While a growing body of research has identified the crossmodal correspondences between nonlinguistic pitch and vertical space, little is known about the non-arbitrariness of linguistic tones within the language domain. This study investigates the vertical mappings of Cantonese lexical tones (high rising vs. low falling) in two crossmodal domains: (1) spatial motions (upward vs. downward) and (2) emotional valence (positive vs. negative). Results showed that Cantonese speakers could reliably map tones based on pitch and motion trajectories (i.e., rising contour is upward, and falling contour is downward), but the vertical mapping between tones and valence (i.e., rising contour is positive, and falling contour is negative) was much weaker and modulated by vowels. In general, these findings suggest that lexical tones can be conceptualized and understood in the vertical dimension, even in the language where speakers do not talk about tones using spatially relevant terms. In addition, tonal iconicity, and sound symbolism in general, is differentiated according to the mechanisms in which they are grounded. Specifically, sound symbolism that relies on a direct structural parallel produces a more consistent mapping, whereas correspondence grounded in metaphorical mediation is more easily influenced by other linguistic factors. Thus, this research provides evidence for the sound symbolic nature of lexical tones, suggesting their conceptual mapping in tonal languages.
It is natural to express some beliefs in binary terms (P is either believed or not, e.g., "bribery is immoral"), and others in terms of a subjective probability (P has some probability of being true, e.g., "there's an 80% chance a Democrat will win the next election"). Construing a belief in probabilistic terms (other than 0% or 100%) implicitly conveys a recognition of uncertainty (one's ignorance) and the possibility of being wrong (one's fallibility), which are key characteristics of intellectual humility (IH). Across two studies, we explore the hypothesis that probabilistic belief is associated with greater IH. Although we fail to find this association across participants, comparing their trait-level IH, in either a student sample (Study 1, N = 279) or two nationally representative samples (Study 2a, N = 269; Study 2b, N = 270), we do find the predicted association within participants, analyzing belief-specific IH (Studies 2a and 2b): political and moral beliefs that are classified as probabilistic rather than binary are associated with a more intellectually humble attitude toward those beliefs, even controlling for the extremity of belief. These findings support the psychological reality of a distinction between binary and probabilistic belief, and further raise the possibility that promoting probabilistic belief could foster greater IH, including in consequential domains such as politics and morality.
Low-level occipital regions, including the primary visual cortex (V1), are known to play a pivotal role in generating the Ebbinghaus size illusion, in which a circle surrounded by small inducers looks larger than an identical circle surrounded by large inducers. However, a definitive answer is still lacking as to whether the low-level visual cortex plays its role at the feedforward stage of processing. To resolve this issue, we measured event-related potentials (ERPs) evoked by two equal-size circles flashed on the left and right sides of human participants, with one side constantly displaying large surrounding inducers and the other side small inducers. Experiment 1 unveiled a significant lateralization in the evoked C1 component, a well-established ERP index of afferent V1 activity, only when the circle within small inducers was perceived as larger but not when the circle within large inducers was perceived as larger. This illusion-specific C1 lateralization, which closely resembled the C1 modulation evoked by physical differences in size, was further replicated in Experiment 2 even when the attentional bias toward small inducers was invalidated as an alternative account. Furthermore, Granger causality analysis revealed a one-way transfer of information from V1-representative electrodes to extrastriate-representative electrodes during the C1 lateralization. Together, these findings offer the strongest evidence to date that the Ebbinghaus illusion begins at the feedforward stage of cortical visual processing, consistent with the prediction of the contour interaction theory.
Much like with statistical learning, there is a tendency in the cognitive and language science literature to treat chunking as a unitary, domain-general ability that operates fairly similarly across domains of cognition. An alternative position suggests chunking in language and cognition involves multiple mechanisms tuned to the specific perceptual properties and statistical regularities of the target structures of learning. To test these hypotheses, we compared performance on two different forms of chunking in English: (1) multiword chunking as assessed by a recall task involving multiple high- and low-frequency three-word sequences, and (2) clausal chunking in which participants intuitively insert chunk boundaries on a touchscreen tablet while listening to speech extracts and viewing the corresponding transcripts. Results indicated no correlations between multiword chunking and clausal chunking, providing initial support for the notion that chunking processes may likely not operate equally across or even within the same cognitive domain, casting doubt on a unitary view of chunking. Despite limitations in the tasks used to assess chunking, they nonetheless provide a useful starting point for future studies looking into the nature of chunking across different levels of linguistic representation.
