Neural noise is an unavoidable byproduct of brain function. Rather than noise being merely an unavoidable limitation, accumulating evidence suggests it may serve a functional role in maintaining optimal neural processing. Some theoretical accounts suggest that the brain may modulate its noise level to achieve this; however, experimental and computational evidence that illustrates this dynamic nature of neural noise is limited. Here, we investigated whether the brain aims to optimise its neural noise levels in response to task demands using a visual letter identification task with varying stimulus contrast. Computational modelling predicted conditions where counterintuitive increases in neural noise should lead to optimal performance. We tested this prediction experimentally, estimating neural noise using an N-pass paradigm. Results revealed that lower stimulus contrasts (i.e., more difficult perceptual conditions) were associated with higher neural noise, suggesting that the brain adapts and modulates its internal variability depending on sensory demands to enhance performance. Overall, our results support the notion that the brain's neural noise is contextually flexible and may serve an adaptive role, which reflects the brain's ability to adapt to changing perceptual demands.
Distinguishing objects from the background, a process known as Object Individuation (OI), is fundamental for us to interact with the environment and relies critically on location information across sensory modalities. Nonetheless, it remains unclear and contested in the literature whether the enumeration of tactile and visual events relies on the OI process (especially given spatial constraints), or if the representation of numerosity is governed by a modality-independent mechanism common to both visual and tactile systems. In this study, we used a cross-modal enumeration and a working memory dual-task paradigm to investigate whether OI processes in tactile and visual modalities draw upon a shared cognitive resource. We implemented two experiments. In Experiment 1, we combined a tactile working memory (WM) task with visual enumeration, and in Experiment 2, we used a visual WM task with tactile enumeration. Both experiments revealed that the task-irrelevant WM load significantly modulated subitizing performance (enumeration of small quantities) in the target modality. Under high WM load, participants showed increased error rates and reduced subitizing capacity compared to low load. This modulation is selective to the subitizing range and cannot be attributed to general dual-task costs, ruling out general dual-tasking effects. The data shows that visual and tactile working memory and enumeration ("subitizing") share a common OI process that operates on location, independent of the sensory modality. This finding is consistent with existent neuroimaging evidence that highlights the modality-shared role of frontoparietal brain regions (e.g., IPS, LPFC) in enumeration and working memory.
Children typically prefer the so-called "basic"-level (e.g., dog) versus narrower subordinate (e.g., dalmatian) meanings in acquisition. Subordinate conjectures are facilitated by contrast; for example, when shown a dalmatian and a corgi each labelled "dax" and "wug," learners take "dax" to mean dalmatian, as opposed to dog. But what is the nature of this effect? On a pragmatic account, contrast is helpful because it highlights a communicative pressure on the speaker to be more informative than usual when describing the target referent, given relevant alternatives. Another possibility, however, is that contrastive forms create a simple lexical heuristic. We raise and test two predictions that follow from the pragmatic, but not the lexical, account: in learning subordinate nouns, children use contrastive information even when there are no competing lexical labels (Exp. 1); furthermore, they consider the speaker's knowledge of contrasting alternatives during word learning, as a fully Gricean rich-computation model expects (Exp. 2). Results support a pragmatic (over a purely lexical) account of subordinate conjectures. Thus subordinate noun learning offers an early demonstration of children's ability to integrate informativeness with the speaker's epistemic state during word learning.
Adaptive control-the ability to modulate attention based on contextual demands-is typically assumed to be a domain-general cognitive function. The function's actual domain-generality, however, remains unclear, as confound-controlled evidence for adaptive control is presently limited to the interference tasks used to examine it (typically, Stroop tasks) and the stimuli used in those tasks (typically, words and colors). We addressed these limitations by conducting a systematic examination of adaptive control in the three main interference-task types in the cognitive-control literature (i.e., Stroop, Stroop-like, and Simon, which are defined in terms of the degree of overlap between the targets, distractors, and responses used in those tasks) while varying the stimuli used in the tasks. The adaptive-control index we used was the Proportion-Congruent (PC) effect-the finding that, on confound-controlled trials, the effect of the distractor's congruency with the target on response time and accuracy is larger when the two are frequently congruent (e.g., RED in red) than when they are frequently incongruent (e.g., RED in blue). Across five experiments and seven different tasks with young adults (N = 432), we found robust PC effects in Stroop and Simon task types (in both types the distractor-associated response can conflict with the target-associated response) but not in Stroop-like task types (in which response conflict plays a weaker role). These findings establish the domain-generality of adaptive control but also suggest that response conflict is important for triggering adaptive control, a boundary condition that challenges extant adaptive-control models and will need consideration when revising them.
