 2001). Weshow that the model achieves a good fit with existing data atthe group level, and correctly identifies two distinct cognitivemechanisms held to underlie two distinct types of error.However, at the individual level, the correlations amongstother error types produced by the model differ from thoseobserved in the human data. To address this, we clusterparticipant performance into distinct groups and show, byfitting each group separately, how the model can account forthe empirically observed correlations between error types.Methodologically, this demonstrates the importance ofmodelling participant performance at the sub-group orindividual level, rather than modelling group performance.We also discuss implications of the model for the WCSTperformance of elderly participants and Parkinson’s patients.

In this study, results of computational simulations on Englishchild-directed speech are presented to uncover what distribu-tional properties of words make it easier to group them intolexical categories. This analysis provides evidence that wordsare easier to categorize when (i) they are hard to predict giventhe contexts they occur in; (ii) they occur in few different con-texts; and (iii) their contextual distributions have a low entropy,meaning that they tend to occur more often in one of the con-texts they occur in. This profile fits that of content words, espe-cially nouns and verbs, which is consistent with developmentalevidence showing that children learning English start by form-ing a noun and a verb category. These results further charac-terize the role of distributional information in lexical categoryacquisition and confirm that it is a robust, reliable, and devel-opmentally plausible source to learn lexical categories.

In many domains, people are able to transfer abstract knowl-edge about objects, events, or contexts that are superficiallydissimilar, enabling striking new insights and inferences. Weprovide evidence that this ability is naturally explained as theaddition of new primitive elements to a compositional mentalrepresentation, such as that in the probabilistic Language OfThought (LOT). We conducted a transfer-learning experimentin which participants learned about two sequences, one afterthe other. We show that participants’ ability to learn the secondsequence is affected the first sequence they saw. We test twoprobabilistic models to evaluate alternative theories of how al-gorithmic knowledge is transferred from the first to second se-quence: one model rationally updates the prior probability ofthe primitive operations in the LOT based on what was used inthe first sequence; the other stores previously likely hypothesesas new primitives. Both models perform better than baselinesin explaining behavior, with the human subjects appearing totransfer entire hypotheses when they can, and otherwise updat-ing the prior on primitives.

Word learning research has shown that learners constrain thehypothesis space for word meanings by using multiplesources of information, such as cross-situational regularitiesof word-context co-occurrences or syntactic cues, like thenumber of arguments. These studies typically focus on wordmeaning development where these cues can be helpful but notnecessary. As such, it sheds little light on the acquisition ofanaphors, which requires tracking syntactic dependenciesacross situations. To test whether or how learners track thisinformation, we conducted a novel anaphor learningexperiment with English and Japanese speakers, manipulatingcross-situational regularities in anaphors and their syntacticdependencies. Results show both English and Japanesespeakers closely track the frequency of interpretivepossibilities for novel anaphors. However, they demonstratedifficulties learning long-distance reflexives, which arecompatible with either local or non-local antecedents. Thissuggests that successful anaphor learning requires more thancross-situational regularities of interpretive possibilities.

The choice to enter and leave a romantic relationship can be framed as a decision-making problem based on expected utility of the partnership over time, akin to a forager deciding whether to stay in a particular patch based on the amount of resources it provides. We examined the temporal trajectory of three traits that may correspond to resources in romantic relationships—trust, love, and satisfaction—to determine whether they behave like depleting or replenishing patches from a foraging perspective. All three rise over time in intact relationships—suggesting replenishment—but plateau or fall in dissolved relationships—suggesting depletion. Survival analysis demonstrated that higher ratings of all three quality variables decreased the risk of romantic dissolution. The results suggest that these cues are lower in dissolved relationships, indicating individuals could potentially use them as cues for leaving an unsatisfactory relationship patch via aspiration-level cognitive mechanisms.

Sensitivity to the causal structure underlying phenomena iscritical to expert understanding. Fostering such understandingin learners is therefore a key goal in education. Wehypothesized that observing analogical gestures—whichrepresent relational information in visuospatial format—would lead learners to notice and reason about underlyingcausal patterns, such as positive and negative feedback.Participants watched brief video lectures about the humanbody and the plant kingdom, which were delivered along withgestures representing either: 1) visuospatial details (iconicgesture condition); or 2) relational structure (analogicalgesture condition). In a subsequent classification task, relativeto participants who saw iconic gestures, participants who sawanalogical gestures were more likely to sort the phenomenadescribed in the videos—as well as novel phenomena—bytheir causal structure (e.g., positive feedback). The resultssuggest that analogical gestures can be harnessed to fostercausal understanding.

Spatial construction—creating or copying spatialarrangements—is a hallmark of human spatial cognition.Spatial construction appears early in development, predictslater spatial and mathematical skills, and is used throughoutlife. Despite its importance, we know little about the cognitiveprocesses underlying skilled construction. Construction tasksare highly complex but analyses have tended to focus onbroad-stroke measures of end-goal accuracy. In this paper weintroduce a novel behavioral coding formalism to characterizean individual’s entire construction process, examine manyindividuals’ processes in aggregate, and summarize patternsthat emerge. The results show high consistency at certainpoints occurring throughout the construction, but also indicateflexibility in the interim paths that lead to and diverge fromthese points. Our approach offers a new method that can moreprecisely describe the behavioral patterns observed duringconstruction in order to reveal the underlying cognitiveprocesses engaged, and capture individual differences inbuilding expertise.

Recent research has identified three invariants or identities thatappear to hold in people’s probabilistic decision making: theaddition law identity, the Bayes rule identity, and the QQidentity (Costello and Watts, 2014, Fisher and Wolfe, 2014,Costello and Watts, 2016b, Wang and Busemeyer, 2013, Wanget al., 2014). Each of these identities represent specific agree-ment with the requirements of normative probability theory;strikingly, these identities seem to hold in people’s probabilityjudgments despite the presence of strong and systematic bi-ases against the requirements of normative probability theoryin those very same judgments. We assess the degree to whichtwo formal models of probabilistic reasoning (the ‘probabilitytheory plus noise’ model and the ‘quantum probability’ model)can explain these identities and biases in probabilistic reason-ing.

Olfactory imagery is disputed to exist in novices, but is reported to be easier for smell experts. It plays an importantrole in wine expertise. Previous research shows experts’ superior cognitive abilities do not transfer beyond their domain ofexpertise. This leads to two questions: do wine experts have more vivid imagery for the multisensory experience of wine? Andhow general is wine experts’ olfactory imagery? Wine experts and novices completed a questionnaire measuring the vividnessof imagery for the color, smell, and flavor of wine. In addition, all participants completed a questionnaire on general smellimagery. Wine experts were better than novices at imagining wines in all modalities, but not better at imagining smells ingeneral. Novices reported the strongest imagery for the appearance of wine, but experts showed no difference between thesenses. So mental imagery becomes more vivid with expertise; but only for imagery directly expertise related.

The inference of or-introduction, p, therefore p or q, is fundamental in classical logic and probability theory. Yet traditional research in the psychology of reasoning found that people did not endorse this inference as highly as other one- premise valid inferences. A radical response to this finding is to claim that or-introduction is in fact invalid. This response is found in the recent revision of mental model theory (MMT). We argue that this revision of the theory leads to a number of logical problems and counterintuitive consequences for valid inferences, and present an experiment extending recent studies showing that people readily accept or-introduction under probabilistic instructions. We argue for a pragmatic explanation of why the inference is sometimes considered odd. The inference is not odd when people reason from their degrees of belief.

The delay discounting perspective, which assumes an alternative-wise processing of attribute information, has long dominated research on intertemporal choice. Recent studies, however, have suggested that intertemporal choice is based on attribute-wise comparison. This line of research culminated in the tradeoff model (Scholten & Read, 2010; Scholten, Read, & Sanborn, 2014), which can accommodate most established behavioral regularities in intertemporal choice. One drawback of the tradeoff model, however, is that it is static, providing no account of the dynamic process leading to a choice. Here we develop a dynamic tradeoff model that can qualitatively account for empirical findings in intertemporal choice regarding not only choices but also response times. The dynamic model also outperforms the original, static tradeoff model when quantitatively fitting choices from representative data sets, and even outperforms the best-performing dynamic model derived from Decision Field Theory in Dai and Busemeyer (2014) when fitting both choices and response times.

Many intuitive notions of “learning” do not support thediverse kinds of learning across different situations andlearners. In this paper I offer a functional definition oflearning from a cognitive science perspective, which attemptsto account for the presence of learning in different physicalsubstrates. The definition is that a particular event should beconsidered a good example of “learning” to the degree towhich the following characteristics describe it: 1) a systemundergoes change to its informational state or processing 2)the change is for the purpose of more effective future action,3) the change is in response what the system experiences, and4) the system executes the change, rather than some outsideforce. Episodes are better examples of learning according tohow many of these characteristics they have. I discussbenefits and limitations of this characterization.

Re-representation is a crucial component of structure mappingtheory, allowing individuals to notice structural commonalitiesbetween situations that do not initially have identical relationalrepresentations. Despite its theoretical importance, however, thisconcept has been the subject of very little empirical work. In twoexperiments, we find that a case’s participation in one comparisonsystematically changes its perceived similarity to new cases, in apattern consistent with re-representation. Additional work rules outalternative explanations based on relational priming.

When semantic information is activated by a context prior to new bottom-up input (i.e. when a word is predicted), semantic processing of that incoming word is typically facilitated, attenuating the amplitude of the N400 event related potential (ERP) – a direct neural measure of semantic processing. This N400 modulation is observed even when the context is just a single semantically related “prime” word. This so-called “N400 semantic priming effect” is sensitive to the probability of seeing a related prime-target pair within experimental blocks, suggesting that participants may be adapting the strength of their predictions to the predictive validity of their broader experimental environment. We formalize this adaptation using an optimal Bayesian learner, and link this model to N400 amplitudes using an information-theoretic measure, surprisal. We found that this model could account for the N400 amplitudes evoked by words (whether related or unrelated) as adaptation unfolds across individual trials. These findings suggest that comprehenders may rationally adapt their semantic predictions to the statistical structure of their broader environment, with implications for the functional significance of the N400 component and the predictive nature of language processing.

Structured imagination refers to reliance upon prior knowledge when generating novel examples of a provided category. Yet studies supporting this tenet use experimental designs where the stimuli themselves cue exemplars based on culturally relevant items. The present study combined exemplar generation with abstract stimuli as a means of attenuating instructional bias. Participants were shown a group of abstract shapes identified as a single category and instructed to generate another member of this category. We additionally examined whether the introduction of a cognitive conflict (by including an anomalous category member) and self-explanation during generation affected the level of imaginative responses. Contrary to expectations, the presentation of a conflicting category member did not result in more imaginative responses when compared to more homogenous stimuli sets. However, a significantly greater degree of imaginative responses was observed from participants who were required to explain their thinking prior to and whilst constructing their exemplars.

Participants wore a smartphone, which collected GPS, audio,accelerometry and image data, in a pouch around their necksfor a period of two weeks. After a retention interval of oneweek, they were asked to judge the specific day on whicheach of a selection of images was taken. To account forpeople’s judgements, we proposed a mixture model of fourprocesses - uniform guessing, a signal detection process basedon decaying memory strength, a week confusion process anda event confusion process in which the sensor streams wereused to calculate the similarity of events. A model selectionexercise testing all possible subsets of the processes favoureda model that included only the event confusion model. GPSsimilarities were found to be the most significant predictors,followed by audio and accelerometry similarities and thenimage similarities.

The standard model is that word identification in reading isa holistic process, most efficient when words are centered inthe visual field. In contrast, rational models of reading predictword identification to be a constructive process, where readersefficiently gather visual information about a word, and maycombine visual information about different parts of the wordobtained across multiple fixations to identify it. We tease apartthese accounts by arguing that rational models should predictthat the most efficient place in a word to make a second fixa-tion (refixation) depends on the visual information the readerhas already obtained. We present evidence supporting this pre-diction from an eye movement corpus. Computational modelsimulations confirm that a rational model predicts this find-ing, but a model implementing holistic identification does not.These results suggest that refixations can be well understoodas rationally gathering visual information, and that word iden-tification works constructively.

We present a rational analysis of curiosity, proposing that peo-ple’s curiosity is driven by seeking stimuli that maximize theirability to make appropriate responses in the future. This per-spective offers a way to unify previous theories of curiosityinto a single framework. Experimental results confirm ourmodel’s predictions, showing how the relationship between cu-riosity and confidence can change significantly depending onthe nature of the environment.

Distributional models of semantics assume that the meaningof a given word is a function of the contexts in which itoccurs. In line with this, prior research suggests that a word’ssemantic representation can be manipulated – pushed towarda target meaning, for example – by situating that word indistributional contexts frequented by the target. Left open toquestion is the role that order plays in the distributionalconstruction of meaning. Learning occurs in time, and it canproduce asymmetric outcomes depending on the order inwhich information is presented. Discriminative learningmodels predict that systematically manipulating a word’spreceding context should more strongly influence its meaningthan should varying what follows. We find support for thishypothesis in three experiments in which we manipulatedsubjects’ contextual experience with novel and marginallyfamiliar words, while varying the locus of manipulation.

A central goal of typological research is to characterizelinguistic features in terms of both their functional role andtheir fit to social and cognitive systems. One longstandingpuzzle concerns why certain languages employ grammaticalgender. In an information theoretic analysis of German nounclassification, Dye et al. (2017) enumerated a number ofimportant processing advantages gender confers. Yet thisraises a further puzzle: If gender systems are so beneficial toprocessing, what does this mean for languages that make dowithout them? Here, we compare the communicative functionof gender marking in German (a deterministic system) to thatof prenominal adjectives in English (a probabilistic one),finding that despite their differences, both systems act toefficiently smooth information over discourse, making nounsmore equally predictable in context. We examine whyevolutionary pressures may favor one system over another,and discuss the implications for compositional accounts ofmeaning and Gricean principles of communication.

In the study of recognition memory, a mirror effect iscommonly observed for word frequency, with low frequencyitems yielding both a higher hit rate and lower false alarm ratethan high frequency items. The finding that LF itemsconsistently outperform HF items in recognition was oncetermed the “frequency paradox”, as LF items are less wellrepresented in memory. However, recognition is known to beinfluenced both by ‘context noise’—the prior contexts inwhich an item has appeared—and ‘item noise’—interferencefrom other items present within the list context. In a typicalrecognition list, HF items will suffer more interference thanLF items. To illustrate this principle, we deliberatelymanipulated both the contexts in which critical items hadbeen encountered prior to study, and the confusability oftargets and distractors. Our results suggest that when noisesources are balanced, the mirror effect disappears.

We report the results of a large-scale (N=1571) experiment toinvestigate whether spoken words can emerge from the processof repeated imitation. Participants played a version of the chil-dren’s game “Telephone”. The first generation was asked toimitate recognizable environmental sounds (e.g., glass break-ing, water splashing); subsequent generations imitated the im-itators for a total of 8 generations. We then examined whetherthe vocal imitations became more stable and word-like, re-tained a resemblance to the original sound, and became moresuitable as learned category labels. The results showed (1) theimitations became progressively more word-like, (2) even af-ter 8 generations, they could be matched above chance to theenvironmental sound that motivated them, and (3) imitationsfrom later generations were more effective as learned cate-gory labels. These results show how repeated imitation cancreate progressively more word-like forms while retaining asemblance of iconicity.

We present a connectionist model of event knowledge that istrained on examples of sequences of activities that are notexplicitly labeled as events. The model learns co-occurrencepatterns among the components of activities as they occur inthe moment (entities, actions, and contexts), and also learns topredict sequential patterns of activities. In so doing, the modeldisplays behaviors that in humans have been characterized asexemplifying inferencing of unmentioned event components,the prediction of upcoming components (which may or maynot ever happen or be mentioned), reconstructive memory,and the ability to flexibly accommodate novel variations frompreviously encountered experiences. All of these behaviorsemerge from what the model learns.

Speakers of Mandarin, Spanish, and Yucatec Maya watched videos of two actors involved in a causal chain initiated by one of them. After watching each video, participants divided 10 tokens into piles indicating their assignment of responsibility for the resulting event. There was a significant interaction between intentionality and population: causer and causee intentionality made a significant difference only for the Spanish and Yucatec participants, but not for the Chinese participants. This is in line with previous findings suggesting that internal dispositions play a lesser role in responsibility attribution in societies in which attention to individual agency is far more common than attention to group agency.

Many of our cognitive capacities are shaped by enculturation. Enculturation is the temporally extended transformative acquisition of cognitive practices such as reading, writing, and mathematics. They are embodied and normatively constrained ways to interact with epistemic resources (e.g., writing systems, number systems). Enculturation is associated with significant changes of the organization and connectivity of the brain and of the functional profiles of embodied actions and motor programs. Furthermore, it has a socio-culturally structured dimension, because it relies on cumulative cultural evolution and on the socially distributed acquisition of cognitive norms governing the engagement with epistemic resources. This paper argues that we need distinct, yet complementary levels of explanation and corresponding temporal scales. This leads to explanatory pluralism about enculturated cognition, which is the view that we need multiple perspectives and explanatory strategies to account for the complexity of enculturation.

We developed a cognitive-pharmacokinetic computational(CPC) model to understand how pharmacoactive substances,such as caffeine and toluene, modulate cognition. In this in-tegrated model, dynamic physiological mechanisms are sim-ulated to predict concentrations of the solvent toluene in thebrain, which modulates specific cognitive systems in a dose-response fashion over multiple hours. We used our CPC modelto reanalyze the results from prior research that documented anincrease in reaction time following exposure to toluene in sev-eral laboratory tasks with no change in accuracy. Our analysisprovides tentative evidence that toluene affects motor execu-tion, rather than attention or declarative memory.

People often use spatial language to talk about time, and thisis known to both reflect and shape how they think about it.Despite much research on the spatial grounding of temporallanguage and thought, little attention has been given to howspatial metaphors influence reasoning about real events,especially those in the future. In a large online study(N=2362), we framed a discussion of climate change usingspatial metaphors that varied on reference-frame (ego- vs.time-moving), speed of movement (fast vs. slow), and timehorizon (near, medium, or far future). We found thatdescribing climate change as approaching (time-movingframe) – versus something we approach – made the issueseem more serious, but also more tractable, at least when therate of motion was fast (e.g., “it’s rapidly approaching”).These findings offer novel insights into the relationshipbetween spatial metaphors and temporal reasoning and howwe communicate about uncertain future events.

Colour word learning has traditionally been viewed as a diffi-cult task. Previous accounts have focussed on infants’ ability toshow an adult-like understanding of colour terms. Here we ex-amine whether infants understand colour terms at a basic level,using two different methods: first, evidence from parental re-ports that British infants can comprehend colour terms early,second from experimental data using eye-tracking. These find-ing show that colour word learning is a process that beginsmuch earlier than previously thought, and develops slowly asinfants learn where the boundaries of each term are located.Due to their abstract properties, colour words present a uniqueopportunity to assess category learning in infants, as well asthe mechanisms that control word learning in general.

Identifying the visual referent of a spoken word – that a partic-ular insect is referred to by the word “bee” – requires both theability to process and integrate multimodal input and the abil-ity to reason under uncertainty. How do these tasks interactwith one another? We introduce a task that allows us to ex-amine how adults identify words under joint uncertainty in theauditory and visual modalities. We propose an ideal observermodel of the task which provides an optimal baseline. Modelpredictions are tested in two experiments where word recogni-tion is made under two kinds of uncertainty: category ambigu-ity and distorting noise. In both cases, the ideal observer modelexplains much of the variance in human judgments. But whenone modality had noise added to it, human perceivers system-atically preferred the unperturbed modality to a greater extentthan the ideal observer model did.

While some word meanings, like “spotted,” depend on in-tersubjectively accessible properties of the world, others like“pretty” invoke speakers’ subjective beliefs. We explored chil-dren and adults’ sensitivity to the subjectivity of a range ofadjectives, including words like “spotted” and “pretty,” butalso words like “tall,” which are evaluated relative to a stan-dard. Participants saw two speakers who had independentlyexperienced sets of exemplars of a novel object kind disagreeabout whether a critical exemplar was, e.g., “tall,” “pretty,” and“spotted.” In Experiments 1 and 3, speakers had seen distinctsets of exemplars, while in Experiments 2 and 4, the sets wereidentical. Adults always judged disagreements over words like“pretty” as faultless—indicating that both speakers “could beright”—and permitted less faultless disagreement for ones like“tall” when the speakers had experienced identical sets of ex-emplars. Strikingly, children did not respond in an adult-likemanner until age 8 or 9, but their explanations for speakers’conflicting assertions suggested some sensitivity to the kindsof knowledge relevant for evaluating different adjectives.

It has been claimed that larger semantic distance between thewords of a sentence, as quantified by a distributional seman-tics model, increases both N400 size and word-reading time.The current study shows that the reading-time effect disap-pears when word surprisal is factored out, suggesting that theearlier findings were caused by a confound between semanticdistance and surprisal. This absence of a behavioural effectof semantic distance (in the presence of a strong neurophysi-ological effect) may be due to methodological differences be-tween eye-tracking and EEG experiments, but it can also beinterpreted as evidence that eye movements are optimized forreading efficiency.

A great deal of reasoning research indicates that individualsare often biased by intuitive heuristics. However,contemporary results indicate that individuals seem sensitiveto their biases; they seem to detect conflict with reasoningnorms. One of the key remaining questions is whether thisconflict sensitivity is domain general. To address thisquestion, we administered a battery of five classical reasoningtasks to a large sample of subjects and assessed their conflictdetection efficiency on each task by measuring their responseconfidence. Results indicate that conflict detection is, in mostsenses, not domain general, though there are compellingexceptions.

We studied a simple binary prediction task and discovered that,when making predictions, humans display sequential effectssimilar to those in reaction time. Moreover, we found that thereare considerable individual differences in sequential effects inprediction, again similarly to reaction time studies. We discussour results in light of the view that sequential effects are thetrace of an attempt at detecting a pattern in the sequence, aswell as the possible influence of randomness perception in ourresults. We conclude that the same pattern detection mech-anism is likely to underlie sequential effects in reaction timeand prediction.

Much of our knowledge is transmitted socially rather thanthrough firsthand experience. Even our memories depend onrecollections of those around us. Surprisingly, when people re-call memories with others, they do not reach the potential num-ber of items they could have recalled alone. This phenomenonis called collaborative inhibition. Recently, Luhmann and Ra-jaram (2015) analyzed the dynamics of collaborative inhibitionat scale with an agent-based model, extrapolating from previ-ous small-scale laboratory experiments. We tested their modelagainst human data collected in a large-scale experiment andfound that participants demonstrate non-monotonicities not ev-ident in these predictions. We next analyzed memory transmis-sion beyond directly interacting agents by placing agents intonetworks. Contrary to model predictions, we observed highsimilarity only within directly interacting pairs. By comparingbehavior to model predictions in large-scale experiments, wereveal unexpected results that motivate future work in eluci-dating the algorithms underlying collaborative memory.

In this paper we introduce the hypothetical simulation model(HSM) of physical support. The HSM predicts that peoplejudge physical support by mentally simulating what wouldhappen if the object of interest were removed. Two experi-ments test the model by asking participants to evaluate the ex-tent to which one brick in a tower is responsible for the restof the bricks staying on a table. The results of both experi-ments show a very close correspondence between hypotheticalsimulations and responsibility judgments. We compare threeversions of the HSM which differ in how they model people’suncertainty about what would happen. Participants’ selectionsof which bricks would fall are best explained by assumingthat hypothetical interventions only lead to local changes whileleaving the rest of the scene unchanged.

Learning to read is often considered the most important skill taughtin school because reading is a gateway to other learning. Manychildren struggle to acquire this fundamental skill. Suboptimaldesign of books for beginning readers may contribute to thedifficulties children experience as close proximity between textand illustrations could promote attentional competition hamperingliteracy skills. The present work utilized eye-tracking technologyto examine how beginning readers allocate attention and whetherthese patterns are related to fluency (Experiment 1) andcomprehension (Experiment 2). Results suggest when readingbooks in which text and illustrations are in close proximity,children frequently shift attention away from the text. This patternof attention was negatively associated with fluency, but notassociated with comprehension. This line of research aims toprovide theoretical insights about design principles for readingmaterials that can be employed to optimize instructional materialsand promote literacy development in young children.

We study game theoretic decision making using a bidirectional evidence accumulation model. Our model represents both preferences for the strategies available to the decision maker, as well as beliefs regarding the opponent’s choices. Through sequential sampling and accumulation, the model is able to intelligently reason through two-player strategic games, while also generating specific violations of Nash equilibrium typically observed in these games. The main ingredients of accumulator models, stochastic sampling and dynamic accumulation, play a critical role in explaining these behavioral patterns as well as generating novel predictions.

A question-answering system needs to be able to reason aboutunobserved causes in order to answer questions of the sort thatpeople face in everyday conversations. Recent neural networkmodels that incorporate explicit memory and attention mecha-nisms have taken steps towards this capability. However, thesemodels have not been tested in scenarios for which reasoningabout the unobservable mental states of other agents is nec-essary to answer a question. We propose a new set of tasksinspired by the well-known false-belief test to examine howa recent question-answering model performs in situations thatrequire reasoning about latent mental states. We find that themodel is only successful when the training and test data bearsubstantial similarity, as it memorizes how to answer specificquestions and cannot reason about the causal relationship be-tween actions and latent mental states. We introduce an ex-tension to the model that explicitly simulates the mental rep-resentations of different participants in a reasoning task, andshow that this capacity increases the model’s performance onour theory of mind test.

Previous studies have suggested that children possess cognitiverepresentations of multi-word units (MWUs) and that MWUscan facilitate the acquisition of smaller units contained withinthem. We propose that the formation of MWU representationsprecedes and facilitates the formation of single-word represen-tations in children. Using different computational methods,we extract MWUs from two large corpora of English child-directed speech. In subsequent regression analyses, we use ageof first production of individual words as the dependent and thenumber of MWUs within which each word appears as an in-dependent variable. We find that early-learned words appearwithin many MWUs – an effect which is neither reducible tofrequency or other common co-variates, nor to the number ofcontext words contained in the MWUs. Our findings supportaccounts wherein children acquire linguistic patterns of vary-ing sizes, moving gradually from the discovery of MWUs tothe acquisition of small-grained linguistic representations.

When solving arithmetic problems, semantic factors influence the representations built (Gamo, Sander & Richard, 2010). In order to specify such interpretative processes, we created structurally isomorphic word problems that could be solved with two distinct algorithms. We tested whether a distinction between cardinal and ordinal quantities would lead solvers, due to their daily-life knowledge, to build different representations, influencing their strategies as well as the nature of their drawings. We compared 5th grade children and adults in order to assess the validity of this hypothesis with participants of varying arithmetic proficiency. The results confirmed that the distinction between cardinal and ordinal situations led to different solving strategies and to different drawings among both age groups. This study supports the ontological distinction of cardinal versus ordinal quantities and calls for the consideration of the role of daily-life semantics when accounting for arithmetic problem solving processes.

In Mandarin Chinese, the space-time word “前/qian” is used to express both the spatial concept of front/forward and the temporal concept of early/before (e.g., “前天/qian- tian”, literally front day, meaning the day before yesterday). This is consistent with the fact that Mandarin speakers can gesture to the front of the body to refer to a past event, and more generally can have past-in-front space-time mappings. In Chinese Sign Languages, however, the spatial front/forward and the temporal early/before are signed differently as the sign for spatial front is only used for the spatial concept of forward, and the sign for before/past is directed to the back. In this study we investigate whether the Mandarin sagittal spatial metaphors for time influence Chinese deaf signers’ spatio-temporal reasoning. In two experiments, we found that Chinese deaf signers with higher Mandarin proficiency were more likely to interpret the Mandarin word “前/qian” as the temporal conception of past (Study 1), and to perform past-in-front space-time mappings (Study 2) as opposed to signers with lower Mandarin proficiency. The findings of the study not only provide within-culture evidence for the influence of language on thought, but also demonstrate that even cross- modal space-time metaphors can have an impact on deaf- signers’ spatio-temporal reasoning.

Demonstratives are among the most frequent words in all languages, but demonstrative systems vary considerablybetween languages. In two experiments, we tested demonstrative use and the influence of demonstratives on spatial memory inJapanese and English – languages with purportedly very different demonstrative systems. Participants engaged in a ‘memorygame’, tapping their use of demonstratives to describe objects located on a sagittal plane (Experiment 1) and the influenceof demonstratives on memory for object location (Experiment 2). In addition to distance from speaker, the experiments alsomanipulated the position of a conspecific (next to or opposite participants). Distance and position of conspecific both affecteddemonstrative choice and memory in Japanese, with similar effects in English even though English does not explicitly encodethe position of a conspecific. We discuss possible universals underlying demonstrative systems and the influence of languageon memory.

We present a computational model that can learn event tax-onomies online from the continuous sensorimotor informationflow perceived by an agent while interacting with its environ-ment. Our model implements two fundamental learning bi-ases. First, it learns probabilistic event models as temporal sen-sorimotor forward models and event transition models, whichpredict event model transitions given particular perceptual cir-cumstances. Second, learning is based on the principle of min-imizing free energy, which is further biased towards the detec-tion of free energy transients. As a result, the algorithm formsconceptual structures that encode events and event boundaries.We show that event taxonomies can emerge when the algo-rithm is run on multiple levels of precision. Moreover, weshow that generally any type of forward model can be used,as long as it learns sufficiently fast. Finally, we show that thedeveloped structures can be used to hierarchically plan goal-directed behavior by means of active inference.

The ability to reason about the difficulty of novel tasks is criti-cal for many real-world decisions. To decide whether to tacklea task or how to divide labor across people, we must estimatethe difficulty of the goal in the absence of prior experience.Here we examine adults’ and preschoolers’ inferences aboutthe difficulty of simple block-building tasks. Exp.1 first estab-lished that building time is a useful proxy for difficulty. Exp.2asked participants to view the initial and final states of variousblock-building tasks and judge their relative difficulty. Whileadults were near-ceiling on all trials, children showed varyinglevels of performance depending on the nature of the dimen-sions that varied across structures. Exp. 3 replicated the pat-tern. These results suggest that children can reverse-engineerthe process of goal-directed actions to infer the relative diffi-culty of novel tasks, although their ability to incorporate morenuanced factors may continue to develop.

Although a common assumption in models of perceptual dis-crimination, most models of categorization do not explicitlyaccount for uncertainty in stimulus measurement. Such un-certainty may arise from inherent perceptual noise or externalmeasurement noise (e.g., a medical test that gives variable re-sults). In this paper we explore how people decide to gatherinformation from various stimulus properties when each sam-ple or measurement is noisy. The participant’s goal is to cor-rectly classify the given item. Across two experiments we findsupport for the idea that people take category structure intoaccount when selecting information for a classification deci-sion. In addition, we find some evidence that people are alsosensitive to their own perceptual uncertainty when selectinginformation.

Sequential categorization of perceptual stimuli typically shows contrast from one trial to the next. Using familiar categories of animals and faces, contrast effects were dissociated from assimilation effects. Two independent main effects were observed: contrast to the preceding stimulus, and assimilation to the previous response. It is argued that contrast and assimilation may reflect different processes in categorization.

Tony Chemero advances the radical thesis thatcognition and consciousness is actually the samething. He draws this conclusion from his understandingof cognition as an extended process. I question thisconclusion because this view expands cognition beyondbeing the sort of natural kind to which one can tiephenomenal experience. Moreover, because cognition hasbeen radically inflated, despite Chemero’s claim to thecontrary, embodied cognition does not solve any of thehard problems associated with consciousness.

What cognitive mechanisms support the emergence of linguis-tic conventions from repeated interaction? We present re-sults from a large-scale, multi-player replication of the clas-sic tangrams task, focusing on three foundational propertiesof conventions: arbitrariness, stability, and reduction of ut-terance length over time. These results motivate a theory ofconvention-formation where agents, though initially uncertainabout word meanings in context, assume others are using lan-guage with such knowledge. Thus, agents may learn aboutmeanings by reasoning about a knowledgeable, informativepartner; if all agents engage in such a process, they success-fully coordinate their beliefs, giving rise to a conventionalcommunication system. We formalize this theory in a compu-tational model of language understanding as social inferenceand demonstrate that it produces all three properties in a sim-plified domain.

We outline and test a Bayesian model of the effects of evidencesampling on property induction. Our model assumes thatpeople are sensitive to the effects of different sampling framesapplied to sampled evidence. Two studies tested the model bycomparing property generalization following exposure tosamples selected because they belong to the same taxonomiccategory or because they share a salient property. Both studiesfound that category-based sampling led to broadergeneralization than property-based sampling. In line withmodel predictions, these differences were attenuated when amixture of positive and negative evidence was presented(Experiment 1) and when category-property relations wereprobabilistic rather than deterministic (Experiment 2).

Dating back as far as 100 ka, the Blombos ochre and the Diepkloof ostrich egg engravings are considered among theearliest fossilized evidence of human symbolic behavior. Of special interest to this study is the temporal trajectory spanningmore than 30 thousand years from earlier simpler parallel line patterns to later complex cross-hatchings suggesting adaptivecompositional development. Through a series of three psychophysical experiments we test the hypotheses that the line engrav-ings at each site evolved to become 1) more salient to the human perceptual system, 2) more discriminable from each other, and3) increasingly associated with symbolic intent. Our findings suggest that just as instrumental tools have been found to undergocumulative refinements in adaptation to their function, the ochre and egg shell engravings evolved adaptively to become morefit for their cognitive function as signs.

The ability to monitor epistemic uncertainty is critical for self-directed learning. However, we still know little about youngchildren’s ability to detect uncertainty in their mental repre-sentations. Here we asked whether a spontaneous informationgathering behavior – social referencing – is driven by uncer-tainty during early childhood. Children ages 2-5 completed aword-learning task in which they were presented with one ortwo objects, heard a label, and were asked to put the labeledobject in a bucket. Referential ambiguity was manipulatedthrough the number of objects present and their familiarity. InExperiment 1, when there were two novel objects and a novellabel, the referent was ambiguous; when there were two famil-iar objects, or only one novel or familiar object, the referentwas known or could be inferred. In Experiment 2, there wereeither two novel objects, two familiar objects, or one familiarand one novel object; in the latter case the referent could be in-ferred by excluding the familiar object. To further manipulatethe availability of referential cues, the experimenter gazed ateither the target or the center of the table while labeling the ob-ject. In both experiments, children looked at the experimentermore often while making their response when the referent wasambiguous. In Experiment 2, children also looked at the ex-perimenter more when there was one familiar and one novelobject, but only when the experimenter’s gaze during label-ing was uninformative. These results suggest that children’ssocial referencing is a sensitive index of graded epistemic un-certainty.

We explore how adults learn counterintuitive causal relation-ships, and whether they discover hypotheses by revising theirbeliefs incrementally. We examined how adults learned a noveland unusual causal rule when presented with data that initiallyappeared to conform to a simpler, more salient rule. Adultswatched a video of several blocks placed sequentially on ablicket detector, and were then asked to determine the under-lying causal structure. In the near condition the true rule wascomplex, but could be found by making incremental improve-ments to the simple and salient initial hypothesis. The distantcondition was governed by a simpler rule, but to adopt that ruleparticipants had to set aside their initial beliefs, rather thanrevising them incrementally. Adults performed better in thenear condition, despite this rule being more complex, provid-ing some of the first evidence for an explore-exploit trade-offin inference, analogous to the trade-off in active learning.

Complex uncertainty expressions such as probably likely andcertainly possible naturally occur in everyday conversations.However, they received much less attention in the literaturethan simple ones. We propose a probabilistic model of the useand interpretation of complex uncertainty expressions based onthe assumption that their predominant function is to communi-cate factual information about the world, and that further layersof uncertainty are pragmatically inferred. We collected empir-ical data on the use and interpretation of these expressions anduse it for detailed model criticism.

Categorization has a large impact on how people perceive theworld, especially when used to make inferences about uncer-tain features of new objects. While making these inferences,people tend to draw information from only one possible cate-gorization of a new object; in addition, people are sensitive topre-existing correlations between features. Here, we explainthese trends of feature inference using a priming-based cogni-tive process model, and show that our model is distinguishedin that it can explain not only these two main trends, but alsocases where people seem to reverse the first trend and base in-ferences on information from multiple categories.

Past theoretical studies on word learning have offeredsimple sampling models as a means of explaining realword learning, with a particular goal of addressing thespeed of word learning: people learn tens of thousandsof words within their first 18 years. The present studyrevisits past theoretical claims by considering a more re-alistic word frequency distribution in which a large num-ber of words are sampled with extremely small probabil-ities (e.g., according to Zipf’s law). Our new mathemati-cal analysis of a recently-proposed simple learning modelsuggests that the model is unable to account for wordlearning in feasible time when the distribution of wordfrequency is Zipfian (i.e., power-law distributed). Toameliorate the difficulty of learning real-world word fre-quency distributions, we consider a type of active, self-directed learning in which the learner can influence theconstruction of contexts from which they learn words.We show that active learners who choose optimal learn-ing situations can learn words hundreds of times fasterthan passive learners faced with randomly-sampled situ-ations. Thus, in agreement with past empirical studies,we find theoretical support for the idea that statisticalstructure in real-world situations–potentially structuredfor learning by both a self-directed learner, and by abeneficent teacher–is a potential remedy for the patho-logical case of learning words with Zipf-distributed fre-quency.

Three experiments investigated the frame of reference used in memory to represent vertical spatial layouts perceiv-able from a single viewpoint. We tested for the selection of three different reference systems: the body orientation, the visualvertical of the surrounding room, and the direction of gravity. Participants learned and retrieved differently colored objects ona vertical board with body and room orientations varying in relation to gravity and each other systematically. Across all threeexperiments participants were quicker or more accurate in memory recall when they saw the vertical spatial layout in the sameorientation in relation to their body vertical as during learning, irrespective of the direction of gravity or visual room upright.These results indicate that spatial long-term memories for small-scale vertical relations are mainly defined in an egocentricreference system with respect to the body vertical despite the availability of alternative highly salient allocentric referencedirections.

When teaching, people often intentionally intervene on alearner while it is acting. For instance, a dog owner mightmove the dog so it eats out of the right bowl, or a coach mightintervene while a tennis player is practicing to teach a skill.How do people teach by intervention? And how do thesestrategies interact with learning mechanisms? Here, we ex-amine one global and two local strategies: working backwardsfrom the end-goal of a task (backwards chaining), placing alearner in a previous state when an incorrect action was taken(undoing), or placing a learner in the state they would be in ifthey had taken the correct action (correcting). Depending onhow the learner interprets an intervention, different teachingstrategies result in better learning. We also examine how peo-ple teach by intervention in an interactive experiment and finda bias for using local strategies like undoing.

SOB-CS is an interference-based computational model ofworking memory that explains findings from simple and com-plex span experiments. According to the model’s mechanismof interference by superposition, high similarity between mem-ory items and subsequently processed distractors is beneficialbecause the more a distractor is similar to an item, the morethey share similar units, leading to less distortion of the mem-ory item. When time allows, SOB-CS removes interfering dis-tractors from memory by unbinding them from their context.The combination of these two mechanisms leads to the predic-tion that when free time is long enough to remove the distrac-tors entirely, similarity between items and distractors shouldno longer be beneficial to memory performance. The aim ofthe present study was to test this prediction. Adult participantsperformed a complex-span task in which the free time follow-ing each distractor and the similarity between items and dis-tractors were varied. As predicted by the model, we observeda positive effect of the similarity between items and distrac-tors, and a negative effect of pace on the mean working mem-ory performance. However, we did not observe the predictedinteraction. An analysis of the errors produced during recallshowed that longer free time reduced the tendency of distrac-tors to intrude in recall much less than the model predicted.The SOB-CS model accounted well for the data after a sub-stantial reduction of the removal-rate parameter.

The current study investigated the relationship betweenchildren’s spatial ability and their scientific knowledge, skills andunderstanding. Children aged 7-11 years (N=123) completed abattery of five spatial tasks, based on a model of spatial ability inwhich skills fall along two dimensions: intrinsic-extrinsic; static-dynamic. Participants also answered science questions fromstandardised assessments, grouped into conceptual topic areas.Spatial scaling (extrinsic static spatial ability) and mental folding(intrinsic dynamic spatial ability) each emerged as predictors oftotal science scores, with mental folding accounting for morevariance than spatial scaling. Mental folding predicted bothphysics and biology scores, whereas spatial scaling accounted foradditional variance only in biology scores. The embeddedfigures task (intrinsic static spatial ability) predicted chemistryscores. The pattern was consistent across the age range. Thesefindings provide novel evidence for the differential role ofdistinct aspects of spatial ability in relation to children’s scienceperformance.

When people move together, as they dance, march or flirt, itincreases affiliation between them. But what about ‘movingtogether’ produces affiliation: the movements themselves, orthe social context of moving ‘together’? We instructed pairs ofparticipants to listen to music and move their arms or legsaccording to shapes appearing on screen. They either carriedout the same movements, or when one moved their arms theother moved their legs. They either saw shapes on one laptop,or each had their own laptop. Surprisingly, participants did notlike each other more if they carried out the same movements,but affiliation did increase if they danced looking at the samescreen. Rather than their movements, instructions, intentionsor perceptual experiences, here it is the social context of theactions that produces affiliation, a surprising finding that is noteasily accounted for by the dominant theories of mimicry andbehavioural synchrony.