Early verbal dialogue emerges during the second year of life, despite children's still-developing linguistic, pragmatic, and cognitive abilities. This raises a puzzle: how can young children produce contingent responses under the real-time demands of turn-taking? In this study, we articulate a unified scaffolding account and test whether three caregiver-side mechanisms help explain this early emergence: lexical specificity in initiating utterances, interactive routines that constrain next-intent selection, and short caregiver realizations that can serve as models for children's own intent expression. Using recently developed automatic annotation tools for communicative intents and response contingency, we investigated this question in large-scale naturalistic corpora of child-caregiver dialogue, spanning 40 corpora and involving more than 600 children across more than 2500 transcripts. Across corpora, each mechanism was associated with higher child contingency, and all three remained predictive in a joint model controlling for age, child utterance length, and frequency-based factors. These findings are consistent with the view that early dialogue participation is supported not only by children's emerging capacities, but also by caregivers' structuring of interaction in ways that reduce processing demands at multiple steps of response formation. More broadly, the results provide quantitative support for a scaffolding-based account of early dialogue emergence and help connect developmental evidence to broader theories of real-time communication.
There is an asymmetry between the origin (Source; move from X…) and end-point (Goal; …to Y) of motion events in both language and memory: Goals are mentioned more often and remembered more accurately than Sources. Here, we pursue the hypothesis that this Goal bias emerges as an online attentional bias that occurs during event apprehension and itself connects to event memory and linguistic measures. We record participants' eye movements as they view or prepare to describe motion events and later test their memory of Goals or Sources. We find an online attentional bias for Goals over Sources during initial encoding of events. This bias is stronger during free inspection compared to speech planning. Moreover, the extent of the attentional Goal bias is systematically related to both language production and memory: When the bias is the greatest, the Source is not later mentioned during production and/or not remembered later at test. In sum, we provide the first evidence that the Goal bias appears during visual encoding of motion events and identify a direct relation between event apprehension and later event description and memory.
Thinking about things as instances of kinds and thinking about kinds are fundamental aspects of cognition. Current theories of conceptual representation specify the qualitative character of a kind via a definition, a prototype, exemplars, shared causal-explanatory structure, or the kind's inferential role in a broader theory, but they do not represent the quantitative dimension of kind representations. Four experiments investigated two characteristics of the quantitative dimension of kind and instance-of-kind representations. Experiments 1 and 2 found evidence that we understand kinds to be constituted of an unlimited number of instances of that kind. Experiments 3 and 4 provided evidence that whereas two instances of a given kind may be distinguished merely numerically, two kinds must also be qualitatively distinct. These, and other characteristics of the quantitative dimension of kind concepts must be accommodated by any viable theory of kind concepts.
Pronouns are a ubiquitous part of discourse, but unusual in that their meaning is almost entirely determined by context. While early theorists hoped to explain pronouns based on a small number of simple principles, the last half-century of research has revealed a cornucopia of influences at the syntactic, semantic, discourse, and pragmatic levels. While there are currently a few popular theories, evaluating them is complicated by the complexity of the empirical situation, which is compounded by the fact that many popular experimental methods are incommensurate and are uninterpretable under one theory or another. Moreover, with a few notable examples, research has focused on English, and so the generalizability of results is uncertain. Here, we take a step toward a clear empirical foundation for theory, with a tightly controlled study of comprehension of overt and null pronouns in Turkish. We show that pronoun resolution in Turkish is influenced by verb type, word order, and referential form, though not always in ways predicted by existing theories. Our findings highlight the need for further cross-linguistic research and more careful experimental control in order to refine models of pronoun interpretation and better account for the interaction of syntactic and discourse factors.
