Search

How do humans adapt when others exploit patterns in their behavior? When can people modify such patterns and when are they simply trapped? The present work explores these questions using the children's game of rock, paper, scissors (RPS). Adult participants played 300 rounds of RPS against one of eight bot opponents. The bots chose a move each round by exploiting unique sequential regularities in participant move choices. In order to avoid losing against their bot opponent, participants needed to recognize the ways in which their own behavior was predictable and disrupt the pattern. We find that for simple biases, participants were able to recognize that they were being exploited and even counter-exploit their opponents. However, for more complex sequential dependencies, participants were unable to change their behavior and lost reliably to the bots. Results provide a quantitative delineation of people's ability to identify and alter patterns in their past decisions.

How do humans adapt when others exploit patterns in their behavior? When can people modify such patterns and when are they simply trapped? The present work explores these questions using the children's game of rock, paper, scissors (RPS). Adult participants played 300 rounds of RPS against one of eight bot opponents. The bots chose a move each round by exploiting unique sequential regularities in participant move choices. In order to avoid losing against their bot opponent, participants needed to recognize the ways in which their own behavior was predictable and disrupt the pattern. We find that for simple biases, participants were able to recognize that they were being exploited and even counter-exploit their opponents. However, for more complex sequential dependencies, participants were unable to change their behavior and lost reliably to the bots. Results provide a quantitative delineation of people's ability to identify and alter patterns in their past decisions.

This is a modern horror story about an innocently misbehaving projector, and why we beseech everyone to report minimal col- orimetric data about stimulus displays. We present anecdotal experience of configuring a projector to display video stimuli in a high-tesla MRI room, along with all the gotchas, (broken) technical assumptions, and theoretical rehashings that should be considered by every scientist who uses computers to dis- play visual stimuli. The moral of our story is: (1) check that your monitor/projector is actually showing the colors and lu- minances that you think it is, (2) make explicit assumptions regarding the physical/perceptual space of your stimuli and how they relate to any model analysis you will perform. This is especially important when modeling non-human animals, since most equipment and data formats implicitly assume hu- man perception. We show that innocent changes to display settings such as brightness reduction can cause dangerously unexpected results. Understanding and reporting colorimetric data in scientific publications is important for two reasons: (1) reproducibility, and (2) model fidelity.

This is a modern horror story about an innocently misbehaving projector, and why we beseech everyone to report minimal col- orimetric data about stimulus displays. We present anecdotal experience of configuring a projector to display video stimuli in a high-tesla MRI room, along with all the gotchas, (broken) technical assumptions, and theoretical rehashings that should be considered by every scientist who uses computers to dis- play visual stimuli. The moral of our story is: (1) check that your monitor/projector is actually showing the colors and lu- minances that you think it is, (2) make explicit assumptions regarding the physical/perceptual space of your stimuli and how they relate to any model analysis you will perform. This is especially important when modeling non-human animals, since most equipment and data formats implicitly assume hu- man perception. We show that innocent changes to display settings such as brightness reduction can cause dangerously unexpected results. Understanding and reporting colorimetric data in scientific publications is important for two reasons: (1) reproducibility, and (2) model fidelity.

Deep reinforcement learning (Deep RL) methods can train artificial agents (AAs) to reach or exceed human-level performance. However, in multiagent contexts requiring competitive behavior or where the aim is to use AAs for human training, the qualitative behaviors AAs adopt may be just as important as their performance to ensure representative training. This paper compares human behaviors and performance when competing against either a human expert or an AA opponent trained using Deep RL on a 2-dimensional version of Pong. Results show that participants were not sensitive to the movement differences between the human expert and AA. Further, the participants did not alter their behaviors, except to compensate for differences in the environmental states caused by the opponents. The paper concludes with discussion on the potential impacts of AA training on human behavior with regard to representative design in the areas of skill development and team training.

This paper explores previous arguments that language evolved not from animal communication, as naturally assumed by most scholars, but instead out of animal cognition. It is proposed here that additional support for this argument comes from Cognitive Linguistics, an interdisciplinary branch of linguistics. Cognitive Linguistics mediates communication and cognition but, as with Cognitive Discourse Analysis, studies language use in terms of what it demonstrates about underlying cognitive processes and concepts. The paper demonstrates key examples of animal cognition’s links to language, with Cognitive Linguistics support, as well as the approach’s application to animal cognition in terms of domain-general symbolism, beyond verbal language. However, because communication remains a major function of language, the communicative aspect ought to be maintained in the explanation of language evolution.