In order to successfully learn the meaning of words such asbin or pin, language learners must not only perceive relevantdifferences in the speech signal but also learn mappings fromwords to referents. Prior work in native (Stager & Werker,1997) and second (Pajak, Creel, & Levy, 2016) language ac-quisition has found that the ability to perceptually discrimi-nate between words does not guarantee successful word learn-ing. Learners fail to utilize knowledge that they can otherwiseuse in speech perception. To explore possible mechanisms ac-counting for this phenomenon, we developed a probabilisticmodel that infers both a word’s phonetic form and its asso-ciated referent. By analyzing different versions of the modelfitted to experimental results from Pajak et al. (2016), we ar-gue that a mechanism for spoken word learning needs to incor-porate both perceptual uncertainty as well as additional, task-specific sources of uncertainty.

The words of different languages partition the world in strikingly different ways. Yet many people are unaware ofsuch differences, believing that some of the words of their native language pick out discrete categories based in nature. Weinvestigated whether knowledge of cross-linguistic semantic diversity—putatively inherent to bilingualism—can reduce suchessentialist beliefs. In three experiments, we found (a) that bilinguals were less likely than monolinguals to judge membershipfor animal categories in essentialist terms, (b) that explicit exposure to cross-linguistic semantic diversity, independent ofbilingualism, yielded similar effects, and (c) that this manipulation reduced essentialist beliefs about social categories as well.Together, our findings suggest that learning about how languages differ in their semantic systems—a form of metalinguisticknowledge—can lead people to think about categories more flexibly. Implications for research on language and thought, andfor ameliorating the negative consequences of social essentialism, are discussed.

We examined decisions based on verbal probability phrases,such as "small chance," "likely," or "doubtful" (we call thesephrases verbal probabilities). Verbal probabilities havecommunicative functions called directionality and can becategorized into positive (e.g., "likely" or "probable") ornegative (e.g., “unlikely,” “doubtful”) phrases in terms oftheir directionality. Previous studies have shown that thedirectionality of phrases affects decisions. Although suchdecisions seem biased, we argue that they are not. Wehypothesize that since a speaker has the option to choose thedirectionality used during communication, the selecteddirectionality becomes relevant information to a decisionmaker, and is taken into account in making decisions. Wemodeled these processes using the Decision by BeliefSampling (DbBS) model. We found that the observed datacould be well explained by our hypothesis, and that the DbBSmodel could be one of the best potential models for decisionsbased on verbal probability information.

Children who read more tend to be better readers than children who read less. Reading exposure captures not onlythe number of times words are experienced but also the breadth of the contexts words appear in. Using a large children’s corpusof written language, we quantified the former as word frequency and the latter as Semantic Diversity (SemD) (Hoffman etal., 2013). SemD was indexed using Latent Semantic Analysis to calculate the degree of semantic dissimilarity between thecontexts in which each appeared. We selected 300 words that varied in SemD for a visual lexical decision and naming task with9-year-old children (N=114). Results showed that both frequency and SemD were associated with performance, independentlyaccounting for variation in speed and accuracy. Those words high in frequency and high in SemD were read more efficiently.These findings show that factors beyond frequency are important in determining children’s word reading.

Attempts to connect individual differences in statisticallearning with broader aspects of cognition have receivedconsiderable attention, but have yielded mixed results. Apossible explanation is that statistical learning is typicallytested using the two-alternative forced choice (2AFC) task.As a meta-cognitive task relying on explicit familiarityjudgments, 2AFC may not accurately capture implicitlyformed statistical computations. In this paper, we adapt theclassic serial-recall memory paradigm to implicitly teststatistical learning in a statistically-induced chunking recall(SICR) task. We hypothesized that artificial languageexposure would lead subjects to chunk recurring statisticalpatterns, facilitating recall of words from the input.Experiment 1 demonstrates that SICR offers more fine-grained insights into individual differences in statisticallearning than 2AFC. Experiment 2 shows that SICR hashigher test-retest reliability than that reported for 2AFC. Thus,SICR offers a more sensitive measure of individualdifferences, suggesting that basic chunking abilities mayexplain statistical learning.

When we reason about others’ behavior, there are often manyequally-plausible explanations. If Bob climbs a tree to get anapple, we may be unsure if Bob found climbing difficult butreally wanted an apple; if he found climbing easy and was notparticularly excited about the apple; or if he found climbingintrinsically fun and just got the apple because it was conve-nient. Past research suggests that we solve this problem byobtaining repeated observations about the agent and about theworld. Here we argue that, beyond allowing us to sharpen ourinferences about agents and the world, covariation data alsoenables us to do one-shot inferences about novel agents. Weshow that given minimal covariation data, people can infer ob-jective and subjective properties of a new agent from a singleevent. We show that a model that assumes that agents maxi-mize utilities matches participant judgments with quantitativeprecision.

The development of the ubiquitous logical connectives and andor provides a window into the role of semantics and pragmat-ics in children’s linguistic development. Previous research hassuggested that adults and children might differ in their interpre-tation of or in two ways. First, unlike adults, children mightinterpret or as logical conjunction, akin to and. Second, chil-dren might interpret or as inclusive disjunction while adultsinterpret it as exclusive. We report experimental studies thatprobe interpretations of and and or in adults and children us-ing truth value judgements as well as children’s spontaneouslinguistic feedback. Both truth judgements and linguistic feed-back showed that four-year-olds do not interpret or as and.While children’s truth judgments suggested that they did notderive exclusivity implicatures, however, their corrective feed-back showed signs of sensitivity to the implicature, suggestingthat the truth value judgement task could have underestimatedchildren’s pragmatic competence. More generally, four-year-olds’ interpretation of logical connectives may not be as differ-ent from adults as previously supposed.

Research into risk communication has commonly highlighted the disparity between the meaning intended by the communicator and what is understood by the recipient. Such miscommunications will have implications for perceived trust and expertise of the communicator, but it is not known whether this differs according to the communication format. We examined the effect of using verbal, numerical and mixed communication formats on perceptions of credibility and correctness, as well as whether they influenced a decision to evacuate, both before and after an ‘erroneous’ prediction (i.e. an ‘unlikely’ event occurs, or a ‘likely’ event does not occur). We observed no effect of communication format on any of the measures pre-outcome, but found the numerical format was perceived as less incorrect, as well as more credible than the other formats after an ‘erroneous’ prediction, but only when low probability expressions were used. Our findings suggest numbers should be used in consequential risk communications.

The problem of stability has long been a limiting factor in de-veloping neural networks that can grow in size and complex-ity. Outside of particular, narrow parameter ranges, changesin activity can easily result in total loss of control. Humancognition must have reliable means of acting to stay withinthe stable ranges of sensitivity and activation. Learning is onesuch mechanism, and population dynamics are another. Here,we focus on another, often overlooked stability mechanism:cellular homeostasis through metabolism dynamics. We ran avisual change detection experiment designed to strain networkstability while minimizing any learnable patterns. We fit thedata using models with and without cellular energy levels as afactor, finding that the model influenced by its past history ofenergy use was a closer fit to the human data.

Bounded and unbounded events differ in whether they include an inherent endpoint (Bach, 1986). Even though this distinction can be important for the way events are identified and processed, the literature on event cognition has not focused on such abstract aspects of event structure. In the present study, we asked whether viewers are sensitive to the distinction between bounded and unbounded events in a category learning task. Our results show that people were more successful in forming the category of bounded events than that of unbounded events. We discuss implications of this finding for event cognition.

Deep understanding of mathematical equivalence is criticalfor later mathematical understandings. However, researchstudies and national test results have repeatedly demonstratedthat many students fail to develop adequate understanding ofequivalence. Recent work from McNeil and colleaguesproposes that this failure is partly due to the format oftraditional instruction and practice with highly similarproblems. Specifically, the change-resistance account(McNeil & Alibali, 2005) proposes that students struggle withequivalence because they have developed overgeneralized“rules” that affect how they process and approach mathproblems, (e.g., the operators are always on the left side, theequal sign means to “do something” or “give the answer”)and fail to see equations having two separate sides that arebeing related to one another. Extensive practice withproblems in a similar format (e.g., those that present allarithmetic operations on the left side of the equal sign)encourages students to develop ineffective mental models ofproblem types. We replicate and extend prior work that bringscognitive science research to the classroom. Our findingsindicate that applying research-based design principles toarithmetic practice improves student understanding ofmathematical equivalence enough to support transfer to novelproblem types.

People often prefer simpler explanations because they havehigher prior probability. However, simpler explanations arenot always normatively superior because they often do notfit the data as well as complex explanations. How dopeople negotiate this trade-off between prior probability(favoring simplicity) and goodness-of-fit (favoringcomplexity)? Here, we argue that people use opponentheuristics—relying on simplicity as a cue to priorprobability but complexity as a cue to goodness-of-fit(Study 1). We also examine factors that lead one or theother heuristic to predominate in a given context. Study 2finds that people have a stronger simplicity preference indeterministic rather than stochastic contexts, while Study 3finds that people have a stronger simplicity preference forphysical rather than social causal systems. Together, weargue that these cues and contextual moderators act aspowerful constraints that help to specify otherwise ill-defined hypothesis comparison problems.

Mathematics is critical for making sense of the world. Yet,little is known about how people evaluate mathematicalexplanations. Here, we use an explanatory reasoning taskto investigate the intuitive structure of mathematics. Weshow that people evaluate arithmetic explanations bybuilding mental proofs over the conceptual structure ofintuitive arithmetic, evaluating those proofs using criteriasimilar to those of professional mathematicians.Specifically, we find that people prefer explanationsconsistent with the conceptual order of the operations(“9÷3=3 because 3 ́3=9” rather than “3 ́3=9 because9÷3=3”), and corresponding to simpler proofs (“9÷3=3because 3 ́3=9” rather than “9÷3=3 because 3+3+3=9”).Implications for mathematics cognition and education arediscussed.

People use a variety of strategies for evaluating causalclaims, including mechanistic strategies (seeking a step-by-step explanation for how a cause would bring about itseffect) and statistical strategies (examining patterns of co-occurrence). Two studies examine factors leading one orthe other of these strategies to predominate. First, generalcausal claims (e.g., “Smoking causes cancer”) areevaluated predominantly using statistical evidence,whereas statistics is less preferred for specific claims (e.g.,“Smoking caused Jack’s cancer”). Second, social andbiological causal claims are evaluated primarily throughstatistical evidence, whereas statistical evidence is deemedless relevant for evaluating physical causal claims. Weargue for a pluralistic view of causal learning on which amultiplicity of causal concepts lead to distinct strategies forlearning about causation.

Typing on a smartphone is an everyday activity that involves various cognitive and behavioural processes. This papermodels touch screen text entry as cognitively bounded rationality. The model aims to maximise error-free text throughput, whilebeing constrained by its architecture and task environment. Empirical data are used to calibrate the model, which demonstratesadequate fit. The model is used to explore how strategic choices under given constraints affect text entry performance. Thepreliminary model presented here serves as a confirmation that touch screen text entry can be modelled as cognitively boundedrationality. Future extensions by integration into richer cognitive architectures are outlined.

Much research effort has been expended improving police lineup procedures used in collecting eyewitness identification evidence. Sequential presentation of lineup members, in contrast to simultaneous presentation, has been posited to increase witness accuracy, though analyses based in Signal Detection Theory (SDT) have challenged these claims. A possible way to clarify the effect of presentation format on witness accuracy is to develop SDT-based measurement models, which characterise decision performance in terms of psychologically-relevant parameters, particularly discrimin- ability and response bias. A model of the sequential lineup task was developed with a “first-above-criterion” decision rule, alongside a simultaneous model with a “maximum familiarity” decision rule. These models were fit to a corpus of data comparing simultaneous and sequential lineup performance. Results showed no difference in discriminability between the procedures and more conservative responding for the sequential lineup. Future work will examine criterion setting in the sequential lineup and model alternative decision rules.

The semantic fluency task has been used to understand the ef-fects of semantic relationships on human memory search. Avariety of computational models have been proposed that ex-plain human behavioral data, yet it remains unclear how mil-lions of spiking neurons work in unison to realize the cogni-tive processes involved in memory search. In this paper, wepresent a biologically constrained neural network model thatperforms the task in a fashion similar to humans. The modelreproduces experimentally observed response timing effects,as well as similarity trends within and across semantic cate-gories derived from responses. Three different sources of theassociation data have been tested by embedding associationsin neural connections, with free association norms providingthe best match.

Serial stay-or-search problems are ubiquitous across manydomains, including employment, internet search, mate search,and animal foraging. For instance, in patch foragingproblems, animals must decide whether to stick with adepleting reward vs search for a new source. The optimalstrategy in patch foraging problems, described by theMarginal Value Theorem (MVT; Charnov, 1976), is to leavethe depleting patch when the local reward rate within a patchmatches the overall long-run reward rate. Many species ofanimals, ranging from birds to rodents, monkeys, andhumans, adhere to this policy in important respects, but tendto overharvest, or stick with the depleting resource too long.Here we attempt to determine the cognitive biases thatunderlie overharvesting in one of these species (the rat). Wecharacterized rat behavior in response to two basicmanipulations in patch foraging tasks: to travel time betweenpatches and depletion rate, and two novel manipulations tothe foraging environment: the size of reward and length ofdelays, and placement of delays (pre- vs. post-reward). Inresponse to the basic manipulations, rats qualitativelyfollowed predictions of MVT, but stayed in patches for longerthan is predicted. In the latter two manipulations, rats deviatedfrom predictions of MVT, exhibiting changes in behavior notpredicted by MVT. We formally tested whether four separatecognitive biases – subjective costs, decreasing marginal utilityfor reward discounting of future reward, and ignoring post-reward delays – could explain overharvesting in the formertwo manipulations and deviations from MVT in the latter two.All of the biases tested explained overharvesting behavior inthe former contexts, but only one bias – in which rats ignorepost-reward delays – also explained deviations from MVTdue to larger rewards with longer delays and due tointroduction of a pre-reward delay. Our results show thatmultiple biases can explain certain aspects of overharvestingbehavior, and, while foraging behavior may be the result ofthe use of multiple biases, inaccurate estimation of post-reward delays likely contributes to overharvesting.

Zipf (1935) observed that word length is inversely proportionalto word frequency in the lexicon. He hypothesised that thiscross-linguistically universal feature was due to the Principleof Least Effort: language-users align form-meaning mappingsin such a way that the lexicon is optimally coded for efficientinformation transfer. However, word frequency is not the onlyreliable predictor of word length: Piantadosi, Tily, and Gib-son (2011) show that a word’s predictability in context is infact more strongly correlated with word length than word fre-quency. Here, we present an artificial language learning studyaimed at investigating the mechanisms that could give rise tosuch a distribution at the level of the lexicon. We find thatparticipants are more likely to use an ambiguous short form inpredictive contexts, and distinct long forms in surprising con-texts, only when they are subject to the competing pressures tocommunicate accurately and efficiently. These results supportthe hypothesis that language-users are driven by a least-effortprinciple to restructure their input in order to align word lengthwith information content, and this mechanism could thereforeexplain the global pattern observed at the level of the lexicon.

A randomized-controlled study compared the preparatory effects of problem-posing on learning from subsequentinstruction. Students engaged in problem-posing either with solution generation (where they generated problems and solutionsto a novel situation) or problem-posing without solution generation (where they generated only problems) prior to learning anovel math concept. Problem-posing with solution generation prior to instruction resulted in significantly better conceptualknowledge, without any significant difference in procedural knowledge and transfer. These findings suggest that althoughsolution generation prior to instruction plays a critical role in the development of conceptual understanding, and generatingproblems can further enhance transfer.

Collaborative spatial problem solving is an important yet not thoroughly examined task. Participants navigatedindividually and in dyads through virtual cities of varying complexity. They only saw the environment part visible from theircurrent location from a bird’s eye view map perspective. We recorded missed target locations, overall trajectory length andsearch time per person until self-indicating whole coverage. Our results show a general increase in missed locations, trajectorylength, and search time with the complexity of the environment. These increases differed due to individual and collaborativesearch. For complex, but not for simple environments individual participants navigated shorter distances, finished earlier, butalso missed more target locations than when searching the same environments in collaboration. These results indicate that incomplex environments collaborative search is less error prone than individual search, but takes longer. Such initial findings willconstrain future theorizing about collaborative spatial problem solving.

Experimental studies of the cultural evolution of languagehave focused on how constraints on learning andcommunication drive emergence of linguistic structure. Yetlanguage is typically transmitted by experts who adjust theinput in ways that facilitates learning by novices, e.g. throughchild-directed speech. Using iterated language learning ofbinary auditory sequences, we explored how language changeis affected by experts’ intention to teach the language tonovices. Comparison between teaching chains and simpletransmission chains revealed that teaching was associatedwith a greater rate of innovation which led to emergence ofmore expressive languages consisting of shorter signals. Thisis the first study to show that during cultural transmission,teaching can modify, and potentially optimise, functionalcharacteristics of language.

The generative capacity of language entails an ability to flexibly combine concepts with each other. Conceptual combination can occur either by using an attribute of one concept to describe another (attributive combination) or by forming some relation between two concepts to create a new one (relational combination). Prior research has addressed whether common or distinct processes support these two putatively different types of combinations. We turn the question around and ask whether the consequences of these combination types on our conceptual system might differ, by comparing semantic memory networks before and after participants perform either attributive or relational conceptual combinations. We find a general effect on the semantic networks: the structure of network decreases after participants conceptually combine some of the concepts in the network. However, the relational combination manipulation has a greater effect. Furthermore, only the relational combination manipulation leads to an increase in the network’s connectivity.

People reason about possibilities routinely, and reasoners caninfer “modal” conclusions, i.e., conclusions that concern whatis possible or necessary, from premises that make no mentionof modality. For instance, given that Cullen was born in NewYork or Kentucky, it is intuitive to infer that it’s possible thatCullen was born in New York, and a recent set of studies onmodal reasoning bear out these intuitions (Hinterecker,Knauff, & Johnson-Laird, 2016). What explains the tendencyto make modal inferences? Conventional logic does not applyto modal reasoning, and so logicians invented manyalternative systems of modal logic to capture valid modalinferences. But, none of those systems can explain theinference above. We posit a novel theory based on the ideathat reasoners build mental models, i.e., iconic simulations ofpossibilities, when they reason about sentential connectivessuch as and, if, and or (Johnson-Laird, 2006). The theoryposits that reasoners represent a set of conjunctivepossibilities to capture the meanings of compound assertions.It is implemented in a new computational process model ofsentential reasoning that can draw modal conclusions fromnon-modal premises. We describe the theory andcomputational model, and show how its performance matchesreasoners’ inferences in two studies by Hinterecker et al.(2016). We conclude by discussing the model-based theory inlight of alternative accounts of reasoning.

Despite a heated debate regarding a cognitive mechanism of magnitude representation, little has been done to di-rectly compare numerical and non-numerical estimation and provide a unified account of the two processes. In the current study,we examined estimation of numerical and non-numerical quantities on a continuum using various psychophysical functions.Inconsistent with the proportion reasoning and measurement skills accounts, estimates of both numerical and non-numericalquantities were better predicted by the logarithmic-linear model than by cyclic power models. Furthermore, individual dif-ferences in the degree of logarithmic compression was highly correlated over tasks, whereas bias measures from competingmodels did not show such associations. These findings suggest that estimation of both numerical and non-numerical magnitudeis processed via shared representation systems that are logarithmically or linearly constructed.

Selective sustained attention (SSA) is an importantcognitive process that enables everyday functioning and taskperformance by allowing us to: 1) choose components of ourenvironment to process at the exclusion of others and 2)maintain focus on those components over time. AlthoughSSA is known to undergo rapid and marked changes duringthe preschool and early primary school children years, therehas been a paucity of behavioral data on these years ofdevelopment due to a lack of child-appropriate testingparadigms. TrackIt is a paradigm that was recently developedto fill the previously existing measurement gap for SSA inthese years. In this study, we analyzed errors that children(aged 3-7) make when performing TrackIt, to betterunderstand what factors drive improvement in theirperformance over age. In addition, we manipulatedparameters within TrackIt to place varying levels of demandon children’s SSA, and measured behavioral performanceover age, with the goal of measuring and characterizingdevelopmental trends during these years. Since TrackIt is stilla recent paradigm, our results also help suggest appropriateparameter settings for calibrating the task to different agegroups.

Successful and repeated cooperation requires fairly sharingthe spoils of joint endeavors. Fair distribution is often doneaccording to preferences for equitable outcomes even thoughstrictly equitable outcomes can lead to inefficient waste. In ad-dition to preferences about the outcome itself, decision makersare also sensitive to the attributions others might make aboutthem as a result of their choice. We develop a novel mathemat-ical model where decision makers turn their capacity to inferlatent desires and beliefs from the behavior of others (theory-of-mind) towards themselves, anticipating the judgments oth-ers will make about them. Using this model we can construct apreference to be seen as impartial and integrate it with prefer-ences for equitable and efficient outcomes. We test this modelin two studies where the anticipated attribution of impartialityis ambiguous: when one agent is more deserving than the otherand when unbiased procedures for distribution are made avail-able. This model explains both participants’ judgments aboutthe partiality of others and their hypothetical decisions. Ourmodel argues that people avoid inequity not only because theyfind it inherently undesirable, they also want to avoid beingjudged as partial.

When learners explain to themselves as they encounter new in-formation, they recruit a suite of processes that influence sub-sequent learning. One consequence is that learners are morelikely to discover exceptionless rules that underlie what theyare trying to explain. Here we investigate what it is about ex-ceptionless rules that satisfies the demands of explanation. Areexceptions unwelcome because they lower predictive accuracy,or because they challenge some other explanatory ideal, suchas simplicity and breadth? To compare these alternatives, weintroduce a causally rich property explanation task in whichexceptions to a general rule are either arbitrary or predictable(i.e., exceptions share a common feature that supports a “ruleplus exception” structure). If predictive accuracy is sufficientto satisfy the demands of explanation, the introduction of a ruleplus exception that supports perfect prediction should blockthe discovery of a more subtle but exceptionless rule. Acrosstwo experiments, we find that effects of explanation go beyondattaining perfect prediction.

People often perceive their in-groups as more heterogeneousthan their out-groups. We propose an information samplingexplanation for this in-group heterogeneity effect. We analyzea model in which an agent forms beliefs and attitudes aboutsocial groups from her experience. Consistent with robust evi-dence from the social sciences, we assume that people are morelikely to interact again with in-group members than with out-group members. This implies that people obtain larger sam-ples of information about in-groups than about out-groups. Be-cause estimators of variability tend to be right-skewed, but lessso when sample size is large, sampled in-group variability willtend to be higher than sampled out-group variability. This im-plies that even agents that process information correctly – evenif they are naive intuitive statisticians – will be subject to thein-group heterogeneity effect. Our sampling mechanism com-plements existing explanations that rely on how informationabout in-group and out-group members is processed.

Previous research on the effects of probability and delay ondecision-making has focused on examining each dimensionseparately, and hence little is known about when these dimen-sions are combined into a single choice option. Importantly,we know little about the psychological processes underlyingchoice behavior with rewards that are both delayed and proba-bilistic. Using a process-tracing experimental design, we mon-itored information acquisition patterns and processing strate-gies. We found that probability and delay are processed se-quentially and evaluations of risky delayed prospects are de-pendent on the sequence of information acquisition. Amongchoice strategies, directly comparing the values of each dimen-sion (i.e., dimension-wise processing) appears to be most fa-vored by participants. Our results provide insights into the psy-chological plausibility of existing computational models andmake suggestions for the development of a process model forrisky intertemporal choice.

A growing body of evidence supports the hypothesis that hu-mans infer future states of perceived physical situations bypropagating noisy representations forward in time using ratio-nal (approximate) physics. In the present study, we examinewhether humans are able to predict (1) the resting geometryof sand pouring from a funnel and (2) the dynamics of threesubstances—liquid, sand, and rigid balls—flowing past obsta-cles into two basins. Participants’ judgments in each experi-ment are consistent with simulation results from the intuitivesubstance engine (ISE) model, which employs a Material PointMethod (MPM) simulator with noisy inputs. The ISE outper-forms ground-truth physical models in each situation, as wellas two data-driven models. The results reported herein expandon previous work proposing human use of mental simulation inphysical reasoning and demonstrate human proficiency in pre-dicting the dynamics of sand, a substance that is less commonin daily life than liquid or rigid objects.

Team cognition can be defined as the ability that humans haveto coordinate with others through a complex environment.Sports offer exquisite examples of this dynamic interplayrequiring decision making and other perceptual-cognitiveskills to adjust individual decisions to the team self-organization and vice-versa. Considering players of a team asperiodic phase oscillators, synchrony analyses can be used tomodel the coordination of a team. Nonetheless, a mainlimitation of current models is that collective behavior iscontext independent. In other words, players of a team can behighly synchronized without this corresponding to ameaningful coordination dynamics relevant to the context ofthe game. Considering these issues, the aim of this study wasto develop a method of analysis sensitive to the context forevidence-based measures of team cognition.

Previous research has suggested that first-born infants acquire words faster than their later-born peers (Berglund etal., 2005), but may have some disadvantages in other aspects of syntactic and socio-communicative development (e.g. Hoff,2006). Here we analyzed infants’ early lexical development alongside their caregiver input from 6-18 months, in relation to howmany siblings they have. We find that having more siblings (rather than being first- or later-born) has a gradient and negativerelationship with infants’ language development. This affect appears to be manifested in caregiver input: across three differentmeasures of input quality/quantity, disadvantages were found for infants with more siblings. Having a larger number of siblingsdiminished the quality of the input and led to slower overall lexical development. Implications for language development andlearning within dyadic and multi-member contexts are discussed.

Visual memory can be understood as an inferential process thatcombines noisy information about the world with knowledgedrawn from experience. Biases can arise during encoding ofinformation from the outside world into internal representa-tions, or during retrieval. In this work, we use the methodof serial reproduction, in which information is passed along achain of participants who try to recreate what they observed.We apply this method to the study of visual perception in thecontext of spatial memory biases for the remembered positionof dots inside different geometric shapes. We present the re-sults of non-parametric kernel density estimation of the end re-sult of serial reproduction to model visual biases. We confirmprevious findings, and show that memory biases revealed withour method are often more intricate and complex than whathad previously been reported, suggesting that serial reproduc-tion can be effective for studying perceptual priors.

When it comes to decision making, the dominant viewsuggests that engaging in a detailed analytical thought processis more beneficial than deciding based on one’s feelings.However, there seems to be a tradeoff, as the complexity andamount of elements on which to base the decision increases,decisions based on affect seem to be more accurate thandecisions based on a thorough analytical process in specificcontexts. In those last cases, an affective modulation ofmemory may help to make better decisions in complex tasksthat exceed human’s limited cognitive capacities. Some dualprocess accounts, ‘‘deliberation-without-attention’’hypothesis (Dijksterhuis et al., 2006), oppose a cognitive (i.e.,conscious) route to an affective (i.e., unconscious) route.Since most dual process accounts suggest one type of processis better than the other, the interaction and integration ofaffective and more conscious analytical processes in decisionmaking have been understudied. To address this issue, wepropose an explanation of the dynamics and interaction ofcognitive (i.e., explicit) and affective (i.e., implicit) encodingand retrieval of elements in memory, using a unified theorybased on core affect (Russell, 2003), in the shape of acognitive model in the ACT-R cognitive architecture.

In this study, we focused on the Generic Overgeneralisation(GOG) effect (Leslie, Khemlani, and Glucksberg 2011) andtested the relevance of context and an explanation based onquantifier domain restriction for the pattern of judgement dataobserved. Participants judged generic majority characteristicstatements like tigers have stripes or statements withuniversal quantifiers that have different sensitivity to context(‘all’, ‘all the’, ‘each’) preceded by one of three levels ofcontext: a) neutral, where the information in the context doesnot interact with the truth value of the critical statement, b)contradictory, where it presents an exception which shouldrule out a universally quantified statement, and c) supportive.Our results suggest that proponents of the generics-as-defaultview ruled out context prematurely and that in fact context isa viable alternative explanation for much of the so-calledGOG effect.

Using real-time eye-movement measures, we asked how a fantastical discourse context competes with stored representations of semantic and world knowledge to influence children's and adults' moment-by-moment interpretation of a story. Seven-year- olds were less effective at bypassing stored semantic and world knowledge during real-time interpretation than adults. Nevertheless, an effect of discourse context on comprehension was still apparent.

We replicate and extend work demonstrating that choice and probability estimation can be dissociated through the coexistence of contradictory reactions to rare events. In the context of experience-based risky choice, we find the simultaneous underweighting of rare events in choice and their overestimation in probability judgement. This tendency persisted in the presence of accurate descriptions of rare event occurrence (Experiment 1), but was attenuated by incentivizing accurate probability estimates (Experiment 2). The implications of these results for popular models of risky choice are briefly discussed.

What is the optimal way to make a decision given that yourtime is limited and your cognitive resources are bounded? Toanswer this question, we formalized the bounded optimal de-cision process as the solution to a meta-level Markov deci-sion process whose actions are costly computations. We ap-proximated the optimal solution and evaluated its predictionsagainst human choice behavior in the Mouselab paradigm,which is widely used to study decision strategies. Our compu-tational method rediscovered well-known heuristic strategiesand the conditions under which they are used, as well as novelheuristics. A Mouselab experiment confirmed our model’smain predictions. These findings are a proof-of-concept thatoptimal cognitive strategies can be automatically derived as therational use of finite time and bounded cognitive resources.

The distinction between integral and separable dimensions isof central importance to understanding how humans integrateinformation from multiple stimulus sources. One approach tocharacterizing stimulus integrality is through a set of speededcategorization tasks most closely associated with the work ofWendell Garner. These tasks demonstrate that integral dimen-sions result in marked speed up or slow down in respondingwhen there is correlated or irrelevant variation, respectively,compared with a baseline task. Little, Wang & Nosofsky(2016) recently found that the slow down or interference canbe largely explained by a reduction in the number of direct rep-etitions in a modified Garner filtering task. In this paper, weexamine a large sample of subjects tested on either separable orintegral dimensions to determine the extent of and individualdifferences in the overall and sequential effects in the standardGarner tasks.

Children and adults are guided by verb-specific syntactic like-lihoods, or verb bias, in language comprehension and produc-tion. Recent reports showed that verb bias can be altered bynew linguistic experience. We investigated the mechanismsunderlying this verb bias learning or adaptation. Specifically,we asked whether verb bias learning, like abstract syntacticpriming, is driven by error-based implicit learning. We reportthree experiments in which we altered the biases of familiardative verbs in children’s and adults’ sentence production, viatraining trials that induced participants to produce each verbconsistently in either double-object or prepositional-object da-tive structures. Participants’ syntactic choices in later test trialsreflected the expected adaptation of verb bias to the trainingexperience. In addition, the magnitude of the training effectvaried with the likelihood of each sentence structure and withpre-existing verb bias: Unexpected verb-structure combina-tions resulted in larger training effects, suggesting the opera-tion of error-based implicit learning.

Previous research has shown that repeated interactions cancause iconicity in signals to reduce. However, data from sev-eral recent studies has shown the opposite trend: an increase iniconicity as the result of repeated interactions. Here, we dis-cuss whether signals may become less or more iconic as a re-sult of the modality used to produce them. We review severalrecent experimental results before presenting new data frommulti-modal signals, where visual input creates audio feed-back. Our results show that the growth in iconicity presentin the audio information may come at a cost to iconicity inthe visual information. Our results have implications for howwe think about and measure iconicity in artificial signallingexperiments. Further, we discuss how iconicity in real worldspeech may stem from auditory, kinetic or visual information,but iconicity in these different modalities may conflict

Infants understand that people act in order to achieve their goals, but how can they tell what goals people findworthwhile? Here, we explore the thesis that human infants solve this problem by building a mental model of action planning,taking into account the costs of acting and the rewards actions bring. Consistent with this thesis, we found that 10-month-oldinfants, after viewing an agent approach two objects equally often, inferred that the agent preferred the object whose attainmentrequired a costlier action. Infants’ responses generalized across changes in perceptual variables that distinguished one actionfrom another (e.g. path length, angle of incline), suggesting that an abstract cost metric based on force or effort supported theirjudgments. These findings suggest that infants’ knowledge about agents may be expressed as a generative model for actionplanning, which can then be inverted to identify the probable hidden causes for observed actions.

Algorithms for approximate Bayesian inference, such asMonte Carlo methods, provide one source of models of howpeople may deal with uncertainty in spite of limited cognitiveresources. Here, we model learning as a process of sequentialsampling, or ‘particle filtering’, and suggest that an individ-ual’s working memory capacity (WMC) may be usefully mod-elled in terms of the number of samples, or ‘particles’, that areavailable for inference. The model qualitatively captures twodistinct effects reported recently, namely that individuals withhigher WMC are better able to (i) learn novel categories, and(ii) flexibly switch between different categorization strategies.

Natural languages vary widely in the degree to which they make use of hierarchical composition in their grammars,in particular, the degree to which syntactic versus morphologi- cal means of composition are utilized. Languages historicallyspoken in small communities develop much deeper levels of morphological embedding than those spoken by larger groups, anobservation confirmed by a statistical analysis of the World Atlas of Language Structures. However, beyond population alone,social networks change in topological structure as they grow, and it may be the pattern of connectivity rather than numberof speakers driving these differences. To examine mechanistically this connection between social and linguistic structure, wepropose an agent-based model of grammatical change using complex network methods. We identify global transitivity asa physical parameter of social networks critical for developing morphological structure, and hubs associated with scale-freenetworks as inhibitory, encouraging syntactic composition instead.

The cognitive system readily extracts regularities in terms of object co-occurrences over space and time through statistical learning. However, how does learning such relationships influence the memory representations of individual objects? Here we used a false memory paradigm to examine the impact of statistical learning on memory representations of individual objects. Observers were exposed to a temporal sequence (Experiment 1) or spatial arrays (Experiment 2) of objects which contained object pairs (e.g., A-B). In a subsequent recognition phase, observers viewed a sequence or an array containing only one member of the original pair, and judged whether either the presented object or the missing object in the original pair was present. We found that statistical learning not only sharpened the detection of the presented object, but also induced a false memory of the missing object. This reveals a novel consequence of statistical learning: learning of regularities can create illusory memories.

Language comprehension in grounded contexts involves in-tegrating visual and linguistic information through decisionsabout visual fixation. But when the visual signal also con-tains information about the language source – as in the caseof written text or sign language – how do we decide where tolook? Here, we hypothesize that eye movements during lan-guage comprehension represent an adaptive response. Usingtwo case studies, we show that, compared to English-learners,young signers delayed their gaze shifts away from a languagesource, were more accurate with these shifts, and produced asmaller proportion of nonlanguage-driven shifts (E1). Next,we present a well-controlled, confirmatory experiment, show-ing that English-speaking adults produced fewer nonlanguage-driven shifts when processing printed text compared to spokenlanguage (E2). Together, these data suggest that people adaptto the value of seeking different information in order to in-crease the chance of rapid and accurate language understand-ing.

In cooperative activities, all parties have a shared goalbut may not have the same set of skills. The currentstudy considers whether preschoolers are sensitive toprobable differences in individuals’ competence whenallocating roles. We found that 3.5- to 5.5-year-olds userelative competence, as indexed by the age of theirintended partner, to determine who should do the harderand easier of two tasks in a cooperative interaction. Asecond experiment demonstrated that children allocateroles differently in a competitive context. Youngchildren infer differences in others’ ability and candivide labor efficiently to achieve their goals.

Studies suggest that children’s exploratory behavior is sensitive to uncertainty; however, few have approached thiswith sufficient precision to model quantitatively. Across three experiments, children (mean age=70 months) were asked to shakea box to identify which of two sets of marbles, differing in numerosity, were hidden inside. The sets’ numerosities varied in theirdiscriminability indices – the degree to which listeners can distinguish the sets based on the acoustic information generated.The time children spent shaking the box varied systematically with the discriminability of the alternative hypotheses they wereasked to distinguish, even though they heard only one set for each contrast. This suggests that children represent the uncertaintyin their own perceptual discrimination abilities (an ability we refer to as an intuitive psychophysics) and their exploratorybehavior is precisely calibrated to their degree of uncertainty about alternative hypotheses that might explain unobserved causesof perceptual data.

Averaging the estimates of a number of individuals has beenshown to produce an estimate that is generally more accuratethan those of the individuals themselves. Similarly, averagingresponses from a single individual can also lead to a moreaccurate answer. How can we best combine estimates withinand between individuals to create an accurate group estimate?We report empirical results from a general knowledge rank-ordering experiment and demonstrate that individuals thatprovide more consistent answers across repeated elicitationsare also more accurate. We develop a consistency weightingheuristic and show that repeated elicitations within anindividual can be used to improve group accuracy. We alsodevelop a Thurstonian cognitive model which assumes adirect link between the process that explains the accuracy ofan individual and response consistency and show how themodel can infer accurate group answers.

A widely observed phenomenon in children’s word-extensionsand generalizations is the characteristic-to-defining shift,whereby young children initially generalize words based ontypical properties and gradually transition into generalizingwords using abstract, logical information. In this paper, wepropose a statistically principled model of conceptual devel-opment grounded in the trade-off between simplicity and fit tothe data. We run our model based on informant-provided fam-ily trees and the real-life characteristic features of people onthose trees. We demonstrate that the characteristic-to-definingshift does not necessarily depend on discrete change in rep-resentation or processes. Instead, the shift could fall out nat-urally from statistical inference over conceptual hypotheses.Our model finds that the shift occurs even when abstract logicalrelations are present from the outset of learning as long as char-acteristic features are informative but imperfect in their abilityto capture the underlying concept to be learned—a property ofour elicited features.

People have the capability to process text three times fasterthan they would naturally read it, yet many current theories ofsentence processing rely on natural reading times as a proxyfor processing difficulty. How can people read material soquickly in spite of information processing limitations sug-gested by sentence processing theories? One possibility is thatsurprisal effects on reading time, the hallmark of processingdifficulty under sentence processing theories, might arise fromperceptual processing, implying no relation between surprisaland sentence processing difficulty. In this paper, we conducteda novel self-paced rapid serial visual presentation (RSVP) ex-periment, which controlled perceptual processes to probe forsentence processing related surprisal effects. We further testedhow readers might compensate for information processing lim-its during RSVP. We find support for sentence processing re-lated surprisal effects, the pattern of which is consistent with aFirst-In, First-Out (FIFO) buffer model.

Learning to read has a substantial effect on the representationsof spoken and meaning forms of words. In this paper weassess literacy effects beyond representational changes,focusing on adaptations to the architecture of the readingsystem that maps between these representations. We present aconnectionist model of reading that predicted distinctprocessing of pre- and post-literacy acquired words. Forreading for meaning, words learned prior to literacy wereprocessed more indirectly via phonological representations,whereas for post-literacy acquired words, processing wasmore direct along the orthography to semantics pathway. Thismore computationally intensive route was prioritised becauseindirect phonology to semantics mappings were unavailable.Such an effect was less apparent for naming, because learningdirect orthography to phonology mappings is lesscomputationally intensive. These results were confirmed in ananalysis of naming and lexical decision behavioural data. Theeffect of literacy onset remains an observable artefact in adultreading.

Learning how words refer to aspects of the environment is acomplex task, but one that is supported by numerous cueswithin the environment which constrain the possibilities formatching words to their intended referents. In this paper wetested the predictions of a computational model of multiplecue integration for word learning, that predicted variation inthe presence of cues provides an optimal learning situation. Ina cross-situational learning task with adult participants, wevaried the reliability of presence of distributional, prosodic,and gestural cues. We found that the best learning occurredwhen cues were often present, but not always. The effect ofvariability increased the salience of individual cues for thelearner, but resulted in robust learning that was not vulnerableto individual cues’ presence or absence. Thus, variability ofmultiple cues in the language-learning environment providedthe optimal circumstances for word learning.

In everyday life, before deciding what to do, one has to think about what could be done. We investigate option generation from a developmental perspective, testing the predictions of the Take-The-First-heuristic (TTF). Moreover, we examine the influence of time limitation on decision-making processes. Using soccer as a testbed, 6- to 13-year-old children (N = 97) were tested in a video-based option-generation paradigm. Children’s performance was aligned with predictions of TTF: Children generated a mean of 2.21 options, did so in a meaningful way and selected the first as final option in 74%. With shorter time, children generated fewer and higher quality options, selected better options and more often the first option as final decision. Further, with age, an increase of the number of options generated and an increase in quality of the final decisions emerged. This age effect was more pronounced with shorter time. Implications for real-life decision-making are discussed.