Studies on synesthesia have revealed some advantages on discrimination, categorization, and identification tasks. There have also been studies on language's ability to boost performance on these same tasks. Are these two effects related? In this paper, I argue that a plausible explanation of language's ability to boost performance on such tasks can be extended to synesthesia. In particular, I argue that category labels can reveal trends in perceptual data that allow for representational space to be dimensionally reduced. The transformation of representational space makes representations more categorical, facilitating lexical access and ultimately boosting performance in categorization, identification, and discrimination tasks. This account can be extended to explain how synesthetic color associations boost performance. Synesthetic colors also reveal trends in perceptual data that allow for dimensional transformation. The upshots of such a hypothesis are several. For one, it helps explain synesthesia's cognitive advantages. I supply experimental designs to test predictions made by the account. Second, and more broadly, this account extends cases of nonlinguistic categorical effects on representations. I close with a discussion of some implications for the debate over language and thought.
Processing state change verb phrases (e.g., fill the cookie jar) during language comprehension appears to require the activation of representations encoding both the initial (an empty jar) and resultant states (a filled jar) of event participants. Given that the transitions between these states can generally be inferred (on the basis of experiential semantic knowledge), these boundary states may be the only states of event participants that are activated during sentence processing or at event boundaries. However, simulation-based accounts of representation and linguistic analyses of state change predicates suggest that, in addition to initial and resultant object states, intermediary object states should also be activated during sentence comprehension. To compare these two alternatives, we investigated the activation of initial, intermediary, and end object states by sentences describing completed events (e.g., Jasmine filled/has filled the cookie jar). Our results suggest that perceptual features associated with objects at intermediary points in an event are indeed activated and maintained in memory through at least the end of the event description. These findings support an account in which representations of the entire event sequence-not just the boundaries-are activated during language processing.
Learners often encounter situations where explicit rules fail to account for novel cases, leading to an underdetermination problem. One way to explore how learners navigate this uncertainty is to see if their inferences align with closure principles, through which a learner assumes that anything not explicitly prohibited is permitted, and anything not explicitly permitted is prohibited. Beyond permission and prohibition, normative inferences must also be made from and about a learner's obligations-not only what they can do, but what they must do. Contrary to adult learners, young children may make more restrictive inferences about how permissible or obligatory a novel action is, aligning with their well-documented "strict normativity". Across two studies, we explore inferences from adults (N = 115, Mage = 34.63) and 4- to 6-year-old children (N = 120, Mage = 5.48). Our findings suggest that while both adults and children rationally learn closure principles consistent with deontic logic, children's reasoning about prohibitions undergoes developmental changes. Contrary to adults, children taught explicit prohibitions were restrictive in their permissibility judgments until age 5. Thus, the framing of explicit rules leads to differing inferences about novel, unspecified cases across age and rule type.
People reliably associate the meanings of both abstract and concrete words with colors distributed over color space, a phenomenon that influences aspects of visual cognition ranging from object recognition to interpreting information visualizations. Prior research has hypothesized that color-concept associations arise from the cross-modal statistical structure of experience, but it remains unclear whether natural environments contain such structure or whether learning systems can discover it without strong prior constraints. To address these questions, we investigated whether GPT-4, a multimodal large language model, can estimate color-concept association ratings that approximate those made by people. We tested 71 colors spanning perceptual color space and a variety of concepts varying in abstractness. GPT-4 ratings correlated strongly with human ratings across a range of prompting strategies, outperforming prior state-of-the-art methods for automatically estimating color-concept associations from images. In an empirical study assessing people's ability to interpret the meanings of colors in information visualizations, palettes generated from GPT-4's rating data were not only interpretable but, in some cases, more effective than those based on human ratings. Taken together, our results suggest that high-order covariance between language and perception, present in web-scale data, provide sufficient information to learn color-concept associations without initial constraints, and that machine-derived associations can support the optimization of information visualizations for visual communication.
The standard model of theory of mind (ToM) development holds that children normally acquire a representational ToM by about age 4 at the latest. Perceptual Access Reasoning (PAR) theory presents a fundamental challenge to the standard model by arguing that representational ToM does not develop until about age 7. One set of predictions of PAR theory concerns children's errors on certain true belief (TB) tasks. Schidelko and Rakoczy subscribe to the standard model, and report a test of the PAR account of TB errors versus their pragmatic account. The authors' preregistered predictions were not confirmed, and their publicly available data show that, contrary to their claims, the findings actually support the PAR account.