The origins of abstract concepts remain a central question in cognitive science. Empiricist theories propose that such concepts can be acquired through a set of domain-general learning mechanisms, whereas nativist accounts maintain that most abstract concepts are not learned. To investigate these contrasting perspectives, we focus on number, and specifically on how human minds perceive number in visual scenes, even without counting. Building on a recent empiricist theory inspired by generative deep neural networks - the "emergentist" account - we tested two predictions derived from model simulations: (a) number perception should be superior for stimuli reflecting properties of real-world scenes, i.e., those containing real-life correlations between number and cumulative area, and those containing real-life negative power-law distributions; and (b) these superiority effects should be especially strong in early childhood. Adult and 5- to 9-year-old participants completed a number estimation task in which stimuli either did or did not follow these natural properties. Contrary to the predictions of the emergentist models, we find little-to-no evidence for superiority for stimuli obeying natural statistics, and no evidence for predicted developmental effects. We conclude that the currently existing emergentist models of number perception do not represent number following the same mechanisms as human minds.
The majority of Chinese characters convey meaning explicitly via their semantic radical and studies investigating isolated character identification and character acquisition have shown them to be an important source of sublexical semantic information. However, to date, it is unclear whether Chinese readers recruit and utilize semantic cues from radicals when encountering, learning and representing entirely novel characters in natural reading - an important aspect of vocabulary growth. To address this question, we created sixteen semantic-phonetic compound single pseudo-character words as novel targets, whose semantic radicals were either consistent (i.e., transparent) or inconsistent (i.e., opaque) with high-constraint sentential contexts, and embedded them within nine sentence frames to support the formation of new lexical representations. The eye movements of 113 undergraduates were recorded as they read sentences. The results revealed that reading times for transparent novel words were significantly shorter relative to opaque novel words. This effect diminished over exposures, indicating a familiarity, or learning effect. When participants were categorized into faster and slower readers, we found that the semantic transparency effect was greater for faster than slower readers suggesting that faster readers utilized sublexical semantic information associated with semantic radicals to enhance word learning to a greater degree than slower readers. We consider our results in the context of Chen and Zhang's (2012) model of sublexical semantic radical processing and Share's self-teaching hypothesis. Our results are the first to demonstrate effects of sub-lexical semantic transparency during novel word learning in Chinese and highlight the important and distinctive role of semantic radicals in Chinese orthographic learning.
Stimulus probability manipulations typically increase the proportion of responses associated with the more likely stimulus and reduce response times, especially when stimulus discriminability is low. Sequential sampling models primarily attribute these asymmetries to shifts in the starting point of evidence accumulation, but possible modulations at the motor execution stage and their links to upstream processing remain poorly understood. To address this gap, we conducted electromyographic recordings of muscle activity during a random dot motion task with manual responses, varying stimulus probability and stimulus discriminability. We then used a computational framework linking decision and motor processes, the gated cascade diffusion model, to account for behavior and muscle activity. Stimulus probability strongly modulated the time from stimulus onset to response-related muscle activation: on correct trials, muscle onsets occurred earlier for expected than unexpected stimuli, whereas this pattern reversed on incorrect trials. By contrast, it had little impact on the time from muscle activation to response completion. Stimulus probability also modulated subthreshold muscle activations: when the first activation occurred in the response channel, accuracy was higher for expected than unexpected stimuli, whereas this effect reversed when the first activation occurred in the opposite channel. The model accounted for these effects through an additive bias to the rate of evidence accumulation that favored expected stimuli, and an evidence-independent urgency signal at the motor preparation level that was engaged earlier for the response alternative linked to expected stimuli. These analyses advance our understanding of how stimulus probability influences processing across the decision-motor cascade.