One of the most salient and well-recognized features of humangoal-directed behavior is our limited ability to conduct mul-tiple demanding tasks at once. Previous work has identifiedoverlap between task processing pathways as a limiting fac-tor for multitasking performance in neural architectures. Thisraises an important question: insofar as shared representationbetween tasks introduces the risk of cross-talk and thereby lim-itations in multitasking, why would the brain prefer shared taskrepresentations over separate representations across tasks? Weseek to answer this question by introducing formal considera-tions and neural network simulations in which we contrast themultitasking limitations that shared task representations incurwith their benefits for task learning. Our results suggest thatneural network architectures face a fundamental tradeoff be-tween learning efficiency and multitasking performance in en-vironments with shared structure between tasks.

Analogical and causal reasoning theories both seek to explainpatterns of inductive inference. Researchers have claimed thatreasoning scenarios incorporating aspects of both analogicalcomparison and causal thinking necessitate a new model of in-ductive inference (Holyoak, Lee, & Lu, 2010; Lee & Holyoak,2008). This paper takes an opposing position, arguing that fea-tures of analogical models make correct claims about infer-ence patterns found among causal analogies, including analo-gies with both generative and preventative relations. Experi-ment 1 demonstrates that analogical inferences for these kindsof causal systems can be explained by alignment of relationalstructure, including higher-order relations. Experiment 2 fur-ther demonstrates that inferences strengthened by matchinghigher-order relations are not guided by the transfer of prob-abilistic information about a cause from base to target. Weconclude that causal analogies behave like analogies in gen-eral—analogical mapping provides candidate inferences whichcan then be reasoned about in the target.

Research in analogical transfer suggests a simple type of transfer that occurs due to the activation of key relational concepts. Analysis on mental structured representations indicates that this transfer may act differently depending upon structural and perceptual features of the priming task. Two hundred eight participants were assigned to three experimental groups where they received a structure-priming, tested once and afterwards they received a perceptual-priming and tested again. As predicted, the effect of structure-priming was found across conditions whereas the effect of perceptual- priming (a six-second animation) was detected only in subjects with high levels of cognitive reflectiveness. These individual differences are interpreted as evidence that only highly reflective subjects were able to process visuospatial cues in the animation and to extract their structural features, hence activating relational concepts that influenced their interpretations of subsequent tasks.

How does the process of information transmission affect thecultural products that emerge from that process? This questionis often studied experimentally and computationally via iter-ated learning, in which participants learn from previous partic-ipants in a chain. Much research in this area builds on math-ematical analyses suggesting that iterated learning chains con-verge to people’s priors. We present three simulation studiessuggesting that when the population of learners is heteroge-neous, the behavior of the chain is systematically distorted bythe learners with the most extreme biases. We discuss implica-tions for the use of iterated learning as a methodological tooland for the processes that might have shaped cultural productsin the real world.

Children can use the statistical regularities of their environ-ment to learn word meanings, a mechanism known as cross-situational learning. We take a computational approach to in-vestigate how the information present during each observationin a cross-situational framework can affect the overall acqui-sition of word meanings. We do so by formulating variousin-the-moment learning mechanisms that are sensitive to dif-ferent statistics of the environment, such as counts and con-ditional probabilities. Each mechanism introduces a uniquesource of competition or mutual exclusivity bias to the model;the mechanism that maximally uses the model’s knowledge ofword meanings performs the best. Moreover, the gap betweenthis mechanism and others is amplified in more challenginglearning scenarios, such as learning from few examples. Key-words: cross-situational word learning; computational model-ing; word learning biases

Vector-space models of semantics represent words ascontinuously-valued vectors and measure similarity based onthe distance or angle between those vectors. Such representa-tions have become increasingly popular due to the recent de-velopment of methods that allow them to be efficiently esti-mated from very large amounts of data. However, the ideaof relating similarity to distance in a spatial representationhas been criticized by cognitive scientists, as human similar-ity judgments have many properties that are inconsistent withthe geometric constraints that a distance metric must obey. Weshow that two popular vector-space models, Word2Vec andGloVe, are unable to capture certain critical aspects of humanword association data as a consequence of these constraints.However, a probabilistic topic model estimated from a rela-tively small curated corpus qualitatively reproduces the asym-metric patterns seen in the human data. We also demonstratethat a simple co-occurrence frequency performs similarly toreduced-dimensionality vector-space models on medium-sizecorpora, at least for relatively frequent words.

Human communication has a remarkable capacity todescribe events that occurred elsewhere and at othertimes. In particular, when describing complex narratives,speakers must communicate temporal structure using amixture of words (e.g., “after”), gestures (e.g., pointingrightward for a later event), and discourse structure (e.g.,mentioning earlier events first). How do listenersintegrate these sources of temporal information to makesense of complex narratives? In two experiments, wesystematically manipulated gesture, speech, and order-of-mention to investigate their respective impacts oncomprehension of temporal structure. Gesture had asignificant effect on interpretations of temporal order.This influence of gesture, however, was weaker than theinfluence of both speech and order-of-mention. Indeed,in some cases, order-of-mention trumped explicitdescriptions in speech; for instance, if ‘earlier’ eventswere mentioned second, they were sometimes thought tohave occurred second. Listeners integrate multiplesources of information to interpret what happened when.

The actualization of affordances can often be accomplished in numerous, equifinal ways. For instance, an individual could discard an item in a rubbish bin by walking over and dropping it, or by throwing it from a distance. The aim of the current study was to investigate the behavioral dynamics associated with such metastability using a ball-to-bin transportation task. Using time-interval between sequential ball-presentation as a control parameter, participants transported balls from a pickup location to a drop-off bin 9m away. A high degree of variability in task-actualization was expected and found, and the Cusp Catatrophe model was used to understand how this behavioral variability emerged as a function of hard (time interval) and soft (e.g. motivation) task dynamic constraints. Simulations demonstrated that this two parameter state manifold could capture the wide range of participant behaviors, and explain how these behaviors naturally emerge in an under-constrained task context.

Within the literature of psychological and decision sciences, there is a critical difference in the way recognition is defined and studied experimentally. To address this difference, the current experiment examines and attempts to disentangle the influence of two recognition judgment sources (from within an experiment and from an individual’s prior life experiences) upon two different recognition judgments. By presenting participants with a set of related stimuli that vary naturally in environmental occurrence and by manipulating exposure within an experimental context, this experiment allows for a broader and more ecologically valid assessment of recognition memory. Contrasting with the typical word- frequency effect, the results reveal an overall bias to judge high-frequency items as studied on an episodic recognition test. Additionally, the results underscore the role of context by showing that a single study exposure increases the probability that individuals will judge stimuli as presented outside the laboratory.

A recent meta-analysis (Khemlani & Johnson-Laird, 2012)about psychological experiments of syllogistic reasoningdemonstrates that the conclusions drawn by human reasonersstrongly deviate from conclusions of classical logic. Moreover,none of the current cognitive theories predictions fit reliablythe empirical data. In this paper, we show how humansyllogistic reasoning can be modeled under a new cognitivetheory, the Weak Completion Semantics. Our analysis basedon computational logics identifies seven principles necessaryto draw the inferences. Hence, this work contributes to acomputational foundation of cognitive reasoning processes.

Sign languages stand out in that there is high prevalence ofconventionalised linguistic forms that map directly to theirreferent (i.e., iconic). Hearing adults show low performancewhen asked to guess the meaning of iconic signs suggestingthat their iconic features are largely inaccessible to them.However, it has not been investigated whether speakers’gestures, which also share the property of iconicity, mayassist non-signers in guessing the meaning of signs. Resultsfrom a pantomime generation task (Study 1) show thatspeakers’ gestures exhibit a high degree of systematicity, andshare different degrees of form overlap with signs (full,partial, and no overlap). Study 2 shows that signs with fulland partial overlap are more accurately guessed and areassigned higher iconicity ratings than signs with no overlap.Deaf and hearing adults converge in their iconic depictionsfor some concepts due to the shared conceptual knowledgeand manual-visual modality.

This paper introduces a formal method to model the level of de-mand on control when executing cognitive processes. The costof cognitive control is parsed into an intensity cost which en-capsulates how much additional input information is requiredso as to get the specified response, and an interaction costwhich encapsulates the level of interference between individ-ual processes in a network. We develop a formal relationshipbetween the probability of successful execution of desired pro-cesses and the control signals (additive control biases). Thisrelationship is also used to specify optimal control policies toachieve a desired probability of activation for processes. Weobserve that there are boundary cases when finding such con-trol policies which leads us to introduce the interaction cost.We show that the interaction cost is influenced by the relativestrengths of individual processes, as well as the directionalityof the underlying competition between processes.

Frequent words tend to be short, and many researchers haveproposed that this relationship reflects a tendency towards ef-ficient communication. Recent work has sought to formalizethis observation in the context of information theory, which es-tablishes a limit on communicative efficiency called the chan-nel capacity. In this paper, I first show that the compositionalstructure of natural language prevents natural language com-munication from getting close to the channel capacity, but thata different limit, which incorporates probability in context,may be achievable. Next, I present two corpus studies in threetypologically-diverse languages that provide evidence that lan-guages change over time towards the achievable limit. Theseresults suggest that natural language optimizes for efficiencyover time, and does so in a way that is appopriate for compo-sitional codes.

A key goal in both education and higher-order cognitionresearch is to understand how relational concepts are bestlearned. In the current work, we present a novel approach forlearning complex relational categories – a low-support,interactive discovery interface. The platform, which allowslearners to make modifications to exemplars and see thecorresponding effects on membership, holds the potential toaugment relational learning by facilitating self-directed,alignably-different comparisons that explore what the learnerdoes not yet understand. We compared interactive learning to anidentification learning task. Participants were assessed on theirability to generalize category knowledge to novel exemplarsfrom the same domain. Although identification learners wereprovided with seven times as many positive examples of thecategory during training, interactive learners demonstratedenhanced generalization accuracy and knowledge of specificmembership constraints. Moreover, the data suggest thatidentification learners tended to overgeneralize categoryknowledge to non-members – a problem that interactive learnersexhibited to a significantly lesser degree. Overall, the resultsshow interactive training to be a powerful tool forsupplementing relational category learning, with particularutility for refining category knowledge. We conclude withimplications of these findings and promising future directions.

This paper examines the use of iconic words in earlyconversations between children and caregivers. Thelongitudinal data include a span of six observations of 35children-parent dyads in the same semi-structured activity.Our findings show that children’s speech initially has a highproportion of iconic words, and over time, these wordsbecome diluted by an increase of arbitrary words. Parents’speech is also initially high in iconic words, with a decreasein the proportion of iconic words over time – in this casedriven by the use of fewer iconic words. The level anddevelopment of iconicity are related to individual differencesin the children’s cognitive skills. Our findings fit with thehypothesis that iconicity facilitates early word learning andmay play an important role in learning to produce new words

Shepard’s Universal Law of Generalization offered a com-pelling case for the first physics-like law in cognitive sciencethat should hold for all intelligent agents in the universe. Shep-ard’s account is based on a rational Bayesian model of general-ization, providing an answer to the question of why such a lawshould emerge. Extending this account to explain how humansuse multiple examples to make better generalizations requiresan additional assumption, called the size principle: hypothesesthat pick out fewer objects should make a larger contributionto generalization. The degree to which this principle warrantssimilarly law-like status is far from conclusive. Typically, eval-uating this principle has not been straightforward, requiringadditional assumptions. We present a new method for evaluat-ing the size principle that is more direct, and apply this methodto a diverse array of datasets. Our results provide support forthe broad applicability of the size principle.

We present a formal measure of argument strength, whichcombines the ideas that conclusions of strong arguments are (i)highly probable and (ii) their uncertainty is relatively precise.Likewise, arguments are weak when their conclusion proba-bility is low or when it is highly imprecise. We show howthe proposed measure provides a new model of the Ellsbergparadox. Moreover, we further substantiate the psychologi-cal plausibility of our approach by an experiment (N = 60).The data show that the proposed measure predicts human in-ferences in the original Ellsberg task and in corresponding ar-gument strength tasks. Finally, we report qualitative data takenfrom structured interviews on folk psychological conceptionson what argument strength means.

Causal judgments are well-known to be sensitive to violationsof both prescriptive moral and descriptive statistical norms.There is ongoing discussion as to whether both effects arebest explained through changes in the relevance of counter-factual possibilities, or if moral norm violations should be in-dependently explained through a potential polysemy whereby‘cause’ may simply mean ‘is morally responsible for’. Insupport of the latter view, recent work has pointed out thatmoral norm violations affect judgments of agents, but not inan-imate objects, and that these effects are moderated by agents’knowledge states. We advance this debate by demonstratingthat judgments of counterfactual relevance exhibit preciselythe same patterns, and that judgments of inanimate objects areactually highly sensitive to whether the object violated a pre-scriptive norm by malfunctioning. The latter finding is difficultto account for through polysemy, but is predicted by changes inthe relevance of counterfactual alternatives. Finally, we showthat direct (non-moral) interventions on the the relevance ofcounterfactual alternatives affect causal judgments in preciselythe same way as functional and moral norm violations.

Increasingly, cognitive scientists have demonstrated interest inapplying tools from deep learning. One use for deep learning isin language acquisition where it is useful to know if a linguisticphenomenon can be learned through domain-general means.To assess whether unsupervised deep learning is appropriate,we first pose a smaller question: Can unsupervised neural net-works apply linguistic rules productively, using them in novelsituations? We draw from the literature on determiner/nounproductivity by training an unsupervised, autoencoder networkmeasuring its ability to combine nouns with determiners. Oursimple autoencoder creates combinations it has not previouslyencountered and produces a degree of overlap matching adults.While this preliminary work does not provide conclusive evi-dence for productivity, it warrants further investigation withmore complex models. Further, this work helps lay the foun-dations for future collaboration between the deep learning andcognitive science communities.

In online networks, the polarization of opinions (e.g., regarding presidential elections or referenda) has been associated with the creation of “echo-chambers” of like- minded peers, secluded from those of contrary viewpoints. Previous work has commonly attributed such phenomena to self-regarding preferences (e.g., confirmation bias), individual differences, and the pre-dispositions of users, with clusters forming over repeated interactions. The present work provides a proof of concept Agent-Based Model that demonstrates online networks are susceptible to echo-chambers from a single opinion cascade, due to the spatiotemporal order induced by lateral transmission. This susceptibility is found to vary as a function of degree of interconnectivity and opinion strength. Critically, such effects are found despite globally proportionate levels of opinions, equally rational agents (i.e. absent conformity, confirmation bias or pre-disposition architecture), and prior to cyclical interactions. The assumptions and implications of this work, including the value of Agent-Based Modelling to cognitive psychology, are discussed.

Humans regularly invest time towards activities characterized by dramatic success or failure outcomes, where criti-cally, the outcome is uncertain ex-ante. How should people allocate time between such make-or-break activities and other safealternatives, where rewards are more predictable (e.g., linear) functions of time? We present a formal framework for studyingtime allocation between these two types of activities, and explore (optimal) behavior in both one-shot and dynamic versions ofthe problem. In the one-shot version, we illustrate the striking discontinuous relation between peoples skill and optimal timeallocation to the make-or-break task. In the dynamic version, we formulate both fully rational and boundedly rational strategies,both defined by a giving up threshold, which adaptively dictates when one should cease pursuit of the make- or-break goal.Comparing strategies across environments, we investigate the cost of sidestepping the computational burden of full rationality.

A key challenge for cognitive neuroscience is to decipher therepresentational schemes of the brain. A recent class of decodingalgorithms for fMRI data, stimulus-feature-based encodingmodels, is becoming increasingly popular for inferring thedimensions of neural representational spaces from stimulus-feature spaces. We argue that such inferences are not always valid,because decoding can occur even if the neural representationalspace and the stimulus-feature space use different representationalschemes. This can happen when there is a systematic mappingbetween them. In a simulation, we successfully decoded the binaryrepresentation of numbers from their decimal features. Sincebinary and decimal number systems use different representations,we cannot conclude that the binary representation encodes decimalfeatures. The same argument applies to the decoding of neuralpatterns from stimulus-feature spaces and we urge caution ininferring the nature of the neural code from such methods. Wediscuss ways to overcome these inferential limitations.

We present evidence for a novel relational luring effect (RLE) in recognition memory. Participants performeda continuous associative recognition task in which they had to discriminate between new, old and recombined word pairs.Participants made more false alarms and responded more slowly to lures (TABLE CLOTH) that were relationally similarto studied pairs (FLOOR CARPET). RTs and false alarms for lures increased linearly as the number of previously studieddifferent exemplars of the relation increased (e.g., 0 to 4 previous exemplars). The RLE effect was stronger for relations thatwere represented by exemplars that were more typical of the relation. These results suggest that semantic relations exist asindependent representations in LTM, and that during associative recognition these representations can be a spurious source offamiliarity. The RLE has implications for models of semantic and episodic memory, unitization in associative recognition,analogical reasoning, and constructive memory research.

We found an unexpected positive effect of target-to-distractorsimilarity (TD) in a visual search task, despite overwhelmingevidence in the literature that TD similarity hurts visual searchperformance. Participants with no prior knowledge of Chineseperformed 12 hour-long sessions over 4 weeks, where they hadto find a briefly presented target character among a set ofdistractors. At the beginning of the experiment, TD similarityhurt performance, but the effect reversed during the firstsession and remained positive throughout the remainingsessions. We present a simple connectionist model thataccounts for that reversal of TD similarity effects on visualsearch and we discuss possible theoretical explanations.

Newen and Vetter (2016) argue that cognitive penetration (CP) ofperceptual experience is the most possible account of the evidence.They target both the weak impenetrability thesis that only some earlyvisual processes are cognitively impenetrable (CI), and the strongimpenetrability thesis that all perceptual processes are CI. Since Iagree that perceptual processing as a whole is CP, I will concentrateon their arguments against the weak CI thesis. In attacking weak CI,the authors take aim at Raftopoulos’ arguments supporting the CI ofearly vision. Their main argument comes from studies that, Newenand Vetter think, show that early vision is CP by demonstrating theexistence of cognitive effects on early vision. I examine the sameempirical evidence that Newen and Vatter discuss and argue thissame evidence strongly supports the view that early vision is CI.

In Wason’s selection task, participants select whichever offour cards could provide evidence about the truth or falsity ofa conditional rule. As our meta-analysis of hundreds of ex-periments corroborates, participants tend to overlook one ofthe cards that could falsify the rule. 15 distinct theories aimto explain this phenomenon and others, but many of thempresuppose that cards are selected independently of one an-other. We show that this assumption is false: Shannon’s en-tropy for selections is reliably redundant in comparison withthose of 10,000 simulated experiments using the same fourindividual probabilities for each real experiment. This resultrules out those theories presupposing independent selections.Of the remaining theories, only two predict the frequenciesof selections, one (due to Johnson-Laird & Wason, 1970a)provides a better fit to the experimental data than the other(due to Klauer et al., 2007). We discuss the implications ofthese results.

Words frequently acquire new senses, but the mental processthat underlies the historical emergence of these senses is oftenopaque. Many have suggested that word meanings develop innon-arbitrary ways, but no attempt has been made to formalizethese proposals and test them against historical data at scale.We propose that word meaning extension should reflect a drivetowards cognitive economy. We test this proposal by exploringa family of computational models that predict the evolution ofword senses, evaluated against a large digitized lexicon thatdates back 1000 years in English language history. Our find-ings suggest that word meanings not only extend in predictableways, but also that they do so following an historical path thattends to minimize cognitive cost - through a process of nearest-neighbor chaining. Our work contributes a formal approach toreverse-engineering mental algorithms of the human lexicon.

When the interests of interlocutors are not aligned, either partymay wish to avoid truthful disclosure. A sender wishing toconceal the truth from a receiver may lie by providing falseinformation, mislead by actively encouraging the receiver toreach a false conclusion, or simply be uninformative by provid-ing little or no relevant information. Lying entails moral andother hazards, such as detection and its consequences, and isthus often avoided. We focus here on the latter two strategies,arguably more pernicious and prevalent, but not without theirown drawbacks. We argue and show in two studies that whenchoosing between these options, senders consider the level ofsuspicion likely to be exercised on the part of the receiver andhow much truth must be revealed in order to mislead. Extend-ing Bayesian models of cooperative communication to includehigher level inference regarding the helpfulness of the senderleads to insight into the strategies employed in non-cooperativecontexts.

The exaggerated intonation and special rhythmic properties ofinfant-directed speech (IDS) have been hypothesized to attractinfant’s attention to the speech stream. However, studiesinvestigating IDS in the context of models of attention arefew. A number of such models suggest that surprising ornovel perceptual inputs attract attention, where novelty can beoperationalized as the statistical predictability of the stimulusin a context. Since prosodic patterns such as F0 contours areaccessible to young infants who are also adept statisticallearners, the present paper investigates a hypothesis that pitchcontours in IDS are less predictable than those in adult-directed speech (ADS), thereby efficiently tapping into thebasic attentional mechanisms of the listeners. Results fromanalyses with naturalistic IDS and ADS speech show that IDShas lower overall predictability of intonation acrossneighboring syllables even when the F0 contours in bothspeaking styles are normalized to the same frequency range.

A proof by mathematical induction demonstrates that ageneral theorem is necessarily true for all natural numbers. Ithas been suggested that some theorems may also be provenby a ‘visual proof by induction’ (Brown, 2010), despite thefact that the image only displays particular cases of thegeneral theorem. In this study we examine the nature of theconclusions drawn from a visual proof by induction. We findthat, while most university-educated viewers demonstrate awillingness to generalize the statement to nearby cases notdepicted in the image, only viewers who have been trained informal proof strategies show significantly higher resistance tothe suggestion of large-magnitude counterexamples to thetheorem. We conclude that for most university-educatedadults without proof-training the image serves as the basis ofa standard inductive generalization and does not provide thedegree of certainty required for mathematical proof.

This paper proposes a Bayesian account of asymmetriesfound in speech perception: In many languages, listenersshow greater sensitivity if a non-coronal sound (/b/, /p/, /g/,/k/) is changed to coronal sounds (/d/, /t/) than vice versa. Thecurrently predominant explanation for these asymmetries isthat they reflect innate constraints from Universal Grammar.Alternatively, we propose that the asymmetries could simplyarise from optimal inference given the statistical properties ofdifferent speech categories of the listener’s native language.In the framework of Bayesian inference, we examined twostatistical parameters of coronal and non-coronal sounds:frequencies of occurrence and variance in articulation. In thelanguages in which perceptual asymmetries have been found,coronal sounds are either more frequent or more variable thannon-coronal sounds. Given such differences, an ideal observeris more likely to perceive a non-coronal speech signal as acoronal segment than vice versa. Thus, the perceptualasymmetries can be explained as a natural consequence ofprobabilistic inference. The coronal/non-coronal asymmetryis similar to asymmetries observed in many other cognitivedomains. Thus, we argue that it is more parsimonious toexplain this asymmetry as one of many similar asymmetriesfound in cognitive processing, rather than a linguistic-specific, innate constraint.

Natural language dialogue between multiple participants re-quires conversational grounding, a process whereby interlocu-tors achieve a shared understanding. However, the mecha-nisms involved in the grounding process are under dispute.Two prominent models of dialogue between multiple partici-pants are: interactive alignment, a simpler model that relieson automatic priming processes within individuals, and in-terpersonal synergy, a more complicated model emphasizingcoordinated interaction across participants. Using recurrenceanalysis methods, Fusaroli and Tyl ́en (2016) simultaneouslyevaluated both models and showed that alignment is an insuf-ficient explanation for grounding or for the teams’ task per-formance. However, their task and resulting dialogues lackthe typical complexity of conversations or teamwork. Further-more, the interpersonal synergy model was not clearly differ-entiated from other coordination-focused models of groundingwith explicit foundations in strategy and intentionality (i.e., au-dience design, joint activity, perspective taking). Here we testrecurrence-based models in a collaborative task that stressedthe grounding process. Results support a coordination modelof dialogue over the alignment model as a predictor of perfor-mance. Content-based mediation analyses showed that the co-ordination recurrence model includes critical aspects of strate-gic design and is not purely interpersonal synergy.

Recent work suggests that language understanding is the result of rational inference over a noisy channel. Uponperceiving a sentence, listeners decode the speaker’s intended sentence from the prior probability that a speaker would say thatsentence and the probability that it would be corrupted to the perceived sentence by noise. Here we examine the listener’snoise model. Readers were asked to correct sentences if they thought they contained an error. We manipulated context suchthat participants corrected exposure sentences containing either deletion, insertion, swap, mixed, or no errors (e.g., swap: Abystander was rescued by the fireman in the time of nick.). Test sentences were syntactically licensed but implausible (e.g., Thebat swung the player). On test sentences, participants’ corrections differed by exposure condition. This suggests participantstrack the type of errors that have a higher likelihood and make inferences about the intentions of the speaker accordingly.

Human languages contain very little unconditioned variation.In contexts where language learners are exposed to input thatcontains inconsistencies, they tend to regularise it, either byeliminating competing variants, or conditioning variant use onthe context. In the present study we compare regularisationbehaviour across linguistic levels, looking at how adult learn-ers respond to variability in morphology and word order. Ourresults suggest similar strengths in regularisation between lin-guistic levels given input languages whose complexity is com-parable.

Humans are not only adept in recognizing what class an in-put instance belongs to (i.e., classification task), but perhapsmore remarkably, they can imagine (i.e., generate) plausibleinstances of a desired class with ease, when prompted. Inspiredby this, we propose a framework which allows transformingCascade-Correlation Neural Networks (CCNNs) into proba-bilistic generative models, thereby enabling CCNNs to gen-erate samples from a category of interest. CCNNs are a well-known class of deterministic, discriminative NNs, which au-tonomously construct their topology, and have been successfulin accounting for a variety of psychological phenomena. Ourproposed framework is based on a Markov Chain Monte Carlo(MCMC) method, called the Metropolis-adjusted Langevin al-gorithm, which capitalizes on the gradient information of thetarget distribution to direct its explorations towards regionsof high probability, thereby achieving good mixing proper-ties. Through extensive simulations, we demonstrate the effi-cacy of our proposed framework. Importantly, our frameworkbridges computational, algorithmic, and implementational lev-els of analysis.

Human behavior is frequently guided by social and moralnorms; in fact, no societies, no social groups could exist with-out norms. However, there are few cognitive science ap-proaches to this central phenomenon of norms. While therehas been some progress in developing formal representationsof norm systems (e.g., deontological approaches), we do notyet know basic properties of human norms: how they arerepresented, activated, and learned. Further, what computa-tional models can capture these properties, and what algo-rithms could learn them? In this paper we describe initial ex-periments on human norm representations in which the contextspecificity of norms features prominently. We then provide aformal representation of norms using Dempster-Shafer Theorythat allows a machine learning algorithm to learn norms un-der uncertainty from these human data, while preserving theircontext specificity.

The outcomes of intertemporal choices indicate that people discount rewards by their delay. These outcomes are well described by discounting functions. However, to fully understand the decision process one needs models describing how the process of decision-making unfolds dynamically over time. Here, we validate a recently published attractor model that extends discounting functions through a description of the dynamics leading to a final choice outcome within and across trials. We focus on the decision dynamics across trials. We derive qualitative predictions for the inter-trial dynamics of sequences of decisions that are unique to this type of model. We test these predictions in a delay discounting game where we sequentially manipulated subjective values of options across all attribute dimensions. Results confirm the model’s predictions. We discuss future challenges on integrating attractor models towards a general attractor model of delay discounting to enhance our understanding of the processes underlying delay discounting decisions.

How do people explore in order to gain rewards in uncer-tain dynamical systems? Within a reinforcement learningparadigm, control normally involves trading off between ex-ploration (i.e. trying out actions in order to gain more knowl-edge about the system) and exploitation (i.e. using currentknowledge of the system to maximize reward). We study anovel control task in which participants must steer a boat ona grid, aiming to follow a path of high reward whilst learninghow their actions affect the boat’s position. We find that partic-ipants explore strategically yet conservatively, exploring morewhen mistakes are less costly and practicing actions that willbe required later on.

To make inferences about the frequency of events in theworld (e.g., the prevalence of diseases or the popularity ofconsumer products), people often exploit observations ofrelevant instances sampled from their personal social network.How does this ability to infer event frequencies by searchingand relying on personal instance knowledge develop fromchildhood to adulthood? To address this question, weconducted a study in which children (age 8–11 years) andadults (age 19–34 years) judged the relative frequencies offirst names in Germany. Based on the recalled instances of thenames in participants’ social networks, we modeled theirfrequency judgments and the underlying search process with aBayesian hierarchical latent-mixture approach encompassingdifferent computational models. We found developmentaldifferences in the inference strategies that children and adultsused. Whereas the judgments of most adults were bestdescribed by a noncompensatory strategy that assumes limitedand sequentially ordered search (social-circle model), thejudgments of most children were best described by acompensatory strategy that assumes exhaustive search andinformation aggregation (availability-by-recall). Our resultshighlight that already children use instance knowledge to inferevent frequencies but they appear to search more exhaustivelyfor instances than adults. One interpretation of these results isthat the ability to conduct ordered and focused search is acentral aspect in the development of noncompensatoryinstance-based inference.

When learners are exposed to inconsistent input, do they reproduce the probabilities in the input (probability match-ing), or produce some variants disproportionately often (regularization)? Laboratory results and computational models ofartificial language learning both argue that the learning mechanism is basically probability matching, with regularization aris-ing from additional factors. However, these models were fit to aggregated experimental data, which can exhibit probabilitymatching even if all individuals regularize. To assess whether learning can be accurately characterized as basically probabilitymatching or systematizing at the individual level, we ran a large-scale experiment. We found substantial individual variation.The structure of this variation is not predicted by recent beta-binomial models. We introduce a new model, the Double ScalingSigmoid (DSS) model, fit its parameters on a by-participant basis, and show that it captures the patterns in the data. Priorexpectations in the DSS are abstract, and do not entirely represent previous experience.

Can grammatical gender influence how people conceptualize the referents of nouns? Using an implicit measure, we investigated whether such an effect could be found in a task where neither grammatical nor biological gender is highlighted. In the current study, conducted in English, speakers of French, German and Romanian with knowledge of English were asked to generate adjectives they associate with referents of nouns. Afterwards, the gender valence of the adjectives was measured. The results showed that participants generated more feminine adjectives for nouns with majority feminine translations compared to nouns with majority masculine translations. We found a stronger effect of grammatical gender for some semantic categories than for others. Significant effects of grammatical gender were present starting with the 2nd adjective generated by participants (effects were stronger for adjectives generated 2nd and 3rd by participants, as opposed to the 1st adjective).

People are adept at perceiving interactions from movementsof simple shapes but the underlying mechanism remains un-known. Previous studies have often used object movementsdefined by experimenters. The present study used aerial videosrecorded by drones in a real-life environment to generate de-contextualized motion stimuli. Motion trajectories of dis-played elements were the only visual input. We measuredhuman judgments of interactiveness between two moving el-ements, and the dynamic change of such judgments over time.A hierarchical model was developed to account for human per-formance in this task, which represents interactivity using la-tent variables, and learns the distribution of critical movementfeatures that signal potential interactivity. The model providesa good fit to human judgments and can also be generalized tothe original Heider-Simmel animations (1944). The model canalso synthesize decontextualized animations with controlleddegree of interactiveness, providing a viable tool for studyinganimacy and social perception.

Comparison supports the development of children’s analogicalreasoning. The evidence for this claim comes from labora-tory studies. We describe spontaneous comparisons producedby 24 typically developing children from 26 to 58 months.Children tend to express similarity before expressing differ-ence. They compare objects from the same category beforeobjects from different categories, make global comparisons be-fore specific comparisons, and specify perceptual features ofsimilarity/difference before non-perceptual features. We theninvestigate how a theoretically interesting subset of children’scomparisons – those expressing a specific feature of similar-ity or difference – relates to analogical reasoning as measuredby verbal and non-verbal tests in 6th grade. The number ofspecific comparisons children produce before 58 months pre-dicts their scores on both tests, controlling for vocabulary at54 months. The results provide naturalistic support for experi-mental findings on comparison development, and demonstratea strong relationship between children’s early comparisons andtheir later analogical reasoning.

It is commonly believed that children are able to learn through play. Recent studies have found that children are able to learn causal rules through free play (Sim & Xu, in press). One such study found that children learned how to correctly activate machines, using either a block that was the same shape or the same color as the machine, when given five minutes to play with them. However, would children be able to learn a more complex causal rule through free play as well and would their performance be comparable to children who were didactically taught the same causal rule? In the current study, we show that children are able to learn more complex causal rules through free play. We also show that children perform significantly better when learning these rules through free play or by first engaging in free play and then observing, as opposed to solely through observation.

In this manuscript we study individual variation in the interpretation of conditionals by establishing individual profiles of the participants based on their behavioral respon- ses and reflective attitudes. To investigate the participants’ reflective attitudes we introduce a new experimental paradigm called the Scorekeeping Task, and a Bayesian mixture model tailored to analyze the data. The goal is thereby to identify the participants who follow the Suppositional Theory of condi- tionals and Inferentialism and to investigate their performance on the uncertain and-to-if inference task.

Turn-taking in conversation is a cognitively demanding process that proceeds rapidly due to interlocutors utilizing a range of cues to aid prediction. In the present study we set out to test recent claims that content question words (also called wh-words) sound similar within languages as an adaptation to help listeners predict that a question is about to be asked. We test whether upcoming questions can be predicted based on the first phoneme of a turn and the prior context. We analyze the Switchboard corpus of English by means of a decision tree to test whether /w/ and /h/ are good statistical cues of upcoming questions in conversation. Based on the results, we perform a controlled experiment to test whether people really use these cues to recognize questions. In both studies we show that both the initial phoneme and the sequential context help predict questions. This contributes converging evidence that elements of languages adapt to pragmatic pressures applied during conversation.

Humans make complex inferences on faces, ranging from ob-jective properties (gender, ethnicity, expression, age, identity,etc) to subjective judgments (facial attractiveness, trustworthi-ness, sociability, friendliness, etc). While the objective as-pects of face perception have been extensively studied, rela-tively fewer computational models have been developed forthe social impressions of faces. Bridging this gap, we de-velop a method to predict human impressions of faces in 40subjective social dimensions, using deep representations fromstate-of-the-art neural networks. We find that model perfor-mance grows as the human consensus on a face trait increases,and that model predictions outperform human groups in cor-relation with human averages. This illustrates the learnabilityof subjective social perception of faces, especially when thereis high human consensus. Our system can be used to decidewhich photographs from a personal collection will make thebest impression. The results are significant for the field of so-cial robotics, demonstrating that robots can learn the subjectivejudgments defining the underlying fabric of human interaction.

Existing theories of stereotype change have often made use ofcategorisation principles in order to provide qualitative expla-nations for both the revision and maintenance of stereotypicalbeliefs. The present paper examines the quantitative methodsunderlying these explanations, contrasting both rational andheuristic models of stereotype change using participant dataand model fits. In a comparison of three models each simulat-ing existing descriptions of stereotype change, both empiricaldata and model fits suggest that stereotypes are updated usingrational categorisation processes. This presents stereotype useas a more rational behaviour than may commonly be assumed,and provides new avenues of encouraging stereotype changeaccording to rational principles.

There are a number of claims regarding why linguistic com-plexity varies, for example: i) different types of societalstructure (e.g. Wray & Grace, 2007), ii) population size (e.g.Lupyan & Dale, 2010), and iii) the proportion of child vs. adultlearners (e.g. Trudgill, 2011). This simple model of interact-ing agents, capable of learning and innovation, partially sup-ports all these accounts. However, several subtle points arise.Firstly, differences in the capacity or opportunity to learn deter-mine how much complexity can remain stable. Secondly, smallpopulations are susceptible to large amounts of drift and sub-sequent loss, unless innovation is frequent. Conversely, largepopulations remain resilient to change unless there is too muchinnovation, which leads to a collapse in complexity. Next, ifadult learners are prevalent, we can instead expect less sus-tained complexity in large populations. Finally, creolisationdoes not imply simplification in smaller populations.

Rational accounts of decision-making are incompatible withthe prevalence and success of ubiquitous marketing strategies.In this paper, we demonstrate, using computational experi-ments, how an ideal Bayesian observer model of preferencelearning is compatible with the manipulation of purchasingdecisions via a number of well-known marketing techniques.The ability of this model to predict the effects of both famil-iar and novel marketing interventions suggests it as a plausiblecandidate theory of consumer marketing. Simultaneously, byclarifying the logic underneath the interplay between environ-mental exposure and preference distortions seen in economicdecisions, this model rationalizes the seemingly irrational sus-ceptibility of consumers to marketing.

Spatial language is often used metaphorically to describe other domains, including time (a long sound) and pitch (a high sound). How does experience with these metaphors shape the associations we make across disparate domains? Here, we tested 3- to 6-year-old English-speaking children and adults with a cross-domain matching task that assessed space-time and space-pitch mappings. We tested spatial relations that are expressed in English metaphors for time and pitch, as well as metaphors that are unfamiliar to English speakers, but expressed in other languages. Participants performed a perceptual matching task, in which they matched pictures and sounds, and a linguistic matching task, in which they matched pictures or sounds to verbal labels. Adults readily matched between space and time and between space and pitch, using relations expressed by both familiar and unfamiliar metaphors. Children showed an advantage for linguistic matching compared to perceptual matching, but their performance was similarly unaffected by metaphor familiarity. Together, these results suggest that spatial language promotes the development of cross-domain associations, and that experience with particular spatial metaphors is not required to produce this benefit.

Causal queries about singular cases are ubiquitous, yet thequestion of how we assess whether a particular outcome wasactually caused by a specific potential cause turns out to bedifficult to answer. Relying on the causal power approach,Cheng and Novick (2005) proposed a model of causal attribu-tion intended to help answering this question. We challengethis model, both conceptually and empirically. The centralproblem of this model is that it treats the presence of sufficientcauses as necessarily causal in singular causation, and thus ne-glects that causes can be preempted in their efficacy. Also, themodel does not take into account that reasoners incorporateuncertainty about the underlying causal structure and strengthof causes when making causal inferences. We propose a newmeasure of causal attribution and embed it into our structure in-duction model of singular causation (SISC). Two experimentssupport the model.

Consider the following causal explanation: The ball wentthrough the goal because the defender didn’t block it. Thereare at least two problems with citing omissions as causal ex-planations. First, how do we choose the relevant candidateomission (e.g. why the defender and not the goalkeeper). Sec-ond, how do we determine what would have happened in therelevant counterfactual situation (i.e. maybe the shot wouldstill have gone through the goal even if it had been blocked).In this paper, we extend the counterfactual simulation model(CSM) of causal judgment (Gerstenberg, Goodman, Lagnado,& Tenenbaum, 2014) to handle the second problem. In two ex-periments, we show how people’s causal model of the situationaffects their causal judgments via influencing what counterfac-tuals they consider. Omissions are considered causes to theextent that the outcome in the relevant counterfactual situationwould have been different from what it actually was.

When asked to assess the deductive validity of an argument, people are influenced by their prior knowledge of the content. Recently, two competing explanations for this belief bias effect have been proposed, each based on signal detection theory. Under a response bias explanation, people set more lenient decision criteria for believable than for unbelievable arguments. Alternatively, believable and unbelievable arguments may differ in subjective argument strength for both valid and invalid items. Two experiments tested these accounts by asking participants to assess the validity of categorical syllogisms and rate their confidence. Conclusion- believability was manipulated either within- or between- groups. A two-step signal detection model was applied to examine the effects on the relative location of the decision threshold and the distributions of argument strength. Equivalent belief bias effects were found when believability was manipulated within- and between-groups, supporting the view that the belief bias effect is due to response bias.

Speakers can be taken to be committed to utterance con-tent even when that content is contributed in the scope ofan entailment-canceling operator, like negation (e.g., Chier-chia & McConnell-Ginet, 1990). We develop a probabilisticmodel of this phenomenon, called ‘projection’, that relies onthe prosodic realization of utterances. We synthesize exist-ing theoretical claims about prosody, information structure andprojection into a model that assumes a rational speaker (Frank& Goodman, 2012) who produces utterances with prosodicmelodies that can signal which utterance content she is com-mitted to. Predictions of the probabilistic model are comparedto the responses of an experiment designed to test the effect ofprosody on projection in manner adverb utterances. Key be-haviors of the model are borne out empirically, and the quan-titative fit is surprisingly good given that the model has onlyone free parameter. Our findings lend support to analyses ofprojection that are sensitive to the information structure of ut-terances (e.g., Simons, Beaver, Roberts, & Tonhauser, 2017).

Finger gnosis (the ability to identify which finger has beentouched) and magnitude comparison (the ability to determinewhich of two numbers is larger) are surprisingly correlated.We present a spiking neuron model of a common componentthat could be used in both tasks: an array of pointers. Weshow that if the model's single tuned parameter is set to matchhuman accuracy performance in one task, then it also matcheson the other task (with the exception of one data point). Thisprovides a novel explanation of the relation, and proposes acommon component that could be used across cognitive tasks.

The representation of navigable space, consisting of multiple interconnected spaces, yet is not well understood. Weexamined different levels of integration within memory (local, regional, global). Participants learned two distinctive regions ofa virtual environment that converged at a common transition-point. Subsequently, we tested their memory with a pointing task,varying body alignment during pointing, corridor distance to and regional belonging of the target. Pointing latency increasedwith increasing distance to the target and when pointing into the other region. Further, alignment with local, regional and globalreference frames were found to facilitate pointing latency. These findings suggest that participants memorized local corridors,clustered corridors into regions, and also formed global reference frames, thus, represented the environment on multiple levelsof integration. They are inconsistent with conceptions of spatial memory for navigable environments based either on exclusiverepresentation within a single reference frame or exclusive reliance on local reference frames.