Human causal judgments frequently deviate from normative Bayesian expectations, particularly with respect to conditional independence and explaining away. Rather than interpreting these deviations as reasoning errors, recent computational accounts suggest they may emerge from principled approximations to ideal Bayesian inference. We evaluate four leading frameworks: Bayesian sampler (BS), mutation sampler (MS), Bayesian mutation sampler (BMS), and Bayesian uncertainty model (BUM), which each formalize different cognitive constraints, including limited sampling, prototype anchoring, prior regularization, and uncertainty over causal structure. These models were systematically compared across 33 reported experimental conditions (N = 1154) spanning diverse causal structures, including common cause, common effect, and chain networks with generative and inhibitory relations. All four models had at least some success reproducing the positive and negative Markov violations observed in common cause and chain structures. In common effect structures with inhibitory or mixed links, only MS and BMS captured the observed patterns more consistently, although BMS's additional parameter offered limited improvement over MS. BS and BUM showed poorer fits in those conditions, though they may still offer plausible accounts under different assumptions. We discuss these findings in light of prior work on causal reasoning, emphasizing that deviations from normative inference may reflect adaptive strategies shaped by structural uncertainty and cognitive constraints. This supports a pluralistic and resource-rational view of causal reasoning and underscores the need for targeted experiments probing inhibitory causal relations and model uncertainty.
Anchoring occurs when a quantitative estimate is biased toward an initially presented value (the anchor). Anchoring occurs both in high-level explicit estimation of numeric quantities and in lower-level perceptual tasks and persists even when reliable information about the quantity being estimated is directly available at the point of judgment. This suggests anchoring might derive from generic processing underpinning estimation. Such tasks, however, are almost exclusively lab-based, and the information required to complete the task is rarely available to the participant in a way that reflects how such information is sampled in real-life. To address this, across two experiments, we immersed participants in a virtual world on a platform that could be placed at any height. After an anchor was presented, participants subsequently estimated their height naturally, by interpreting sensory and cognitive cues from the 3D environment in which they were immersed. We manipulated the amount of information directly available to the participant for making their judgment as well as the offset between the anchor and the actual height. To modulate the extent to which task-relevant information was acquired naturally from the environment, we also contrasted anchoring effects in and out of VR. Although participants clearly used the available perceptual and cognitive information to make height estimates, anchoring effects were evident, displayed similar properties to those reported in previous lab-based studies, and were consistent both in and out of VR. Our design also allowed us to recover a novel subjective anchoring measure that facilitated a particularly parsimonious descriptive model of our anchoring data. We conclude that anchoring is a generic feature of all estimation tasks, even when task-relevant information is acquired naturally, as in real-life estimation. These results emphasize the potential for this cognitive bias to have a significant impact on performance in any real-world task requiring quantitative estimation.
Icon arrays are graphical displays in which a subset of shapes are filled to represent the probability of an outcome (e.g., the probability of side-effects from a medical treatment). Prior work has shown that the perceptions of probabilities can be more accurate with icon arrays compared to other formats. As a result, they are now widely used to communicate information about risk and uncertainty to the general public. However, little is known about how the design of icon arrays-in particular, the perceptual characteristics of icons and their spatial arrangement-affect perceptions of risk. The present study builds on research on visual perception which suggests that variation in icon shape may affect their perceived numerosity. Three experiments were conducted using a proportion judgment task with icon arrays that varied in (a) whether icons were organized in grouped or random configurations, and (b) whether there was variability in the icon shapes used to represent the target and nontarget proportions. The results show that proportion judgments are highly accurate for grouped arrays, with no effect of shape variability regardless of the distribution of shapes across the target and nontarget categories. For random spatial arrangements, however, irrelevant shape variability in one (but not both) of the categories leads to biased proportion judgments, increasing the perceived quantity of whichever category has elements with the same shape. These findings show that perceptual variation in shape-while irrelevant to the proportion judgment task-can alter how people perceive quantities depicted by icon arrays, providing new insight into how these visualizations should be designed to communicate information about risk.
