Cognition最新文献

A central question about our shared capacity for language is how it is integrated with other cognitive systems. One important debate focuses on the extent to which the form of linguistic expressions is grounded in their communicative function: Can all constraints on linguistic form be attributed to the way constructions package information, or is linguistic form autonomous of meaning and function? One area of disagreement involves islands: phrases which block the formation of long-distance filler-gap dependencies (Ross, 1967). Grammatical subjects are considered islands, since questioning a sub-part of a subject results in an ill-formed sentence, e.g., "Which topic did the article about inspire you?". Autonomous syntactic approaches to islands attribute this ungrammaticality to the abstract movement dependency between the wh-phrase and the subject-internal position with which it is associated. An alternative developed in Abeillé et al. (2020) suggests that subjects' island status is specific to the information structure of wh-questions, suggesting that subjects are not islands for movement, but for focusing, due to their discourse-backgroundedness. This predicts that other constructions that involve movement but not focusing should not create a subject island effect. We test this in three acceptability studies, using a factorial design to isolate subject island violations across three constructions: wh-questions, relative clauses and topicalization. We find a subject island effect in each case, despite only wh-questions introducing what Abeillé et al. (2020) call "a clash in information structure". We argue that this motivates an account of islands in terms of syntactic representations shared across constructions, independent of communicative function.

We are intimately familiar with liquids in our visual experience, yet the computational basis of liquid perception remains underexplored. This is an important knowledge gap because liquids, with their mutable shapes and complex intrinsic dynamics, differ remarkably from the commonly studied categories in computational vision, such as rigid objects or non-rigid solids. To understand the computational basis of liquid perception, we implemented different models of this ability and tested them in a new behavioral study. The models realize two distinct theoretical possibilities for the visual perception of liquid viscosity. The first possibility, and the focus of most existing work, explains the representation of liquid viscosity as a consequence of high-level image and motion statistics discriminative of the gradations of this physical property. A second, much different possibility is that the perceptual representations of liquids functionally map the physical processes of how viscosity and external forces (e.g., gravity, rigid surfaces) shape the way liquids move. We task these models and humans in a new behavioral task: making similarity judgments of liquid viscosity across pairs of animations depicting qualitatively different scenarios - e.g., a metal ball falling into a liquid container vs. liquid pouring over a non-flat surface. We find that a new model, Ripple, which builds and manipulates physics-based representations of liquid viscosity from sensory inputs, explains substantial variance in human judgments beyond powerful, previously behaviorally validated, statistical representations of viscosity. Moreover, statistical representations of viscosity across vastly different model architectures - a task-specific DNN and a general video foundation model - converge with one another, while remaining equally differentiated from Ripple. These results suggest that liquid perception extends beyond image statistics to also involve simulation-based intuitive physics.

Visual attention can be guided away from objects known to be irrelevant to the current task. These negative templates (specifying distractor features) can be maintained in visual working memory (VWM) and in long-term memory (LTM). LTM-based negative templates allow for direct suppression of to-be-avoided feature values, observable in the earliest selective operations during search (i.e., implemented proactively). However, there is mixed evidence regarding whether VWM-based negative templates are likewise implemented directly and proactively. Here, we contrasted LTM- and VWM-based negative guidance within the same visual search experiment. There were two broad lines of findings. First, the two sources of guidance dissociated on several measures of oculomotor orienting during visual search, including: a) the polarity of initial guidance, b) the latency of initial orienting, and c) the pattern of guidance across an extended search trial. We conclude that the two forms of guidance are implemented by fundamentally different mechanisms. Second, we created conditions in which the two forms of guidance were potentially operational within the same trial, testing their interaction. Both were expressed within a trial when they specified different sets of objects. However, VWM-based biases dominated when the two biases were placed in competition, indicating that online attentional sets tend to overshadow learned biases in the computation of priority.

People find it harder to switch from one task to another than to repeat a task. One common explanation is that lingering activation of the just-executed task facilitates repetitions and impairs switching. However, beyond specific task sets, it is also conceivable that switching a task increases more abstract switch readiness, whereas repeating a task reduces switch readiness. To investigate switch readiness independent of task-set activation, we used consecutive chunks, each consisting of two tasks, with self-paced breaks between chunks. This way, the salient task transition happens within a chunk, independent of the task transition between chunks. In four experiments, we applied a (hybrid) task-switching paradigm with a mixture of forced choice (only one task presented) and free choice (participants can decide which task to perform). We expected an increased ability and willingness to switch (i.e., switch readiness) in the current chunk when the previous chunk entailed a task switch rather than a repetition. In line with a switch-readiness account, Experiments 1 and 2 showed reduced switch costs and increased voluntary switch rates (VSR) after a switch within the previous chunk. Furthermore, this effect transferred to new task pairs (only descriptively in Experiment 3, significantly in Experiment 4). Taken together, the present study uncovered a novel property of sequential control during task switching.