People’s causal judgments exhibit substantial variability, but the processes that lead to this variability are not currently understood. In this paper, we use a repeated-measures design to study the within-participant variability of conditional probability judgments in common-cause networks. We establish that these judgments indeed exhibit substantial within-participant variability. This variability differs by inference type and is related to the extent to which participants commit Markov violations. The consistency and systematicity of this variability suggest that it may be an important source of evidence for the cognitive processes that lead to causal judgments. The systematic study of both within- and between-person variability broadens the scope of behavior that can be studied in causal cognition and promotes the evaluation of formal models of the underlying process. The data and methods provided in this paper provide tools to enable the further study of within-participant variability in causal judgment.

This paper studies social interaction in newborns and of various species, whose behavioral developments attributed to the circular sensory motor Stage 3 of behavior development. The article has associated 16 exciting facts of the social behavior in groups of different species, highlighting three questions: (1) how can newborns successfully classify social phenomena that are abstract or absent from their reality; (2) whether emotional contagion can appear through cues of body language that subjects cannot consciously perceive; (3) how do organisms distinguish identical stimuli by their importance (value) without perceptual driver stimuli. The analysis suggests that organisms can interact at initial stages of development by distinguishing cues of a similar modality by their significance (their value). This ability contributes to the assimilation of knowledge about the initial social phenomena. This paper proposes the model of Coherent intelligence, which is strictly based on experimental evidence in modern literature and laws of physics.

It has been shown that in hand-written transcription tasks temporal micro-behavioural chunk signals hold promise as measures of competence in various domains (e.g., Cheng, 2014). But data capture under that an approach requires the use of graphics tablets which are relatively uncommon. In this theoretical paper we propose and explore an alternative method – Competence Assessment by Stimulus Matching (CASM). This new method uses simple mouse-driven interfaces to produce temporal chunk signals as measures of learner’s ability. However, it is not obvious what features of CASM will produce effective competence measures and the design space of CASM tasks is large. Thus, this paper uses GOMS modelling in order to explore the design space to find factors that will maximize the discrimination of chunk measures of competence. The modelling results of this paper show that CASM has potential in using chunk signals to measure competence in the domain of English language.

Deep reinforcement learning (Deep RL) methods can train artificial agents (AAs) to reach or exceed human-level performance. However, in multiagent contexts requiring competitive behavior or where the aim is to use AAs for human training, the qualitative behaviors AAs adopt may be just as important as their performance to ensure representative training. This paper compares human behaviors and performance when competing against either a human expert or an AA opponent trained using Deep RL on a 2-dimensional version of Pong. Results show that participants were not sensitive to the movement differences between the human expert and AA. Further, the participants did not alter their behaviors, except to compensate for differences in the environmental states caused by the opponents. The paper concludes with discussion on the potential impacts of AA training on human behavior with regard to representative design in the areas of skill development and team training.

This paper studies social interaction in newborns and of various species, whose behavioral developments attributed to the circular sensory motor Stage 3 of behavior development. The article has associated 16 exciting facts of the social behavior in groups of different species, highlighting three questions: (1) how can newborns successfully classify social phenomena that are abstract or absent from their reality; (2) whether emotional contagion can appear through cues of body language that subjects cannot consciously perceive; (3) how do organisms distinguish identical stimuli by their importance (value) without perceptual driver stimuli. The analysis suggests that organisms can interact at initial stages of development by distinguishing cues of a similar modality by their significance (their value). This ability contributes to the assimilation of knowledge about the initial social phenomena. This paper proposes the model of Coherent intelligence, which is strictly based on experimental evidence in modern literature and laws of physics.

It has been shown that in hand-written transcription tasks temporal micro-behavioural chunk signals hold promise as measures of competence in various domains (e.g., Cheng, 2014). But data capture under that an approach requires the use of graphics tablets which are relatively uncommon. In this theoretical paper we propose and explore an alternative method – Competence Assessment by Stimulus Matching (CASM). This new method uses simple mouse-driven interfaces to produce temporal chunk signals as measures of learner’s ability. However, it is not obvious what features of CASM will produce effective competence measures and the design space of CASM tasks is large. Thus, this paper uses GOMS modelling in order to explore the design space to find factors that will maximize the discrimination of chunk measures of competence. The modelling results of this paper show that CASM has potential in using chunk signals to measure competence in the domain of English language.