Child-directed speech (CDS) is a talking style adopted by caregivers when they talk to toddlers (Snow, 1995). We consider the role of distributional semantic features of CDS in language acquisition. We view semantic structure as a manifold on which words lie. We compare the semantic structure of verbs in CDS to the semantic structure of child speech (CS) and adult-directed speech (ADS) by measuring how easy it is to align the manifolds. We find that it is easier to align verbs in CS to CDS than to align CS to ADS, suggesting that the semantic structure of CDS is reflected in child productions. We also find, by measuring verbs vertex degrees in a semantic graph, that a mixed initialized set of verbs with high degrees and medium degrees has the best performance among all alignments, suggesting that both semantic generality and diversity may be important for developing semantic representations.

Digital-tablets distribute cognition through visual, auditoryand haptic interactivity. We designed a tutor-game thatexplored how narratives ((S)trong/(W)eak) and gestures((I)conic/(D)eictic) could be combined to situate embodiedlearning. Students played seven levels of a fractions gamedesigned to teach them how to create and compare fractions.One hundred thirty-one students (N=131, age x̄ =8.78 yrs,52.6% Female) were randomly assigned to one of four groups(SI, SD, WI, WD) in a 2x2 factorial experiment. Studentscompleted pre/post direct and transfer assessments and tutor-game log data was mined to explore characteristics ofstudents learning. Results revealed a significant interactionbetween narrative and gesture moderated by studentproficiency. In effect, students new to fractions performedbetter in an abstract environment using deictic (pointing)gestures. However, as students' proficiencies improved, theylearned better using iconically enactive gestures in strongnarrative with setting, characters and a plot. This hasimportant implications for designing adaptive learningplatforms and curricula for teaching fractions.

Educational software based on teachable agents has repeatedly proven to have positive effects on students’ learning outcomes. The strongest effects have been shown for low-performers. A number of mechanisms have been proposed to explore this outcome, in particular mechanisms that involve attributions of social agency to teachable agents. Our study examined whether an expression of high versus low self-efficacy in a teachable agent would affect low- performing students with respect to their learning outcomes and with respect to a potential change in their own self- efficacy. The learning domain was mathematics, specifically the base-ten system. Results were that the learning outcomes of low-performers who taught a low self-efficacy agent were significantly better than the learning outcomes of low- performers who taught a high self-efficacy agent. There were no effects from the manipulation of self-efficacy expressed by the teachable agent on changes of the low-performing students’ own self-efficacy.

The current study manipulated how frequently different prob-lems were practiced during a first day of practice, with themore frequent items being more closely spaced. Fitting thedata to a skill acquisition model, we find that greater spac-ing between items is associated with an increased probabilityof transitioning to more efficient phases of performance, butwith a shallower speedup within each phase. Three days aftertraining, we find that performance is predicted not by the prac-tice frequency during training, but rather by the phase of skillacquisition attained during training. Thus, it is type of pro-cessing achieved not the amount and spacing of practice, thatdetermines retention. Spacing, however, promotes learning bydriving changes in cognitive processing.

Speakers often refer to context only implicitly when using lan-guage. The utterance “it’s warm outside” could signal it’swarm relative to other days of the year or just relative to thecurrent season (e.g., it’s warm for winter). Warm vaguely con-veys that the temperature is high relative to some contextualcomparison class, but little is known about how a listener de-cides upon such a standard of comparison. Here, we formalizehow world knowledge and listeners’ internal models of speechproduction can drive the resolution of a comparison class incontext. We introduce a Rational Speech Act model and de-rive two novel predictions from it, which we validate using aparaphrase experiment to measure listeners’ beliefs about thelikely comparison class used by a speaker. Our model makesquantitative predictions given prior world knowledge for thedomains in question. We triangulate this knowledge with afollow-up language task in the same domains, using Bayesiandata analysis to infer priors from both data sets

It is well known that, all things being equal, the accuracy ofmammalian timing and memory decays gradually with the pas-sage of time. The gradual decay of temporal accuracy is alsoobserved in single-unit neural recordings. Here we review re-cent modeling work describing a specific mechanism for tim-ing and memory and relevant neural data. The model describesa neural mechanism that can give rise to a logarithmically com-pressed representation of the recent past. We examine the spe-cific predictions of the model, in particular that the elapse oftime is represented by sequentially activated cells which firefor a circumscribed period of time. Such cells, called timecells, have been observed in neural recordings from severalbrain regions in multiple species. As predicted by the model,the cells show accuracy that decreases with time.

English-speaking adults recruit a left-to-right “mentaltimeline” (MTL) when thinking about time. The origins of theMTL are debated, with some arguing that it is a culturalconstruct and others arguing that it is rooted in innateassociations between time and space. Here we ask whetherpreschoolers, with limited experience with cultural practicesthought to shape the MTL, prefer conventional linearrepresentations of temporal events. English-speakingpreschoolers and adults were told stories and asked to choosewhich of two visual representations best illustrated the story.As expected, adults overwhelmingly preferred images thatwere linearly ordered from left-to-right. Five-year-olds alsopreferred left-to-right to right-to-left series, but were equallylikely choose left-to-right and top-to-bottom. By contrast,3-year-olds chose at random, apparently insensitive to thespatial ordering of event-denoting images. These resultssuggest that attention to the ordinal structure of visualrepresentations of time increases across early childhood, andthat adults’ preference for horizontal space-time mappingsresults from increased cultural conditioning.

Research on lexical development in Down syndrome (DS) has emphasized a dissociation between language comprehension and production abilities, with production of words being relatively more impaired than comprehension. Current theories stress the role of associative learning on lexical development. However, there have been no attempts to explain the atypical lexical development in DS based on atypical associative learning. The long-term potentiation (LTP) and long-term depression (LTD) of synapses, underlying associative learning, are altered in DS. Here we present a neural network model that instantiates notions from neurophysiological studies to account for the disparities between lexical comprehension and production in DS. Our simulations show that an atypical LTP/LTD balance affects comprehension and production differently in an associative model of lexical development.

According to a classic view of reasoning, intuition is fast but fallible, while reflection is slow but reliable. Biases,therefore, emerge when a reasoner’s intuitions are wrong and they fail to notice. Recent evidence, however, suggests thatpeople may be aware when their intuitions are incorrect. A possible explanation reason for this is that both correct and incorrectresponses are cued in parallel, but the strongly-cued incorrect response is given unless people can inhibit it. We tested thisexplanation using base rate neglect problems, and recorded participants’ mouse cursor movements as they chose betweenpossible answers under time pressure. Descriptions affected both participants’ early movements and ultimate responses, andinterfered with their use of the base rates, while base rates rarely interfered with participants’ use of descriptions, and then onlyat a later point in time. Thus, despite suggestive findings elsewhere, our results support the classic of view reasoning.

Recent results demonstrate that inducing an abstractrepresentation of target analogs at retrieval time aids access toanalogous situations with mismatching surface features (i.e.,the late abstraction principle). A limitation of currentimplementations of this principle is that they either require theexternal provision of target-specific information or demandvery high intellectual engagement. Experiment 1 demonstratedthat constructing an idealized situation model of a targetproblem increases the rate of correct solutions compared toconstructing either concrete simulations or no simulations.Experiment 2 confirmed that these results were based on anadvantage for accessing the base analog, and not merely on anadvantage of idealized simulations for understanding the targetproblem in its own terms. This target idealization strategy hasbroader applicability than prior interventions based on the lateabstraction principle, because it can be achieved by a greaterproportion of participants and without the need to receivetarget-specific information.

Real-time sentence processing involves connecting linguistic inputwith knowledge. Here, we ask how variability in semantic memory(specific domain knowledge) may influence semantic access in real-time sentence processing. We recorded EEG while participantsmore/less knowledgeable about the narrative world of Harry Potter(HP) read sentences. In Experiment 1, all participants showed N400predictability effects for general-knowledge sentences, but onlythose with high HP knowledge showed predictability effects forsentences about Harry Potter. This effect was driven by graded brainresponses to predictable endings as a function of knowledge.Experiment 2 revealed greater semantic activation (inferred fromN400 effects) for HP items participants reported knowing. High-knowledge participants also showed greater semantic activation foritems they reported not knowing/remembering. These findingssuggest that amount and/or functional organization of knowledgehas real-time consequences on written sentence processing andimplicate implicit/partial access to domain knowledge for expertswhen information is not explicitly recalled.

Using referential context in language (e.g., saying “blue pen” when two different-colored pens are visible) makes communication efficient. But it is still unclear which general cognitive processes support the use of context in conversation. Research on pragmatic use in language implicates working memory and inhibitory control; however, no studies have shown evidence of a shared cognitive mechanism in both production and comprehension within an individual. The current study asked a) whether referential context use is supported by the same cognitive mechanisms in production and comprehension, b) which processes are implicated, and c) whether the nature of the context itself affects processing. Participants completed a referential communication eye-tracking task in which a disambiguating adjective was either necessary or over-informative, as well as a cognitive test battery. The results implicated inhibitory control in both production and comprehension (although the comprehension results were more variable), suggesting a shared underlying cognitive mechanism across domains.

We present a dynamic process model for workload, developedaccording to a conducted experiment, which recorded the pupildilation during an air traffic controller simulation. We describehow we built such a dynamic system based on the collecteddata. Logged events that happened in our simulation were usedas system input and the recorded pupil dilation as output. Af-terwards, we used the MATLAB system identification toolboxto identify the transfer function between input and output. Theidentified model is validated with a validation data set that hasbeen excluded from the identification process. Results showthat we are able to explain nearly 50% of the variance of therecorded pupil dilation data in the air traffic controller simu-lation. Moreover, the model explains some contrary results ofthe statistical analysis from our experiment.

When making decisions humans often violate the principlesof rational choice theory. Recent experiments, involving rapidexperiential decisions, uncovered a mechanism that is respon-sible for various rationality violations. According to this se-lective gating mechanism, incoming value samples are accu-mulated across time, but prior to their accumulation they areweighted in proportion to their momentary rank-order. Here,using a data-driven approach, I present a dynamic extensionof this mechanism, which involves potentially asymmetric in-hibition between the inputs. As a result, and contrary to theprevious selective gating implementation, the vigour of gatingis modulated by the difference between two value samples (adistance effect) as well as by the absolute magnitude of thesamples (a magnitude effect). This extension offers a supe-rior explanation to existing and new data; and links high-leveldecision phenomena with computational principles previouslydescribed in theories of selective attention and visual search.

A large literature shows that language influences cognition.Yet, we know very little about when and how linguisticinfluences on cognition become important in development.Here we test the proposal that one pathway by whichlanguage affects cognition is by activating categoryinformation which influences visual processing, and that thisinfluence starts early. Across two experiments, we show thatcategory information affects visual processing and that wordscan activate category information in young children.

Semantic knowledge influences various higher-ordercognitive processes; therefore, it is important to understandhow it changes with development. The Match-to-Sample taskis perhaps the most common paradigm for studying changesin semantic knowledge over development, yet this paradigmhas a number of limitations. Here we provide initial evidencevalidating a Visual Search paradigm as a measure of semanticknowledge in preschoolers, and discuss the potential of thisparadigm to address the limitations posed by the Match-to-Sample task to study semantic knowledge development.

In the triangle of coins problem coins are arranged tocreate a triangle pointing down and the solution involvesmoving a few coins to change its orientation. The taskecology can be designed such that participants can workon it in a low interactivity environment, maintaining amental representation of simulated moves, or in a highinteractivity environment, thinking with and through aphysical model of the problem. These task ecologiesinvolve working memory to a different degree: Problemsolving draws more on working memory the lower thedegree of physical interaction. Participants first engagedin a writing task that required vigilance to inhibitcommon word choices, a degree of self regulationdesigned to induce a so-called ego depletion;participants then worked on the ToC problem in either alow or high interactivity environment. Solution rateswere determined by level of interactivity; the precedingdepletion experience did not impact performance.

How do people plan ahead in sequential decision-makingtasks? In this article, we compare computational models of hu-man behavior in a challenging variant of tic-tac-toe, to inves-tigate the cognitive processes underlying sequential planning.We validate the most successful model by predicting choicesduring games, two-alternative forced choices and board evalu-ations. We then use this model to study individual skill differ-ences, the effects of time pressure and the nature of expertise.Our findings suggest that people perform less tree search un-der time pressure, and that players search more as they improveduring learning.

The lexical frequency of an upcoming word affects read-ing times even when the upcoming word is masked fromreaders (Angele et al., 2015). One explanation for thisobservation is that readers may slow down if there is highuncertainty about upcoming material. In line with thishypothesis, this study finds a positive correlation be-tween predictive entropy and self-paced reading times.This study also demonstrates that such predictive en-tropy can be effectively approximated by the surprisalof upcoming observations and that this future surprisalestimate is more predictive of reading times when thegrammar is more granular, which would be prohibitivelyexpensive for predictive entropy. These results suggestreaders engage in fine-grained predictive estimations ofcertainty about upcoming lexical and syntactic material,that such predictions influence reading times, and thatestimating that uncertainty can be done less expensivelyand more robustly with information-theoretic surprisal.

Representations of social categories help us make sense of thesocial world, supporting predictions and explanations aboutgroups and individuals. Here we explore whether children andadults are able to understanding category-property associationsin structural terms, locating an object of explanation within alarger structure and identifying structural constraints that acton elements of the structure. We show that children as young3-4 years of age show signs of structural thinking, but that thiscapacity does not fully develop until after 7 years of age. Thesefindings introduce a viable alternative to internalist accounts ofsocial categories, such as psychological essentialism.

We show that publishing results using the statistical signif-icance filter—publishing only when the p-value is less than0.05—leads to a vicious cycle of overoptimistic expectationof the replicability of results. First, we show analytically thatwhen true statistical power is relatively low, computing powerbased on statistically significant results will lead to overesti-mates of power. Then, we present a case study using 10 exper-imental comparisons drawn from a recently published meta-analysis in psycholinguistics (J ̈ager et al., 2017). We show thatthe statistically significant results yield an illusion of replica-bility. This illusion holds even if the researcher doesn’t con-duct any formal power analysis but just uses statistical signifi-cance to informally assess robustness (i.e., replicability) of re-sults.

We present a case-study demonstrating the usefulness ofBayesian hierarchical mixture modelling for investigating cog-nitive processes. In sentence comprehension, it is widely as-sumed that the distance between linguistic co-dependents af-fects the latency of dependency resolution: the longer thedistance, the longer the retrieval time (the distance-based ac-count). An alternative theory, direct-access, assumes that re-trieval times are a mixture of two distributions: one distribu-tion represents successful retrievals (these are independent ofdependency distance) and the other represents an initial failureto retrieve the correct dependent, followed by a reanalysis thatleads to successful retrieval. We implement both models asBayesian hierarchical models and show that the direct-accessmodel explains Chinese relative clause reading time data betterthan the distance account.

We cannot see others’ mental states, so we infer them by watching how people behave. Bayesian inference in a model of rational action – called inverse planning – captures how humans infer desires from observable actions. These models represent desires as simple associations between agents and world states. In this paper we show that by representing desires as probabilistic programs, an inverse planning model can infer complex desires underlying complex behaviors—desires with temporal and logical structure, which can be fulfilled in different ways. Our model, which combines basic desires via logical primitives, is inspired by recent probabilistic grammar- based models of concept learning. Through an experiment where we vary behaviors parametrically, we show that our model predicts with high accuracy how people infer complex desires. Our work sheds light on the representations underlying mental states, and paves the way towards algorithms that can reason about others’ minds as we do.

English-speakers whose access to number language isartificially compromised by verbal interference and the Pirahã(an Amazonian tribe without exact number words) appear torely on analog magnitude estimation for representing non-symbolic exact quantities greater than 3. Here, 16 participantswith aphasia performed the 5 counting tasks from theseprevious studies. Performance was poorest when targets werenot visible during response (70% correct, task 4; 71% correct,task 5) and best when targets were presented as subitizablegroups of 2 and 3 (98% correct, task 2). Western AphasiaBattery-Revised subtest scores correlated with taskperformance, suggesting diverse forms of languageimpairment may contribute to errors. Coefficients of variationfor tasks and significant correlations of target magnitude witherror rate (r 2=.88) and error size (r 2=.87) across tasks suggestparticipant use of analog magnitude estimation. Experimentsinvolving people with aphasia may further refine ourunderstanding of how language and thought interact.

Humans constantly receive sensory input from several sensorymodalities. Via the process of multisensory integration, this inputis often integrated into a unitary percept. Researchers haveinvestigated several factors that could affect the process ofmultisensory integration. However, in this field of research, socialfactors (i.e., whether a task is performed alone or jointly) havebeen widely neglected. Using an audiovisual crossmodalcongruency task we investigated whether social factors affectaudiovisual integration. Pairs of participants received congruent orincongruent audiovisual stimuli and were required to indicate theelevation of these stimuli. We found that the reaction time cost ofresponding to incongruent stimuli (relative to congruent stimuli)was reduced significantly when participants performed the taskjointly compared to when they performed the task alone. Theseresults extend earlier findings on visuotactile integration byshowing that audiovisual integration is also affected by socialfactors.

The ability to represent same-different relations is a condition for abstract thought. However, there is mixed evidence for when this ability develops, both ontogenetically and phylogenetically. Apparent success in relational reasoning may be evidence for conceptual understanding or may be due to low-level, perceptual strategies. We introduce a method to discriminate these possibilities by pitting two conditions that are perceptually matched but conceptually different: in a “fused” condition, same and different objects are joined, creating single objects that have the same perceptual features as the pairs in the “relational” condition. However, the “fused” objects do not provide evidence for the relation. Using this method in a causal task provides evidence for genuine conceptual understanding. This novel technique offers a simple manipulation that may be applied to a variety of existing match-to-sample procedures used to assess same-different reasoning to include in future research with non-human animals, as well as human infants.

Learning syntax requires determining relations between the grammatical categories of words in the language, but learning those categories requires understanding the role of words in the syntax. In this study, we examined how this chicken and egg problem is resolved by learners of an artificial language comprising nouns, verbs, adjectives and case markers following syntactic rules. We found that the language could be acquired through cross-situational statistical correspondences with complex scenes and without explicit feedback, and that knowledge was maintained after 24 hours. Results also showed that verbs and word order were the first to be acquired, followed by nouns, adjectives and finally case markers. Interdependencies in learning were found for word order and verbs, and also for nouns, adjectives and case markers. Grammar and vocabulary can be acquired simultaneously, but with distinctive patterns of acquisition – grammar and the role of verbs first, then the vocabulary of other lexical categories.

Radical Enactivism is a position in the philosophy of cognitive science that aims to displace representationalism, the dominant position in cognitive science for the last 50-60 years. To accomplish this aim, radical enactivism must provide an alternative explanation of cognition. Radical enactivism offers two alternative explanations of cognition. The first I call the dynamical explanation and the second I call the historical explanation. The mechanists have given us reasons for doubting that the first alternative makes for a good explanation. The historical explanation does not hit the right explanatory target without the introduction of a proximate mechanism, but the proximate mechanisms suggested by radical enactivism are associationist mechanisms, the limitations of which led to the initial widespread endorsement of representationalism. Therefore, radical enactivism cannot displace representationalism in cognitive science.

In recent years deep reinforcement learning (RL) systems have attained superhuman performance in a number ofchallenging task domains, but are constrained by a demand for large training sets. A critical present objective is thus to developdeep RL methods that can adapt rapidly to new tasks. In the present work we introduce a novel approach to this challenge,which we refer to as deep meta-reinforcement learning. Previous work has shown that recurrent networks can support meta-learning in a fully supervised context. We extend this approach to the RL setting. What emerges is a system that is trainedusing one RL algorithm, but whose recurrent dynamics implement a second, quite separate RL procedure. This second, learnedRL algorithm can differ from the original one in arbitrary ways and exploit structure in the training domain. We unpack thesepoints in five proof-of-concept experiments to examine key aspects of deep meta-RL.

In three experiments, adults attempted to learn words with subordinate-level meanings (dalmatian) by sampling thereferent world cross-situationally. Xu & Tenenbaum, 2007 predicted that encountering three uses of a word, each referringto a dalmatian would evoke “suspicious coincidence” inferencing, leading to the subordinate meaning (dalmatian). Exp. 1found little evidence for this; cross-situational exposure led to a basic-level bias. This bias was unchanged even when thesample was increased to five subordinate exemplars (Exp. 2). Exp. 3 encouraged semantic contrast by simultaneously teachingeach subject a word for the subordinate-level and the basic-level category within the same semantic domain (dap=dalmatian;blit=dog). Participants now showed non-basic level learning, but more in line with mutual exclusivity: they may think “dap”means dalmatian but “blit” means all-dogs-except-dalmatians. We conclude that the basic-level interpretation is powerful andcannot be removed by the mere observation of exemplar items over multiple word instances.

By integrating theories and methodologies from a diverserange of scientific disciplines (e.g., physics, neuroscience,cognitive science, psychology and robotics engineering) thepresent work is aimed at harnessing self-organizedanticipatory synchronization in order to advance human-robotic interaction (HRI). This phenomenon is characterizedby the emergence of anticipatory behavior by one systemcoupled to the chaotic behavior of another, following theintroduction of short self-referential delays in the coordinatingsystem. The current set of studies involved the creation of anartificial agent based on a time-delayed, low-dimensionaldynamical model capable of behaving prospectively during aninteraction with a human actor performing complex,unpredictable behaviors. By achieving characteristics similarto those observed during natural human interaction andcoordination, the time-delayed modeling approachedadvocated here provides the potential for considerable futureadvancements in HRI.

Motor adaptation displays a structure-learning effect: adapta-tion to a new perturbation occurs more quickly when the sub-ject has prior exposure to perturbations with related structure.Although this ‘learning-to-learn’ effect is well documented, itsunderlying computational mechanisms are poorly understood.We present a new model of motor structure learning, approach-ing it from the point of view of deep reinforcement learning.Previous work outside of motor control has shown how recur-rent neural networks can account for learning-to-learn effects.We leverage this insight to address motor learning, by import-ing it into the setting of model-based reinforcement learning.We apply the resulting processing architecture to empiricalfindings from a landmark study of structure learning in target-directed reaching (Braun et al., 2009), and discuss its implica-tions for a wider range of learning-to-learn phenomena.

We investigated children’s understanding of mental life byanalyzing attributions of perceptual, cognitive, affective, andother capacities. 200 children (7-9y) and 200 adults evaluatedthe mental capacities of beetles or robots. By assessing whichcapacities traveled together when participants disagreed aboutthese controversial “edge cases,” we reconstructed the latentstructure underlying mental capacity judgments from thebottom up—a novel approach to elucidating conceptualstructure among children. For both children and adults, factoranalyses revealed a distinction between social-emotional,physiological, and perceptual-cognitive capacities, hinting atthree fundamental ways of explaining and predicting others’actions: as social partners, biological creatures, and goal-directed agents (each involving related forms of both“experience” and “agency”; Gray et al., 2007). Relative toadults, children attributed greater social-emotional capacitiesto beetles and robots, suggesting that intuitive ontologies ofmental life could be critical for making sense of children’sdeveloping understanding of the social world.

Previous work has demonstrated that the visual complexity ofletter-shapes is processed differently by naïve and expertobservers. Specifically, fluent readers of the Arabic alphabetwere found to discriminate complex letters more readily thanless complex letters, whereas naïve observers exhibited theopposite effect. This “complexity benefit”, wherein complexletters confer a processing advantage to expert observers, isnot yet well understood. In a new study, we investigatewhether this effect generalizes across scripts, and whether itis unique to individuals with biscriptal experience (knowledgeof reading two different scripts). The results of the threeexperiments confirm that the complexity benefit ischaracteristic of expert monoscriptal and biscriptal readers,and that, furthermore, there may be a biscriptal advantage inprocessing visual complexity.

During collaborative problem solving (CPS), coordination occurs at different spatial and temporal scales. This multiscale coordination should, at least on some scales, play a functional role in facilitating effective collaboration outcomes. To evaluate this, we conducted a study of computer-based CPS with 42 dyads. We used cross-wavelet coherence as a way to examine the degree to which movement coordination is evident at a variety of scales and tested whether the observed coordination was greater than both the amount expected due to chance and due to task demands. We found that coordination at scales less than 2s was greater than expected due to chance and at most scales (except 16s, 1m, and 2m) was greater than expected due to task demands. Lastly, we evaluated whether the degree of coherence at scales less than 2s, and the form of coordination (in terms of relative phase), were predictive of CPS performance. We found that .25s and 1s scales were predictive of performance. When including relative phase, our results suggest that higher in-phase movement coordination at the 1s scale was the strongest predictor of CPS performance. We discuss these findings and detail their relevance for expanding our knowledge on how coordination facilitates CPS.

When choosing between multiple alternatives, people usuallydo not have ready-made preferences in their mind but ratherconstruct them on the go. The 2N-ary Choice Tree Model(Wollschlaeger & Diederich, 2012) proposes a preference con-struction process for N choice options from description, whichis based on attribute weights, differences between attribute val-ues, and noise. It is able to produce similarity, attraction,and compromise effects, which have become a benchmark formulti-alternative choice models, but also several other contextand reference point effects. Here, we present a new and math-ematically tractable version of the model – the Simple ChoiceTree Model – which also explains the above mentioned effectsand additionally accounts for the positive correlation betweenthe attraction and compromise effect, and the negative correla-tion between these two and the similarity effect as observed byBerkowitsch, Scheibehenne, and Rieskamp (2014).

We introduce the spatially correlated multi-armed banditas a task coupling function learning with the exploration-exploitation trade-off. Participants interacted with bi-variatereward functions on a two-dimensional grid, with the goal ofeither gaining the largest average score or finding the largestpayoff. By providing an opportunity to learn the underly-ing reward function through spatial correlations, we modelto what extent people form beliefs about unexplored payoffsand how that guides search behavior. Participants adapted toassigned payoff conditions, performed better in smooth thanin rough environments, and—surprisingly—sometimes per-formed equally well in short as in long search horizons. Ourmodeling results indicate a preference for local search options,which when accounted for, still suggests participants werebest-described as forming local inferences about unexploredregions, combined with a search strategy that directly tradedoff between exploiting high expected rewards and exploring toreduce uncertainty about the spatial structure of rewards.

We investigate children’s ability to use social display rules toinfer agents’ otherwise under-determined desires. InExperiment 1, seven-to-ten-year-olds saw a protagonistexpress one emotional reaction to an event in front of hersocial partner (the Social Context), and a different expressionbehind her social partner’s back (the Nonsocial Context).Children were able to use the expression in the Social Contextto infer the social partner’s desire and the expression in theNonsocial Context to infer the protagonist’s desire. Thisability increased between ages seven and ten (Experiment 1).When task demands were reduced (Experiment 2), seven-to-eight-year-olds, but not five-to-six-year-olds, succeeded.These results suggest that although it is not easy for observersto infer emotions masked by social display rules, changingemotional expressions between social and non-social contextsallow even children to recover not only the desire of theperson displaying the emotions, but also that of the audience.

Human interaction involves the organization of a collection ofsensorimotor systems across space and time. The study ofhow coordination develops in child-parent interaction hasprimarily focused on understanding the development ofspecific coordination patterns from individual modalities.However, less work has taken a systems view andinvestigated the development of coordination among multipleinterdependent behaviors. In the present work, we usedGranger causality as a mathematical model to constructdyadic causal networks of multimodal data collected from alongitudinal study of child-parent interaction. At a group-level, we observed increases in the number of causal links andin the strength of such links in dyadic interaction from 9-months to 12-months. At an individual-level, we observedhigh variability in the types of causal links that emergedacross developmental ages. We discuss these results in termsof a multicausality hypothesis for the development of humancoordination.

The enormous scale of the available information and productson the Internet has necessitated the development of algorithmsthat intermediate between options and human users. These al-gorithms do not select information at random, but attempt toprovide the user with relevant information. In doing so, thealgorithms may incur potential negative consequences relatedto, for example, “filter bubbles.” Building from existing al-gorithms, we introduce a parametrized model that unifies andinterpolates between recommending relevant information andactive learning. In a concept learning paradigm, we illustratethe trade-offs of optimizing prediction and recommendation,show that there is a broad parameter region of stable perfor-mance that optimizes for both, identify a specific regime thatis most robust to human variability, and identify the cause ofthis optimized performance. We conclude by discussing im-plications for the cognitive science of concept learning and thepractice of machine learning in the real world.

Human perceptual grouping of sequential auditory cues hastraditionally been modeled using a mechanistic approach. Theproblem however is essentially one of source inference – aproblem that has recently been tackled using statisticalBayesian models in visual and auditory-visual modalities.Usually the models are restricted to performing inference overjust one or two possible sources, but human perceptualsystems have to deal with much more complex scenarios. Tocharacterize human perception we have developed a Bayesianinference model that allows an unlimited number of signalsources to be considered: it is general enough to allow anydiscrete sequential cues, from any modality. The model uses anon-parametric prior, hence increased complexity of thesignal does not necessitate more parameters. The model notonly determines the most likely number of sources, but alsospecifies the source that each signal is associated with. Themodel gives an excellent fit to data from an auditory streamsegregation experiment in which the pitch and presentationrate of pure tones determined the perceived number ofsources.

We outline a simple way of representing sets of non-normativejudgements that makes them look as similar as possible to nor-mative ones. This representation allows us to view certaintypes of non-normative judgments, such as conjunction falla-cies, as arising from a misestimation of the correlation betweenevents, that might arise when decision-makers have no priorinformation about the frequency of co-occurrence. We sug-gest that decision-makers use the perceived similarity betweenevents to make inferences about correlation, and we describethe results of an experiment showing that judged correlationand violations of independence in probabilistic judgments arestrongly influenced by the perceived similarity between events.

From a quick glance or the touch of an object, our brains map sensory signals to scenes composed of rich anddetailed shapes and surfaces. Unlike the standard approaches to perception, we argue that this mapping draws on internalcausal and compositional models of the physical world and these internal models underlie the generalization capacity of humanperception. Here, we present a generative model of visual and multisensory perception in which the latent variables encodeintrinsic (e.g., shape) and extrinsic (e.g., occlusion) object properties. Latent variables are inputs to causal models that outputsense-specific signals. We present a recognition network that performs efficient inference in the generative model, computingat a speed similar to online perception. We show that our model, but not alternatives, can account for human performance in anoccluded face matching task and in a visual-to-haptic face matching task.

In risky decision-making, expected utility (EU) theory is widely used to examine people's risk attitude and choice behavior. However, it is unknown how risk attitude relates to attention and information search. In this paper, we explore the relationship between risk attitude (as measured by a variant of EU) and eye movement patterns (which serve as a proxy for attention and information search). Participants made choices between gambles presented perceptually as flickering grids in which monetary values were indicated by colors and probabilities by color proportions. To explore attention and information search patterns, we investigated eye movement patterns when faced with different gambles and correlated these patterns with the parameters of EU. We observed that people who are more risk-seeking (as determined by modeling) tend to look at risky options more often. These results bridge choice behaviors conceptualized by EU and information search strategies under risky decision-making revealed by eye movements.

Bilinguals’ need to monitor and inhibit non-relevantlanguages over a relevant one confers advantage in cognitivecontrol. No studies have demonstrated that the dual-languagecontrol process directly contributes to the bilingual cognitiveadvantage. We utilized a novel language control manipulationparadigm where 83 English-Chinese bilingual adultscompleted a reading and comprehension task in either single-language (low-language-control) or dual-language (high-language-control) prior to performing nonverbal executivecontrol tasks (Stroop, task-switching, and n-back). Resultsshowed that language control had significant effects onsubsequent cognitive performance, depending on whether theparticipants were regular dual language users or not. In thedual-language condition, but not the single-languagecondition, participants who used both languages regularlydemonstrated a smaller mixing cost in task-switching and agreater sensitivity in n-back detection compared toparticipants who did not. This suggests that dual languagecontrol utilizes similar resources as executive function andfrequent dual language use enhances this resource.

Biased sampling of participants presents a major limiting factor for the generalizability of findings from behavioral studies. This effect may be especially pronounced in developmental studies, where parents serve as both the primary environmental input and decide whether their child participates in a study. To estimate the effects of parental non-consent, we coupled naturalistic observations of parent-child interactions with a behavioral test. Results showed that one particular parenting practice, the tendency to use questions to teach, associated with both children’s behavior in the test and parents’ tendencies to participate. Exploiting these associations with a model-based multiple imputation, we estimated that the means of the consented and not-consented groups could differ as much as 0.2 standard deviations for five of the seven test measurements we used, and standard deviations are likely underestimated. These results suggest that ignoring the role of consent may lead to systematic biases when generalizing beyond lab samples.

Sustained visual attention is crucial to many developmentaloutcomes. We demonstrate that, consistent with thedevelopmental systems view, sustained visual attentionemerges from and is tightly tied to sensory motor coordination.We examined whether changes in manual behavior altertoddlers’ eye gaze by giving one group of children heavy toysthat were hard to pick up, while giving another group ofchildren perceptually identical toys that were lighter, easy topick up and hold. We found a tight temporal coupling betweenthe dynamics of visual attention and the dynamics of manualactivities on objects, a relation that cannot be explained byinterest alone. In the Heavy condition, toddlers looked atobjects just as much as did toddlers in the Light condition butdid so through many brief glances, whereas in Light conditionlooks to the objects were longer and sustained. We discuss theimplication of hand-eye coordination in the development ofvisual attention.

It is often assumed that the socio-cultural context positively influences mindreading performances. Among the available theories, mindshaping is proposed to consist of cultural mechanisms that make the social domain homogeneous and, hence, easier to interpret. Proponents of the mindshaping hypothesis claim that homogeneity is responsible for the computational tractability of mindreading, which is otherwise intractable. In this paper, we examine this core claim of mindshaping and investigate how homogeneity influences mindreading tractability. By taking action understanding as a case-study for mindreading, we formally operationalize mindshaping homogeneity in different ways with the goal of bridging the gap between informal claims and formal (in)tractability results. The analysis shows that only specific combinations of homogeneity may lead to tractable mindreading, whilst others do not. Additionally, the analysis reveals the possibility of a yet undiscovered mindshaping mechanism.

Computerised paradigms have enabled decision making re-searchers to gather rich data on human behaviour, includinginformation on motor execution of a decision, e.g., by track-ing mouse cursor trajectories. As the number and complexityof mouse-tracking studies rapidly increase, more sophisticatedmethodology is needed to analyse the decision trajectories.Here we present a new computational approach to generat-ing decision landscape visualisations based on mouse-trackingdata. Decision landscape is an analogue of energy potentialfield mathematically derived from velocity of mouse move-ment during a decision. Visualised as a 3D surface, it pro-vides a comprehensive overview of motor evolution of deci-sions. Employing the dynamical systems theory framework,we develop a new method for generating decision landscapesbased on arbitrary number of trajectories. The decision land-scape visualisation have potential to become a novel tool foranalysing mouse trajectories during decision execution, whichcan provide new insights into the dynamics of decision mak-ing.

Insomniacs were found to have compromised perception of facial expressions. Through eye movement examinations, here we test the hypothesis that this effect is due to impaired visual attention functions for retrieving diagnostic features in facial expression judgments. 23 individuals with insomnia symptoms and 23 non-insomniac controls completed a task to categorize happy, sad, fearful, and angry faces. The insomniacs were less accurate to recognize angry faces and made more “fearful” mistakes than controls. A hidden Markov modeling approach for eye movement data analysis revealed that when recognizing angry faces, more insomniacs adopted an eye movement pattern focusing on the mouth while more controls adopted a pattern attending to both the eyes and the mouth. This result is consistent with previous findings that the primary diagnostic feature for recognizing angry faces is the eyes suggesting that impaired information selection through visual attention control may account for the compromised emotion perception in insomniac individuals.

The study of multi-cue judgment investigates how decisionmakers integrate cues to predict the value of a criterionvariable. We consider a multi-cue judgment task in whichdecision makers have prior knowledge of inter-cuerelationships but are ignorant of how the cues correlate withthe criterion. In this setting, a naive judgment strategyprescribes an equal weight for each cue. However, we findthat many participants appear to use a weighting schemebased on a low-dimensional representation of the cue space.The use of such a representation is consistent with coreinsights in semantic memory research and has importantoptimality properties concerning judgment accuracy.

There is long history of studies documenting that some neurons respond to images of objects, faces, and scenes in a highly selective manner. This includes neurons in the human hippocampus (e.g., the famous example of a neuron responding to images of the actress Jennifer Aniston) and neurons in high-level visual cortex in monkey (for reviews see Bowers, 2009; Ison, Quian Quiroga, & Fried, 2015). These findings have led to a growing interest in the claim that some neurons code for information in a localist (‘grandmother cell’) manner, as reflected in the many contributions to a recent special issue on this topic in the journal Language, Cognition, & Neuroscience (Bowers, 2017). By contrast, it is only recently that interest in characterizing the selectivity of single units in connectionist networks has gathered speed. Critically, these studies also show that networks learn highly selective representations under a number of conditions, as detailed below. In this talk I will summarize recent research in my lab that explores the conditions in which artificial networks learn selective codes, and research comparing the responses of selective neurons and localist representations used in cognitive models. These findings suggest when and why some neurons in cortex respond in a highly selective manner, and highlight the biological plausibility of localist models in psychology.

Concepts are the basis of the human cognitive system, andthe question of what constitutes the content of these mentalrepresentations has long occupied the cognitive sciences.Work in psychology, linguistics and cognitive neurosciencehas converged on the idea that we develop our conceptualrepresentations through our perception of and interactionwith our environment. To date, such research has typicallyrestricted consideration to the perceptual modalities ofvision, touch, sound, taste, and smell. However, there isanother major modality of perceptual information that isdistinct from these traditional five senses; that is,interoception, or sensations within the body. In this paper,we explore the role of interoception in the perceptualgrounding of concepts.Recently, modality-specific measures of the strength ofperceptual experience (Lynott & Connell 2009, 2013) haveproven themselves important predictors of human behaviourin a range of conceptual tasks including word recognitionand reading (Connell & Lynott, 2010, 2012, 2014a, 2014b,2015, 2016). In a megastudy of over 32,000 words fromacross the abstract-concrete spectrum, we asked people toprovide modality-specific ratings of perceptual strength forsix modalities: the usual five (auditory, haptic, gustatory,olfactory, visual) plus the new category of interoceptivestrength. We found that interoceptive information dominatesthe perceptual profile of a sizeable number of concepts (9%;e.g., hangover, eternal, remorse), less than the proportion ofconcepts dominated by vision (74%; e.g., book) or sound(12%; e.g., melody), but more than are dominated by touch(3%; e.g., silky), gustation (2%; e.g., candy), or olfaction(<1%; e.g., bleach). Using principal components analysis toexamine how interoception relates to the other perceptualmodalities, we found that it tends to be strongly loadedagainst visual and haptic strength (i.e., that which is sensedwithin the body can be neither seen nor touched) but isrelatively distinct from sound, taste, and smell.Finally, we tested whether interoceptive strength offersvaluable information to conceptual content by examining itsrole in semantic facilitation of word recognition. Maximumperceptual strength (i.e., strength in the dominant modality)has previously been shown to predict word recognitionperformance better than concreteness or imageability(Connell & Lynott, 2012). We therefore compared thepredictive ability of two different versions of maximumperceptual strength: the original measure based on fivetraditional modalities, and a new version based on sixmodalities including interoceptive strength. In a regressionanalysis of lexical decision and word naming performance,interoceptive information considerably improved theefficacy of maximum perceptual strength in predicting bothresponse time and accuracy (Bayes Factors ranged fromBF 10 = 3.303×10 7 to BF 10 = 3.059×10 16 ). That is,perceptually strong words were recognized more quicklyand accurately than perceptually weak words, andinteroceptive strength was a valuable component in thisperceptual facilitation. Overall, these findings suggest thatinteroception has comparable status to other modalities incontributing to the perceptual grounding of concepts.

Notions of entropy and uncertainty are fundamental tomany domains, ranging from the philosophy of science tophysics. One important application is to quantify theexpected usefulness of possible experiments (or questions ortests). Many different entropy models could be used;different models do not in general lead to the sameconclusions about which tests (or experiments) are mostvaluable. It is often unclear whether this is due to differenttheoretical and practical goals or are merely due to historicalaccident. We introduce a unified two-parameter family ofentropy models that incorporates a great deal of entropies asspecial cases. This family of models offers insight intoheretofore perplexing psychological results, and generatespredictions for future research.

Word meaning priming has become a key method to studyhow listeners (and readers) retune their lexical semanticrepresentations in response to their linguistic environment inorder to facilitate access to word meanings. We present asummary of recent findings using this method that help toconstrain our theories of how this important form of lexical-semantic learning occurs.