Language control is a cognitive ability that bilinguals use to suppress interference from the language they are not currently using to accurately select and use the intended language. Adaptive language control underpins language switching and enables bilinguals to flexibly switch between languages according to context. Reinforcement learning, which models how individuals update their strategies based on reward prediction errors, provides a computational framework for studying adaptive behavior in changing environments. To investigate how bilingual language control is shaped by reward signals in social interactions, we used dual-electroencephalography (EEG) to measure the performance of bilinguals who alternated between active and observational learner roles in voluntary language switching tasks. Computational modeling results indicated that the dual-sensitivity model best captured behavior which showed that bilinguals adaptively updated values by assigning distinct weights to feedback from themselves and others. EEG analyses revealed that bilinguals relied on expected values during active learning and on prediction errors during observational learning to modulate delta band activity. Taken together, these findings reveal how rewards dynamically modulate language control through expected values and prediction errors, providing new evidence for the adaptability of bilingual control during social interaction.
Scholars propose some new directions for researching statistical learning (SL), including the need to adopt stimuli with greater ecological validity. The language sciences are moving in these directions. Studies investigating adult SL after short exposure to an unfamiliar (spoken or signed) language show that SL can occur from richer, continuous, multimodal input, suggesting that learners are able to track multiple statistics. This makes it more plausible that SL operates in naturalistic, interactive situations. This paradigm shift can potentially extend our understanding of the exact ways in which SL is deployed in the service of learning languages, thereby refining it theoretically and clarifying its place in cognitive science.
Prosody is an intrinsic element of language production, linking together multiple levels of linguistic representation to shape both the structure and interpretation of utterances. However, common theories of prosodic phrasing in spoken language often fail to capture factors associated with planning and recovery, as well as performance-based effects related to working memory. Much of what we know about prosody, whether it be the features speakers are thought to generate or the ones listeners are believed to process, is based on forms that are atypical in spoken language. Recent developments in data analysis methods, however, allow for the efficient study of unrehearsed spoken language. The current work aims to develop more ecologically valid theories of prosody and its relationship to syntactic structure through the analysis of unrehearsed scene descriptions. Data from unrehearsed speech collected across four different studies showed only a weak to moderate relationship between prosodic phrasing and syntactic structure, such that the likelihood of a prosodic phrase boundary occurring at the end of a syntactic phrase was only slightly above chance. Additionally, correlations between occurrences of prosodic phrase boundaries and speech rate revealed that individuals who speak more slowly are likely to insert more prosodic phrase boundaries, indicating a relationship between prosodic phrasing and speech planning. The findings challenge some categorical approaches to prosody and suggest that prosodic phrasing may be a consequence of planning and recovery in language production, rather than a complement to syntactic phrasing. These results have implications for theories of language production and comprehension, formal theories of phonological structure, and computational tools for generating and interpreting language.
Statistical learning (SL) is believed to enable humans to assimilate a range of statistical structures, and thus plays a role in many cognitive functions. There is also a growing interest in how SL interacts with basic cognitive processes, including perception and attention. Here, we ask how the extent to which stimuli predict, and are predicted by, other elements in a continuous stream affects their perception (i.e., encoding and representation) and the attention they attract. In two experiments, participants were first exposed to a stream of structured pairs of visual shapes (e.g., AB and CD). Then, they completed a target detection task to test if stimulus detection speed is influenced by a target's predictability during exposure (indexing representation) or by whether the shape preceding the target reliably predicted elements in the input (indexing attention). In Experiment 1 (N = 86), Reaction Times (RTs) were faster for second elements from structured pairs (i.e., cued elements) than first elements (i.e., cue elements), even when they appeared in new configurations (e.g., AD and CB). However, in Experiment 2 (N = 89), which orthogonally manipulated the target and the preceding shape properties, RTs were influenced solely by whether the preceding element was a cue for other elements, but not by the target's predictability. Thus, in contrast to previous studies using the same paradigm, our results do not provide evidence for an effect of SL on representation. Instead, our findings highlight how attention is guided by knowledge of statistical regularities, pointing to SL as a system that helps minimize uncertainty in structured environments.