This paper explores previous arguments that language evolved not from animal communication, as naturally assumed by most scholars, but instead out of animal cognition. It is proposed here that additional support for this argument comes from Cognitive Linguistics, an interdisciplinary branch of linguistics. Cognitive Linguistics mediates communication and cognition but, as with Cognitive Discourse Analysis, studies language use in terms of what it demonstrates about underlying cognitive processes and concepts. The paper demonstrates key examples of animal cognition’s links to language, with Cognitive Linguistics support, as well as the approach’s application to animal cognition in terms of domain-general symbolism, beyond verbal language. However, because communication remains a major function of language, the communicative aspect ought to be maintained in the explanation of language evolution.

People’s causal judgments exhibit substantial variability, but the processes that lead to this variability are not currently understood. In this paper, we use a repeated-measures design to study the within-participant variability of conditional probability judgments in common-cause networks. We establish that these judgments indeed exhibit substantial within-participant variability. This variability differs by inference type and is related to the extent to which participants commit Markov violations. The consistency and systematicity of this variability suggest that it may be an important source of evidence for the cognitive processes that lead to causal judgments. The systematic study of both within- and between-person variability broadens the scope of behavior that can be studied in causal cognition and promotes the evaluation of formal models of the underlying process. The data and methods provided in this paper provide tools to enable the further study of within-participant variability in causal judgment.

Dual-process accounts posit that human learning can occur as a consequence of both associative and propositional processes. This can be contrasted with single process accounts that suggest learning is entirely propositional. In this paper, we offer evidence for both associative and propositional processes using a within-subjects two alternative forced choice discrimination paradigm. Stimuli that varied concurrently along two dimensions were created and each participant’s awareness was directed toward one, facilitating rule induction (i.e., propositional processing) on that dimension. Performance on the other dimension was then used to assess associatively-based performance. We report results that are initially inconsistent with both single process and dual-process accounts of discrimination learning. However, we then show how an associative network, that represents stimuli integrally, can predict the performance shown by participants in the experiment, providing evidence for a dual-process account.

Predictive Processing (PP) has been suggested to account for early cognitive development (Köster et al., 2020). In this paper, we extend its application to early social coordination in parent-infant interactions. Interpersonal neural synchrony in parent-child interactions is hypothesized to be a function of the coupling of internal generative models for mutual prediction. By aligning these internal models and reducing prediction errors, social uncertainty is reduced, and interpersonal neural synchrony is enhanced. Support for this hypothesis is provided by assessing neural synchrony during mother-infant interaction.

In a recent theoretical paper, Birsch, Schnell & Clayton (2020) introduced a multidimensional framework of animal consciousness. In two online studies, we adopted their classification system and asked which of these dimensions contribute most to moral concern for non-human beings. Participants placed moral value on mental attributes more than physical similarity to humans in biology, appearance, and size. Specifically, behavioural indications of rich and complex visual processing had strong effects on both consciousness ratings and moral concern, more so than indications of self-awareness. Furthermore, moral worth was highly correlated with consciousness ratings, across items and participants. We discuss our findings in light of the philosophical debate over the moral significance of functional aspects of consciousness (Carruthers, 2019; Danaher, 2020; Levy, 2014), and in relation to the relevance of the scientific study of consciousness to ethics.

The block design test (BDT), in which a person has to recreate a visual design using colored blocks, is notable among cognitive assessments because it makes so much of a person's problem-solving strategy ``visible'' through their ongoing manual actions. While, for decades, numerous pockets of research on the BDT have identified certain behavioral variables as being important for certain cognitive or neurological hypotheses, there is no unifying framework for bringing together this spread of variables and hypotheses. In this paper, we identify 25 independent and dependent variables that have been examined as part of published BDT studies across many areas of cognitive science and present a sample of the research on each one. We also suggest variables of interest for future BDT research, especially as made possible with the advent of advanced recording technologies like wearable eye trackers.

Hindsight bias is exhibited when knowledge of an outcome (i.e., an anchor) affects subsequent recollections of previous predictions (i.e., an estimate). Hindsight bias usually leads to estimates being remembered as closer to the anchor than they actually were. The exact amount of hindsight bias exhibited depends on the anchor value and the anchor plausibility, with experimental results showing that hindsight bias is elicited only when the anchor is perceived to be plausible. In this paper we present a Bayesian model that captures the relationship between hindsight bias and anchor plausibility. This model provides a rational account of hindsight bias by considering memory recall as a statistical problem, where the goal is to reconstruct the original estimate using the anchor as new evidence. Simulations show that the modeled trends align closely with previously published human data.