Implementing interventions that are supposed to enhance students’general learning skill and overall cognitive ability is still acommon practice in education. The basic idea on which thisapproach relies is that improving domain-general skills providesbenefits for a broad range of domain-specific areas, such asacademic disciplines. Thus, it is assumed that there is far transfer –i.e., the generalization of a set of skills between domains looselyrelated to each other. In recent years, chess instruction, musicinstruction, and working memory training have been claimed to beable to train domain-general abilities (e.g., fluidreasoning/intelligence) which, in turn, generalize to other cognitiveand academic skills (e.g., mathematics). We tested these claims inthe population of healthy children via meta-analysis. The resultsshowed small to moderate overall far-transfer effects in all theoutcome measures of the three meta-analyses. However, the effectsizes were inversely related to the design quality (e.g., presence ofactive control groups), which casts doubts on the effectiveness ofthe three activities. We discuss the theoretical and practicalimplications of these findings for education and expertise andextend the debate to another type of training, video games training.

There is contradictory evidence on whether speech production gets impaired or enhanced when people are restrained from gesturing. There is also very little research on how this effect can change with aging. The present study sought evidence for these by asking young and elderly adults to describe two different routes on a map in spontaneous speech and when gestures were prohibited. We found that elderly adults produced more spatial language when they were restricted to use gestures compared to their spontaneous speech, whereas young adults produced comparable levels of spatial language in both conditions. Young and elderly adults used comparable levels of gestures in their spontaneous route descriptions. Yet, only young adults’ gesture use correlated positively with their spatial language production. Thus, the results of gesture prohibition on speech production are different for young and elderly adults.

Subjective duration estimates are positively related to the magnitude of various non-temporal stimuli (e.g. Xuan et al., 2007). Our study investigated whether temporal and spatial magnitude information conveyed by linguistic stimuli would affect perceived duration in a temporal reproduction task. We used time-related words referring to different exact durations (e.g. second; Experiment 1), and spatial-temporal metaphors (e.g. long), referring to indistinct temporal as well as spatial magnitudes (Experiment 2). In both experiments, participants over-reproduced the shorter target duration (2.4 s) and under- reproduced the longer target duration (4.8 s). In Experiment 1, participants under-reproduced the longer target duration more when they saw “week” in the training and “year” in the reproduction. Yet, we did not observe the same semantic magnitude effect in other word pairs either in Experiment 1 or 2. Overall, we did not find supporting evidence for magnitude information conveyed by language affecting subjective time estimates.

Eye tracking research on situated language comprehension hasshown that participants rely more on a recent event than on aplausible future event during spoken sentence comprehension.When people saw a recent action event and then they listenedto a German (NP1-Verb-Adv-NP2) past or futuric present tensesentence, they preferentially looked at the recent event targetover another plausible target object (that might be involvedin a future action) independent of tense. This preferential in-spection persisted even when future events and futuric presentsentences were much more frequent within the experiment,or when a gaze cue biased towards the future action target.The present experiments extend this line of research by intro-ducing incongruence (in Experiment 1 a past tense verb mis-matched the recently seen action and in Experiment 2 an actorgaze cue mismatched the past tense sentence condition). Canthe verb-action and the gaze-sentence mismatches eliminatethe recent-event inspection preference? Would participants re-call information in post-experimental memory tests better formatches (the futuric present tense condition) than mismatches(the past tense condition)? Results revealed inspection of therecent event target as participants processed the verb-actionmismatch (Exp 1) and actor gaze incongruence (Exp 2). How-ever, the gaze (but not the verb-action) incongruence elimi-nated the overall recent event preference in the NP2 region.The memory tests also showed some evidence for a reversal ofthe recent-event preference.

The study of the coordination of attention, a term called jointattention (JA), has resulted in a better understanding of thedynamics and development of communication. Despite theimportant insights gained from studying JA, there is littleconsensus regarding the specific components that are included inoperationalizing JA. The present work explored a parameter spaceof JA during a dyadic naturalistic toy play task between 9-month-old infants and their parents. We systematically measured thetemporal properties of two components commonly used tooperationalize JA: the duration of continuous alignment of parentand infant visual fixations and the flexibility of fluctuations ofattention. The results show that very brief bouts of JA areimportant predictors for vocabulary development. The results fromthis work provide new insights into the specific properties used tooperationalize JA and point to the importance of consideringmultiple timescales of behavior that make up JA.

For a language learner, any new word is a pseudoword. Apseudoword is a string of of letters or phonemes that soundslike an existing word in a language, though it has no meaningin the lexicon. On the other hand, speakers are well aware ofpermissible phonemes, their frequencies and collocations intheir language due to the phonotactics inherent in thelanguage. For example, saktal is a pseudoword in Turkish,whereas szyan is not, due to Turkish phonotactics. This studyinvestigates the relationship between pseudoword letterformation and eye movement characteristics in reading. Inparticular, we examine the role of Turkish vowel harmony,middle-word consonant collocation, and word-initial andword-final consonants on eye movements with adult nativespeakers reading sentences that involve predesigned Turkishpseudowords. The results of an experiment with 34participants are indicative of the role of pseudowordformation on a set of eye movement parameters.

In both adults and school-age children, volitional controlover the presentation of stimuli during study leads to en-hanced recognition memory. Yet little is known abouthow very young learners choose to allocate their timeand attention during self-directed study. Using a recog-nition memory task, we investigate self-directed study inlow-income preschoolers, who are at an age when atten-tion, memory, and executive function skills rapidly de-velop and learning strategies emerge. By pre-exposingchildren to some items before self-directed study, weaimed to discover how familiarity modulates their studystrategies. We found that children showed a preferencefor studying pre-exposed items. Overall, items stud-ied longer led to increased recognition of those items attest. We also compared recognition task performanceand strategies with measures of cognitive control skills,finding that children’s selective attention skills supportrecognition performance. These findings may informboth theory and educational intervention.

Recent Machine Learning systems in vision and languageprocessing have drawn attention to single-word vector spaces,where concepts are represented by a set of basic features orattributes based on textual and perceptual input. However,such representations are still shallow and fall short fromsymbol grounding. In contrast, Grounded Cognition theoriessuch as CAR (Concept Attribute Representation; Binder etal., 2009) provide an intrinsic analysis of word meaning interms of sensory, motor, spatial, temporal, affective andsocial features, as well as a mapping to corresponding brainnetworks. Building on this theory, this research aims tounderstand an intriguing effect of grounding, i.e. how wordmeaning changes depending on context. CAR representationsof words are mapped to fMRI images of subjects readingdifferent sentences, and the contributions of each worddetermined through Multiple Linear Regression and theFGREP nonlinear neural network. As a result, the FGREPmodel in particular identifies significant changes on theCARs for the same word used in different sentences, thussupporting the hypothesis that context adapts the meaning ofwords in the brain. In future work, such context-modifiedword vectors could be used as representations for a naturallanguage processing system, making it more effective androbust.

Social media platforms provide a source for transmittinginformation that can become widely accepted. However, inthis process of transmission, information becomes susceptibleto distortion. In this study, we assessed people’s semantic(i.e., prior expectations) and recognition memory for popculture content, as a function of confidence and perceivedinformation source. In Experiment 1, we investigatedsemantic memory for ubiquitous movie quotes (e.g., thefamous Star Wars quote “Luke I am your father”). Notablythis quote is incorrect, but we found that a majority ofparticipants accepted these lure quotes as true with highconfidence and indicated they had experienced the originalsource. In Experiment 2, participants viewed the originalmovie sources before a recognition test of the quotes. Wefound that while there was some improvement, people stillpreferred the lure quote with high confidence. We discuss thefindings in terms of the strength of people’s prior expectationswhen reconstructing events from memory.

The paper folding task is commonly used for the evaluation ofnonverbal, spatial reasoning skills. In this paper, we presenta computational model that attempts to use visual-imagery-based representations and operations to solve this task. Themodel was tested against all problems from the standard pa-per folding task and achieved a perfect score, illustrating thatvisual-imagery-based representations and operations are suf-ficiently expressive to capture at least one successful solutionstrategy. Although the model does not closely resemble humancognitive processing, and thus should not be considered in itscurrent form to be a plausible psychological model of humantask performance, the assumptions made and their implicationsfor our understanding of human cognition on the paper foldingtask point to fruitful lines of future work towards this goal.

In psychology, we tend to follow the general logic of fal-sificationism: we separate the ‘context of discovery’ (howwe come up with theories) from the ‘context of justification’(how we test them). However, when studying human interac-tion, separating these contexts can lead to theories with lowecological validity that do not generalize well to life outsidethe lab. We propose borrowing research practices from for-mal inductive methodologies during the process of discover-ing new regularities and analyzing natural data without beingled by theory. From the perspective of experimental psychol-ogy, this approach may appear similar to the ‘questionable re-search practice’ of HARKing (Hypothesizing After The Re-sults are Known). We argue that a carefully constructed formof HARKing can be used systematically and transparently dur-ing exploratory research and can lead to more robust and eco-logically valid theories.

We present the Retention and Recognition model (R&R), aprobabilistic exemplar model that accounts for segmentationin Artificial Language Learning experiments. We show thatR&R provides an excellent fit to human responses in threesegmentation experiments with adults (Frank et al., 2010),outperforming existing models. Additionally, we analyze theresults of the simulations and propose alternative explanationsfor the experimental findings.

Scandinavian languages like Swedish employ tonal accent as a lexical phonological feature, where suprasegmentalinformation can be the sole factor differentiating between words. Using cross-modal semantic fragment priming we testedthe following: (a) Do monolingual speakers of Swedish use tonal accent information during lexical access? (b) Do bilingualspeakers, who grew up with one tonal (Swedish) and one non-tonal language, treat this feature the same way as monolinguals?Our results show that for monolinguals, accent mispronunciations eliminate priming effects, implying that tone is used duringlexical access. For bilinguals, by contrast, mispronunciation sensitivity depends on both the accent type and its distributionacross the linguistic input, as well as on the lexical neighbourhood.

Semantic networks generated from different word corporashow common structural characteristics, including high de-grees of clustering, short average path lengths, and scale freedegree distributions. Previous research has disagreed aboutwhether these features emerge from internally- or externally-driven properties (i.e. words already in the lexicon vs. regu-larities in the external world), mapping onto preferential at-tachment and preferential acquisition accounts, respectively(Steyvers & Tenenbaum, 2005; Hills, Maouene, Maouene,Sheya, & Smith, 2009). Such accounts suggest that inherentsemantic structure shapes new lexical growth. Here we ex-tend previous work by creating semantic networks using theSEEDLingS corpus, a newly collected corpus of linguistic in-put to infants. Using a recently developed LSA-like approach(GLoVe vectors), we confirm the presence of previously re-ported structural characteristics, but only in certain ranges ofsemantic similarity space. Our results confirm the robustnessof certain aspects of network organization, and provide novelevidence in support of preferential acquisition accounts.

This research tests whether analogical processing ability is present in 3-month-old infants. Infants are habituated to a series of analogous pairs, instantiating either same (e.g., AA, BB, etc.) or different (e.g., AB, CD, etc.), and then tested with further exemplars of the relations. If they can distinguish the familiar relation from the novel relation, even with new objects, this is evidence that for analogical abstraction across the study pairs. In Experiment 1, we did not find evidence of analogical abstraction when 3-month-olds were habituated to six pairs instantiating the relation. However, in Experiment 2, infants showed evidence of analogical abstraction after habituation to two alternating pairs (e.g., AA, BB, AA, BB...). Further, as with older groups, rendering individual objects salient disrupted relational learning. These results demonstrate that 3-month-old infants are capable of analogical comparison and abstraction. Our findings also place limits on the conditions under which these processes are likely to occur. We discuss implications for theories of relational learning.

A preliminary meta-analysis using the p-curve method(Simonsohn, Nelson, & Simmons, 2014) was performed on asubset of the learned categorical perception literature toexplore the robustness of the phenomenon. Only studies usingnovel visual categories and behavioral measures wereincluded. The results strongly suggest that the phenomenon isrobust but that the studies are somewhat underpowered. Weargue that this is problematic because it renders bothstatistically significant and nonsignificant results verydifficult to interpret, which impedes progress inunderstanding the learned CP phenomenon, for example, whyexpansion vs. compression is observed, or boundary vs.dimensional effects. Fortunately, there is a clear solution:conduct studies with greater statistical power.

A reading skill test to diagnose basic language skills is introduced. The test is designed to measure six component skills relevant to reading in comparison with those of state-of- the-art natural language processing technologies. The results of the first large-scale experiments using the test are reported. Surprisingly, almost half of Japanese junior high school students do no better than machines in dependency analysis. More than half of 7th grade students do no better than making random choices on questions involving inferences and definition understanding.

Students cheating on an exam in an academic setting createsan environment where one person (the student) must reasonabout the perception of another (the teacher). In exploring thestudent’s mindset, trends concerning how humans make deci-sions based on their understanding of another human’s inten-tions and knowledge can be uncovered. In this work, we studyhuman cheating behavior through simulated examinations invirtual reality, showing that the teacher’s animacy and orienta-tion plays a large part in the student’s reasoning of the teacher’sawareness. By utilizing a virtual classroom setting and accu-rately tracking a users behavior (through head tracking, eyemovement, etc.), we have also demonstrated how a novel vir-tual reality approach can be used for such experiments involv-ing human behavioral observations, which can be further ex-plored in other cognitive science research experiments.

A central assumption in joint action research is that in order toexplain how individuals act as part of a group, we must firstexplain how the group comes into existence. This assump-tion has led to an unnecessarily narrow research programme:research has focussed largely on interpersonal coordinationmechanisms. I outline an alternative approach predicated ona dynamic conception of the ecosystem. On this view, thereis no need to assume that actors must first constitute a groupagent with their fellows before entering into coordinated ac-tion. Such coordination can be more efficiently explained byrecognizing that all actions perturb the structure of the ecosys-tem itself in a manner that can alter the action possibilitiesavailable to neighbouring actors. This move allows us to over-come entrenched debates over the nature of shared intention-ality, and to instead focus on practical interventions in multi-actor settings.

Emotional state influences nearly every aspect of human cog-nition. However, coding emotional state is a costly processthat relies on proprietary software or the subjective judgmentsof trained raters, highlighting the need for a reliable, automaticmethod of recognizing and labeling emotional expression. Wedemonstrate that machine learning methods can approach near-human levels for categorization of facial expression in natural-istic experiments. Our results show relative success of modelson highly controlled stimuli and relative failure on less con-trolled images, emphasizing the need for real-world data forapplication to real-world experiments. We then test the poten-tial of combining multiple freely available datasets to broadlycategorize faces that vary across age, race, gender and photo-graphic quality.

Does the ability to reason well make one less likely to engage in motivated reasoning? Following a paradigm used by Kahan, Peters, Dawson, and Slovic (2013), this study aims to replicate, extend, and explain the surprising finding that those most likely to process politicized data in a biased manner are those who score highest on a measure of numerical proficiency. Although our study found general effects of motivated reasoning, we failed to replicate Kahan et al.’s “motivated numeracy effect”. However, our study did find that, when forced to consider competing statistical interpretations of the data before responding, highly numerate participants were more likely than less numerate ones to choose a correct but belief-contradicting interpretation of data. These results suggest that while numerate participants were biased when generating responses, they were not when evaluating reasons to justify their responses

Studies of how evidence affects beliefs sometimes show be-lief polarization in response to mixed evidence. However, thenature of the mental processes leading to change in opinion isup for debate. Different accounts of how people process evi-dence and then update their beliefs make different predictions,especially about one-sided evidence, which is rarely examined.We presented subjects with multiple text arguments regardingsocio-political topics as one-sided or mixed evidence. Partici-pants rated arguments differently according to their extant be-liefs, which is consistent with accounts of motivated reason-ing. They did not polarize afterward, instead showing evi-dence of belief updating according to Bayesian principles: be-lief change is sensitive to prior opinions and to the directionand quality of the evidence presented. These data support re-thinking some of the mental processes underlying incorpora-tion of evidence into a personal belief structure.

Nowadays, several of the situations in which we have to makedecisions are in digital form. In a first experiment (N=1010)we showed that people’s moral judgments depend on theDigital Context (Smartphone vs. PC) in which a dilemma ispresented, becoming more utilitarian (vs. deontological) whenusing Smartphones. To provide additional evidence, we ran asecond (N=250) and a third experiment (N=300), where weintroduced time constraints and we manipulated timeinstructions. Our results provide an extended perspective onDual-Process Models of Moral Judgment, as we showed thatthe use of smartphones, often assumed to be hurried whichwould be consistent with gut-feeling decision-making,increased the likelihood of utilitarian responses and decreaseddeontological ones. This is the first study to look at theimpact of the digital age on moral judgments and the resultspresented have consequences for understanding moral choicein our increasingly virtualized world

Two hypothesized functions of working memory –coordination (ability to maintain unrelated storage loads duringprocessing) and integration (ability to integrate multipleelements into a relation) – were explored and compared to fluidintelligence. In Experiment 1, 130 participants completed amodified Latin-Square Task (LST) which experimentallyadded or reduced storage load. Results suggested that pureintegration (with no storage load) could predict Gf, but nodifference was found between coordination and integration.Experiment 2 employed the Arithmetic Chain Task (ACT),again with modifications to storage load. Results supportreplication of LST findings, though a distinction was foundbetween coordination and integration when storage materialcould not be easily rehearsed. Findings from both experimentssupport a distinction between coordination and integrationtasks in understanding the WM-Gf association.

In this study, we wanted to investigate whether the processing of semantic information is easier when mapping names to pictures or is it the other way around. In order to test this hypothesis, we ran a behavioural and an ERP (Event Related Potential) study, with specific interest in the N400 component as an indicator of semantic processing. We compared three groups of participants who did a match/mismatch task with the only difference being that the labels would appear before, after or simultaneously with the pictures. Not surprisingly, the hardest condition was the one where the two information were presented simultaneously. The amplitude of the N400 was more prominent in the condition where labels were presented after the pictures in comparison to the condition where labels preceded picture presentation, suggesting that this second experimental situation led to smaller violation of expectation for our participants (word to picture condition) in comparison to mapping pictures to words.

Executive function is a fundamental component of the human cognitive architecture. Here, we investigate the rela-tionship between executive function and scientific reasoning. Eighth graders completed measures of three executive functions(EFs): shifting, inhibiting, and updating. They also completed a measure of cognitive flexibility, the Wisconsin Card Sort Task(WCST), that has predicted scientific reasoning in prior studies. Scientific reasoning was measured by a standardized test ofscience achievement. A principal components analysis found that the three EFs were separable. Different EFs predicted dif-ferent aspects of cognitive flexibility; notably, participants with poor shifting ability made more perseverative errors. Both EFand WCST predicted science achievement. Of note was the finding that better updating (i.e., working memory) was associatedwith higher science scores. These findings illuminate the role of EF in cognitive flexibility and scientific reasoning, and pointthe way to future studies of the effect of training EF on science achievement.

Omissions figure prominently in causal reasoning fromdiagnosis to ascriptions of negligence. One philosophicalproposal posits that omissions are accompanied by acontrasting alternative that describes a case of orthodox (non-omissive) causation (Schaffer, 2005; Bernstein, 2014). Apsychological hypothesis can be drawn from this contrastview of omissions: by default, humans should interpretomissive causations as representing at least two possibilities,i.e., a possibility representing the omission and a possibilityrepresenting a contrast. The theory of mental models supposesthat reasoners construct only one possibility (the omission) bydefault, and that they consider separate alternativepossibilities in sequential order. Two experiments test thecontrast hypothesis against the model theory, and findevidence in favor of the model-theoretic account.

People often use spatial vocabulary to describe temporal rela-tions, and this has increasingly motivated attempts to mapspatial frames of reference (FoRs) onto time. How people as-sign FRONT to time and to temporal entities depends on cul-tural conventions, and is crucial for diagnosing which tem-poral FoR a person actually adopts. Here, we report findingsfrom a survey with speakers of Norwegian that aimed at as-sessing the cultural conventions involved in FRONT assign-ment. Data on temporal movements of events, on the temporalorder of events, and on explicit FRONT assignments to events,time units, and “time itself” suggest that participants use dif-ferent principles for describing fixed relations (static time)versus moving events (dynamic time).

In many ecologically situated cognitive tasks, participants en-gage in self-selection of the particular stimuli they choose toevaluate or test themselves on. This contrasts with a traditionalexperimental approach in which an experimenter has completecontrol over the participant’s experience. Considering thesetwo situations jointly provides an opportunity to understandwhy participants opt in to some stimuli or tasks but not toothers. We present here a Bayesian model of cognitive andmetacognitive processes that uses latent contextual knowledgeto model how learners use knowledge to make opt-in decisions.We leverage the model to describe how performance on self-selected stimuli relates to performance on true experimentaltasks that deny learners the opportunity for self-selection. Weillustrate the utility of the approach with an application to ageneral-knowledge answering task.

We tested whether conceptual processing is modality-specific by tracking the time course of the Conceptual Modality Switch effect. Forty-six participants verified the relation between property words and concept words. The conceptual modality of consecutive trials was manipulated in order to produce an Auditory-to-visual switch condition, a Haptic-to- visual switch condition, and a Visual-to-visual, no-switch condition. Event-Related Potentials (ERPs) were time-locked to the onset of the first word (property) in the target trials so as to measure the effect online and to avoid a within-trial confound. A switch effect was found, characterized by more negative ERP amplitudes for modality switches than no- switches. It proved significant in four typical time windows from 160 to 750 milliseconds post word onset, with greater strength in posterior brain regions, and after 350 milliseconds. These results suggest that conceptual processing may be modality-specific in certain tasks, but also that the early stage of processing is relatively amodal.

We propose a theoretical framework for understanding how everyday choice objects are represented and how decisions involving these objects are made. Our framework combines insights regarding object and concept representation in semantic memory research with multiattribute choice rules proposed by scholars of decision making. We also outline computational techniques for using our framework to quantitatively predict naturalistic multiattribute choices. We test our approach in two-object and three-object forced choice experiments involving common books, movies, and foods. Despite using complex naturalistic stimuli, we find that our approach achieves high predictive accuracy rates, and is also able to provide a good account of decision time distributions.

This study investigated the social transmission of memoriesand skills collected from a collaborative cooking task (ravioli-making) and across transmission chains. The transmissionover three generations of pairs of participants occurred undertwo conditions. In the interactive condition, transmissionsover generations occurred in face-to-face conversations,whereas in the non-interactive condition, generations video-recorded their instructions to the next generations. Weanalyzed the effects of verbal and embodied features ofinformational transfer on task performance. Our results showthat performances improved over generations regardless ofinteractivity. In the discussion we suggest that tools (likecooking utensils) may have operated as cultural affordancesencapsulating and transmitting important cultural knowledgefor the successful completion of the task.

Ambiguity resolution requires high-level interpretation processes, at least some of which are subserved by theinferior frontal gyrus (IFG), a region that is susceptible to modulation by task demands. This fMRI study investigates the extentto which ambiguity-related activation in IFG is modulated by the specific cognitive-linguistic demands posed by the modality inwhich a sentence is presented. In the present study, ambiguous sentences and matched unambiguous sentences were presentedin three conditions: listening, reading, and rapid serial visual presentation (RSVP). The RSVP modality elicited strongerambiguity-related haemodynamic responses than the other two modalities, particularly in left anterior IFG. This indicatesthat the RSVP modality cannot be used as a simple substitute for natural reading without taking into account the additionalprocessing resources it requires.

Analysis of language provides important insights into the un-derlying psychological properties of individuals and groups.While the majority of language analysis work in psychologyhas focused on semantics, psychological information is en-coded not just in what people say, but how they say it. Inthe current work, we propose Conversation Level Syntax Simi-larity Metric-Group Representations (CASSIM-GR). This toolbuilds generalized representations of syntactic structures ofdocuments, thus allowing researchers to distinguish betweenpeople and groups based on syntactic differences. CASSIM-GR builds off of Conversation Level Syntax Similarity Metricby applying spectral clustering to syntactic similarity matricesand calculating the center of each cluster of documents. Thisresulting cluster centroid then represents the syntactical struc-ture of the group of documents. To examine the effectivenessof CASSIM-GR, we conduct three experiments across threeunique corpora. In each experiment, we calculate the cluster-ing accuracy and compare our proposed technique to a bag-of-words approach. Our results provide evidence for the ef-fectiveness of CASSIM-GR and demonstrate that combiningsyntactic similarity and tf-idf semantic information improvesthe total accuracy of group classification.

This paper presents an eyetracking during reading experimentthat investigated the role of supportive context on processingaspectual coercion. Coercion sentences in need of aspectualenrichment were embedded in discourse contexts providingthe necessary information for successful interpretation. Thefindings of the reported experiment show that context infor-mation can be used immediately without disrupting reading ofcoercion sentences. The lack of coercion costs in supportivediscourse contexts provides experimental evidence for the pro-posed Composition in Context Hypothesis and against theoriesthat view semantic composition as largely encapsulated fromcontext. Furthermore, the present experiment investigated therole of inter-individual differences in verbal working memorycapacity on the immediate use of contextual information incomputing coerced interpretations.

The current study examined the impact of agency on college students’ emotions and learning during gameplay with CRYSTAL ISLAND, a game-based learning environment designed to foster microbiology learning. 96 undergraduate students (59% female) from a large North American university participated in the study. Participants were randomly assigned to one of three experimental conditions (i.e., full agency, partial agency, no agency), based on the level of control granted during gameplay, and were asked to uncover the source, identity, and best treatment for a mysterious illness. Results revealed participants in the partial agency condition achieved the highest (pre- to post-test) proportional learning gain (PLG), even when controlling for session duration. Additionally, there was a positive correlation between evidence scores of four emotions (anger, fear, confusion, and frustration) and PLG within the partial agency condition—meaning the higher the evidence of the above emotions, the higher the PLG. Further, a stepwise multiple regression showed anger as the sole predictor of PLG. Results from this study have important implications for understanding the role of autonomy and emotions during learning and problem solving with GBLEs designed to foster scientific thinking in STEM. The current study suggests that although GBLEs offer significant learning benefits, they also induce several emotions that can facilitate or inhibit learning gains, requiring further examination.

Recent studies in the perception of numerosity have indicatedthat subitizing (the rapid and accurate enumeration of smallquantities) requires attention. We present a novel computa-tional model of enumeration in which attention unifies dis-tinct processes of numerosity approximation, subitizing, andexplicit counting. We demonstrate how this model accountsfor both the reaction time results from the subitizing literatureand the effects of attentional load on subitizing accuracy.

Two factors seem to play a major role in the cultural evolutionof language. On the one hand, there is functional pressure to-wards efficient transfer of information. On the other hand, lan-guages have to be learned repeatedly and will therefore showtraces of systematic stochastic disturbances of the transmissionof linguistic knowledge. While a lot of attention has been paidto the effects of cognitive learning biases on the transmissionof language, there is reason to expect that the class of possiblyrelevant transmission perturbations is much larger. This papertherefore explores some potential effects of transmission noisedue to errors in the observation of states of the world. We lookat three case studies on (i) vagueness, (ii) meaning deflation,and (iii) underspecified lexical meaning. These case studiessuggest that transmission perturbations other than learning bi-ases might help explain attested patterns in the cultural evolu-tion of language and that perturbations due to perceptual noisemay even produce effects very similar to learning biases.

Recursive loops in informal algorithms are difficult to formulate, even for na ̈ıve adults (Khemlani et al., 2013).Children can formulate algorithms that do not require loops (Bucciarelli et al., 2016), and anecdotal evidence suggests thatthey can understand loops. As there were no previous studies, we examined how they made deductions of the consequencesof loops, and how they abduced loops in creating informal algorithms in everyday language. We therefore tested fifth-gradechildren’s ability carry out both these tasks in algorithms that rearrange the order of cars on a toy railway track with onesiding. Experiment 1 showed that they could deduce rearrangements from algorithms containing loops, and Experiment 2showed that they could formulate at least some algorithms that contained loops. These abilities are the likely precursors to thecomprehension of recursion and to computer programming.

Sentence production relies on the activation of both semanticinformation (e.g. noun animacy) and syntactic frames thatspecify an order for grammatical functions (e.g. subjectbefore object; Levelt, Roelofs & Meyers, 1999). However, itis unclear whether these semantic and syntactic processesinteract (Gámez & Vasilyeva, 2015), and if this changesdevelopmentally. We thus examined the extent to whichanimacy-semantic role mappings in dative prime sentencesand target scenes influenced choice of syntactic structure. 143participants (47 three year olds, 48 five year olds and 48adults) alternated with the experimenter in describinganimations. Animacy mappings for themes and goals wereeither prototypical or non-prototypical and either matched ormismatched across the experimenter’s prime scenes andparticipants’ target elicitation scenes. Prime sentences wereeither double-object datives (e.g. the girl brought the monkeya ball) or prepositional datives (e.g. the girl brought the ballto the monkey). Participants’ target sentences were coded forsyntactic form. All age groups showed a main structuralpriming effect. For the youngest group, animacy-semanticrole mappings facilitated prepositional dative priming. Noanimacy facilitation was found for the older groups. Ourresults demonstrate the changing influence of animacy cueson sentence production through interactions with syntacticstructure over the course of development. The theoreticalimplications of our findings are discussed.

The Monty Hall Dilemma (MHD) is a well-known cognitive illusion. It is often claimed that one reason for the incorrect answers is that people apply the equiprobability principle: they assume that the probability of the two remaining options must be equal. An alternative explanation for assigning the same probabilities to options is that they had the same prior probabilities and people perceive no significant change. Standard MHD versions do not distinguish these possibilities, but a version with unequal prior probabilities could. Participants were given an unequal probabilities version of MHD and told that either the high or low probability option had been eliminated. This affected participants’ choices and their posterior probabilities. Only 14% of participants’ responses were consistent with applying the equiprobability principle, but 51% were consistent with a “no change” principle. Participants were sensitive to the implications of the prior probabilities but did not appear to use Bayesian updating.

In the history of cognitive science, there have been two com-peting philosophies regarding how people reason about theworld. In one, people rely on rich, generative models to makepredictions about a wide range of scenarios; while in the other,people have a large “bag of tricks”, idiosyncratic heuristics thattend to work well in practice. In this paper, we suggest thatrather than being in opposition to one another, these two ideascomplement each other. We argue that people’s capacity formental simulation may support their ability to learn new cue-based heuristics, and demonstrate this phenomenon in two ex-periments. However, our results also indicate that participantsare far less likely to learn a heuristic when there is no logical orexplicitly conveyed relationship between the cue and the rele-vant outcome. Furthermore, simulation—while a potentiallyuseful tool—is no substitute for real world experience

A model of sentence production is presented, which imple-ments a strategy that produces sentences with more uniformsurprisal profiles, as compared to other strategies, and in accor-dance to the Uniform Information Density Hypothesis (Jaeger,2006; Levy & Jaeger, 2007). The model operates at the al-gorithmic level combining information concerning word prob-abilities and sentence lengths, representing a first attempt tomodel UID as resulting from underlying factors during lan-guage production. The sentences produced by this modelshowed indeed the expected tendency, having more uniformsurprisal profiles and lower average word surprisal, in compar-ison to other production strategies.

Recent studies demonstrate a puzzling decline in relationalreasoning during development. Specifically, 3-year-olds failin a relational match-to-sample (RMTS) task, while youngerchildren (18-30 months) succeed (Walker, Bridgers, &Gopnik, 2016). Hoyos, Shao, and Gentner (2016) propose thatolder children fail because of a bias toward individual objectproperties induced by “avid noun learning.” If this is the case,children learning a language with a stronger emphasis onverbs, like Mandarin Chinese, may show an attenuateddecline in relational reasoning. We first test this possibility byreproducing the causal RMTS task in China, and find thatMandarin-speaking 3-year-olds outperform their English-speaking peers in the U.S. In a second experiment, we showthat Mandarin speakers exhibit a corresponding bias towardrelational solutions while English speakers prefer object-based solutions in an ambiguous context. We discuss possiblemechanisms through which language and culture maypromote (or hinder) the early development of relationalreasoning.

Language provides an important source of information to pre-dict human personality. However, most studies that have pre-dicted personality traits using computational linguistic meth-ods have focused on lexicon-based information. We investigateto what extent the performance of lexicon-based and grammar-based methods compare when predicting personality traits. Weanalyzed a corpus of student essays and their personality traitsusing two lexicon-based approaches, one top-down (Linguis-tic Inquiry and Word Count (LIWC)), one bottom-up (topicmodels) and one grammar-driven approach (Biber model), aswell as combinations of these models. Results showed thatthe performance of the models and their combinations demon-strated similar performance, showing that lexicon-based top-down models and bottom-up models do not differ, and neitherdo lexicon-based models and grammar-based models. More-over, combination of models did not improve performance.These findings suggest that predicting personality traits fromtext remains difficult, but that the performance from lexicon-based and grammar-based models are on par.

Hindi/Urdu (HU) numbers 10–99 are highly irregular, unlikethe transparent systems of most languages. I investigate themorphological decomposability of HU numbers using a seriesof computational models. While these models classify mostforms accurately, problems are encountered in high-frequencyforms of low cardinality, suggesting that some HU numbers aremore transparent (i.e., morphologically decomposable) thanothers. These results are compatible with a dual-route accessmodel proposed for the processing of numeral forms.

Early vocabularies in most languages tend to contain more nouns than verbs. Yet, the strength of this noun bias hasbeen observed to vary across languages and cultures. Two main hypotheses have aimed at explaining such variations; eitherthat the relative importance of nouns vs. verbs is language- specific, or that socio-cultural influences shape early vocabularystructures. The present study compares the relative distribution of verbs and nouns, in English, between two groups of bilingualinfants and toddlers; Malay-English and Mandarin- English. We found that early English lexicons of Mandarin- Englishbilinguals contained more verbs than in the English lexicon of Malay-English bilinguals, in both comprehension and production.We discuss the potential role of socio-cultural influences on the vocabulary structure in young users of a language.

To become a proficient reader, children have to learnmappings between print, sound and meaning. There isdebate over whether reading instruction should focus on therelations between print and sound, as in phonics, or on therelationship between print and meaning, as in sight wordreading. In a study where participants learned a novelartificial orthography, Taylor, Davis and Rastle (2017)compared print to sound focused or print to meaningfocused reading training, demonstrating that sound trainingwas superior for learning to read. However, a benefit fromsound focused training is likely dependent on prioracquisition of effective sound to meaning relations of words.To explore this issue, we developed a connectionist modelof reading. We exposed the model to a sound or a meaningfocused training, but varied the model’s pre-acquired orallanguage skills. The simulation results showed thatproficiency in oral language is a determinant of theadvantage of print to sound focused reading training,suggesting that reading training should address both orallanguage skills and print to sound mappings.

Vector-space representations provide geometric tools for rea-soning about the similarity of a set of objects and their relation-ships. Recent machine learning methods for deriving vector-space embeddings of words (e.g., word2vec) have achievedconsiderable success in natural language processing. Thesevector spaces have also been shown to exhibit a surprising ca-pacity to capture verbal analogies, with similar results for nat-ural images, giving new life to a classic model of analogies asparallelograms that was first proposed by cognitive scientists.We evaluate the parallelogram model of analogy as applied tomodern word embeddings, providing a detailed analysis of theextent to which this approach captures human relational sim-ilarity judgments in a large benchmark dataset. We find thatthat some semantic relationships are better captured than oth-ers. We then provide evidence for deeper limitations of the par-allelogram model based on the intrinsic geometric constraintsof vector spaces, paralleling classic results for first-order simi-larity.

Distributional learning research has established that humanscan track the frequencies of sequentially presented stimuli inorder to infer the probabilities of upcoming events (e.g., Hasher& Zacks, 1984). Here, we set out to explore anticipation of astimulus after implicit distributional learning. We hypothesizethat as people learn the category frequency informationimplicitly, response times will scale according to the relativefrequency of the stimulus category. Twelve adult participantsviewed photographs of faces, tools, and buildings whileperforming a simple classification task. We found that responsetimes significantly decreased with greater frequencies in thedistribution of stimulus categories. This result suggested thatdistributional information about the internal representations ofthe stimuli could be learned and indicated the possibility thatparticipants anticipated the stimuli proportional to theprobability of the category appearing and thereby reducedresponse times for the more frequent categories.

The present paper examines what domain-general causalknowledge reasoners need for at least some outcome-variabletypes to construct useable content-specific causal knowledge.In particular, it explains why it is essential to have analyticknowledge of causal-invariance integration functions:knowledge for predicting the expected outcome assuming thatthe empirical knowledge acquired regarding a causal relationholds across the learning context and an application context.The paper reports two studies that support the hypothesis thatpreschool children have such knowledge regarding binarycauses and effects, enabling them to generalize acrosscontexts rationally, favoring the causal-invariance hypothesisover alternative hypotheses, including interaction (e.g., linear)integration functions, heuristics, and biases.

Fairness, or the ability to distribute resources in a manner thataccords with societally recognized principles of justice, is ahallmark of human cooperation. Young children rapidlydevelop the ability to enact fairness, but the cognitiveunderpinnings of this ability remain unknown. The presentstudy investigated 4-7-year-olds’ acquisition of threeprinciples of fairness -- equality (the principle that all partiesshould have the same), merit (the principle that those whowork harder should get more), and starting opportunity (theprinciple that those who started with less should get more) --in relation to their emerging cognitive control and memory fornumerical information (numerical accuracy). Cognitivecontrol predicted children’s equal sharing, whereas numericalaccuracy predicted merit-based sharing. Children up throughthe oldest age we tested ignored starting opportunities. Theresults suggest that different principles of fairness may beunderpinned by distinct cognitive processes.

Prior studies on older adults’ risk taking have paid little attention to the healthcare domain or social influences on decision making. This study examined age-related differences in medication risk taking and the effects of a collaborative decision-making experience on individuals’ tendency to take risks. We recruited 24 younger (mean age = 19.50, SD = 1.41) and 24 older adults (mean age = 70.54, SD = 2.30), and asked them to choose between hypothetical medications that differed in probabilities and outcomes of treatment success. To investigate the effects of risk-neutral versus risk- advantageous trials, participants chose between a risky option and a sure option that had equal expected values (risk-neutral) or between a risky option and a sure option that had a lower expected value (risk-advantageous). Participants completed the decision task first individually (the pre-collaboration phase), then in dyads (the collaboration phase), and once again individually (the post-collaboration phase). During the pre-collaboration phase older adults showed a smaller increase in risk-taking tendency in response to risk- advantageous trials compared to younger adults. The pre-and post-collaboration data showed that older adults’ risk preferences converged towards their partner’s preference to a greater extent following collaboration relative to younger adults. These findings highlight the importance of designing decision aids to encourage older adults to take risks when risk taking is beneficial, and considering how social processes influence patients’ medication decisions.

Finding creative new ideas requires both release from fixationand a productive search mindset. Recent research has shownthat messy desks, walking, and mind-wandering can lead tomore new uses for old objects. Here we show that a human-centric mindset is superior to mind-wandering for generatingmore alternative uses and more creative uses because itprovides both release from fixation and an effective searchstrategy. A human-centric mindset entails perspective-taking,and perspective-taking is likely to be an effective generalstrategy for enhancing creativity, problem-solving andinnovation.

Perceptual learning is a key perceptual skill that people possess, in particular, it contributes to their ability to distinguish between faces thus recognize individuals. Recently, we showed that anodal transcranial Direct Current Stimulation (tDCS) at Fp3 abolishes the inversion effect (that would otherwise exist) for familiar checkerboards created from a prototype. Because of the close analogy between the inversion effect obtained with checkerboards, which we use as a marker for perceptual learning, and the traditional face inversion effect (upright faces recognized better than inverted ones), we investigated the effects of anodal tDCS at Fp3 during an old/new recognition task for upright and inverted faces. Results showed that stimulation significantly reduced the face inversion effect compared to controls. The effect was strongest in reducing recognition performance to upright faces. This result supports our account of perceptual learning and its role as a key factor in face recognition.

What is the best way of figuring out the structure of a causalsystem composed of multiple variables? One prominent ideais that learners should manipulate each candidate variable inisolation to avoid confounds (known as the “Control of Vari-ables” strategy). Here, we demonstrate that this strategy is notalways the most efficient method for learning. Using an opti-mal learner model which aims to minimize the number of tests,we show that when a causal system is sparse, that is, whenthe outcome of interest has few or even just one actual causeamong the candidate variables, it is more efficient to test mul-tiple variables at once. In a series of behavioral experiments,we then show that people are sensitive to causal sparsity whenplanning causal experiments.

Languages differ in the way they convey paths. S-languagesconveying manner of motion directly in a main verb, while V-languages require a separate verb. This difference has beenshown to influence the conceptualization and narration ofmotion events. We therefore asked: would this difference arisein the paths that people draw, particularly in visual narratives?We annotated the representations of path information (source,trajectory, goal) in a corpus of 35 comics from S- and V-languages. We found that panels from S-languages depicted thepath of an action more often than those from V-languages,consistent with previous research on increased motion eventsalience for S-languages. These findings suggest that theconceptualization of paths from spoken language mayinfluence the graphic depiction of paths.