While performing distractor-interference tasks, people exhibit smaller congruency effects (i.e., less interference from distractors) after incongruent trials than after congruent trials. The influential conflict-monitoring model (Botvinick et al., 2001) posits that such congruency sequence effects (CSEs) reflect "conflict adaptation": increased control following heightened conflict in a previous trial. An alternative view posits that objective trial congruency (i.e., incongruent vs congruent) triggers increased control independent of conflict. Support for the latter view, however, comes from atypical tasks that lack an overall congruency effect because the distractor appears well before the target. Here, we seek to distinguish between these two views with a more typical task design. First, we simulate, in the context of a confound-minimized flanker task, (1) the conflict monitoring model, wherein conflict triggers control processes to shift attention toward the target and away from the distractors in the next trial, and (2) a proposed new variant of this model, wherein objective trial congruency triggers control processes instead of conflict. Our simulations show that response time correlates with internal conflict in these models and statistically mediates the relationship between congruency and conflict. Thus, the conflict model predicts the congruency effect varies with the previous trial's response time and not its congruency (each controlling for the other), whereas our proposed congruency model predicts the reverse. Second, we show that human data from the same task (n = 48) exhibit the latter data pattern and not the former one, thereby favoring the congruency model of the CSE over the conflict monitoring model.
How do people decide whether one event caused another? While previous research has focused on visual and auditory cues in causal perception, the role of touch remains underexplored. Here, we investigate how haptic feedback contributes to causal judgments across three psychophysical experiments. In Experiment 1, we introduced force-based haptic feedback to a visual launching paradigm and found that haptic information increased causal judgments compared to visual feedback alone. Experiment 2 combined vision, audio, force-based haptics, and vibrotactile haptics, revealing that additional sensory cues increase causal judgments with diminishing returns—the largest benefit comes from adding a second modality. In Experiment 3, we examined how both the number and physical realism of multisensory cues affect causal perception, finding that both factors boosted causal judgments. We present a Bayesian multimodal inference model that captures human judgments by integrating visual, auditory, and haptic information based on their relative timing, uncertainty, and realism. Taken together, these experiments show that haptic information contributes to causal judgments by shaping the multisensory evidence observers use when deciding whether one event physically caused another, including how realistic and physically coherent the event appears. More broadly, we find that temporal alignment and cross-modal coherence are key constraints for how multisensory evidence shapes causal judgments, with implications for virtual reality, robotics, and human-computer interaction systems.
Human perception, despite its high precision, consistently exhibits systematic biases. Interestingly, perceptual biases measured from perceptual estimation tasks can be either towards (attractive) or away from (repulsive) a feature reference, depending on specific visual features and experimental design. Extensive studies have been conducted to investigate computational mechanisms underlying these biases, many of which have introduced a Bayesian framework to integrate the influence of current input with prior knowledge. This framework is practically successful, especially with constraints of efficient coding of the current visual input. However, most models developed under this framework are mainly data-driven, requiring a large number of free parameters to flexibly fit empirical patterns. This approach, while enabling accurate fitting, often compromises model interpretability. In the present study, we aim to establish a unified model that reconciles repulsive and attractive biases using only two naturally presumed processes—efficient coding and Kalman filtering—along with a minimal set of free parameters. We tested the model by fitting data from three perceptual estimation studies exhibiting distinct bias patterns. The results showed that both attractive and repulsive biases were well captured, supporting the model’s validity and adaptability. Together, these findings suggest that the integration of efficient encoding and recursive Bayesian inference through Kalman filtering provides a parsimonious yet powerful account of a broad spectrum of perceptual biases. The model offers insight into a mechanistic explanation for diverse empirical patterns across perceptual domains.
Delay discounting characterized the devaluation of a reward’s subjective value as a function of temporal delay in receiving it and has been shown to be associated with a range of neural and psychological factors. However, our understanding of how multi-domain factors contribute to individual differences in delay discounting remains unclear. To address this question, we performed network analyses to disentangle the complex relationships between multi-domain factors and delay discounting utilizing data from the Young Adult cohort of the Human Connectome Project (HCP). Exploratory factor analysis was first conducted to parse 114 behavioral phenotype measures into 19 factors and 88 regional gray matter volume (GMV) variables into 14 factors. Subsequently, we used Gaussian Graphical Model and directed acyclic graph to estimate both the undirected and directed networks. Across network analyses, the crystallized intelligence, GMV of the cerebellum, and cigarette use emerged as critical nodes that directed contribute to individual differences in delay discounting. Notably, the externalizing behaviors appear to be a downstream consequence of delay discounting. Our findings identify crystallized intelligence, cerebellum, and cigarette use as potentially crucial mechanisms underlying dividual differences in delay discounting, which suggest that interventions specifically targeting these factors may effectively facilitate future-oriented decision-making.