Block-building is an early-developing spatial skill in which multiple spatial actions unfold over time. The processes underlying block-building have eluded our understanding, but existing data show that children’s and adults’ construction paths are highly systematic and selective. In this paper, we investigate whether manipulating the stability of block models to be built affects the character of construction paths. We asked participants to build models with either strong or weak support. We measured the step-by-step actions taken and used eye-tracking to assess differences between adults and children in how they collect and use visual information while building. We find that children and adults are highly selective in their construction paths, but models with weak support are more difficult to build, reflected in the specifics of their paths and the eye movements made during building. These results suggest that the stability of a structure drives information-gathering and building strategies in block construction.

Signalling systems of various species (humans and non-human animals) as well as our world both exhibit discrete and continuous properties. However, continuous meanings are not always expressed using continuous forms but instead frequently categorised into discrete symbols. While discrete symbols are ubiquitous in communication, the emergence of discretisation from a continuous form space is not well understood. We investigate the emergence of discrete symbols by simulating the learning process of two agents that acquire a shared signalling system. The task is formalised as a reinforcement learning problem in continuous form and meaning space. We identify two central causes for the emergence of discretisation: 1) sub-optimal signalling conventions and 2) a topological mismatch between form and meaning space. A long version of this paper has been accepted for publication in Cognition (International Journal of Cognitive Science).

A good deal of recent research has examined children’s norm learning across a wide range of novel contexts. The typical interpretation of these findings is that children’s norm learning is driven by group-based biases. In this paper, we present an alternative interpretation and corresponding meta-analyses that cast the current body of evidence in a rather different light. First, we argue the extant literature uses an ill-suited standard for assessing bias. Rather than comparing children’s judgments to what is expected under random chance (a ‘random standard’), bias is better assessed by comparing children’s judgments to what is most probable, given their total evidence (an ‘evidential standard’). Next, we report a meta-analysis of the known findings to date (k = 40 effect sizes; N = 1,369 in total; ages 4- to 13-years-old) to compare children’s norm learning against an appropriate evidential standard. Meta-analytic estimates reveal that children’s norm learning is not restrictively biased toward narrow-scope inferences on account of group-based factors. Rather, the findings to date are consistent with children’s norm learning being rational (i.e., statistically appropriate, given their evidence) or even inclusively biased toward making the wide-scope inference that a novel norm applies to everyone in a population. We conclude with brief discussion of implications for current understanding and future research on norm acquisition.

People think and learn abstractly and compositionally. These two key properties of human cognition are shared with natural language: we use a finite, composable vocabulary of nameable concepts to generate and understand a combinatorially large space of new sentence. In this paper, we present a domain of compositional reasoning tasks and an artificial language learning paradigm designed to probe the role language plays in bootstrapping learning. We discuss results from a language-guided program learning model suggesting that language can play an important role in bootstrapping learning by providing an important signal for search on individual problems, and a cue towards named, reusable abstractions across the domain as a whole. We evaluate adults on the same domain, comparing learning performance between those tasked with jointly learning language and solving reasoning tasks, and those who only approach the domain as a collection of inductive reasoning problems. We find that adults provided with abstract language prompts are better equipped to generalize and compose concepts learned across a domain than adults solving the same problems using reasoning alone.

This paper proposes a formalized approach to the specification of experimental tasks in cognitive science. Put briefly, the proposal is to represent the structure of a task by a logic program that accepts only valid experimental event logs for the chosen paradigm. It is argued that the proposed approach stands to benefit the research process at various stages as it involves the creation of executable documentation for experimental tasks, which may facilitate the communication, validation, implementation, and analysis of experimental tasks. A worked example is presented in detail and some potential new directions of research at the intersection of psychology and computer science are discussed.

Reasoning about what other people know is an important cognitive ability, known as epistemic reasoning, which has fascinated psychologists, economists, and logicians. In this paper, we propose a computational model of humans’ epistemic reasoning, including higher-order epistemic reasoning—reasoning about what one person knows about another person’s knowledge—that we test in an experiment using a deductive card game called “Aces and Eights”. Our starting point is the model of perfect higher-order epistemic reasoners given by the framework of dynamic epistemic logic. We modify this idealized model with bounds on the level of feasible epistemic reasoning and stochastic update of a player’s space of possibilities in response to new information. These modifications are crucial for explaining the variation in human performance across different participants and different games in the experiment. Our results demonstrate how research on epistemic logic and cognitive models can inform each other.