The predictive performance equation (PPE) is a mathematicalmodel of learning and retention that uses regularities seen inhuman learning to predict future performance. Previous research(Collins, Gluck, Walsh Krusmark & Gunzelmann, , 2016) foundthat prior data could be used to inform PPE’s free parameterswhen generating predictions of a group’s aggregate performance,allowing for more accurate initial performance predictions. Herewe investigate an extension of this methodology to predictperformance of individuals, rather than aggregate samples. Thispaper documents the results of that investigation, which is on thecritical path to the use of this cognitive technology in educationand training.

This study investigates how judgments of explanatory power are affected by (i) the prior credibility of a potential explanation, (ii) the causal framing used to describe the explanation, and (iii) the generalizability of the explanation. We found that the prior credibility of a causal explanation plays a central role in explanatory reasoning: first, because of the presence of strong main effects on judgments of explanatory power, and second, because of the gate-keeping role prior credibility has for other factors. Highly credible explanations were not susceptible to causal framing effects. Instead, highly credible hypotheses were sensitive to the generalizability of an explanation. While these results yield a more nuanced understanding of the determinants of judgments of explanatory power, they also illuminate the close relationship between prior beliefs and explanatory power and the relationship between abductive and probabilistic reasoning.

Generating new concepts is an intriguing yet understudiedtopic in cognitive science. In this paper, we present a novelexemplar model of category generation: PACKER (ProducingAlike and Contrasting Knowledge using Exemplar Representa-tions). PACKER’s core design assumptions are (1) categoriesare represented as exemplars in a multidimensional psycholog-ical space, (2) generated items should be similar to exemplarsof the same category, and (3) generated categories should bedissimilar to existing categories. A behavioral study revealsstrong effects of contrast- and target-class similarity. Theseeffects are novel empirical phenomena, which are directly pre-dicted by the PACKER model but are not explained by existingformal approaches.

Managing disagreement in conversation requires subtlelinguistic and pragmatics skills. One key dimension is thedegree of ‘knowingness’ with which people present theirstance on an issue. It has been hypothesised that framingstances as ‘knowing’, i.e. with higher implied levels ofspeaker certainty limits the potential for challenge by others.We present the first experimental test of this hypothesis. Usinga text based chat-tool paradigm and a debating task we areable to systematically manipulate how ‘knowing’ people’sturns appear to one-another. The results show that ‘knowing’stances tend to close off discussion leading to less carefullyformulated, truncated turns, but do not reliably affect the rangeof solutions considered. Unknowing stances, by contrast,do not affect turn length or formulation but do encouragemore deliberation and include more signals of certainty in themessage contents.

Current theories of relational learning on structure mappingemphasize the importance of compositional representations,based on the concept’s components and the relations amongthem. We consider the possibility that relational concepts canalso be represented unitarily, whereby the concept is aproperty of the stimulus as a whole. The distinction betweencompositional and unitary representations of relationalconcepts is a natural consequence of structure-mappingtheory, but its psychological implications have not beenexplored. We report two experiments in which we examinehow encouraging subjects to represent relational conceptscompositionally versus unitarily affects learning onclassification- and inference-based category learning tasks.Our findings show that unitary representations lead to betterlearning than compositional representations, especially for theinference task. We conclude that unitary representations incurless cognitive load than structural alignment of compositionalrepresentations, and thus may be the default for everydayrelational reasoning.

Conversation partners’ assumptions about each other’sknowledge (their partner models) on a subject are importantin spoken interaction. However, little is known about whatinfluences our partner models in spoken interactions withartificial partners. In our experiment we asked people to name15 British landmarks, and estimate their identifiability to aperson as well as an automated conversational agent of eitherBritish or American origin. Our results show that people’sassumptions about what an artificial partner knows are relatedto their estimates of what other people are likely to know -but they generally estimate artificial partners to have moreknowledge in the task than human partners. These findingsshed light on the way in which people build partner models ofartificial partners. Importantly, they suggest that people useassumptions about what other humans know as a heuristicwhen assessing an artificial partner’s knowledge.

Influential theories in social psychology, philosophy, andlinguistics assume that ordinary people judge many mentalstates as outside voluntary control, yet few studies have directlyinvestigated these claims. We report four studies suggestingthat, contrary to several prominent models, ordinary peopleattribute at least moderate intentional control to others over awide variety of mental states. Furthermore, it appears thatperceived control may vary systematically according to mentalstate type (e.g. emotions vs. desires vs. beliefs). These resultspoint to several important directions for future research inbehavior explanation and moral judgment.

Theories about the role of emotions in moral cognition makedifferent predictions about the relative speed of moral andaffective judgments: those that argue that felt emotions arecausal inputs to moral judgments predict that recognition ofaffective states should precede moral judgments; theoriesthat posit emotional states as the output of moral judgmentpredict the opposite. Across four studies, using a speededreaction time task, we found that self-reports of felt emotionwere delayed relative to reports of event-directed moraljudgments (e.g. badness) and were no faster than person-directed moral judgments (e.g. blame). These results pose achallenge to prominent theories arguing that moraljudgments are made on the basis of reflecting on affectivestates.

The present study examined throughout three experiments the nature of stimulus-response compatibility (SRC) effects related to affordance perception in situations wherein object affordances and response effectors are irrelevant to each other. In the first experiment, using a foot-press response dispositive, we found a SRC effect between the orientation of the graspable part of the presented object and the laterality of the response. In Experiment 2a, we showed that constraining the subject hands in a given position (i.e., a Lego hand shape) during the same task interfered with the SRC effect. In Experiment 2b, participants performed a short training phase with their hands constrained before performing the experiment. This resulted in an inversion of the direction of the SRC effect previously observed. We discuss these results and provide arguments in favor of a specific motor activation account.

We trust that the uncertainty regarding the outcome of a cointoss makes it a fair procedure for making a decision. Smalldifferences in the force used to toss a coin should not affect thisuncertainty. However, the voluntary movement involved intossing a coin is subject to motivational influences arising fromthe anticipation of the value of the outcome of the toss.Presented here are measurements of hand velocities duringcoin tossing when the outcomes entail monetary gains andlosses. Finger position measurements show that hand velocitiesare proportional to the amount of money at stake. Coin tossmovements are faster and larger for higher stakes than forsmaller monetary stakes.

We investigated whether speakers’ referential communication benefits from an explicit focus on addressees’ perspective. Dyads took part in a referential communication game and were allocated to one of three experimental settings. Each of these settings elicited a different perspective mindset (none, self- focus, other-focus). In the two perspective settings, speakers were explicitly instructed to regard their addressee’s (other- focus) or their own (self-focus) perspective before construing their referential message. Results indicated that eliciting speakers’ self- versus other-focus did not influence their reference production. We did find that speakers with an elicited egocentric perspective reported a higher perspective-taking tendency than speakers in the other two settings. This tendency correlated with actual referring behavior during the game, indicating that speakers who reported a high perspective-taking tendency were less likely to make egocentric errors such as leaking information privileged to speakers themselves. These findings are explained using the objective self-awareness theory.

How do covert emotional stimuli affect decision-making? We investigated this question by exposing par-ticipants to subliminal visual stimuli during a computer-ized version of the Iowa Gambling Task (IGT) to assesswhether different categories of images (negative, neutral,or positive emotional evaluations) would influence deci-sion-making behavior. Results did show sex-group inter-actions for IGT scores. In decision learning model simu-lations, it was found that different models were more ap-propriate to explain the task performance for differentsex-group pairs. Overall, women showed more of anability to integrate the additive negative signals from thestimuli to make more advantageous decisions than themen; consequently, this made the men more resilient tothe negative effects of the positive stimuli on task-performance. When taken with existing research, the re-sults indicate that subliminal emotional stimuli can havesubtle, potentially sex-dependent, effects on behaviorduring the decision-making process.

To familiarize herself with a user-interface of a software system, a user needs practice. With practice, a user's think time gradually decreases—the novice to expert transition. We propose a queueing model that accounts for this transition in analyzing the performance of a distributed software system. We solve the model using deterministic simulation. Our model captures system performance in terms of system response time. We use the model to demonstrate how users— who are at various experience levels in the novice to expert continuum—may affect the system response time.

Dialogism provides the grounds for building a comprehensivemodel of discourse and it is focused on the multiplicity ofperspectives (i.e., voices). Dialogism can be present in anytype of text, while voices become themes or recurrent topicsemerging from the discourse. In this study, we examine theextent that differences between self-explanations and think-alouds can be detected using computational textual indicesderived from dialogism. Students (n = 68) read a text aboutnatural selection and were instructed to generate self-explanations or think-alouds. The linguistic features of thesetext responses were analyzed using ReaderBench, anautomated text analysis tool. A discriminant function analysisusing these features correctly classified 80.9% of the students’assigned experimental conditions (self-explanation vs. thinkaloud). Our results indicate that self-explanation promotestext processing that focuses on connected ideas, rather thanseparate voices or points of view covering multiple topics.

Bayesian models of cognition posit that people compute prob-ability distributions over hypotheses, possibly by construct-ing a sample-based approximation. Since people encountermany closely related distributions, a computationally efficientstrategy is to selectively reuse computations – either the sam-ples themselves or some summary statistic. We refer to thesereuse strategies as amortized inference. In two experiments,we present evidence consistent with amortization. When se-quentially answering two related queries about natural scenes,we show that answers to the second query vary systematicallydepending on the structure of the first query. Using a cog-nitive load manipulation, we find evidence that people cachesummary statistics rather than raw sample sets. These resultsenrich our notions of how the brain approximates probabilisticinference.

Although the causal graphical model framework has achievedsuccess accounting for numerous causal-based judgments, akey property of these models, the Markov condition, is con-sistently violated (Rehder, 2014; Rehder & Davis, 2016). Anew process model—the causal sampler—accounts for theseeffects in a psychologically plausible manner by assumingthat people construct their causal representations using theMetropolis-Hastings sampling algorithm constrained to onlya small number of samples (e.g., < 20). Because it assumesthat Markov violations are built into people’s causal represen-tations, the causal sampler accounts for the fact that those vio-lations manifest themselves in multiple tasks (both causal rea-soning and learning). This prediction was corroborated by anew experiment that directly measured people’s causal repre-sentations.

In this paper, we propose a flexible modeling framework forstudying the role of perception in language learning and lan-guage evolution. This is achieved by augmenting some noveland some existing evolutionary signaling game models withexisting techniques in machine learning and cognitive science.The result is a “grounded” signaling game in which agentsmust extract relevant information from their environment viaa cognitive processing mechanism, then learn to communi-cate that information with each other. The choice of cogni-tive processing mechanism is left as a free parameter, allow-ing the model to be tailored to a wide variety of problemsand tasks. We present results from simulations using both aBayesian perception model and a neural network based per-ception model, which demonstrate how perception can “pre-process” environmental data in a way that is well suited forcommunication. Lastly, we discuss how the model can be ex-tended to study other roles that perception may play in lan-guage learning.

The present study provides evidence for far analogicalretrieval, i.e., analogical retrieval across disparate taskdomains, as a result of analogical comparison. Participantsread source stories, which were then retrieved after a filleddelay through abstract letter-string cues that matched therelational form of key parts of stories. They then generatedresponses to an ambiguous letter-string analogy problem.Evidence was found for far analogical retrieval of higher-order relations because 1. comparison of letter-stringanalogies cued source stories specific to the relations showedin the letter-strings, and then 2. those same relations formedthe basis for how subjects solved novel letter-string problems.The experiment offers support for the schema inductionaccount of analogical retrieval, and suggests that people aremore sensitive to relational structures than was previouslythought.

People avoid changing subject abruptly during conversation.There are reasons to think that this constraint is more than asocial convention and is deeply rooted in our cognition. Weshow here that the phenomenon of topic connectedness is anexpected consequence of the maximization of unexpectednessand that it is predicted by Simplicity Theory.

In this manuscript, we summarize the results of our research program aiming at describing the cognitive architecture underlying the representation of recursive hierarchical embedding. After conducting a series of behavioral and fMRI experiments in the visual, musical and motor domains, we found that, behaviorally, the acquisition of recursive rules seems supported by cognitive resources that are general across domains. However, when we test well-trained participants in the fMRI, their representation of recursion seems supported by activating schemas stored in (visual, musical and motor) domain-specific repositories. This suggests that the resources necessary to acquire recursive rules are different from those necessary to utilize these rules after extensive training.

Previous research has found that people are seen as moreattractive when they appear in a group rather than in isolation.The present study asks whether faces that surround us in timealso affect how attractive we appear to be. Participants ratedthe attractiveness of famous female faces presented in asequence of three and in isolation. We found that people dointegrate information about attractiveness over time, but thattemporal context has the opposite effect of static context.People perceived faces as less attractive in a series than inisolation. We also varied the attractiveness of surroundingfaces in order to examine how the serial position of contextualinformation might figure into people’s judgments. We foundthat faces presented earlier in the sequence figured moreheavily into people’s judgment than did faces presented laterin the sequence. These findings highlight the role of temporalcontext in perceptions of attractiveness.

We present the project aimed at creating a database of detailed architectural process models of memory-based decision models. Those models are implemented in the cognitive architecture ACT-R. In creating this database, we have identified commonalities and differences of various decision models in the literature. The model database can provide insights into the interrelation among decision models and can be used in future research to address debates on inferences from memory, which are hard to resolve without specifying the processing steps at the level of precision that a cognitive architecture provides.

Is self-directed speech critical to cognitive processessupporting complex, goal-directed behavior? If so, how? Aninfluential developmental hypothesis is that children talk tothemselves to support cognitive control processes, and thatwith age this speech becomes increasingly covert andstrategic. However, while many studies suggest languagesupports cognitive control, evidence that self-directed speechgradually internalizes has been mixed. Moreover, extraneousfactors that could co-vary with self-directed speech, age, andcognitive control, such as talkativeness, have not beensystematically tested. In this cross-sectional study of 86 5- to7-year-old children we measured overt, partially covert, inner,and strategic speech on four cognitive tasks, along with taskperformance and child talkativeness. We did not findconsistent evidence that self-directed speech changes withage; however, we did find consistent associations betweenself-directed speech and talkativeness. Partially covert andstrategic speech predicted performance on one task, and innerspeech was implicated on another. Self-directed speechtended to correlate across tasks, and these correlations heldcontrolling for talkativeness. Taken together, these findingssuggest 5- to 7-year-old children may use different forms ofself-directed speech to support cognitive control, and that theform this speech takes depends in part on factors beyond age,such as the cognitive demands of a task and childcharacteristics like talkativeness.

Objective Self-awareness Theory (Duval & Wicklund,1972) proposes that self-evaluation increases an individual’sawareness of any discrepancy between their currentperformance and an internal goal. In the current study weprompted self-evaluation throughout an intelligence test(Analysis-Synthesis Test – AST) using confidence ratings(CR). AST performance, the extent to which participantsincidentally learnt task-relevant rules (learning rules wasunnecessary because they were provided), self-efficacy, andgoals, were assessed. The results indicated an effect ofperforming CR on both performance and rule learning, butthe effect depended on self-efficacy. Compared to matchedcontrols (n=45), participants who performed CR (n=41) andhad high self-efficacy performed better on the AST butlearnt fewer rules. Performing CR had no effect onparticipants low in self-efficacy. This suggests that self-evaluation interacts with self-efficacy to modifyparticipants’ goals, specifically CR appear to shiftindividuals high in self-efficacy from a mastery goal to aperformance goal.

The current study incorporates concepts from dynamicalsystems theory (DST) and embodied cognition topropose a novel method of answering traditionalquestions in social psychology. Namely, we wereinterested in understanding postural sway complexityduring the important interpersonal task of disclosing ahidden stigmatized identity (e.g., mental health disorder,history of sexual abuse). Using detrended fluctuationanalysis and multifractal detrended fluctuation analysis,we captured postural activity while people shared theirpersonal secrets to an imagined other. Results suggestthat disclosure context, defined by both disclosureconfidant and antecedent goals, is indeed embodied inour complex postural activity.

Responding to similarity, difference, and relative magnitude isubiquitous in the animal kingdom. However, humans seemunique in the ability to represent relative magnitude andsimilarity as abstract relations that take arguments (e.g.,greater-than (x,y)). While many models use structuredrelational representations of magnitude and similarity, littleprogress has been made on how these representations arise.Models that use these representations assume access tocomputations of similarity and magnitude a priori. We detail amechanism for producing invariant responses to “same”,“different”, “more”, and “less” which can be exploited tocompute similarity and magnitude as an evaluation operator.Using DORA (Doumas, Hummel, & Sandhofer, 2008), theseinvariant responses can serve to learn structured relationalrepresentations of relative magnitude and similarity from pixelimages of simple shapes.

During a music lesson, participants need to co-ordinate boththeir turns at talk and their turns at playing. Verbal and musicalcontributions are shaped by their organisation within the turn-taking system. When lessons are conducted remotely by videoconference, these mechanisms are disrupted by the asymmet-ric effects of delay on the interaction; in effect a “non-mutualreality” comprised of two different conversations at each endof the link. Here we compare detailed case studies of a co-present and a remote music lesson, in order to show how thiseffect arises, and how it impacts conduct during the lesson.

Recent emphases on differences between metaphors andsimiles pose a quandary. The two forms clearly differ instrength, but often seem to require similar interpretations. InExperiment 1 we show that ratings of comprehensibility arehighly correlated across simile and metaphor sentencesdiffering only in the presence or absence of “like”. InExperiment 2 we show that comprehensibility ratings forfigurative forms predict both early (first pass) and late(second pass) fixation durations for metaphor vehicle, butonly late fixation durations for vehicles in similes. Similevehicles appear to initially be processed similarly to literalcomparisons, with figurative interpretation occurring later.These observations are consistent with the different pragmaticstrengths, and similar interpretations of the two forms.

The principal aim of a cognitive model is to infer the processby which the human mind acts on some select set ofenvironmental inputs such that it produces the observed set ofbehavioral outputs. In this endeavor, one of the centralrequirements is that the input to the model be represented asfaithfully and accurately as possible. However, this is ofteneasier said than done. In the study of recognition memory, forinstance, words are the environmental input of choice—yetbecause words vary on many different dimensions, andbecause the problem of quantifying this variation has longbeen out of reach, modelers have tended to rely on idealized,randomly generated representations of their experimentalstimuli. In this paper, we introduce new resources from large-scale text mining that may improve upon this practice,illustrating a simple method for deriving feature informationdirectly from word pools and lists.

Considerable behavioral evidence has been cited in support ofthe COVIS dual-system model of category learning (Ashby &Valentin, 2016). The validity of the inferences drawn fromthese data critically depend on the accurate identification ofparticipants’ categorization strategies. In the COVIS literature,participants’ strategies are identified using a model-based anal-ysis inspired by General Recognition Theory (Maddox, 1999).Here, we examine the accuracy of this analysis in a model-recovery simulation. We find that participants can appear tobe using implicit, procedural strategies when their responseswere actually generated by explicit rule-based strategies. Theimplications of this for the COVIS literature are discussed.

Past studies reported that language-specific color focality hassubstantial influence on the short-term memory (STM) perfor-mance of colors of the speakers of the language, which we callthe ”focality effect.” This study attempts to clarify the contin-uous pattern of this effect, that is, the manner in which correctrecognition possibilities and misrecognition error distances ofcolors, which are two aspects of the STM performance for col-ors, change in a gradual fashion along the continuum of colorfocality. Our experiment, which tests the Japanese language,finds that a U-shaped relationship exists between the focalityand the possibility of correct recognition, and that the mis-recognition error distance increases as the focality decreases.We speculate that the subjects’ frequent and conscious employ-ment of the memorization strategy of coding colors using lin-guistic categories is one important cause of the detected effectpatterns.

This paper experimentally tests the contribution of twodistinct aspects of social interaction to the creation ofshared symbols: behaviour alignment and concurrentpartner feedback. Pairs of participants (N= 120, or 60pairs) completed an experimental-semiotic game,similar to Pictionary, in which they tried to communicatea range of recurring meanings to a partner by drawingon a shared whiteboard (without speaking or usingnumbers of letters in their drawings). The opportunityfor sign alignment and/or concurrent partner feedbackwas manipulated in a full factorial design. Each processmade a distinct contribution to the evolution of sharedsymbols: sign alignment directly influencedcommunication success, and concurrent partnerfeedback drove sign simplification and symbolization.These complimentary processes led to the interactiveevolution of effective and efficient humancommunication systems.

We propose a neural network model that accounts for the emer-gence of the taxonomic constraint in early word learning. Ourproposal is based on Mayor and Plunkett (2010)’s neurocom-putational model of the taxonomic constraint and overcomesone of its limitations, namely the fact that it considers arti-ficially built, simplified stimuli. In fact, while in the originalmodel the visual stimuli are random, sparse dot patterns, in ourproposed solution they are photographic images from the Im-ageNet database. In our model the represented objects in theimage can be of different size, color, location in the picture,point of view, etc.. We show that, notwithstanding the aug-mented complexity in the input, the proposed model comparesfavorably with respect to Mayor and Plunkett (2010)’s model.

The representation of the distribution of knowledge guidesinformation gathering, help seeking, and communication. Theresearch aimed to explore adults’ and 4-year-olds’representation of the distribution of common (conventionaland procedural) knowledge and expert knowledge associatedwith five occupations in their community. In addition, weexamined estimates of occupation-related everyday (non-expert) knowledge. Both groups estimated that commonknowledge is more widely held than expert knowledge, witheveryday knowledge in between. For adults, but not children,the distribution of expert knowledge was correlated withestimates of the proportion of people in each occupation.

Although a rich literature documents pre-literate children’s word learning success from shared storybook reading,a full synthesis of the factors which moderate these word learning effects has been largely neglected. This meta-analysisincluded 37 studies with 2,256 children, reflecting 104 effect sizes, investigating how the number of target words, tokens,story repetitions, reading styles and related factors moderate children’s word comprehension. Dialogic reading styles, tokens,the number of words tested and story repetition all moderated word learning effects. Children’s age, who read, number oftarget words and time between story and test were not moderators. These results provide information to guide researchers andeducators alike to the factors with the greatest impact on improving word learning from shared storybook reading.

Two experiments tested how the number of illustrations instorybooks influences 3.5-year-old children’s word learningfrom shared reading. In Experiment 1, children encounteredstories with either two illustrations, one illustration or onelarge illustration (in the control group) per spread. Childrenlearned significantly fewer words when they had to find thereferent within two illustrations presented at the same time. InExperiment 2 a gesture was added to guide children’sattention to the correct page in the two illustrations condition.Children who saw two illustrations with a guiding gesturelearned words as well as children who had seen only oneillustration per spread. Results are discussed in terms of thecognitive load of word learning from storybooks.

Some studies hypothesize a strong interdependence betweenspeech and tool use development in the first two years of life.To help understand the underlying mechanisms, we presentthe first robotic model learning both speech and tool use fromscratch. It focuses on the role of one important form of bodybabbling where exploration is directed towards self-generatedgoals in free play, combined with imitation learning of a con-tingent caregiver. This model does not assume capabilities forcomplex sequencing and combinatorial planning which are of-ten considered necessary for tool use. Yet, we show that themechanisms in this model allow a learner to progressively dis-cover how to grab objects with the hand, how to use objectsas tools to reach further objects, how to produce vocal sounds,and how to leverage these vocal sounds to use a caregiver asa social tool to retrieve objects. Also, the discovery that cer-tain sounds can be used as a social tool further guides vocallearning. This model predicts that the grounded exploration ofobjects in a social interaction scenario should accelerate infantvocal learning of accurate sounds for these objects’ names.

What do people think of when they think of “math?” We pro-pose that individuals may have very different working defini-tions of the category of math, and that those with broader mathconceptions may have less math anxiety. In Study 1, we intro-duce a method for indexing the “breadth” of individuals’ mathconceptions, and show that there is an inverse relation betweenconception breadth and math anxiety. These results suggestthat math anxiety is related both to how expansive individualsperceive math to be, and how skillful they feel at the activitiesthey think it could involve. Study 2 attempts an intervention onstudents’ conceptions of math with a sample of middle schoolstudents. We find the same inverse relationship in students be-tween math conception breadth and math anxiety as found inadults. We discuss ongoing work that further explores quali-tative variation in math conceptions, and the lessons this mayhold for intervening on math anxiety.

The Associative Read-Out Model (AROM) suggests that associations between words can be defined by the log likelihood that they occur together more often in sentences than predicted by their single-word frequency. Moreover, semantic relations can be defined by associative spreading across many common associates. Here, we addressed developmental effects of associative and semantic priming. Thus, we manipulated sentence-co-occurrence- based direct (syntagmatic) and common (paradigmatic) associations between prime and target words in 2nd and 4th graders. Syntagmatic associations decreased response times and error rates in both, 2nd and 4th graders. Paradigmatic associations increased errors rates in 2nd graders, whereas they decreased errors rates in 4th graders. These results suggest that 2nd graders profit from syntagmatic, i.e. contiguity-based associations, while a benefit from paradigmatic-semantic relationship probably develops from generalizing across many of these simple associations.

Even newborn infants are able to extract structure from a stream of sensory inputs and yet, how this is achieved remains largely a mystery. We present a connectionist autoencoder model, TRACX2, that learns to extract sequence structure by gradually constructing chunks, storing these chunks in a distributed manner across its synaptic weights, and recognizing these chunks when they re-occur in the input stream. Chunks are graded rather than all-or-none in nature. As chunks are learned their component parts become more and more tightly bound together. TRACX2 successfully models the data from four experiments from the infant visual statistical-learning literature, including tasks involving low- salience embedded chunk items, part-sequences, and illusory items. The model also captures performance differences across ages through the tuning of a single learning rate parameter. These results suggest that infant statistical learning is underpinned by the same domain general learning mechanism that operates in auditory statistical learning and, potentially, in adult artificial grammar learning.1

According to the dual systems model, adolescence is a period of imbalance between cognitive and motivational systems that results in increased tendency towards risk. In the study, we investigated the effects of rewards on risk-taking and cognitive control in 90 adolescents (13-16) and 96 adults (18- 35). Our results challenge the assumptions of the model as we observed that rewards lead adolescents to more conservative decisions in one of the risk tasks used in the study. We also observed that in cognitive control tasks, rewards influenced reaction latencies, but not the efficiency of control processes.

Previous studies have shown that one can exploit “wisdom ofcrowds” by oneself. This is achieved by aggregating multiple“quasi-independent” estimates from the same person. However,previous methods were not necessarily easy to utilize. Therefore,we propose an efficient method based on perspective-taking. Theprocedure is as follows: First, one makes her/his own estimation.Second, one estimates again based on a different perspective(“general public”). Then these two estimations were averaged.Two experiments revealed that our method effectively induced thewisdom of crowds by oneself. More importantly, participants inour method made estimations more quickly than those in apreviously proposed method, suggesting that our method requireda relatively diminished cognitive load for participants. Furtherinvestigation suggested that our method was immune to adverseeffects of confidence. Therefore, the present findings show that ourmethod could be effective and efficient method for inducing thewisdom of crowds in one mind.

Graph comprehension requires both bottom-up processing from the graph representation and top-down processing guided by knowledge and attitude. In the current study, we investigated which of the bottom-up process phases: extraction, interpretation, and decision: were affected by the top-down processing derived by the impressions and social attitudes. The experimental results showed that the top-down processing driven by impressions temporarily formed in specific contexts affected both the extraction of information and the following decision phase whereas top-down processing driven by attitude formed over a long time based on social norms affected only the decision phase. In the latter case, a decision was made without any need for bottom-up processing.

A crucial aspect of everyday conversational interactions is ourability to establish and maintain common ground.Understanding the relevant mechanisms involved in suchsocial coordination remains an important challenge forcognitive science. While common ground is often discussedin very general terms, different contexts of interaction arelikely to afford different coordination mechanisms. In thispaper, we investigate the presence and relation of threemechanisms of social coordination – backchannels,interactive alignment and conversational repair – across freeand task-oriented conversations. We find significantdifferences: task-oriented conversations involve higherpresence of repair – restricted offers in particular – andbackchannel, as well as a reduced level of lexical andsyntactic alignment. We find that restricted repair isassociated with lexical alignment and open repair withbackchannels. Our findings highlight the need to explicitlyassess several mechanisms at once and to investigate diversesocial activities to understand their role and relations.

How do speakers coordinate planning and articulation of more than one word at the same time? Here, we test whether they dynamically estimate how long it takes to (i) plan and (ii) articulate the words they intend to produce as a means of achieving such coordination. German speakers named two pictures without pausing, while their eye-movements were recorded. In line with previous reports, after their gaze left the first picture, speakers took longer to start speaking (i.e., the gaze-speech lag was longer) when the name of the first picture was shorter. But while gaze-speech lags were also longer when the second picture was harder to name, the two effects did not interact. We argue that speakers’ flexible planning abilities might be accounted for by reactive, rather than proactive planning mechanisms.

A key goal of cognitive science is to understand and map therelationship between cognitive processes. Previous workshave manually curated cognitive terms and relations,effectively creating an ontology, but do they reflect howcognitive scientists study cognition in practice? In addition,cognitive science should provide theories that informexperimentalists in neuroscience studying implementations ofcognition in the brain. But do neuroscientists and cognitivescientists study the same things? We set out to answer thesequestions in a data-driven way by text-mining and automatedclustering to build a cognitive ontology from existingliterature. We find automatically generated relationships to bemissing in existing ontologies, and that cognitive science doesnot always inform neuroscience. Thus, our work serves as anefficient hypothesis-generating mechanism, inferringrelationships between cognitive processes that can bemanually refined by experts. Furthermore, our resultshighlight the gap between theories of cognition and the studyof their implementation.

This paper explores whether social desirability affects theillusion of explanatory depth (IEOD) by comparing themagnitude of this illusion in topics with different levels ofsocial desirability within several domains. This question waschosen because prior literature shows that social expectationsabout how much a person should know about a certain topicaffect the magnitude of the IOED. Previous research showsalso that social desirability has an effect on a similar illusionrelated to argumentation, and that the IOED is affected by theway a person thinks knowledge is distributed in his or hersocial group. In order to do so, 184 participants were assignedrandomly to three knowledge domains (history, economics,and devices) and in each domain they rated theirunderstanding of a high-desirability and a low-desirabilitytopic following a standard IOED procedure. Results show thatsocial desirability has an effect on the IOED magnitude andthat overestimation of understanding varies among domains.Particularly, participants tend to overestimate theirunderstanding of high desirability topics only. This effect wasstronger in the historical domain.

People often withdraw previously drawn conclusions in light of new information. This defeasible reasoning is also im- portant for law, where judges often have to change their ver- dicts in light of new evidence. Here we investigate defeasibil- ity in the context of conflicting fundamental rights. When, for instance, law to property conflicts with law to information, can one of these rights be “defeated” by the other? We em- bedded conflicting fundamental rights in inference tasks (Ex- periment 1) and in elaborated vignettes (Experiment 2). Re- sults show that people decide between two conflicting funda- mental rights in a rational way. Case by case, participants pro- tected that fundamental right whose violation evoked the highest moral outrage (Experiment 1) or whose violation was considered to be more serious (Experiment 2). We discuss the implications of our findings for law theory and psychology.

One of the hallmarks of human natural language (NL) inter-action is the ability for people to balance a variety of so-cial and communicative goals when choosing how to realizetheir speech actions. These goals can include pragmatic criteriasuch as correctness, informativeness, and brevity (i.e., Griceanconversational maxims) or social factors such as politeness.However, there currently does not exist a general algorithmicmethod to explicitly modulate language generated by artificialagents based on an arbitrary number of pragmatic and socialcriteria. We propose a novel method to accomplish this task,in which rankings of candidate utterances by different prag-matic or social criteria are fused by use of a voting algorithm.We then give a proof-of-concept demonstration of the applica-tion of this method in the context of directive generation forhuman-robot interaction.

This paper examines a process of solving different types of counterfactual arithmetic problems (problems contradicted a visual experience, an experience of temperature, encyclopedic knowledge, etc.) in comparison with their ‘real’ counterparts by different types of subjects (e.g., educated in math and educated in humanities). As a result, a two-stage model of solving arithmetic problems is outlined in the paper.

This work focuses on an issue situated at the intersection of two domains: the oral mode of communication vs. the written mode of communication, and language acquisition. The backbone of this research is a conjecture that, for some age groups (babies, toddlers and preschool-aged children), to explore the acquisition of discourse as a whole (including gestures, facial expressions, prosody, pauses and discursive markers, etc.) is more appropriate than explore the acquisition of language exclusively. “The Pear Film” experimental line underpins the method of this research. The database comprises 74 ‘pear stories’ of Moscow preschool-aged children and high school students. Three parameters of the discourse are of interest for the authors: a logical structure and a coherence of the narrative; gestures and spontaneous movements lost any communicative meaning; discourse words and pauses.

Zoonoses (diseases that enter the human population via animalcontact) are a major global health concern. Because of howzoonoses emerge, understanding human reasoning about therisk factors associated with animal contact is central to com-bating their spread. However, little is known about the factorsthat influence perception of these risks. We present an induc-tive account of zoonosis risk perception, suggesting that it isinfluenced by beliefs about the range of animals that are able totransmit diseases to each other. In Study 1, we find that partic-ipants who endorse higher likelihoods of cross-species diseasetransmission have stronger intention to report animal bites. InStudy 2, adapting real world descriptions of Ebola virus fromthe WHO and CDC, we find that communications conveyinga broader range of animals as susceptible to a disease increaseintentions to report animal bites and decrease perceived safetyof wild game meat. These findings suggest that cognitive fac-tors may be harnessed to modulate zoonosis risk perceptionand combat emerging infectious diseases.

We focused on the controversy whether high-performing readers consistently underestimate their comprehension or are prone to detrimental overestimations as much as less skilled readers are. Therefore, we conducted an experiment (N = 105 university students) to investigate judgment bias as a function of reading skill and text difficulty in terms of text cohesion. Results showed that the easy text produced underestimation of comprehension, whereas the hard text led to overestimation. Furthermore, readers with higher reading skills were less prone to overestimate their comprehension of a hard text than less skilled readers. However, we also found that more skilled readers showed lower sensitivity in discriminating between correct and incorrect answers than less skilled readers. Overall, our results do not support the idea that high-performing readers consistently underestimate their text comprehension. Findings are discussed with respect to readers’ awareness of different text-based judgment cues and their (beliefs about their) reading skill.

Attention plays a fundamental role in higher-level cognition. In this paper we develop a computational model for how auditory spatial attention is distributed in space. Our model builds on the assumption that attentional bias has bottom-up and top-down components. We represent each component and their synthesis as a map, associating a level of attentional bias to locations in space. The maps and their interaction are modeled using an artificial intelligence approach based on constraints. We describe the behavioral task we have designed to measure the attentional bias and discuss the results. We then test different hypotheses on the shape and interaction modalities of the maps in terms of how well they fit our behavioral data. The findings showed that combining top-down and bottom-up spatial attention gradients that differ in their spatial properties produced the best fit to behavioral data, and suggested several novel mechanisms for future testing.

Many different functional roles have been ascribed to themotor system due to its prevalent recruitment in perceptualand cognitive tasks other than motor production. We discussfindings that suggest the motor system might take on multipleroles that vary with context and the brain networks involved.Using single-pulse TMS, we measured the corticospinalexcitability of the FDI muscle in primary motor cortex asparticipants viewed words that were either typed orhandwritten. We observed consistent facilitation ofcorticospinal excitability during reading of handwritten text.Although we observed facilitation in corticospinal excitabilityduring the presentation of typed text, this effect decreasedwith repetitive presentations of stimuli. We suggest that thefacilitation during presentation of typed words is a case ofaction simulation, and that the diminishing facilitation in thecase of typed stimuli is representative of sensory predictionby the motor system. These findings suggest that we shouldconsider multiple roles for motor recruitment during theobservation of visual stimuli, taking context intoconsideration.

Winner-take-all (WTA) mechanisms are an important compo-nent of many cognitive models. For example, they are oftenused to decide between multiple choices or to selectively di-rect attention. Here we compare two biologically plausible,spiking neural WTA mechanisms. We first provide a novelspiking implementation of the well-known leaky, competingaccumulator (LCA) model, by mapping the dynamics onto apopulation-level representation. We then propose a two-layerspiking independent accumulator (IA) model, and compare itsperformance against the LCA network on a variety of WTAbenchmarks. Our findings suggest that while the LCA net-work can rapidly adapt to new winners, the IA network is bet-ter suited for stable decision making in the presence of noise.

In science and philosophy, there is still no general agreement on what ‘consciousness’ is. But how do normalpeople (with no education in psychology or philosophy) use the term in their everyday life? What is the folk understandingof the word “conscious”? We conducted an online study on how the general public uses the word “consciousness” in theirdaily life. Participants (n=445) answered the question “What is consciousness?” in four different formats: they (1) generatedfree definitions in their own words, (2) generated many synonyms, (3) generated one synonym, or (4) selected one alternativedescription in a multiple choice task. The most frequent words were: alertness, clarity, I-sensation, knowledge, perception,reflecting, and thinking. The word perception was provided most often across all formats. There was also a high correlationbetween all response formats. We discuss these findings and their implications for the scientific study of consciousness.

Across languages, back is produced earlier and more frequently than front. This asymmetry has been attributed either to a conceptual/semantic asymmetry in the early meanings of these locatives (with back being more basic than front; conceptual immaturity account) or to the fact that Back configurations are inherently more ‘noteworthy’ than Front configurations (pragmatic account). Here, we tested the two accounts. In Study 1, children and adult speakers of English and Greek described Front/Back motion events. In Study 2, adult speakers of 10 additional languages described the same events. Despite cross-linguistic differences, speakers of all age and language groups typically used more Back than Front adpositions; furthermore, they often encoded Back information in occlusion verbs (e.g. hide) but no such verbs were available for Front. Thus, the front/back asymmetry is not due to children’s conceptual immaturity but should be linked to pragmatic factors that also shape adult spatial language production cross-linguistically.

Magicians often rely on misdirection to fool their audience. Acommon way to achieve this is for the magician to provide aplausible and intuitive (but false) account of how an effect isperformed in order to prevent spectators from uncovering thetruth. We hypothesized that analytical thinkers would be morelikely than intuitive thinkers to seek alternative explanationswhen observing a mental magic effect because generating acoherent explanation requires analytical thought. We foundthat while intuitive thinkers often espoused explanations for amagic trick similar to one provided by the magician,analytical thinkers tended to generate new explanations thatechoed rational principles and relied on physical mechanisms(rather than mental capabilities). This difference was notpredicted by differences in numeracy skills or need forcognition.

An experiment aiming to assess the use of stopping rules ininformation acquisition was performed. Participants wererequested to make a decision in 24 financial scenarios withthe possibility of buying information pieces. Behavioral andEEG data were recorded for analysis. Results showed thatparticipants followed Bayesian calculations in order todetermine a stop on information acquisition and decide.Moreover, the information acquisition strategies wereconsistent with prospect theory, in which participants willweigh information pieces differently and seek more or lessinformation given different manipulations in scenarioprobability and consequences. EEG data suggest SlowCortical Potentials at fronto-central electrodes.

Young children use informal strategies to solve arithmetic word problems. The Situation Strategy First (SSF) framework claims that these strategies prevail even after instruction. The present study was conducted with second grade students in order to investigate the persistence of intuitive, situation- based strategies, on word problems that do not involve dynamic temporal changes. This is challenging for the SSF framework, since the lack of this dimension might bypass intuitive strategies. The results revealed that intuitive strategies persist, are valid for these types of problems, and impact the problems' difficulty. Indeed problems that require the application of arithmetic principles remain hard, even though they have been practiced at school. These findings provide complementary evidence to how mental calculation strategies articulate with arithmetic word problem solving and call for the extension of the SSF framework.

Prior research yielded conflicting findings regarding whether older adults show a greater processing cost than younger adults when encountering unpredicted semantic material during language processing. Here, we investigated whether age-related differences in recovery from prediction error are influenced by increased demands on working memory. We used a dual task design: a primary sentence comprehension task in which semantic predictions were fulfilled or violated, and a concurrent driving task, thought to limit working memory resources in resolving prediction errors. In the dual task, older participants showed an increase in comprehension accuracy for sentences with semantic violations, while demonstrating a decrease in driving accuracy. Thus, when working memory resources were limited, older adults focused exclusively on the language task and neglected the driving task. This could be related to an age-related increase in generating semantic predictions, or to a general inability among older adults to divide attention between two cognitively demanding tasks.

Variance of the outcomes associated with an option often provides a measure of the riskiness of that option. Hence, it is important for organisms are able to detect any sudden changes in outcome variance. In Experiment 1, we presented people with graphs of share price time series or water level time series. In half the graphs, variance (financial or flooding risk) changed at some point. People were better at detecting increases than decreases in risk - maybe because it is more important to detect increases in danger than decreases in it. However, in Experiment 2, people were still better at detecting increases than decreases in variance even when those changes did not reflect altered levels of risk. Our findings may reflect the fact that the actual change in variance exceeds the change needed to identify a regime change in variance by a larger amount for upward than for downward changes.