One benefit of multisensory integration is the acceleration of responses in a perceptual decision task, known as the redundant signals effect (RSE). Based on probability summation, two basic principles have been proposed to explain such benefit: the “principle of congruent effectiveness”, which suggests maximal RSE when unisensory performances are comparable, and the “variability rule”, which proposes that RSE increases when unisensory performances are more variable. Yet, it remains unclear whether these principles extend to dynamic multisensory contexts and how they manifest in the model architectures proposed to explain RSE. To address these questions, we evaluated RSE in a multisensory context featuring transient onsets and offsets using a change detection task. Our results showed that both principles predicted the rank of the RSE in this dynamic setting, with the principle of congruent effectiveness emerging as the dominant one. Model comparison identified two best-performing model architectures that implement distinct crossmodal interactions (enhanced evidence accumulation rates or lowered decision criterion) in the perceptual process when two unimodal signals race towards their decision criterion. Moreover, the key predictors derived from the principles effectively modulate the crossmodal interactions in the two models. Together, these findings demonstrate that the principles underlying RSE generalize to dynamic multisensory contexts and operate via distinct computational mechanisms, shedding light on how perceptual systems flexibly integrate multisensory cues to optimize decision-making.
We humans manifest a remarkable capability of adapting to varying external environments. However, investigations into mechanisms that enable such adaptations raise several fundamental questions in the domain of sensorimotor control—most notably, whether and how we employ internal representations to facilitate more efficient feedforward control. The current study addressed this question by examining how trajectory predictability modulates manual tracking in human participants. A somewhat surprising finding was that no significant improvement was observed in tracking performance due to predictability, regardless of whether people were aware of it. However, analyses using a Kalman filter model revealed that participants did adapt their visuomotor processes by reducing their reliance on real-time visual input when tracking more predictable trajectories. These results suggest that, despite seemingly comparable tracking performances, participants increasingly relied on internal control mechanisms rather than real-time inputs when confronted with higher trajectory predictability. This study provides new insights into human sensorimotor control, highlighting adaptive use of information redundancy from the environments and the balance between feedforward and feedback mechanisms of human beings.
Recent studies have suggested that working memory (WM) representations can be retained through transient changes in synaptic weights, introducing the concept of activity-silent passive memory. This concept has been hypothesized to reflect mechanisms of episodic memory. However, there was no empirical evidence to delineate the role of episodic memory system in the passive memory functioning. To investigate this, we explored whether the passive memory suffered deficits similarly to the episodic memory system when disrupting the episodic memory functioning. It has shown that episodic memory system was damaged near in time to intentional suppression on unwanted events (i.e., amnesic shadow) in a Think/No-Think task, referred to as an amnesic shadow effect. In this study, participants completed Think/No-Think to induce amnesic shadow, and immediately performing a sequential WM task in which the first array was later probed as passive and the second as active. In this task, both active and passive memory representations in WM tasks were exposed to the amnesic shadow. The results showed that only passive memory suffered the amnesic shadow effect, while active memory was not affected. These findings provide the evidence that episodic memory processes contribute to the maintenance of activity-silent passive memory in working memory.
Inductive and deductive inferences have been assumed to rely on two qualitatively distinct processes by dual-process theories. However, studies examining the predictions of this theory have yielded mixed results, with several studies showing that a single process underlies both deductive and inductive judgments. Previous studies have used a range of manipulations, response options, and analytical techniques, which might partly explain the inconclusive findings. In this study, we conducted five experiments (overall N = 614) manipulating the typicality of the category-member relationship (typical vs. atypical pairs) and the quantifier of premises (all/universal vs. most/particular) in reasoning arguments. Dual-process theories predict a double-dissociation pattern in which the quantifier manipulation would impact deductive judgments more than inductive judgments, while typicality would have the reverse effect. To test these predictions, we employed a range of experimental tasks (within- and between-subject), response formats (binary and Likert), and analytical techniques (Bayesian hierarchical regression and state-trace analysis). The results failed to support the dual-process theory in that the predicted double-dissociation effect was not observed in most of the experiments. These findings align with a single-process framework, as proposed by the new paradigm of reasoning, for both deductive and inductive inferences. The implications of these findings for both dual- and single-process accounts are discussed.