Linguistic complexity – manifested in terms of hierarchical recursive structures generated by grammar – is often discussed from the perspective of cross-linguistic comparison (cf. Everett, 2005; Nevins, Pesetsky, & Rodrigues, 2009 on Pirahã). In this paper, we focus instead on the variation in complexity within a single language, English, and on the lower bound of complexity, specifically (cf. Futrell et al., 2016). We report results of two studies, a corpus study (Study 1) and a production experiment (Study 2), that investigate syntactic complexity of expressions that arise in the context of human-computer interaction and compare them to the standard language. The results of both studies show that the expressions generated in the context of human-computer interaction exhibit lesser structural complexity and often violate the norm of the language (cf. margaret mead culture famous research). Our results suggest that such expressions are generated by a qualitatively different type of formal grammar, Linear Grammar (Jackendoff & Wittenberg, 2017), rather than by recursive grammar (Roeper, 1999).

The present paper addresses the study of cross-linguistic phonosemantic correspondences within a deep learning framework. An LSTM-based Recurrent Neural Network is trained to associate the phonetic representation of a word, encoded as a sequence of feature vectors, to its corresponding semantic representation in a multilingual and cross-family vector space. The processing network is then tested, without further training, in a language that does not appear in the training set and belongs to a different language family. The performance of the model is evaluated through a comparison with a monolingual and mono-family upper bound and a randomized baseline. After the assessment of the network's performance, the distribution of phonosemantic properties in the lexicon is inspected in relation to different (psycho)linguistic variables, showing a link between lexical non-arbitrariness and semantic, syntactic, pragmatic, and developmental factors.

The exercise of reopening optional social spaces following the COVID-19 lockdowns presents each individual with a complex problem in determining whether or not to attend these spaces given how the risks of virus transmission scale with crowding. In order to tackle this problem while recognizing individual cognitive capacity limits, this paper used a simulation model based on the El Farol Bar Problem and generated a population of agents relying on simple predictive strategies to determine their attendance to a retail location. It was determined that the more heterogeneous the threshold for crowding among agents was, the less variance there was in overall attendance numbers. This stability in daily attendance comes at the expense of the ability of the most cautious members to enjoy recreational spaces as these locations become the realm of only those least concerned with potential crowding.

Speakers use different language to communicate with partners in different communities. But how do we learn and represent which conventions to use with which partners? In this paper, we argue that solving this challenging computational problem requires speakers to supplement their lexical representations with knowledge of social group structure. We formalize this idea by extending a recent hierarchical Bayesian model of convention formation with an intermediate layer explicitly representing the latent communities each partner belongs to, and derive predictions about how conventions formed within a group ought to extend to new in-group and out-group members. We then present evidence from two behavioral experiments testing these predictions using a minimal group paradigm. Taken together, our findings provide a first step toward a formal framework for understanding the interplay between language use and social group knowledge.

This paper reviews the existing literature on categorical perception of sounds and colors in different animals including humans. We highlight that categorical perception is a combination of nature and nurture; to be specific, categorical perception is innate with a phylogenetic root, but it can also be modified by postnatal experience. We also suggest that language is not wholly the basis for categorical perception, as what Sapir-Whorf hypothesis posits; instead, language is one type of experience that can affect the nurture part of categorical perception across domains and modalities in humans.

Causal inference from observed cases is a central cognitive challenge. There has been some evidence for individual differences in causal learning strategies, but prior work has not examined fine-grained sequences of judgments. In this paper, we report a large-scale model-fitting effort to determine the best-fitting causal inference models for individual participants. We fit a range of different model-types against multiple judgment sequences from each participant, thereby enabling comparisons of learning strategy both between- and within-participant. The model-fitting effort revealed some diversity in learning strategy along both dimensions, though individuals did exhibit some stability. Overall, however, the model fits were worse than expected, particularly when compared to the high accuracy reported for many of the models when used to predict group-level causal judgments. These results thus call into question whether these models might accurately describe the average behavior without accurately describing many (or any) individual’s behaviors.

When encountering objects, we readily perceive not only low-level properties (e.g., color and orientation), but also seemingly higher-level ones--including aspects of physics (e.g., mass). Perhaps nowhere is this contrast more salient than in the perception of soft materials such as cloths: the dynamics of these objects (including how their three-dimensional forms vary) are determined by their physical properties such as stiffness, elasticity, and mass. Here we hypothesize that the perception of cloths and their physical properties must involve not only image statistics, but also abstract object representations that incorporate "intuitive physics". We provide behavioral and computational evidence for this hypothesis. We find that humans can visually match the stiffness of cloths with unfamiliar textures from the way they undergo natural transformations (e.g. flapping in the wind) across different scenarios. A computational model that casts cloth perception as mental physics simulation explains important aspects of this behavior. Full paper can be found at https://www.biorxiv.org/content/10.1101/2021.05.12.443806v1.