The Spatial-Numerical Association effects describe the spatial relationship between number magnitude and response side, with small numbers usually associated with left sided responses and large numbers with right sided responses. Typically, these effects are demonstrated using response time differences in simple key press tasks, where participants are required to process the magnitude (magnitude classification task) or parity (parity judgement task) of a number. The present study investigated whether similar spatial biases (left/right) also occur for decisions that involve more complex movements, namely walking. Using a free response task, presented in a virtual reality environment, participants were shown a number from 1 to 9, that was presented directly in front of them. At the beginning of each trial participants were required to process either the number's magnitude (Experiment 1) or parity (Experiment 2). They were then asked to walk freely in any direction towards a semi-circular target area, while continuing to process information in working memory. The results showed a higher frequency of leftward walking decisions for smaller numbers and rightward walking decisions for larger numbers in both experiments, as well as compatible deviations of walking trajectory. These findings are consistent with previous literature on SNAs. This study highlights that in a free response task both spatial decisions and spontaneous movements are influenced by number magnitude, both when magnitude is task-relevant and when it is task-irrelevant.

The principle of equal human worth is widely endorsed, yet real-world situations often require trade-offs. This raises a fundamental question: Do individuals truly value all human lives equally from an early age, or do they differentiate based on salient attributes? In a cross-sectional study, children aged 5-10 years (N = 253, 47% female) and their mothers made binary life-or-death choices between two individuals differing in age and sex. Results showed that even the youngest children did not value all lives equally. With age, children increasingly prioritized younger individuals, plausibly reflecting a growing understanding that older people have less time left to live, and showed reduced same sex ingroup preference. Machine learning models predicted older children's choices more accurately, suggesting that decision-making becomes more systematic and predictable with development. Mothers prioritized younger and female lives, with the strongest female preference emerging when the two individuals differed in sex but not age. Framing also influenced judgment: saving vs. leaving behind altered the strength of the preference for younger lives. These patterns align with social norms and gender stereotypes (e.g., protection of "vulnerable" groups, gendered expectations of helpfulness and susceptibility to harm). Evolutionary frameworks, such as reproductive value and parental investment, offer potential explanations for why such norms and stereotypes seem pervasive. Overall, the findings indicate that the valuation of human lives is initially not egalitarian, becomes increasingly structured across childhood, and adult priorities may arise from the interplay between evolved caregiving heuristics and fairness norms.

The visual ventral and visual dorsal pathways mediate visual perception and visually guided action, respectively. While perceptual learning is well established, it is still debated whether action learning-training motor responses to visual features-can operate under degraded vision when conscious visual perceptual learning is minimized, and it remains unclear what factors govern its success. These questions strongly limited the application of action learning. Here, we trained 24 adults to grasp low-contrast vertical or horizontal bars rendered highly degraded using continuous flash suppression, testing their contrast thresholds and grasping performance before and after training. We found that at the group level, action learning not only enhanced visual sensitivity but also improved grasping performance despite no improvement in the perceptual visibility of the stimuli. Critically, learning success correlated with gains in contrast sensitivity rather than initial stimulus visibility, suggesting that action learning depends on enhanced sensory encoding rather than conscious perception. Additionally, participants who had no initial learning with low contrast stimuli started to show improvements when stimulus contrast was increased, indicating that action learning requires sufficient (but not necessarily visible) sensory input. These findings demonstrate the dorsal stream's capacity for implicit, action-driven plasticity, even with degraded stimuli, and clarify that successful action learning is governed by improvements in sensory encoding rather than perceptual awareness. Overall, our findings suggest that action-based training could serve as a promising rehabilitative approach for degraded vision, provided stimulus strength is carefully optimized, opening new avenues for rehabilitation in vision deficits (e.g., amblyopia, cortical blindness).

The human body in motion is typically in service of everyday tasks - but it can also be aesthetically pleasing, as in dance. Is there a common property underlying how we perceive and appreciate human motion in both everyday and aesthetic contexts? Here we focus on one of the most basic features of motion: the speed at which it unfolds. In a 'set-the-pace' task, people adjusted the playback speed of a broad range of videos to what was "maximally pleasing" to them. This method revealed three distinct correlates of aesthetic speed preferences across multiple experiments and replications. First, when looking across genres, aesthetically pleasing speeds converged at a sweet spot of amount of motion, computed by a measure of optical flow: faster dances (e.g., Bollywood) were consistently slowed down, and slower dances (e.g., classical ballet or Chinese folk) were sped up. Second, aesthetic speed preferences were domain-specific: responses for dance videos were predicted only by responses for videos involving human (e.g., sports), but not non-human (e.g., object or animal) motion. Finally, what's specific to aesthetic speed preferences for human motion? Separate sets of observers adjusted playback speeds to "where the dancer was maximally expressive" and to "the pace at which people naturally move". Aesthetic speed preferences were highly correlated with both measures, but only "natural" pace ratings ultimately predicted unique variance above and beyond other factors. Thus, aesthetic speed preferences in dance may be tuned not just to the outward expressiveness of the dancer, but also to what we synchronize to, i.e., the pace at which we encounter human motion in everyday life.

Four preregistered studies find that perpetrators who commit moral transgressions are judged less harshly when their transgressions impact individuals who have previously committed immoral actions (i.e., bad victims). In Studies 1 and 2, we used between- and within-participants comparisons to find that perpetrators, and the moral transgressions they committed, were judged less harshly when they affected bad (vs. neutral) victims. In Studies 3 and 4, we replicated these effects and found that bad victims were seen as having less capacity for emotional experience and suffering. Increased leniency towards perpetrators who transgressed against bad victims occurred even though perpetrators in our studies did not know who their victims were, meaning that it does not result from the perception that perpetrators were intentionally punishing bad victims. Instead, moral transgressions against bad victims may be viewed as less wrong because these transgressions are perceived to cause less harm to their victims.