Holistic processing of visual stimuli has long been regarded as unique to faces, or otherwise extended to other object categories given sufficient expertise. We designed novel abstract stimuli that are recognizable strictly by configural information, and matched control stimuli that require the use of only featural information. We then tested four classic markers of holistic perception: inversion, misalignment, part-whole, and composite effects. We found that second-order configuration stimuli elicited robust holistic effects, while featural stimuli do not, showing that such effects emerge specifically when recognition depends on configural information, but not when it relies on featural cues. We further observed that first-order configuration stimuli were also sufficient to induce holistic effects, indicating that holistic processing can emerge from multiple levels of configural information. We also found a significant correlation between individual differences in face recognition ability and holistic processing effects with the second-order configuration stimuli, but not with the first order configuration stimuli, nor the featural stimuli. Lastly, convolutional neural networks trained on the same stimuli reproduced these patterns, strengthening the interpretation that holistic processing arises when configural information must be used to recognize a stimulus. Together, these findings demonstrate that holistic processing is fundamentally rooted in the representation of spatial configuration, with second-order relations providing the critical link to face recognition ability.
Anchoring is a prominent judgment bias which causes people's estimates of uncertain quantities to assimilate towards recently encountered values. Here, we ask whether items can cause anchoring - will the question "Does a handheld flashlight torch cost more or less than a laptop?" induce anchoring in the same way as "Does a handheld flashlight torch cost more or less than £500"? We present evidence from ten studies suggesting that it can, and that perceptions of the value of the anchor item (e.g., the laptop) are also anchored. In other words, estimates for both items being compared assimilate towards each other. We also find that low value items are anchored more strongly than high value items. Overall, there is evidence for a small anchoring-by-items effect (Hedge's g = 0.25), which we suggest previous studies may have been underpowered to detect. Among existing theories of anchoring, Selective Accessibility would appear to provide the best account of the data.
There is a current debate on how time and space are represented in the brain, with some researchers advocating the view that time and space are represented within a generalized magnitude system and others arguing that temporal representations are based on spatial representations. The observation of asymmetric space-time interference, with time perception being more influenced by space than vice versa, has often been interpreted as reflecting a hierarchical representational structure. Here we explore how the factor of speed, which is inherent in many experiments on space-time interference (e.g., growing lines, moving dots), can contribute to the observed asymmetry. This idea is tested in two experiments, directly comparing duration and length judgments for growing and static lines (Experiment 1) and for growing and shrinking lines (Experiment 2). Experiment 1 demonstrates that the introduction of speed generally increases space-time interference, and that this increase is especially pronounced for the space-on-time effect, leading to stronger asymmetry. Moreover, Experiment 2 shows that, when the correlation between line length and speed is reversed (i.e., shorter lengths are coupled with higher stimulus speed), the space-on-time effect reverses as well. We conclude that the observed asymmetric space-time interference in experiments using dynamic stimuli is primarily based on the processing of speed and does not constitute evidence for the idea of a hierarchical representational structure of space and time.
In sentences like "The coach smiled at the player tossed a frisbee," the string "the player tossed a frisbee" cannot be an active subject-verb-object (SVO) clause given the preceding context; yet, comprehenders seem to entertain this incorrect parse, at least momentarily. Behaviorally, this momentary mis-parse is expressed as greater difficulty when the SVO chunk is read. This phenomenon, called a local coherence effect, has important implications for sentence processing theories that treat grammar as a strict filter during incremental sentence processing: Under such a strict filter, local coherence effects should never occur. Although several studies report the existence of local coherence effects, one question remains unanswered: at what moment are local coherence effects triggered, and how quickly - if at all - does grammar override the mis-parse? We investigate the time course of local coherence effects through two experiments in German (self-paced reading and EEG). Our data suggest that the local coherence effect, indexed by longer reading times and a more positive P600, was triggered as soon as the locally coherent chunk was read. However, the locally coherent parse did not linger; it did not continue to cause processing difficulty. Our results are compatible with self-organized parsing and versions of good-enough processing. Our results do not support accounts in which local coherence effects are caused by uncertainty about previous input or by a complete breakdown of algorithmic parsing. A broader implication of our findings is that although grammar is not a strict a-priori filter, it can step in rapidly to correct incremental structure building.