Production of referring expressions (the dog, it, Snoopy) is a complex process regulated by a combination of linguistic and cognitive constraints. In this paper, we explore the impact of world knowledge on the types of references speakers produce, focusing on predictability of event progressions. We argue that speakers are more likely to use a full noun phrase rather than drop the subject or use a pronoun when they describe an event progression they find surprising. In order to avoid the influence of distributional properties of event descriptions, we created an artificial world and trained subjects to recognize typical collision-event progressions within. Speakers then described novel scenes, which either conformed to their expectations or violated them, in a free production experiment. The results reveal that unpredictable event progressions lead to a more frequent production of full noun phrases, in contrast to reduced linguistic expressions (pronouns and null subjects). We conclude that speakers choose more informative descriptions to talk about surprising events.

This paper presents a study of the selection task based on Cognitive Argumentation (CA), a computational framework for dialectic argumentation-based reasoning. CA is built from a theoretical framework of argumentation in AI which is then grounded via cognitive principles from Cognitive Science. The aim is to understand the selection task variations by studying how argumentative reasoning is suitably flexible to uniformly capture the differences among the individuals' selections, the canonical groups and the shifts within the different contexts in which the experiment is carried out. Our approach is assessed with respect to the developed criteria within the meta-analysis of Ragni and Johnson-Laird in 2018.

The use of face masks is one of the measures adopted by the general community to stop the transmission of disease during this ongoing COVID-19 pandemic. This wide use of face masks has indeed been shown to disrupt day-to-day face recognition. People with autism spectrum disorder (ASD) often have predisposed impairment in face recognition and are expected to be more vulnerable to this disruption in face recognition. When faces were initially learned unobstructed, we showed that people with higher autistic traits had a lower face recognition performance for masked faces. In contrast, there was a stronger facilitation in subsequent face recognition of masked faces in typically developing adults, but not in adults with ASD, when masked faces were learned first—this face recognition facilitation is predicted by a higher level of empathy. This paper also discusses how autistic traits and empathy influence processing of faces with and without face masks.

Recent studies have applied network-based approaches to analyze the organization and retrieval of specific semantic categories, with a focus on the animal category. The current study extended previous studies by using network science tools to quantitatively investigate the structural differences of noun and verb categories of various levels of specificity. Specific (animal and body movement) and general noun and verb networks were constructed from four verbal fluency tasks. Common network measures indicated that the two verb networks were more condensed and less modular than the noun networks, supporting that nouns are more well-organized in the mental lexicon than verbs. Comparing the specific and general networks within each lexical category also corroborated lexical semantic studies that nouns have a more clear hierarchical structure. The results of this paper, along with recent semantic network studies, provide converging evidence for the usefulness of network science in semantic memory research.

One of the major debates in the field of word production is whether lexical selection is competitive or not. For nearly half a century, semantic interference effects in picture naming latencies have been claimed as evidence for competitive (relative threshold) models of lexical selection, while semantic facilitation effects have been claimed as evidence for non-competitive (simple threshold) models instead. In this paper, we use a computational modeling approach to compare the consequences of competitive and noncompetitive selection algorithms for blocked cyclic picture naming latencies, combined with two approaches to representing taxonomic and thematic semantic features. We show that although our simple model can capture both semantic interference and facilitation, the presence or absence of competition in the selection mechanism is unrelated to the polarity of these semantic effects. These results question the validity of prior assumptions and offer new perspectives on the origins of interference and facilitation in language production.

Research suggests that people minimize the amount of effort used to generate natural language descriptions of visual scenes. In the case of visual scenes with multiple groups, recent work has found that people tend to generate quantitative descriptions that mention the number and cardinality of groups (e.g., ``two groups of three limes''), but omit the total quantity (e.g., ``six limes''). This finding suggests that people groupitize, that is, they more quickly determine the number of grouped items by rapid enumeration of subgroups, rather than slower item-by-item counting. %This hypothesis predicts that during description, people exert less effort to encode and report only the number and cardinality of groups. A recent proposal predicts that during description, people exert less effort by encoding and reporting only information that is readily available to perception. In previously studied description tasks, people may have omitted the total quantity from their descriptions because of considerations of brevity and informativity. In this paper, we describe a study designed to test how individuals balance effort, brevity, and informativity when evaluating quantified descriptions. The experiment was designed to elicit more fine-grained preferences for descriptions using direct comparisons between two competing descriptive forms. The results suggest that perceptual effort plays a central role in how people describe grouped collections of items.