This study investigated how the expectations of others (i.e.,top-down processes) and actual perceived behavior (i.e.,bottom-up processes) influence negotiations during human-agent interactions. Participants took part in several sessions ofthe ultimatum game; we investigated the bargaining strategiesdirected toward the computer agent. To investigate the influ-ence of top-down and bottom-up processes on performance,we designed an experiment wherein (1) participants expectedtheir partners were humans or agents, and (2) agents used dif-ferent types of algorithmic behavior. Results revealed that ir-rational decisions, which are characteristic of human-humaninteractions, emerged when participants believed their oppo-nents were human and when opponent behaviors were ambigu-ous. Further, we found participants adopted different bargain-ing strategies according to their expectations and the agent’sspecific algorithmic behavior. We discuss interplay of the twotypes of cognitive processing in human-agent interaction.

Previous research conducted by Bergus et al. (2002) identified that treatment evaluations are more negative when risks are presented last. Extending discussion of this order effect, the current studies investigate this effect in tabular style displays, manipulating both order and orientation; and using eye-tracking methodology, explores the effect of these variables on the information search process. Analysis from eye-tracking data revealed a tendency to read information sets sequentially (i.e. read all risk information before transitions to the other set), which is stronger for the vertical orientation where switching between information sets is less common. Further, while balanced search was observed when benefits presented first, when presented with the risks first, search becomes more risk- heavy. Eye-tracking measures did not strongly predict treatment evaluations, although, when holding other variables constant, time proportion spent on benefits positively predicted treatment evaluations.

Criminal behavior and related disorders have been associated with abnormal neural activity when experiencingor anticipating risks and rewards, as well as when exercising inhibition. However, behavioral and neural substrates of riskpreferences and criminality have received scant attention when unconfounded with experience, anticipation, and inhibition. Wetest predictions of fuzzy-trace theory (FTT) in two experiments using a risky-choice framing task. Behavioral results show thatindividuals with a greater history of criminal behavior are less likely to engage in simple meaning-based processing and areless confident when doing so. These findings are supported by fMRI results showing a greater history of criminal behavioris associated with increased activation in regions associated with cognitive control when engaging in simple meaning-basedprocessing. These results provide insight into the cognitive processes and brain mechanisms that are associated with criminalbehavior.

Planning is an everyday activity that is extended in timeand space, yet is frequently studied in the absence ofinteractivity. Successful planning relies on an array ofexecutive functions including self-control. Weinvestigated the effects of interactivity and self-controlon planning using a sequential-task paradigm. Half ofthe participants first completed a video-viewing taskrequiring self-control of visual attention, whereas theother half completed the same task without the self-control constraint. Next, and within each of thesegroups, half of the participants manipulated cards tocomplete their plan (high-interactivity condition); for theother half, plans were made with their hands down (low-interactivity condition). Planning performance wassignificantly better in the high- than in the low-interactivity conditions; however the self-controlmanipulation had no impact on planning performance.An exploration of individual differences revealed thatlong-term planning ability and non-planningimpulsiveness moderated the impact of interactivity onplanning. These findings suggest that interactivityaugments working memory resources and planningperformance, underscoring the importance of aninteractive perspective on planning research.

When faced with a moral dilemma, following your headversus your heart can result in very different decisions. Earlierwork has argued that people who “self-locate” in the headtend to make more rational and less emotional decisions tomoral dilemmas than those who “self-locate” in the heart. Wereplicate this finding, suggest an alternative interpretation ofthe result, and then extend it with a novel experiment. In ametaphor framing task, we manipulated the salience of thehead/heart metaphors—by using them (a) in a single sentence,(b) a more elaborate paragraph, or (c) by emphasizing one incontrast to the other. We found that people who received thehead metaphor made more rational decisions than those whoreceived the heart metaphor, but only in the high saliencecondition that contrasted the two metaphors. This findingillustrates the communicative value of metaphor, which canbe enhanced through comparison.

Languages exhibit striking semantic diversity, but differentlanguages often share core metaphors. Conceptual MetaphorTheory (Lakoff & Johnson, 1980) claims that universalhuman experiences give rise to conceptual representationsthat are then expressed in language. But languages changeslowly, making it difficult to observe implicitconceptualization affecting linguistic convention in realtime. Here, we describe a shared conceptualizationpreviously absent from speech that has now becomeconventionalized in linguistic metaphors. In two studies, wedocument how members of the US military talk about timeusing conventionalized lateral metaphors (e.g., ‘push themeeting right’ to mean ‘move the meeting later’). We showthat military members, unlike civilians, consider suchsentences to be acceptable—sometimes even moreacceptable than more standard phrases. Moreover, militarypersonnel seem unaware that these lateral metaphors arespecific to their linguistic sub-community. Our findingssuggest that implicit mental representations can becomeconventionalized metaphors in language.

Dialog game tools are text chat applications which aim tostructure and promote students' collaborative learning byhaving them select a label and sentence-opener for eachmessage they type to their learning partner. In thisexperiment, we compared students’ learning from discussionsvia a dialog game tool to their learning via a standard freechatapplication. Students discussed topic questions with alearning partner. They then individually completed a multiplechoice test, for assessing knowledge-gain, and a short-answertest, to assess readiness for knowledge-building. Resultssuggest that dialog games applications lead to increasedreadiness for knowledge-building, in the form of integratingdistinct pieces of learned knowledge, than freechatapplications. Follow-up analyses suggest that the degree ofconcept overlap between students' dialog messages and topickeywords, as measured by a "semantic fingerprint" system, isa potentially useful metric for predicting students'knowledge-building. Implications and potential applicationsof our findings are discussed

Past mechanistic accounts of children’s word learningclaim that a simple type of cross-situational learning ispowerful enough to match observed rates of learning,even in quite ambiguous situations. However, a limita-tion in some of these analyses is their reliance on an un-realistic assumption that the learner only hears a word insituations containing the intended referent. This studyanalyzed a more general type of cross-situational learn-ing based on the relative frequency of word-object pairs,and found it to be slower than the simple mechanismanalyzed in prior work. We then analytically exploredwhether relative-frequency learning can be improved byincorporating the mutual exclusivity (ME) principle–an assumption that words map to objects 1-to-1. Ouranalyses show that with a certain type of correlation inword-to-word relationship, ME makes relative frequencylearning as efficient as fast-mapping, which can learn aword in one exposure.

Psychological momentum (PM) and the hot hand are relatedconcepts describing people’s beliefs regarding streaks ofsuperior performance. This study examined the susceptibilityof perceptions of PM to changes in the streakiness ofotherwise equivalent series. Fifty-five male participants (31basketballers and 24 control) completed a ‘hot-cognition’experiment where they rated individual and team momentumand assessed the likelihood of a future shot’s success afterwatching sequences of basketball shots. The experimentalmanipulation of the order of shots strongly affectedparticipants’ ratings of momentum and, less strongly, theprobability they assigned to the future shot (i.e. the hot handeffect). Basketballers showed stronger reactions tomanipulations of order than the controls, which could beattributed to greater investment in the task. The resultsdemonstrate the importance of distinguishing between PMand the hot hand and also provide a valuable extension ofprior work showing such effects into more realistic scenarios.

Computational psychiatry applies advances from computational neuroscience to psychiatric disorders. A core aim is to develop tasks and modeling approaches that can advance clinical science. Special interest has centered on reinforcement learning (RL) tasks and models. However, laboratory tasks in general often have psychometric weaknesses and RL tasks pose special challenges. These challenges must be addressed if computational psychiatry is to capitalize on its promise of developing sensitive, replicable assays of cognitive function. Few resources identify these challenges and discuss strategies to mitigate them. Here, we first overview general psychometric challenges associated with laboratory tasks, as these may be unfamiliar to cognitive scientists. Next, we illustrate how these challenges interact with issues specific to RL tasks, in the context of presenting a case example of preparing an RL task for computational psychiatry. Throughout, we highlight how considering measurement issues prior to a clinical science study can inform study design.

People’s beliefs about what is morally right and wrong vary widely between individuals, contexts, and cultures;however it is thought that they are governed by core latent constructs. While there is evidence that these constructs are reflectedin natural language, this requires further testing. We demonstrate that the structure of moral values in natural discourse can bemodeled by applying factor analyses to distributed representations of morally relevant terms learned by a neural network. Wefirst demonstrate that robust latent constructs can be estimated from the covariance of distributed representations of constructexemplars. We then test whether the factor structure proposed by Moral Foundations Theory (MFT) is reflected in naturallanguage. Finally, we conduct a bottom-up investigation of the structure of moral values in natural language using free-responses reported by participants. Ultimately, we find evidence that the representation of moral values in natural languagepartially corresponds to MFT.

This paper investigates a fundamental conflict in the literatureon people’s probability estimation. Research on ‘perception’of probability shows that people are accurate in their estimatesof probability of various simple events from samples. Equally,however, a large body of research shows that people’s probabil-ity estimates are fundamentally biased, and subject to reliableand striking fallacies in reasoning. We investigate this con-flict in an experiment that examines the occurrence of the con-junction fallacy in a probability perception task where peopleare asked to estimate the probability of simple and conjunc-tive events in a presented set of items. We find that people’sprobability estimates are accurate, especially for simple events,just as seen in previous studies. People’s estimates also showhigh rates of occurrence of the conjunction fallacy. We showhow this apparently contradictory result is consistent with arecent model of probability estimation, the probability theoryplus noise’ model.

Disfluencies such as self-repairs, filled pauses such as ‘um’and silent pauses are pervasive in dialogue, but there is no con-sensus in the literature as to whether they reflect internal pro-duction pressures, or interactive issues – or how their effectsare manifest in dialogue. It is well-known that patients withschizophrenia have problems with language and social cogni-tive skills, yet little research has investigated how these impactinteraction. We report a study on the disfluency behaviours ofpatients with schizophrenia and their interlocutors who wereunaware of the patient’s diagnosis, compared to healthy con-trol groups. Results show that patients use fewer self-repairsthan either their partners or controls and fewer filled pauses(‘er’, ‘um’) than controls. Furthermore, the presence of the pa-tient also affects patients’ partners, who use fewer filled pausesthan controls and more unfilled pauses than both patients andcontrols. This suggests that smooth coordination of turns isproblematic in patients’ dialogues

Mind perception is studied for three different agents: a human,an artificial human, and a humanoid robot. The artificiallycreated agents are presented as being undistinguishable from ahuman. Each agent is rated on 15 mental capacities. Three mindperception dimensions are identified - Experience, Agency,and Cognition. The artificial agents are rated higher on theCognition dimensions than on the other two dimensions. Thehumanoid robot is rated lower than the human on theExperience dimension. These results show that people ascribeto artificial agents some mental capacities more than others. Ina second experiment, the effect of agent’s moral action on mindperception is explored. It is found that when the artificial agentshave undertaken a moral action, they are perceived to besimilar to the human agent. More interestingly, the presentationof the moral action leads to a restructuring of the dimensionsof mind perception.

Previous studies have found a processing difference between unaccusative sentences and unergative sentences. Theyargued that the difference is derived from a syntactic difference, i.e. an unaccusative sentence involves movement whereas anunergative sentence does not. In this study, we are going to show that those studies are uninterpretable due to uncontrolledstimuli and confounds in the materials. After examining their data anlysis, we argue that the effects they found does seem to bestable. This reopens the question whether there is a syntactic difference between unaccusative and unergative verbs.

Existing models of statistical learning involve computation ofconditional probabilities over discrete, categorical items in asequence. We propose an alternative view that learning occursthrough a process of tracking changes along physicaldimensions from one stimulus to the next within a “perceptualsimilarity space.” To test this alternative, we examined asituation where it is difficult or impossible to label stimuli inreal time, and where the two assumptions lead to conflictinghypotheses. We conducted two experiments in which humanparticipants passively listened to a familiarization sequence offrequency-modulated tones and were then asked to makefamiliarity judgments on a series of test bigrams. Behavioralresults were broadly consistent with a conceptualization oflearning as tracking trajectories through perceptual similarityspace. We also trained a neural network that codes stimuli asvalues along two continuous dimensions to predict the nextstimulus given the current stimulus, and show that it capturedkey features of the human data.

Referential success depends on choice of referring expression.The choice of referring expression will depend on contextualfactors as well as factors related to speaker and addresseeknowledge. A shared-learning paradigm was used in whichpartners learned names of objects together and separately beforea referential task. Items differed on commonality, with someindependently rated as more common and some as more rare.Speakers were less likely to use names versus other forms whenitems were rare than common (p<0.001) and less likely to usenames when items were new than learned together (p<0.001).Asymmetry effects showed that speakers were more likely touse a name when the addressee was deemed moreknowledgeable in post-test ratings (p<0.01). Together, we takethis to show speakers choose to use a name versus a descriptionbased on the likelihood that their interlocutor will know thename. Factors affecting the likelihood include prior knowledgeof what a typical addressee will know and shared experience,which includes inferring an interlocutor’s expertise, asdynamically updated during a dialog.

The aim of this paper is to argue that models in cognitivescience based on probabilistic computation should not be re-stricted to those procedures that almost surely (with probabil-ity 1) terminate. There are several reasons to consider non-terminating procedures as candidate components of cognitivemodels. One theoretical reason is that there is a perfect cor-respondence between the enumerable semi-measures and allprobabilistic programs, as we demonstrate here (generalizinga better-known fact about computable measures and almost-surely halting programs). One practical reason is that the linebetween almost sure termination and non-termination is elu-sive, as well as arbitrary. We argue that this matters not onlyfor theorists, but also potentially for a learner faced with thetask of inducing programs from experience.

Behavioral studies demonstrate the influence of speakergaze in visually-situated spoken language comprehension.We present an ERP experiment examining the influence ofspeaker’s gaze congruency on listeners’ comprehension of ref-erential expressions related to a shared visual scene. Wedemonstrate that listeners exploit speakers’ gaze toward ob-jects in order to form sentence continuation expectations:Compared to a congruent gaze condition, we observe an in-creased N400 when (a) the lack of gaze (neutral) does not al-low for upcoming noun prediction, and (b) when the noun vi-olates gaze-driven expectations (incongruent). The later alsoresults in a late (sustained) positivity, reflecting the need to up-date the assumed situation model. We take the combinationof the N400 and late positivity as evidence that speaker gazeinfluences both lexical retrieval and integration processes, re-spectively (Brouwer et al., in press). Moreover, speaker gazeis interpreted as reflecting referential intentions (Staudte &Crocker, 2011).

Studies of dyslexics, whose implicit memory is impaired,suggest that their implicit inference of sound statistics and itsintegration into perception is inefficient. Specifically,dyslexics' implicit memory decays faster and consequentlyonly accumulates information over shorter temporal windows.We now ask whether this abnormal dynamic is domain generalby measuring its cortical distribution. We measure thedynamics of behavioral context effects and the concurrentneural adaptation during fast acquisition fMRI. We find asimilar pattern of fast decay of adaptation across a broad rangeof cortical areas, though most significant effects are found inauditory cortex. This broad neural distribution suggests that therelevant aspect of implicit statistical inferences is not the natureof stimuli, but their temporal distribution.

The study assessed the auditory processing abilities and the cognitive skills in children with specific learning disability. It investigates the top-down or bottom-up influence on auditory processing. Using a test battery approach, the association between cognitive skills (verbal working memory and attention) and auditory processing abilities (auditory closure, binaural integration and temporal processing skills) has been measured. The results revealed that cognitive processes significantly affect the bottom-up auditory perception. The effect of cognition was more evident in speech processing than non-speech signal processing. These findings may be useful in designing appropriate therapeutic protocol for children with specific learning disability.

Difficult tasks should be attempted one at a time, while easytasks can be undertaken in parallel. Reinforcing our previ-ous conclusion that people are surprisingly poor at applyingthis logic, we find people fail to select standing positions thatmaximize their probability of success in throwing a beanbaginto one of two possible hoops. We asked participants to ex-plicitly report their odds of successfully throwing a beanbaginto each hoop from the location they had chosen to stand,and estimates were highly accurate. Nonetheless, participantsfailed to use estimates of success appropriately to maximizesuccess, suggesting a failure of insight, rather than limited orinaccurate information, can account for suboptimal decisionsabout standing position.

Appraising one’s own performance after a task, known as self-assessment, has been studied from a cognitive science perspec-tive in domains such as humor, trivia, and logic. Previous stud-ies have found that participants are systematically poor at judg-ing their own performance, though sometimes self-assessmentvaries based on actual performance. We explored calibrationof self-assessment on algebra problems, a domain where peo-ple have typically received explicit instruction. In this domain,we found that people do not behave as they do in other do-mains previously studied: they are generally well-calibrated injudging their algebra performance. This suggests that in thecourse of learning to solve algebra problems, people have alsolearned to accurately judge their performance, both absolutelyand relative to others.

Previous research showed that individual tendency to believe in conspiracy theories is related to numerous social, personality, and cognitive variables. Moreover, such a tendency may reflect a broader trait for epistemic irrationality, which drives other pseudo- scientific and paranormal beliefs. However, the relationship between conspiracy belief and reasoning ability (fluid intelligence; Gf) was not sufficiently studied to date, even though Gf level strongly influence the way in which individuals think and reason. Using confirmatory factor analysis, we found the robust link between conspiracy belief and other irrational beliefs. All those irrational beliefs were also substantially related to the close- minded cognitive style. However, even though Gf significantly predicted other irrational beliefs, it explained less than 2% of variance in conspiracy belief. This result suggests that effective reasoning cannot prevent even highly intelligent people from endorsing conspiracy theories.

Previous work shows that people often believe, contrary toactual physics, that objects travelling in a curved path througha tube will continue to travel in a curved path after exiting thetube. In the present study, previous work was replicated, butaccuracy increased in a new condition in which people wereasked to catch an actual ball emerging from a tube. That is, inthis case there is a discrepancy between how we believe theworld works, and how our motor system responds to events inthe world. This finding supports the theory that the perceptionand action systems of the brain use different methods topredict how things move in the world, and that the abstractreasoning systems used to explain how the world works areoften in conflict with the action systems.

The focus of this paper is to examine differences in semantic network structure of late talkers and typical talkers to elucidate potential learning strategies used by late talking children. To address this question, we conducted network analysis on the vocabularies of 2,912 children, with 566 of those being late talkers. Contrary to previously reported findings, the results show that late talkers have well-connected vocabularies as measured by median degree, clustering coefficient, and mean distance, with more well-connected networks in some cases than their typical talking peers. Further analysis of word order suggests that late talkers may be selecting based on frequency and connectivity of the words in the learning environment, more so than typical talkers. The language processing difficulties in late talkers appear not to be associated with their semantic network properties. In sum, late talkers may initially benefit from using word frequency and word connectivity strategies to build well-connected vocabularies.

Accounts of spatial language aim to address both the meaningof a spatial term and its usage patterns across diverse cases,but do not always clearly distinguish these from one another.Focusing on the case of English prepositions in and on, we setout to disentangle spatial language meaning from spatiallanguage use by comparing judgments on a series of linguistictasks designed to tap each aspect of spatial language. Wedemonstrate that judgments of truth-conditional meaning andpatterns of naturalistic use show different distributionalsignatures, with judgments of meaning giving rise to a moreuniform distribution than use patterns. We explore a thirdaspect of spatial language: lexical choice, and propose thatchoice is a key factor in shaping the distribution of spatialexpression use. Our analyses reveal that the distribution oflexical choice judgments is highly correlated with thedistribution of expression use in spatial descriptions for thesame spatial scenes, supporting a model of spatial languagethat differs from traditional accounts of meaning andcategorization.

Economic choices depend on our predictions of the future.Yet, at times predictions are not based on all relevantinformation, but instead on the single most likelypossibility, which is treated as though certainly the case—that is, digitally. Two sets of studies test whether thisdigitization bias would occur in higher-stakes economiccontexts. When making predictions about the future assetprices, participants ignored conditional probabilityinformation given relatively unlikely events and reliedentirely on conditional probabilities given the more likelyevents. This effect was found for both financial aggregatesand individual stocks, for binary predictions about thedirection and continuous predictions about expected values,and even when the “unlikely” event explicitly had aprobability as high as 30%; further, it was not moderatedby investing experience. Implications for behavioralfinance are discussed.

The Linguistic Category Model (LCM) was developed as a manual coding scheme for quantifying abstract mindsetsin human language. Previous attempts to computationally automate the LCM have relied primarily on pre-coded semanticfeatures, which fail to incorporate important contextual information integral to the LCM coding scheme. In this paper, weintroduce Syntax-LCM, a novel method for automating LCM coding using syntax and dependency tree features as predictors ofconstrual level. We compare the accuracy of Syntax-LCM to that of two previously used automated methods: LIWC LCM andBrysbaert concreteness ratings. We find support that the Syntax-LCM approximates the hand-coded LCM with higher accuracycompared to both the Brysbaert and the LIWC LCM. We also provide evidence that the syntactic features accounted for bySyntax-LCM mirror the inclusion criteria in the original coding manual and support theoretical relationships between distanceand abstract thinking.

Iconicity, i.e. resemblance between form and meaning, is awidespread feature of natural language vocabulary (Perniss,Thompson, & Vigliocco, 2010), and has been shown tofacilitate vocabulary acquisition (Imai, Kita, Nagumo, &Okada). But what kind of advantage does iconicity actuallygive? Here we use cross-situational learning (Yu & Smith,2007), to address the question for sound-shape iconicity (theso-called kiki-bouba effect, Ramachandran & Hubbard,2001). In contrast to Monaghan, Mattock, and Walker (2012),Experiment 1 suggests that the iconicity advantage comesfrom referential disambiguation rather than more efficientmemory encoding. Experiments 2 and 3 replicate this result,and moreover show that the kiki-bouba effect is roughlyequally strong for sharp and rounded shapes, a property thatclassic experiments were unable to confirm, and which hasimplication for the effect’s mechanism

During mindfulness-based interventions participants can beinvited to bring aversive stimuli to mind while practicingmindfulness. This is thought to help the stimuli become lessaversive. However, the mechanisms underlying this processare not fully understood. In this study we explored these byexamining the effects of mindfulness practice and stimulusvisualization on stimuli associated with electric shocks.Participants were trained on a discrimination between twovisual stimuli using a standard electrodermal conditioningprocedure, in which one stimulus (CS+) was paired withshock and the other (CS-) was not. They then visualized eitherthe CS+ or CS-, while practicing mindfulness or performing acontrol activity. Following a number of extinction trials, theimpact of these manipulations was assessed during areacquisition test-phase. Both mindfulness and visualizationof the CS+ led to slower reacquisition of the CS+/shockassociation, when measured physiologically, and their effectswere additive. Moreover, these effects dissociated fromparticipants’ expectancy of shock. If confirmed in futurework, these findings may have implications for the treatmentof stimulus-specific anxiety.

Creativity is the ability to produce work that is both novel and appropriate. The study, funded by Indian government, analyzed the effect of one-year chess training on the creativity of children. A pretest and posttest with control group design was used, with 31 children in experimental and 32 in control group. The experimental group underwent weekly chess training. Wallach-Kogan Creativity Test (Indian Adaptation) was used. Analysis revealed that only the experimental group had statistically significant gains in total creativity and two nonverbal subtests. The authors conclude that systematic chess training inculcates in the child the ability to think divergently and creatively.

Modeling text document similarity is an important yet challenging task. Even the most advanced computational linguistic models often misjudge document similarity relative to humans. Regarding the pattern of misjudgment between models and humans, Lee and colleagues (2005) suggested that the models’ primary failure is occasional underestimation of strong similarity between documents. According to this suggestion, there should be more extreme misses (i.e., models failing to pick up on strong document similarity) than extreme false positives (i.e., models falsely detecting document similarity that does not exist). We tested this claim by comparing document similarity ratings generated by humans and latent semantic analysis (LSA). Notably, we implemented LSA with 441 unique parameter settings, determined optimal parameters that yielded high correlations with human ratings, and finally identified misses and false positives under the optimal parameter settings. The results showed that, as Lee et al. predicted, large errors were predominantly misses rather than false positives. Potential causes of the misses and false positives are discussed.

Previous research suggests that semi-realistic animation films such as The Polar Express are representative of the uncanny valley (UV) hypothesis, which predicts that highly human- like artificial characters can appear eerie. In the present study, we investigated the extent to which critical film reviews can influence the perceived eeriness of such films. The reviews were adopted from authentic ones and expressed either negative or positive attitudes towards the animation techniques. Audiovisual speech asynchrony, which is known to induce eeriness, was included as an objective manipulation. Our results showed large review tone effects for both implicit and explicit eeriness evaluations. In contrast, speech asynchrony failed to elicit significant effects. These results demonstrate that critical film reviews representing opposite attitudinal poles can elicit consistent changes in the viewers’ evaluations of semi-realistic animations. The present findings cannot, however, be taken as evidence against the UV hypothesis itself in computer-generated characters.

During interaction with others, we perceive and produce so-cial actions in close temporal distance or even simultaneously.It has been argued that the motor system is involved in bothprocesses, but how does it distinguish in this processing be-tween self and other? In this paper, we present a model ofself-other distinction within a hierarchical sensorimotor sys-tem that is based on principles of perception-action couplingand predictive processing. For this we draw on mechanisms as-sumed for the integration of cues to generate sense of agency,i.e., the sense that an action is self-generated. We report re-sults from simulations of different social scenarios, showingthat the model is able to solve the problem of the dual use ofthe sensorimotor system.

This study examined learning and transfer of a simplemathematical concept when learning a symbolic sententialformat versus learning a diagrammatic format. Undergraduatecollege students learned an instantiation of a cyclic group andwere then given a test of a novel isomorphic group of thesame order followed by a test of a novel non-isomorphicgroup of a higher order. The results were that both thesentential and the diagrammatic formats led to successfullearning and transfer to the novel isomorphic group.However, only learning from the diagrammatic representationproduced successful transfer to the non-isomorphic group.These findings suggest that learning a diagrammaticrepresentation of a mathematical concept can have transferadvantages over learning strictly sentential formats.

Recent experimental evidence suggests that adults incorporatespeaker knowledge into the derivation of pragmaticimplicatures. Developmental studies report that 5-year-oldchildren also succeed in taking speaker knowledge intoaccount in implicature computation, but 4-year-olds fail. Thepresent study investigated the pragmatic competence of 4-year-olds, specifically the ability to incorporate speakerknowledge into the derivation of ad hoc scalar implicatures.Using a simple paradigm inspired by referentialcommunication, we found that 4- year-olds are able toincorporate speaker knowledge into implicature derivation.These results have implications for our understanding of thelinguistic, pragmatic, and epistemic abilities of youngchildren.

In production, frequently used words are preferentially extended to new, though related meanings. In comprehension, frequent exposure to a word instead makes the learner confident that all of the word’s legitimate uses have been experienced, resulting in an entrenched form-meaning mapping between the word and its experienced meaning(s). This results in a perception-production dissociation, where the forms speakers are most likely to map onto a novel meaning are precisely the forms that they believe can never be used that way. At first glance, this result challenges the idea of bidirectional form-meaning mappings, assumed by all current approaches to linguistic theory. In this paper, we show that bidirectional form-meaning mappings are not in fact challenged by this production-perception dissociation. We show that the production-perception dissociation is expected even if learners of the lexicon acquire simple symmetrical form-meaning associations through simple Hebbian learning.

Deaf children born to hearing parents are exposed to language input quite late, which has long-lasting effects on language production. Previous studies with deaf individuals mostly focused on linguistic expressions of motion events, which have several event components. We do not know if similar effects emerge in simple events such as descriptions of spatial configurations of objects. Moreover, previous data mainly come from late adult signers. There is not much known about language development of late signing children soon after learning sign language. We compared simple event descriptions of late signers of Turkish Sign Language (adults, children) to age-matched native signers. Our results indicate that while late signers in both age groups are native-like in frequency of expressing a relational encoding, they lag behind native signers in using morphologically complex linguistic forms compared to other simple forms. Late signing children perform similar to adults and thus showed no development over time.

Since its design in 2003, the joint Simon task and corollaryjoint Simon effect (JSE) have been invaluable tools towardsthe study of joint action and the understanding of howindividuals represent the action/task of a co-actor. Thepurpose of this meta-analysis was to systematically andquantitatively review the sizeable behavioural evidence forthe JSE. Google Scholar was used to identify studies citingthe first report of the joint Simon task (Sebanz, Knoblich, &Prinz, 2003) up until June 23, 2015. After screening, thirty-nine manuscripts were included in the meta-analysis, thirteenof which included individual go/no-go (IGNG) control data.Separate random-effects models were conducted for both thejoint Simon and IGNG datasets, and meta-regression modelswere used to assess potential moderators that may impact thestrength of the JSE. The results provide an importantquantitative summary of the literature and serve as afoundation for future research surrounding the JSE.

We propose that probabilistic inference is supported by a men-tal toolbox that includes sampling and symmetry-based rea-soning in addition to several other methods. To flesh out thisclaim we consider a spatial reasoning task and describe a num-ber of different methods for solving the task. Several recentprocess-level accounts of probabilistic inference have focusedon sampling, but we present an experiment that suggests thatsampling alone does not adequately capture people’s infer-ences about our task.

One particularly successful approach to modeling withincognitive science is computational psychology.Computational psychology explores psychological processesby building and testing computational models with humandata. In this paper, it is argued that a specific approach tounderstanding computation, what is called the ‘narrowconception’, has problematically limited the kinds of models,theories, and explanations that are offered withincomputational psychology. After raising two problems for thenarrow conception, an alternative, ‘wide approach’ tocomputational psychology is proposed.

Recently, some authors have begun to raise questions about thepotential unity of 4E (enactive, embedded, embodied, extended)cognition as a distinct research programme within cognitivescience. Two tensions, in particular, have been raised: (i) thatthe body-centric claims embodied cognition militate against thedistributed tendencies of extended cognition and (ii) that thebody/environment distinction emphasized by enactivism standsin tension with the world-spanning claims of extendedcognition. The goal of this paper is to resolve tensions (i) and(ii). The proposal is that a form of ‘wide computationalism’ canbe used to reconcile the two tensions and, in so doing, articulatea common theoretical core for 4E cognition.

The ability to reason about relations is relevant for many spa-tial cognitive processes. This can involve: (i) to represent spa-tial information mentally, (ii) to manipulate the spatial repre-sentation, and (iii) to infer new spatial information. Severalcognitive theories make assumptions and predictions about theunderlying processes. A detailed and systematic overview andanalysis of ireliable effects across studies is missing. This ar-ticle presents a meta-analysis of 35 studies about spatial rela-tional reasoning. Studies were classified according to differentfactors including the ambiguity of the spatial description, i.e.,if it the description allows for more than one representation, thepresentation of information, i.e., if the information has beenpresented auditorily or in a written form, and the task, i.e., ifa conclusion or model of the premises needs to be generatedor verified. Implications of the findings for the mental modeltheory and working memory are discussed.

Recent advances in technology have paved the way for human-agent interactions to become ubiquitous in our daily lives, anddecades worth of research on virtual agents have enhancedthese interactions. However, for the most part, the effect of dif-ferent types of agents on the human brain is unknown, and theneuroscience of human-agent interactions is rarely studied. Inthis study, we examine the underlying neural systems involvedin processing and responding to different types of negotiatingagents. More specifically, we show that different brain patternsare observed for various types of virtual agents; consequently,we can decode the strategy and emotional display of the agentbased on the counterpart’s brain activity. Using fMRI data, weanalyzed participants’ brain activity during negotiations withagents who show three different emotional expressions and usetwo different types of negotiation strategies. We demonstratethat, using Multi-Voxel Pattern Analysis, we can reliably de-code agents’ emotional expressions based on the activity in theleft dorsal anterior insula, and also agents’ strategies based onthe activity in the frontal pole.

People interact differently with humans than they do with com-puters, but there is minimal research on what brings aboutthese differences. Using agents labeled as either “another par-ticipant” or a “computer program”, we investigated the differ-ences in people’s behavior and brain activity during the courseof a negotiation paradigm. Our results indicate that people per-ceive human-labeled agents more human-like than computer-labeled agents, and the level of concession in the negotiationsis dependent on agent type. We have also found that these dif-ferences can be captured in brain activation by showing thatparts of the Theory of Mind neural correlates are activated inhuman-labeled agent conditions, but not in computer-labeledagent conditions. We further demonstrate that brain activitycan predict whether the negotiation agent was introduced asa competing human player or a computer program. Overall,our study suggests that labeling an interaction partner as ei-ther another human or a computer program leads to significantimpacts on one’s decision making.

Dyslexia is a common learning disability, but its core deficit is still under debate. The anchoring deficit hypothesissuggests that dyslexics’ benefit from experimental stimuli statistics is impaired (e.g. Ahissar, 2007). In this study we askedwhether dyslexia is also associated with reduced sensitivity to long-term statistics. Spans for lists of syllables were measured,and indeed, dyslexics benefited less than controls from syllabic frequency. However, dyslexics’ benefit from sequence repetitionwas similar to controls’. In order to dissociate the impact of item familiarity from exposure unrelated factors, native Englishspeakers performed the experiment. They were expected to benefit from repetition, but not from syllabic frequency (in Hebrew).Indeed, that was the case. These data suggest that benefits from long-term distributional statistics are impaired in dyslexia,whereas on-line benefits from sequence repetition are adequate. Moreover, our results suggest different underlying mechanismsfor long-term distribution learning and short-term sequence learning.

Current causal theories aim to incorporate the effect of statistical and prescriptive norms on causal judgements, stating that norm-violating actions are judged as more causal than norm-conforming ones. In this paper, we present two experiments that undermine this claim, showing that people attribute increased causality to agents who conform to the norm of frequent behaviour. Furthermore, we find that the time point at which a moral norm is introduced does not make a difference to causal attributions, but that the frequency of a norm violation further accentuates its causal rating. Because these findings present a challenge to current norm theories of causation, we argue for an extended counterfactual model of causal attribution.

We investigated how people would change and vary inaccepting advice when the effectiveness of advice was unclear.In each trial, participants estimated a monthly rent of anapartment room based on the attribute list. Then, anotherestimate by a real-estate agent was given as advice.Participants made a final estimation, either by taking theadvice fully, partially, or rejecting it totally. They repeated 48estimations without feedback. The weight of advice index,representing how much each participant weighed a givenadvice, gradually decreased as the number of trials increased.Interestingly, the gradual reduction of acceptance was notobserved in participants with high empathy and lowdepressive scores; they kept accepting advice even when theeffectiveness of advice was unclear. These results suggest thatthe willingness of accepting and using advice depends onhistory of advice taking, the individual traits, and mood.

The goal of the present study was to determine whether chronic post-stroke patients with motor aphasia have impaired inner speech abilities and whether they use inner speech in everyday life. To answer these questions, we recruited eight chronic post-stroke aphasic patients and 13 cognitively healthy adults, who underwent testing on a range of evaluative tests and four experiments specifically designed for the purposes of this study. The experimental results suggest that post-stroke patients with motor aphasia have impaired inner speech. However, patients’ subjective reports indicate that they use various types of inner speech, despite the observed deficit. Taken together, our data suggest that impairment of certain aspects of inner speech may still allow a degree of use of other aspects of inner speech, emphasizing a need to extend research on inner speech in aphasia to the variety of its forms.

The dynamics of language evolution and learning in individuals have been extensively studied. Our knowledge ofthe transmission process in particular has been advanced by the iterated learning model. Additionally, work has been done inthe area of population structure and social networks. However, less has been described about the interaction between individual-level transmission and network structures. We present a general framework for representing transmission and learning algo-rithms within social networks. We demonstrate that population structure interacts with the transmission process to influencethe dynamics of change. Taking network effects into account, studies on language evolution will capture a fuller picture of thephenomenon.

When referring to a target object in a visual scene, speakers areassumed to consider certain distractor objects that are visible tobe more relevant than others. However, previous research thathas tested this assumption has mainly applied offline measuresof visual attention, such as the occurrence of overspecificationin speakers’ target descriptions. Therefore, in the current study,we take both online (eye-tracking) and offline (overspecifica-tion) measures of attention, to study how perceptual groupingaffects scene perception, and reference production. We manip-ulated three grouping principles: region of space, type similar-ity, and color similarity. For all three factors, we found effects,either on eye movements (region of space), overspecification(color similarity), or both (type similarity). The results for typesimilarity provide direct evidence for the close link betweenscene perception and reference production.

We discuss here conceptual change and the formation of robustlearning outcomes from the viewpoint of complex dynamicsystems, where students’ conceptions are seen as context de-pendent and multifaceted structures which depend on the con-text of their application. According to this view the conceptualpatterns (i.e. intuitive conceptions) may be robust in a cer-tain situation but are not formed, at last not as robust ones, inanother situation. The stability is then thought to arise dynami-cally in a variety of ways and not so much mirror rigid ontolog-ical categories or static intuitive conceptions. We use compu-tational modelling in understanding the generic dynamic andemergent features of that phenomenon. The model shows howcontext dependence, described here through structure of epis-temic landscape, leads to formation of context dependent ro-bust states. The sharply defined nature of these states makeslearning to appear as a progression of switches from state toanother, given appearance of conceptual change as switch fromone robust state to another.

Individuals with autism spectrum disorders (ASD) are said to have deficits in “theory of mind.” The present paper explores two main accounts of the mechanisms underlying these deficits. On one account, high-functioning adults with ASD struggle to infer others’ mental states. On another account, they lack an ability to integrate those mental states into a coherent understanding of action. We tested these two accounts by making several modifications to the Faux Pas task—a commonly used advanced theory of mind task— including the presentation of explicit mental state information. Surprisingly, in contrast to previous work, individuals on the autism spectrum exhibited both intact integration and intact inference.

Past choices can influence subsequent choices in employee se-lection. Previous approaches rather described similar sequen-tial effects with feedback learning or the misperception of ran-domness. However, in the selection of job candidates also theaccumulation of the moral impact of previous choices mightinfluence subsequent choices. We investigated that questionby making two major contributions to the literature. First, wedeveloped an experimental paradigm for measuring sequentialchoices in employee selection and second, we implementeda widely applicable computational model, the Dependent Se-quential Sampling Model, for explaining sequential effects inchoices. By using this methodological approach, we uncov-ered sequential effects in employee selection. Participants(N=600) were especially motivated to compensate for morallydubious choices, with some participants showing consistentchoice behavior if their previous choices had been morally vir-tuous. These results support the assumption of asymmetriccompensation of morally dubious choices, sometimes referredto as the moral cleansing hypothesis.

In two visual word eye tracking studies, we investigated theinfluence of prosody and case marking on children’s andadults’ thematic role assignment. We assigned an SVO/OVS-biasing (vs. neutral) prosodic contour tounambiguously case marked German subject-verb-object(SVO) and object-verb-subject (OVS) sentencesrespectively. Scenes depicted ambiguous action events(e.g., donkey-paints->elephant-paints->cheetah) but casemarking and prosody could, in principle, disambiguate. Inadults, case marking but not prosody rapidly guidedthematic role assignment. Children did not rely on casemarking but exploited the biasing prosody to enhance theiragent-first interpretation of the sentences. These resultssuggest that in scenes depicting fully ambiguous rolerelations, children’s understanding of case marking at theage of five is not yet robust enough to enable thematic roleassignment. Prosody did not overwrite the SVO preference,it rather enhanced it.

How do we learn to think better, and what can we do to pro-mote such metacognitive learning? Here, we propose that cog-nitive growth proceeds through metacognitive reinforcementlearning. We apply this theory to model how people learn howfar to plan ahead and test its predictions about the speed ofmetacognitive learning in two experiments. In the first experi-ment, we find that our model can discern a reward structure thatpromotes metacognitive reinforcement learning from one thathinders it. In the second experiment, we show that our modelcan be used to design a feedback mechanism that enhancesmetacognitive reinforcement learning in an environment thathinders learning. Our results suggest that modeling metacog-nitive learning is a promising step towards promoting cognitivegrowth.

Searching natural environments, as for example, when foraging or looking for a landmark, combines reasoningunder uncertainty, planning and visual search. Existing paradigms for studying search in humans focus on step-by-step infor-mation sampling, without examining advance planning. We propose and evaluate a Bayesian model of how people search in anaturalistic maze-solving task. The model encodes environment exploration as a sequential process of acquiring informationmodelled by a Partially Observable Markov Decision Process (POMDP), which maximises the information gained. We showthat the search policy averaged across participants is optimal. Individual solutions, however, are highly variable and can beexplained by two heuristics: thinking and guessing. Self-report and inference, a Gaussian Mixture Model over inverse POMDP,consistently assign most subjects to one style or the other. By analysing individual participants’ decision times we show thatindividuals solve partial POMDPs and plan their search a limited number of steps in advance.

ndicative conditionals, that is sentences of the form “If p, thenq,” belong to the most puzzling phenomena of language. Onthe majority of accounts of indicative conditionals, the truth ofp and q suffices for “If p, then q” to be true or highly accept-able. Yet, many conditionals with true clauses, even if there isa meaningful connection between them, sound odd. The mostcommon reaction to this phenomenon is to attribute the oddityof conditionals with true clauses to natural language pragmat-ics. We present an experimental study investigating how thepresence or absence of a connection between the clauses af-fects the assertability of conditionals and conjunction express-ing generic and specific kind of content. The results refute thestandard pragmatic explanation.