How do we manage to understand the minds of others and usefully interact with them? In the last decade, the debate on these issues has developed from unitary to pluralist approaches. According to the latter, we make use of multiple socio-cognitive strategies when predicting, interpretating, and reacting to the behavior of others. This means a departure from the view of mindreading as the main strategy underlying social cognition. In this paper, we address the question of the controversial status of mindreading within such a pluralist framework. Contrary to many other accounts, we ascribe mindreading an equal status in a pluralist framework. Mindreading is required for a variety of central situations in life and importantly underlies the way in which we understand other people. Mindreading is also no less reliable than alternative strategies; reliability is not so much a matter of different competing socio-cognitive strategies, but rather of their complementary use.

Learnability has been a topic of great interest in phonology. In particular is the question of the relative learnability of process interactions. In both historical and experimental domains, researchers have noted that certain kinds of interactions are harder to learn than others. In both domains, however, the results are seemingly in conflict. One potential source of the conflicting outcome is the types of processes involved. In this paper, we investigate the effect of process types on the learnability of different interaction types, using an ideal minimum-description-length learner (MDL). We find that the model indeed predicts different learnability outcomes for each interaction type; however, the asymmetry is largely independent of the process type. This computational model explains certain elements of both the historical as well as some experimental findings of the relative learnability of linguistics process interactions, while contradicting other behavioral findings.

Language is multimodal: non-linguistic cues, such as prosody, gestures and mouth movements, are always present in face-to-face communication and interact to support processing. In this paper, we ask whether and how multimodal cues affect L2 processing by recording EEG for highly proficient bilinguals when watching naturalistic materials. For each word, we quantified surprisal and the informativeness of prosody, gestures, and mouth movements. We found that each cue modulates the N400: prosodic accentuation, meaningful gestures, and informative mouth movements all reduce N400. Further, effects of meaningful gestures but not mouth informativeness are enhanced by prosodic accentuation, whereas effects of mouth are enhanced by meaningful gestures but reduced by beat gestures. Compared with L1, L2 participants benefit less from cues and their interactions, except for meaningful gestures and mouth movements. Thus, in real-world language comprehension, L2 comprehenders use multimodal cues just as L1 speakers albeit to a lesser extent.

Categories such as 'animal' or 'furniture' play a pivotal role in processing, organizing, and communicating world knowledge. Many theories and computational models of categorization exist, but evaluation has disproportionately focused on artificially simplified learning problems (e.g., by assuming a given set of relevant features or small data sets); and on English native speakers. This paper presents a large-scale computational study of category and feature learning. We approximate the learning environment with natural language text, and scale previous work in three ways: We (1) model the full complexity of the learning process, acquiring learning categories and structured features jointly; (2) study the generalizability of categorization models to five diverse languages; and (3) learn categorizations comprising hundreds of concepts and thousands of features. Our experiments show that meaningful representations emerge across languages. We further demonstrate a joint model of category and feature acquisition produces more relevant and coherent features than simpler models, suggesting it as an exploratory tool to support cross-cultural categorization studies.

Understanding a sequence of images as a visual narrative is challenging because it requires not only the understanding of what is shown at a particular moment but also what has changed, been omitted or is out of frame. The human cognitive system makes inferences about the state of the world based on transitions between sequential frames. In this paper, we present a principled analysis of the stylistic differences between two dominant styles of multi-modal narratives, western comics and manga. These two styles differ in terms of screening, ballooning, layout, language, and reading order. We first provide a systematic account of these differences based on an annotated dataset consisting of both comics and manga. We then annotate these datasets with a new feature set and evaluate the contributions of these features through development of a computational model of multi-modal comprehension. The model evaluation is presented through the cloze test that measures the accuracy of the model in predicting unseen next frames given the prior frames in a sequence. Our results provide initial benchmarks and insight into the fundamental challenges that the multi-modal narrative understanding task presents for computational models both for language and vision.

A large body of work has demonstrated the utility of the Bayesian framework for capturing inference in both specialist and everyday contexts. However, the central tool of the framework, conditionalization via Bayes’ rule, does not apply directly to a common type of learning: the acquisition of conditional information. How should an agent change her beliefs on learning that “If A, then C”? This issue, which is central to both reasoning and argumentation, has recently prompted considerable research interest. In this paper, we critique a prominent proposal and provide a new, alternative, answer.

Verbs such as ‘know’ and ‘think’ help people describe mental states, and reasoners without any training in logic can make epistemic inferences about mental states. For instance, verbs such as ‘know’ are factive, i.e., they describe true propositions, and the statement Ora knows that it’s sunny licenses the inference that it’s sunny. Logicians have accordingly developed epistemic logics capable of characterizing valid and invalid epistemic inferences based on operators that serve as analogs to verbs such as ‘know’ and ‘think’. Recent work suggests that no existing logical system can capture the inferences that naïve individuals tend to make. This paper describes a new theory of epistemic reasoning that operates on the assumption that reasoners represent epistemic relations as spatial models. The theory accords with recent theoretical advances, existing data, as well as two novel experiments that show how reasoners cope with nested epistemic verbs, e.g., Ami knows that Ora thinks it’s sunny.

This paper presents a theoretical framework for modeling human visual attention. The framework's core claim is that three mechanisms drive attention: selection, which picks out an item for further processing; engagement, which tags a selected item as relevant or irrelevant to the current task; and enhancement, which increases sensitivity to task-relevant items and decreases sensitivity to task-irrelevant items. Building on these mechanisms, the framework is able to explain human performance on attentionally demanding tasks like visual search and multiple object tracking, and it supports a broad range of predictions about the interactions between such tasks.

Although Artificial Intelligence (AI) is expected to outperform humans in many domains of decision-making, the process by which AI arrives at its superior decisions is often hidden and too complex for humans to fully grasp. As a result, humans may find it difficult to learn from AI, and accordingly, our knowledge about whether and how humans learn from AI is also limited. In this paper, we aim to expand our understanding by examining human decision-making in the board game Go. Our analysis of 1.3 million move decisions made by professional Go players suggests that people learned to make decisions like AI after they observe reasoning processes of AI, rather than mere actions of AI. Follow-up analyses compared the decision quality of two groups of players: those who had access to AI programs and those who did not. In line with the initial results, decision quality significantly improved for the players with AI access after they gained access to reasoning processes of AI, but not for the players without AI access. Our results demonstrate that humans can learn from AI even in a complex domain where the computation process of AI is also complicated.

The N400 component of the event related brain potential is widely used to investigate language and meaning processing. However, despite much research the component’s functional basis remains actively debated. Recent work showed that the update of the predictive representation of sentence meaning (semantic update, or SU) generated by the Sentence Gestalt model (McClelland, St. John, & Taraban, 1989) consistently displayed a similar pattern to the N400 amplitude in a series of conditions known to modulate this event-related potential. These results led (Rabovsky, Hansen, and McClelland,2018) to suggest that the N400 might reflect change in a probabilistic representation of meaning corresponding to an implicit semantic prediction error. However, a limitation of this work is that the model was trained on a small artificial training corpus and thus could not be presented with the same naturalistic stimuli presented in empirical experiments. In the present study, we overcome this limitation and directly model the amplitude of the N400 elicited during naturalistic sentence processing by using as predictor the SU generated by a Sentence Gestalt model trained on a large corpus of texts. The results reported in this paper corroborate the hypothesis that the N400 component reflects the change in a probabilistic representation of meaning after every word presentation. Further analyses demonstrate that the SU of the Sentence Gestalt model and the amplitude of the N400 are influenced similarly by the stochastic and positional properties of the linguistic input.

Our ability to play games like chess and Go relies on both planning several moves ahead and on recognition or gist - intuitively assessing the quality of possible game states without explicit planning. In this paper, we investigate the role of recognition in puzzle solving. We introduce a simple puzzle game to study planning and recognition in a non-adversarial context and a reinforcement learning agent which solves these puzzles relying purely on recognition. The agent relies on a neural network to capture intuitions about which game states are promising. We find that our model effectively predicts the relative difficulty of the puzzles for humans and shows similar qualitative patterns of success and initial moves to humans. Our task and model provide a basis for the study of planning and intuitive notions of fit in puzzle solving that is simple enough for use in developmental studies.

The present paper reports an experiment testing two views of how reasoners learn and generalize potentially complex causal knowledge. Previous work has focused on reasoners’ ability to learn integration functions that best describe how pre-defined candidate causes combine, potentially interactively, to produce an outcome in a domain. This empirical-function learning view predicts that participants would generalize an acquired integration function based on similarity to stimuli they experienced in the domain. An alternative causal-invariance view recognizes that one’s current representation may not yield invariant/useable causal knowledge –– knowledge that holds true when applied to new circumstances. This view incorporates useable causal knowledge as a goal and deviation from causal invariance as a criterion for knowledge revision. It predicts that participants would re-represent causes such that they do not interact with other causes, even when in participants’ experience all (pre-defined) causes in that domain interact. Our results favor the causal-invariance view.