Learning, especially in the case of language acquisition, is notan isolated process; there is ever-present competition betweenwords and objects in the world. Such competition is known toplay a critical role in learning. Namely, the amount andvariability of competing items during word learning havebeen shown to change learning trajectories in young childrenlearning new words. However, very little work has examinedthe interaction of competition amount, competitionvariability, and task demands in adults. The current studyassesses adults’ ability to map new word-referent pairs invarying amounts of competition and competitor variability. Inaddition, the effect of mapping context on retention wasassessed. Results suggest that retention is weak in some casesand importantly, there are cascading effects of competitorvariability in mapping on later retention of new words.Results are discussed in light of associative learningmechanisms and the implications of competition for learning.

Explaining how intelligent systems come to embody knowl-edge of deductive concepts through inductive learning is afundamental challenge of both cognitive science and artificialintelligence. We address this challenge by exploring how adeep reinforcement learning agent, occupying a setting simi-lar to those encountered by early-stage mathematical conceptlearners, comes to represent ideas such as rotation and trans-lation. We first train a Dueling Deep Q-Network on a shapesorting task requiring implicit knowledge of geometric proper-ties, then we query this network with classification and prefer-ence selection tasks. We demonstrate that scalar reinforcementprovides sufficient signal to learn representations of shape cat-egories. After training, the model shows a preference for moresymmetric shapes, which it can sort more quickly than lesssymmetric shapes, supporting the view symmetry preferencesmay be acquired from goal-directed experience.

We propose causal agency models for representing and reason-ing about ethical dilemmas. We find that ethical dilemmas, al-though they appear similar on the surface, differ in their formalstructure. Based on their structural properties, as identified bythe causal agency models, we cluster a set of dilemmas in Type1 and Type 2 dilemmas. We observe that for Type 1 dilemmasbut not for Type 2 dilemmas a utilitarian action does not domi-nate the possibility of refraining from action thereby constitut-ing a conflict. Hence, we hypothesize, based on the model, thatType 1 dilemmas are perceived as more difficult than Type 2dilemmas by human reasoners. A behavioral study where par-ticipants rated the difficulty of dilemmas supports the models’predictions.

Recent work suggests that language production exhibits a biastowards efficient information transmission. Speakers tendto provide more linguistic signal for meaning elements thatare difficult to recover while reducing contextually inferrable(more frequent, probable, or expected) elements. This trade-off has been hypothesized to shape grammatical systems overgenerations, contributing to cross-linguistic patterns. We putthis idea to an empirical test using miniature artificial languagelearning over variable input. Two experiments were conductedto demonstrate that the inferrability of plurality informationinversely predicts the likelihood of overt plural marking, aswould be expected if learners prefer communicatively efficientsystems. The results were obtained even with input frequencycounts of the plural marker counteracting the bias, and thusprovide strong support for a critical role of inferrability ofmeaning in language learning, production, as well as in typo-logically attested variations.

Many common tasks require or are made more efficient by coordinating with others. In this paper we investigate the coordination dynamics of a joint action pick-and-place task in order to identify the behavioral dynamics that underlie the emergence of human coordination. More precisely, we introduce a task dynamics approach for modeling multi-agent interaction in a continuous pick-and-place task where two agents must decide to work together or alone to move an object from one location to another. Our aims in the current paper are to identify and model (1) the relevant affordance dynamics that underlie the selection of the different action modes required by the task and (2) the trajectory dynamics of each actor’s hand movements when moving to grasp, relocate, or pass the object. We demonstrate that the emergence of successful coordination can be characterized in terms of behavioral dynamics models which may have applications for artificial agent design.

When a neural network is trained on multiple analogous tasks,previous research has shown that it will often generate rep-resentations that reflect the analogy. This may explain thevalue of multi-task training, and also may underlie the powerof human analogical reasoning – awareness of analogies mayemerge naturally from gradient-based learning in neural net-works. We explore this issue by generalizing linear analysistechniques to explore two sets of analogous tasks, show thatanalogical structure is commonly extracted, and address somepotential implications.

When participants view and then reproduce simple objects thatvary along a continuous dimension such as length or shade, orwhen they view images of faces that vary in emotional expression,their estimates tend to be biased toward the average value of thepresented objects, a phenomenon that has been modeled as theresult of a Bayesian combination of prior category knowledge withan imprecise memory trace (Corbin, Crawford & Vavra, 2017;Huttenlocher, Hedges & Vevea, 2000). Whereas previous workdescribed a general cognitive strategy based on data aggregatedacross participants, here we examined individual differences instrategy. Thirty-six participants viewed and reproduced 496morphed face stimuli that ranged from angry to happy. We foundsubstantial variation in the bias patterns participants produced.Individuals’ estimates were well fit by a model that positedattraction toward three categories, one at the happy end of therange, one at the angry end, and one that captured the entire rangeof presented stimuli, and by allowing the weight given to eachcategory to vary by participant.

Studies of unconscious plagiarism have reported that peoplemistakenly include a partner’s responses when trying to recalltheir own (recall-own task) and include own responses whentrying to recall their partner’s (recall-partner task). In a simula-tion, we tested if participants’ memory performance at test, in-cluding source errors, can be explained by participants simplyguessing items that come easily to mind. We show that guess-ing alone cannot account for the pattern of data participantsshow at test. Modifying the simulation by including memoryfor self-generated items allows us to replicate the pattern of re-sponding in the recall-own but not the recall-partner task, evenwhen we assume that participants in the recall-partner taskstrategically withhold more fluent items from report. This sug-gests that judgements of items’ memory strength alone cannotexplain performance in the unconscious plagiarism paradigm.

The transformational theory of similarity suggests that whenjudging similarity, people are sensitive to the number of trans-formation operations needed to make two compared repre-sentations match. Although this theory has been influential,little is known about how transformations are learned andto what extent learned transformations affect similarity judg-ments. This paper presents two experiments addressing thesequestions, in which people learned categories defined by atransformation. In Experiment 1, when the transformationswere directly visible, people had no trouble learning and ap-plied their knowledge to similarity and categorization judg-ments involving previously unseen items. In Experiment 2,the task required transformations to be inferred rather than ob-served. People were still able to learn the categories, but inthis more difficult case ratings were less strongly affected bytraining. Overall, this work suggests that newly learned trans-formations can impact similarity judgments but the salience ofthe transformation has a large impact on transfer.

Perseverance, above and beyond IQ, predicts academic outcomes in school age children, however, little is knownabout what factors affect persistence in early childhood. Here, we propose a formal Bayesian model of how children mightlearn how to calibrate effort from observing adult models and then explore this idea behaviorally across two experiments inchildren and infants. Results from Experiment 1 show that preschoolers persist more after watching an adult persist, but onlyif the adult is successful at reaching their goal. Experiment 2 and a pre-registered replication extend these findings, showingthat even infants use adult models to modulate their persistence, and can generalize this inference to novel situations. Theseresults suggest that both preschoolers and infants are sensitive to adult persistence and use it to calibrate their own effort infar-reaching ways.

Usage-based theories of syntax predict that words and syntactic constructions are probabilistically interconnected. If this is true, then words that occur in similar distributions of syntactic constructions should prime each other. These effects should be fine-grained; even small differences between the syntactic distributions of pairs of words of the same grammatical category should cause variation in priming. Prior research from production suggests that this prediction should hold even in tasks without any syntactic requirement. In this study, we introduce a measure of the similarity between the syntactic contexts in which two nouns occur. We show that this similarity measure significantly predicts visual lexical decision priming magnitudes between pairs of nouns. This finding is consistent with the predictions of usage-based theories where fine-grained similarity of syntactic usages between prime-target pairs affects decision latencies, over and above any effects attributable to semantic similarity.

In many natural domains, risks and rewards are inversely re-lated (Pleskac & Hertwig, 2014). We sought to understandhow people might use this relationship in choosing amongrisky gambles. To do so we, manipulated risk-reward struc-tures of monetary gambles to be either negatively or positivelycorrelated, or uncorrelated. After substantial exposure to theseenvironments, participants completed a speeded choice taskamong non-dominated gambles. Eye-tracking data from thistask suggests that participants often shifted their attention tomainly one attribute in the correlated conditions, in which therisk-reward relationship was present. This was an adaptivestrategy that resulted in a similar proportion of expected-valuemaximizing choices, compared to a more compensatory pro-cessing strategy.

Vectors derived from patterns of co-occurrence of words inlarge bodies of text have often been used as representations ofsome aspects of the meanings of different words. Generally,the distance between such vectors is used as a measure of thesemantic similarity between the word meanings theyrepresent. One important way of evaluating the performanceof these vectors has been to use them to answer vocabularymultiple choice questions (MCQs) where the participant isasked to judge which of several choice words is closest inmeaning to a stem word. The existing vocabulary MCQ testsused in this way have been very useful but there are somepractical problems in their use as general evaluationmeasures. Here, we discuss why such tests remain usefulevaluation measures, introduce a new vocabulary test,evaluate several current sets of semantic vectors using thenew test and compare their performance to human data.

How do young children decide to trust testimony thatcontradicts their initial beliefs? The current study examinedwhether children rely on cues to informant credibility (i.e.,history of accuracy) to determine if they would endorse anunexpected label from an informant. Three- and 4-year-olds(N = 60) saw a picture of a hybrid artifact that consisted offeatures of two typical familiar artifacts. Children made initialjudgments about the name of the hybrid object andsubsequently received a different name offered by aninformant who had earlier either accurately or inaccuratelynamed familiar objects. Children were more willing to revisetheir own judgment and accept the unexpected label if it wasfrom a previously accurate informant than if it was fromsomeone who had made obvious naming errors. This suggeststhat preschool-aged children selectively revise their ownknowledge; they are more trusting toward sources provenaccurate than inaccurate.

Music notation and English word reading have similar visualprocessing requirements. It remains unclear how the twoskills influence each other. Here we investigated the modula-tion of music reading expertise on visual processing of Eng-lish words through an ERP study. Participants matched Eng-lish real, pseudo, and non-words preceded by musical seg-ments or novel symbol strings in a sequential matching task.Musicians showed smaller N170 amplitude in response toEnglish non-words preceded by musical segments than bynovel symbol strings in the right hemisphere. This effect wasnot observed in real or pseudo-words, or in any of non-musicians’ responses. Similar to English non-words, musicalsegments do not have morphological rules or semantic infor-mation, giving rise to this modulation effect. This findingsuggested a shared visual processing mechanism in the righthemisphere between music notation and English non-wordreading, which may be related to serial symbol processing assuggested by previous studies.

Metaphor pervades discussions of important socio-politicaltopics. Recent research indicates that metaphorical languagecan influence how people reason about such topics, potentiallyaffecting real-world decision-making. In this study, we reporton research into the effects of metaphor on evaluative judg-ment, another aspect of decision-making that has been lesswell studied than reasoning. We use a cross-linguistic differ-ence in the metaphors used by English and Spanish speakersto discuss economic change to investigate how metaphoricallanguage affects evaluative judgment. We show that the imageschematic information inherent in the semantics of the differ-ent metaphors performs a central role in shaping this process.

Language entails many nested time scales, ranging from therelatively slow scale of cultural evolution to the rapid scale ofindividual cognition. The nested, multiscale nature oflanguage implies that even simple acts of text production,such as typing a sentence, entail complex interactionsinvolving multiple concurrent processes. As such, textproduction may have much in common with other cognitivephenomena thought to emerge from multiplicativeinteractions across temporal scales, namely those that exhibitfractal properties. We investigated the relationship betweenfractal scaling and the quality of produced text. Participants(N=131) wrote essays while their keystrokes were recorded.Fractal analyses were then performed on time series ofinterkeystroke intervals (IKIs). Results showed that fractalproperties characterizing IKIs positively predicted expertratings of essay quality, even after accounting for essaylength. The results support our hypotheses concerningmultiscale coordination and text production.

Visuomotor adaptation plays an important role in motor plan-ning and execution. However, it remains unclear how senso-rimotor transformations are recalibrated when visual and pro-prioceptive feedback are decoupled. To address this question,the present study asked participants to reach toward targets ina virtual reality (VR) environment. They were given visualfeedback of their arm movements in VR that was either con-sistent (normal motion) with the virtual world or reflected (re-versed motion) with respect to the left-right and vertical axes.Participants completed two normal motion experimental ses-sions, with a reversed motion session in between. While re-action time in the reversed motion session was longer than inthe normal motion session, participants showed the learningimprovement by completing trials in the second normal mo-tion session faster than in the first. The reduction in reactiontime was found to correlate with greater use of linear reach-ing trajectory strategies (measured using dynamic time warp-ing) in the reversed and second normal motion sessions. Thisresult appears consistent with linear motor movement plan-ning guided by increased attention to visual feedback. Suchstrategical bias persisted into the second normal motion ses-sion. Participants in the reversed session were grouped intotwo clusters depending on their preference for proximal/distaland awkward/smooth motor movements. We found that partic-ipants who preferred distal-smooth movements produced morelinear trajectories than those who preferred proximal-awkwardmovements.

Whereas adults exchange conversational turns very rapidly, children often respond after long gaps. However, it hasbeen proposed that the infrastructure necessary to take turns develops in infancy. Why are children slow to respond to turns?Adults’ turn-taking skills, it has been argued, rely on an ability to both predict when the current turn will end and prepare aresponse as soon as possible. In two experiments, we ask how these two abilities (prediction and early preparation) develop.Adults and 3-to-5-year-olds answered yes/no questions while playing an iPad-based maze game. Distributional analysis ofanswer latencies suggest that (i) neither children nor adults rely on fine-grained predictions of turn duration and (ii) bothchildren and adults use predictions about turn content to prepare their answer early. In sum, by the age of three, childrenalready have the cognitive architecture necessary to take turns successfully. ,

Evidence that higher cognitive processes are coupled in agraded and time-continuous way to sensory-motor processescomes, in part, from mouse-tracking studies. In these, curvedmouse trajectories toward one of two fixed response locationsreveal the evolution of certainty about a cognitive task that par-ticipants solve. We present a paradigm in which selection ofthe response location is itself the cognitive task. From amongitems in a visual scene, participants select a target that is de-scribed by a spatial relation (e.g.,“the red to the left of thegreen”), where one target item (here, “red”) matches the de-scription better than alternative same-colored targets. In themouse trajectories, we find clear evidence for attraction tothe alternative targets, attraction to the reference item (here“green”), and an early biasing influence of the spatial term.

Probabilistic prediction is a central process in language com-prehension. Properties of probability distributions over predic-tions are often difficult to study in natural language. To obtainprecise control over these distributions, we created artificiallanguages consisting of sequences of shapes. The languageswere constructed to vary the uncertainty of the probability dis-tribution over predictions as well as the probability of the pre-dicted item. Participants were exposed to the languages in aself-paced presentation paradigm, which provides a measureof processing difficulty at each element of a sequence. Therewas a robust pattern of graded predictability: shapes were pro-cessed faster the more predictable they were, as in natural lan-guage. Processing times were also affected by the uncertainty(entropy) over predictions at the point at which those predic-tions were made; this effect was less consistent, however.

How does actively seeking explanations for one’sobservations affect information search over the course oflearning? Generating explanations could plausibly leadlearners to take advantage of the information they havealready obtained, resulting in less exploration. Alternatively,explaining could lead learners to explore more, especiallyafter encountering evidence that suggests their current beliefsare incorrect. In two experiments using a modified observe orbet task, we investigate these possibilities and find support forthe latter: participants who are prompted to explain theirobservations in the course of learning tend to explore more,especially after encountering evidence that challenges acurrent belief.

Conceptual representations in language processing employ both linguistic distributional and embodied information. Here, we aim to demonstrate the roles of these two components in metaphor processing. The linguistic component is captured by linguistic distributional frequency (LDF), that is, how often the constituent words appear together in context. The embodied component, on the other hand, refers to how easy it is to generate an embodied simulation, operationalised by a previous norming study. In the current study, we looked at the interplay of these components in metaphor processing, and investigated their roles at different depths of processing in two experiments. Thus, we required participants to engage in shallow processing (Experiment 1: Sensibility Judgement), or deep processing (Experiment 2: Interpretation Generation). Results showed that the increase of both variables made it more likely to accept a metaphor. However, whereas ease of simulation (EoS) contributed to the speed of processing at both levels of depth, LDF only affected the speed in shallow processing. Specifically, LDF acted as a heuristic, both to speed up responses to accept metaphors as sensible when the frequency is high, and to flag up potentially unsuccessful processing when it is low. Overall, these results support views of language processing that emphasise the importance of both linguistic and embodied components according to task goals.

‘Activity-equivalent’ food labels are believed to encourageconsumers to partake in exercise. This may occur by semanticpriming, where featuring images of physical activity increasesthe mental accessibility of the concept of exercise, making itmore ‘fluent’ and therefore more influential on people’sbehaviour. We tested how the format of labels (image vs.text) and representation of energy (‘activity’ vs. ‘calorie)affected mental accessibility of exercise in a word-fragmentcompletion task and participants’ behavioural intentions forexercise (N = 142). Participants exposed to calorie labelsproduced more exercise-related words and viewed animagined exercise scenario as shorter and more enjoyable.Images led to higher intentions to exercise than text whenthey described activities but they led to lower intentions toexercise than text when they described calories. Findingssuggest that activity labels do not trigger more activity relatedthoughts, but could increase exercise intentions only ifpresented in pictorial format.

Sentence understanding is affected by recent experience. Animportant open question is whether this reflects adaptationto the statistics of the input. Support for this hypothesiscomes from the recent finding that listeners can simultaneouslylearn and maintain the syntactic statistics of multiple talkers(Kamide, 2012). We attempt—and fail—to replicate this find-ing. This calls into questions whether recency effects in sen-tence processing originate in the same adaptive mechanismsoperating during speech perception (for which talker-specificadaptation is well-established).

In cognitive science, computation is largely accompanied with representational theory of mind. Yet, it remains unclear whether this companionship also appears in the realm of sensorimotor control. Grush’s (2004) and Pezzulo’s (2008, 2011) account of anticipatory representations provide a limited answer, as they are only suitable for forward models, but not the entire sensorimotor control. Rescorla’s (2016) representational explanation for sensorimotor psychology addresses several intentional states considered in Bayesian inference and optimal modeling. However, the above accounts does not explain how motor commands are produced and chosen in the course of sensorimotor control for maintaining accuracy of goal-achievement. The present paper aims to explain it with a representational account by considering instrumental representations of sensorimotor control, which appear at the intermediate level and are exemplified by motor commands and costs. Such representations do not presume decouplability, as they need to run on-line in the maintenance of accuracy.

We offer a re-evaluation of the tone-monitoring technique inthe study of parsing. Experiment 1 shows that reaction times(RTs) to tones are affected by two factors: a) processing load,resulting in a tendency for RTs to decrease across a sentence,and b) a perceptual effect which adds to this tendency andmoreover plays a role in neutralising differences between sen-tence types. Experiment 2 successfully discriminates thesetwo factors by registering event-related brain potentials dur-ing a monitoring task, establishing that the amplitudes of theN1 and P3 components —the first associated with temporaluncertainty, the second with processing load— correlate withRTs. Experiment 3 then behaviourally segregates the two fac-tors by placing the last tone at the end of sentences, activating awrap-up operation and thereby both disrupting the decreasingtendency and highlighting structural factors.

According to most recent theories of multisensory integration,weighting of different modalities depends on the reliability ofthe involved sensory estimates. Top-down modulations havebeen studied to a lesser degree. Furthermore, it is still debatedwhether working memory maintains multisensory informationin a distributed modal fashion, or in terms of an integrated rep-resentation. To investigate whether multisensory integrationis modulated by task relevance and to probe the nature of theworking memory encodings, we combined an object interac-tion task with a size estimation task in an immersive virtualreality. During the object interaction, we induced multisen-sory conflict between seen and felt grip aperture. Both, visualand proprioceptive size estimation showed a clear modulationby the experimental manipulation. Thus, the results suggestthat multisensory integration is not only driven by reliability,but is also biased by task demands. Furthermore, multisensoryinformation seems to be represented by means of interactivemodal representations.

We present a computational model exploring goal-directeddeployment of attention during object tracking. Once selected,objects are tracked in parallel, but serial attention can bedirected to an object that is visually crowded and in danger ofbeing lost. An attended object’s future position can beextrapolated from its past motion trajectory, allowing theobject to be tracked even when it is briefly occluded. Usingthe model, we demonstrate that the difficulty of trackingthrough occlusions increases with the number of objectsbecause they compete for serial attention.

From distracted driving, to work focus on a computer,increasing amounts of research is investigating how digitaltechnology influences users’ attention. A couple of widelycited studies have found that the mere presence of cell phonesinterferes with social interactions and cognitive performance,even when not actively in use. These studies have importantimplications but they have not yet been replicated, and alsosuffer from methodological shortcomings and lack ofestablished theoretical frameworks to explain the observedeffects. We improved the methodology used in a previousstudy of phone presence and task performance (Thornton,Faires, Robbins, & Rollins, 2014), while testing an‘opportunity cost’ model of mental effort and attention(Kurzban, Duckworth, Kable, & Myers, 2013). We wereunable to replicate Thornton et al.’s finding that presence ofcell phones reduces performance in a specific cognitive task(additive digit cancellation). Moreover, contrary to ourexpectations, we found that participants who used theirphones more, and who were more attached to them, found thetasks more fun/exciting and effortless, if they completed themwith their phone present.

Predicting the effect of persuasion campaigns is difficult, asbelief changes may cascade through a network. In recentyears, political campaigns have adopted micro-targetingstrategies that segment voters into fine-grained clusters formore specific targetting. At present, there is little evidencethat explores the efficiency of this method. Through anAgent-Based Model, the current paper provides a novelmethod for exploring predicted effects of strategic persuasioncampaigns.The voters in the model are rational and revise their beliefsin the propositions expounded by the politicians inaccordance with Bayesian belief updating through a sourcecredibility model.The model provides a proof of concept and shows strategicadvantages of micro-targeted campaigning. Despite havingonly little voter data allowing crude segmentation, the micro-targeted campaign consistently beat stochastic campaignswith the same reach. However, given substantially greaterreach, a positively perceived stochastic candidate can nullifyor beat a strategic persuasion campaigns.

Conspiracy theories cover topics from politicians to worldevents. Frequently, proponents of conspiracies hold thesebeliefs strongly despite available evidence that may challengeor disprove them. Therefore, conspiratorial reasoning hasoften been described as illegitimate or flawed. In the paper,we explore the possibility of growing a rational (Bayesian)conspiracy theorist through an Agent-Based Model. The agenthas reasonable constraints on access to the total informationas well its access to the global population.The model shows that network structures are central tomaintain objectively mistaken beliefs. Increasing the size ofthe available network yielded increased confidence inmistaken beliefs and subsequent network pruning, allowingfor belief purism. Rather than ameliorating and correctingmistaken beliefs (where agents move toward the correctmean), large networks appear to maintain and strengthenthem. As such, large networks may increase the potential forbelief polarization, extreme beliefs, and conspiratorialthinking – even amongst Bayesian agents.

The dilution effect occurs when the introduction of non-diagnostic information lessens the impact on reasoning ofdiagnostic information despite having no relevance to thehypothesis in question. While the effect has been reproducedin several studies, the psychological basis of the effectremains unclear. Some believe it to be conversational whileothers believe it to be cognitive and social.The paper tests the conversational basis of the effect byminimising pragmatic, conversational influence. To this end,it makes use of a legal setting with witness testimonies. Thestudies replicate the dilution effect, which suggests that thebasis of the results in the original studies is notconversational. However, the credibility of the sourcestrongly influences whether or not the effect occurs. Ifreliable sources provide the non-diagnostic information, theeffect lessens. Conversely, if unreliable sources provide thenon-diagnostic information, we observe a stronger dilutioneffect.

This study focuses on the infuence of a three-dimensional (3D)graphic image and a 3D-printed object on the spatial reasoningof experts and novices in the medical field. The spatial rea-soning task of this study required doctors specializing in di-gestive surgery to infer cross sections of a liver with a 3D im-age and a 3D-printed object in a situation where liver resectionsurgery was simulated. The task performance was comparedwith that of university students who conducted the same task inMaehigashi et al. (2016). The results of the analysis indicatedthat the doctors showed the same task performance when usingthe 3D image and the 3D-printed object. However, the univer-sity students learned faster and inferred the inside of a liverstructure more accurately with the 3D-printed object than withthe 3D image, and they performed equally to the professionaldoctors. Our results are then discussed in relation to previousstudies.

Several recent empirical findings have reinforced the notion that a basic learning and memory skill—chunking—plays a fundamental role in language processing. Here, we provide evidence that chunking shapes sentence processing at multiple levels of linguistic abstraction, consistent with a recent theoretical proposal by Christiansen and Chater (2016). Individual differences in chunking ability at two different levels is shown to predict on-line sentence processing in separate ways: i) phonological chunking ability, as assessed by a variation on the non-word repetition task, predicts processing of complex sentences featuring phonological overlap; ii) multiword chunking ability, as assessed by a variation on the serial recall task, is shown to predict reading times for sentences featuring long-distance number agreement with locally distracting number-marked nouns. Together, our findings suggest that individual differences in chunking ability shape language processing at multiple levels of abstraction, consistent with the notion of language acquisition as learning to process.

Success in Algebra I often predicts whether or not a studentwill pursue higher levels of mathematics and science.However, many students enter algebra holding persistentmisconceptions that are difficult to eliminate, thus, hinderingtheir ability to succeed in algebra. One way to address thesemisconceptions is to implement worked-examples and self-explanation prompts, which have been shown to improvestudents’ conceptual knowledge. However this effect seems tobe greater after a delay. The current study sought to exploresuch time-related effects on algebra conceptual knowledge. Ina year-long random-assignment study, students either studiedworked-examples and answered self-explanation prompts (n =132) or solved typical isomorphic problems (n = 140). Athree-way mixed ANCOVA (pre-algebra knowledge xcondition x time) found a significant condition by time effect.The growth of algebra conceptual knowledge was greater forstudents studying worked-examples than for those solvingtypical problems.

Are different amounts of semantic processing associated with different semantic ambiguity effects? Could this explain some discrepant ambiguity effects observed between and across tasks? Armstrong and Plaut (2016) provided an initial set of neural network simulations indicating this is indeed the case. However, their empirical findings using a lexical decision task were not clear-cut. Here, we use improved methods and five different experimental manipulations to slow responding- --and the presumed amount of semantic processing---to evaluate their account more rigorously. We also expanded the empirical horizon to another language: Spanish. The results are partially consistent with the predictions of the neural network and differ in several important ways from English data. Potential causes of these discrepancies are discussed in relation to theories of ambiguity resolution and cross-linguistic differences.

Humans have a strong “cognitive compulsion” to infer in-tentional agents from violation of randomness and such anagency–nonrandomness link emerges early in development.In two studies, we directly quantified, formalized, and com-pared both ends of this link for the first time. In Experiment1, two groups of participants viewed the same 256 binary se-quences (e.g., AABAAABA) and classified each as generatedby agents/non-agents or by nonrandom/random processes. Wefound a strong correlation between two judgments: sequencesviewed as more agentive also tended to be judged as less ran-dom. In Experiment 2, another two groups were asked toproduce sequences that others might appreciate as agentive ornonrandom. Participant-generated sequences in the two con-ditions had a substantial overlap, indicating common guidingprinciples of agency and nonrandomness generation. Taken to-gether, the present studies provide evidence for a shared cog-nitive basis of agency detection and subjective randomness.

Whereas much is known about how humans categorize andreason based on absolute quantities, research investigating theprocessing of relative quantities, such as proportions, iscomparatively limited. The current study used a Stroop-likeparadigm to examine adults’ automatic processing ofnonsymbolic proportions and how presentation formatsmodulate this processing. Participants were asked to compareindividual components across proportions in six differentpresentation formats. Congruity between component size andoverall proportion affected accuracy of comparison, such thatparticipants were less accurate when proportion (the irrelevantdimension) was incongruent with absolute quantity (therelevant) dimension. Moreover, the congruity effect wasmodulated by the presentation format. These findings serve asevidence that humans automatically access relative quantitywhen presented in nonsymbolic formats and provide evidencethat the strength of this processing is modulated by the formatof presentation.

Participants can adequately take into account several cues regarding the weight they should grant majority opinions,but that they do not consistently take into account cues regarding whether the members of the majority have formed theiropinions independently of each other. We suggest that these conflicting results can be explained by hypothesizing that somecues are evolutionarily valid (i.e. they were present and reliable during human evolution), and others not. Using this frameworkwe derive and test hypotheses about two facets of informational dependency. The first 3 experiments show that participantsadequately take into account cues to informational dependency when they are presented in a simple, evolutionarily valid way.Experiments 4 to 7 show that people consistently take into account shared motivation, but not shared cognitive traits, as a sourceof potential dependency, as predicted by the likely greater importance of differences in motivation during our evolutionaryhistory.

Children’s ability to direct attention to salient stimuli is a key aspect of cognitive functioning. Here we examined the magnitude and lateralization of EEG indices during somatosensory anticipation elicited by a left or right directional cue indicating the bodily location of an upcoming tactile stimulus. In 50 children aged 6-8 years, somatosensory anticipation was accompanied by anticipatory negativity and alpha mu rhythm desynchronization at contralateral central electrode sites (C3 and C4) overlying the hand area of the somatosensory cortex. Individual differences in these contralateral brain responses during somatosensory anticipation were associated with scores on a flanker task of executive function. The results suggest that processes involved in directing attention in the tactile modality may overlap with those involved in broader executive function abilities.

Segmenting observations from an input stream is an impor-tant capability of human cognition. Evidence suggests that hu-mans refine this ability through experiences with the world.However, few models address the unsupervised developmentof event segmentation in artificial agents. This paper presentswork towards developing a computational model of how anintelligent agent can independently learn to recognize mean-ingful events in continuous observations. In this model, theagent’s segmentation mechanism starts from a simple stateand is refined. The agent’s interactions with the environ-ment are unsupervised and driven by its expectation failures.Reinforcement learning drives the mechanism that identifiesevent boundaries by reasoning over a gated-recurrent neuralnetwork’s expectation failures. The learning task is to reduceprediction error by identifying when one event transitions intoanother. Our experimental results support that reinforcementlearning can enable detecting event boundaries in continuousobservations based on a gated-recurrent neural network’s pre-diction error and that this is possible with a simple set of fea-tures.

In this paper, I review the objections against the claim that brains are computers, or, to be precise, information- processing mechanisms. By showing that practically all the popular objections are based on uncharitable (or simply incorrect) interpretations of the claim, I argue that the claim is likely to be true, relevant to contemporary cognitive (neuro)science, and non-trivial.

The standard approach posits that analogical inferences aregenerated by copying unmapped base relations, substitutingbase entities by their corresponding target ones, and generatingslots for unmapped base entities. Contra this account, resultsfrom Experiment 1 revealed that analogical inferences seldominclude relations that resemble the base relation from whichthey were derived. Most of the inferences, however, could becategorized as exemplars of a schema-governed categorycapable of characterizing the base information to be projected.To gather further precision about the criteria that guide inferencegeneration, in Experiment 2 we showed that analogicalinferences tend to match the base information from which theyare derived in values of salient dimensions of the relationalcategory to which they belonged. Our results suggest that therelational constructs employed in modeling analogical inferenceshould move beyond one-term multiplace predicates so as toinclude more complex relational structures.

We investigate how people allocate a limited set of resourcesbetween multiple risky prospects. We found that only a smallpercentage of decisions followed some form of naive diversifi-cation or mean-variance optimization. In general, people wereless mean-variance optimal than a naive 1/N heuristic. As-pects of choice sets, such as domain, skew, and second orderstochastic dominance, affected resource allocation decisions ina similar manner to their influence on single choice gambles.Individual traits traditionally linked to risk propensity seem tomanifest in terms of the degree to which people are inclinedto diversify. Lower risk aversion and higher risk seeking traitsare linked to increasing diversification. Risk congruency, thedegree to which peoples’ self-reported and elicited risk aver-sion matches, moderates how susceptible people are to costframing nudges. We find evidence for heterogeneous clusterswhere people either under-weight or over-weight segregatedcosts, leading to the same nudge producing opposite behav-ioral results within two risk incongruent groups.

Working memory (WM) capacity is critically important for the success in school and complex cognitive activitiesacross the lifespan. Training WM skills has shown to lead to improvements in a variety of important cognitive tasks. One’sperformance on an adaptive and challenging longitudinal WM intervention may serve as an assay of cognitive plasticity. Withover 400 participants having completed a minimum of 15 sessions of WM training, we have a rich dataset that allows investigat-ing individual differences and other factors that might determine training outcome using a novel machine learning techniques.Preliminary results suggest that factors such as age, type of n-back, and baseline abilities significantly impact one’s ability toimprove in training. Other factors such as gender and whether or not training was supervised were not significant. Finally, ourmodel allows prediction of training gain with 78% accuracy.

We investigated people’s assessments of their own personal change over time, comparing predicted, actual, and recalled change in personality, values, and performance. On average, participants underestimated the absolute magnitude of their personal change in both prediction and recall. However, people specifically neglected negative future change, resulting in overly optimistic predictions of improvement. In contrast, recall of positive and negative change was relatively more balanced, such that assessments of past improvement were better calibrated on average. Our findings provide insight into how people think about their own identity over time and address disparate theories in the literature regarding predictions of personal stability versus improvement.

What are the cognitive processes in play when someone uses avisualization tool to interactively explore a new dataset? Here,we focus on one particular type of visualization—the scatterplot—which, despite (or perhaps because of) its simplicity, isstill one of the most frequently used plot types in many data-intensive disciplines. We conducted a pilot study to investigatehow expert astronomers interact with an unfamiliar dataset us-ing a visualization tool called Filtergraph, which supports rapidand easy visualization of large datasets. We present both quali-tative and quantitative results, including observations about thetemporal dynamics of visual data exploration as well as inter-esting behavioral patterns that we saw in our participants, suchas users taking “circular walks” through the data at various lev-els of abstraction.

Though most documented sign languages make use of spaceto denote relationships between predicate arguments, studiesof emerging sign languages suggest that spatial reference doesnot emerge fully-formed but takes time to develop. We presentan artificial sign language learning experiment that expandsthe cultural evolutionary framework to investigate complexlinguistic constructions. Our results demonstrate the gradualemergence of consistent devices to distinguish between sen-tence arguments, some of which rely on iconic spatial con-trasts. These findings mirror data from emerging sign lan-guages and point to the cultural mechanisms that facilitate theevolution of complex linguistic structures.

Successful performance during multimedia learning requiresaccurate metacognitive judgments. However, little researchhas investigated the influence of accurate metacognitivejudgments for different representations of information (e.g.,text and diagram) on performance during multimedialearning. As such, we investigated if participants’metacognitive judgments for text and diagrams (i.e., contentevaluations; CEs) were significantly related to increasedperformance and higher confidence during multimedialearning. Metacognitive judgments and performance measureswere collected from 48 undergraduate participants during 18randomized trials. Results using multilevel modelingindicated that participants’ CEs for text-based content weresignificantly predictive of performance. Results also showedthat accurate CEs for diagrams interacted with accuratemultiple-choice responses to predict higher retrospectiveconfidence judgments (i.e., higher confidence). Identifyingmetacognitive judgments predictive of increased performanceduring multimedia learning has important theoretical,conceptual, and analytical implications.

In the search for an understanding of human communication,researchers often try to isolate listener and speaker roles andstudy them separately. Others claim that it is the intertwined-ness of these roles that makes human communication special.This close relationship between listener and speaker has beencharacterized by concepts such as common ground, backchan-neling, and alignment, but they are only part of the picture. Un-derlying these processes, there must be a mechanism for mak-ing inferences about our interlocutors’ understanding of wordsand gestures that allows us to communicate robustly and effi-ciently without assuming that we take the same words to havethe same meaning. In this paper, I explore this relationship be-tween language and concepts and propose an interactive mech-anism that can facilitate these latent conceptual inferences. Fi-nally, I show how this proposal paves the way for a more pre-cise account of the role of interaction in communication.

The purpose of this study is to demonstrate a constraint relaxation which is followed by the transition to an appropriate representation in insight problem solving. The puzzle game “Tangram” was used as a new insight problem, in which problem-solvers were presented a silhouette and asked to make the same configuration by arranging 7 pieces. At the beginning, problem-solvers had a constraint allocating the pieces into a geometric shape, but then relaxed this to reach the correct configuration at a later stage of problem solving. Participants’ subjective assessments of their confidence to reach the solution predicted neither the constraint relaxation nor the successful problem solving. However, eye-tracking data suggested that the successful problem-solvers tended to search the problem space more widely than the unsuccessful-problem solvers.

Biological systems are capable of acting in a shared environment to produce emergent, self-organized behavior that is the result of the constraints imposed by local interactions– such as bird flocking or ant swarming behavior. These examples present minimal demands for a shared-intention between co-actors, whereas other instances necessitate the formation of a shared goal. In these goal-directed tasks, how much of the observed complexity can be explained by the constraints imposed by both the environment and adherence to the shared task goal? This paper begins to investigate this question by presenting results from a two-person cooperative “shepherding” task first developed in Nalepka et al. (2017) but with fewer constraints. Results provide further evidence that the emergent behavior is the result of the constraints imposed by the task. The included task-dynamic model suggests a general model that can be used to understand multiagent herding behavior in a variety of contexts.

People rely on various frames of reference (FORs), such as egocentric (EFOR) and intrinsic (IFOR), to represent spatial information. The present study examined electroencephalogram profiles on a two-cannon task, which could regulate the conflict of IFOR-IFOR (red cannon, blue cannon) and IFOR-EFOR (target cannon, observer), to elucidate the brain mechanisms of FOR conflict processing by using event-related potentials (ERPs). Results showed that both of the conflicts occurred in the reaction time (RT) and there was an interaction between them. ERP results showed more negative amplitudes on N2 (276-326 ms) and P3 (396- 726 ms) for IFOR-IFOR conflict of the 180° cannon angle condition and EFOR-IFOR conflict of the target cannon point-down condition. What’s more, there was also an interaction between these two conflicts on the P3 amplitudes (561-726 ms). In summary, our findings shed new light on the domain-specific conflict monitoring and domain-general executive control for the IFOR-IFOR and EFOR-IFOR conflicts.

Studies of social cognition often assume a reductionist, computational-representational conceptual framework. Distributed cognition is one of the few extant conceptual frameworks for a nonreductive understanding of social cognition. This concept’s prototypical cases are exclusively of technical-scientific human institutions, including ships, cockpits, and the Hubble Space Telescope. In the first part of the paper, we outline the properties of distributed cognitive systems. We look at the case of wolf (Canis lupus) packs as an instance of distributed cognition in nonhuman systems. Nevertheless, a broad range of social cognitive phenomena across human and animal populations may not fit into this conceptual framework. We present a case study of bird flocks as a counterexample to distributed cognition. We propose “swarm intelligence” as an alternative concept of nonreductive social cognition. This is not to replace distributed cognition as a concept, but to add to and diversify the taxonomy of nonreductive social cognitive systems.

It is well-known that the auditory perception of speech sounds is strongly influenced by the phonetic categorieswhich divide up acoustic space. This paper approaches the problem of modeling categorical perception from the ground up,using a linear model of the tuning properties of auditory neurons – the spectro-temporal receptive field (STRF). An STRFwhich discriminates voiced from voiceless stops was derived from the TIMIT corpus, and two computer simulations wereconducted to investigate its properties. In one simulation, this model neuron was found to exhibit a categorical response to alinear voice-onset-time continuum, closely tracking human behavior. In the second simulation, the STRF was found to exhibita less categorical, more linear response to a stop-voicing continuum, also in line with human behavior. These two simulationsshow that perceptual responses to speech, whether non-linear or veridical, can be modeled by the action of auditory neurons.

Past research has found that mindfulness meditation trainingimproves executive attention and that this effect could bedriven by more efficient allocation of resources on demandingattentional tasks, such as the Flanker Task. However, it is notclear whether these changes depend on long-term practice.We sought to investigate the effects of a brief, 10-minutemeditation session on attention in novice meditators,compared to a control activity. We also tested moderation byindividual differences in Neuroticism. We found thatparticipants randomly assigned to meditate for 10 minutesshowed improved performance on incongruent trials on aFlanker task, with no detriment in reaction times, indicatingbetter allocation of resources. Neuroticism moderated thiseffect, as only those low in Neuroticism showed improvedallocation of attentional resources following meditation.

Understanding proportions is a time-intensive process that does not come cheap during late childhood and early adolescence. It is fostered by learning experiences in which students have opportunities to explore, discuss and experiment with situations involving proportions. Children must undergo many informal learning opportunities before they can gain from direct instruction on proportional reasoning. In this study, we aimed to determine whether physics curricula focusing on the concept of density prepares students for learning from a curriculum on proportional reasoning. A 2x2 design with the factors “physics curricula” (with, without) and “concept used to introduce proportional reasoning” (speed, density) was applied to 253 children from 12 classrooms at the beginning of grade 5. We expected the “density, with physics curriculum” group to outperform the other three groups. However, only the students who scored in the highest quartile on an intelligence measure gained from the prior knowledge they had acquired through the physics curricula. The results show that curricula on proportional reasoning are worthwhile for all students in early adolescence. However, more capable students can boost their proportional reasoning if they have the chance to acquire prior knowledge through a physics curriculum.