CLChujun Lin

Chujun Lin

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Chujun Lin has a total of 48 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    The neurocomputational mechanisms of naturalistic trait impression updating

    by Lindsey Tepfer, Chujun Lin, Mark Allen Thornton

    Making prudent social decisions in everyday life requires people to form impressions of others’ personalities. First impressions, although impactful, are frequently incomplete or inaccurate. This necessitates mechanisms for updating impressions. This research examines types of information and computational operations the brain uses to update impressions. We collected functional neuroimaging data from participants (N = 28) as they viewed naturalistic audiovisual stimuli of people (N = 28) narrating emotional events in their lives. Pre-trained machine learning models annotated these target people’s facial expressions, voice qualities, and speech semantics. Deep neural networks implemented further computational operations on these annotations, including compression, multimodal fusion, temporal integration, and prediction. Independent participants (N = 602) continuously rated how their impressions of each target person’s traits varied across the course of the videos. Using spatiotemporal representational similarity analysis, we tested whether different perceptual modalities and operations explained patterns of brain activity in scanner participants and mediated spontaneous trait impression updating in the brain. We used intersubject representational similarity analysis to test whether participants’ mood, personality, demographics, or social attitudes explained individual differences in neural signatures of trait impression updating. All perceptual modalities and most computational operations were detected in the brain, but only facial expressions, speech semantics, and compression explained neural activity associated with trait impression updating. Social group attitudes were the only individual difference that mediated neural idiosyncrasies in trait updating. These results indicate that changing one’s mind about someone is a multifarious process, reliant upon multiple modalities of information, computational operations, and prior beliefs.

  • Open Access

    Target and perceiver informational determinants of naturalistic trait impression updating

    by Chujun Lin, Mark Allen Thornton

    Understanding others’ personality traits plays an important role in guiding social interactions. In daily life, people update their impressions of others’ traits dynamically based on short encounters. Here, across four studies (N = 1,296), we used naturalistic experimental paradigms and novel computational methods to characterize how targets’ multimodal behavior and perceivers’ own characteristics shaped dynamic trait impression updating during naturalistic video viewing. By examining a comprehensive set of 38 traits, we found that people updated face-based first impressions of many traits after viewing autobiographical audiovisual narratives told by targets (Study 1). By experimentally removing verbal, visual, and vocal cues from the videos, we found that the dynamics of impression updating were predicted by the unique information in what the target said, how they moved, and how they sounded, as well as the shared information among these cues (Study 2). By computationally quantifying multimodal cues as they naturally co-evolved in unedited videos, we confirmed the unique roles of these cues and their sub-features (e.g., facial muscle movements, head movements, accumulated narrative content, energy structure of a voice) in driving the dynamic process of trait impression updating (Studies 3 and 4). We also found that perceivers’ own affect, attitudes, and personality predicted the dynamics of trait impression updating. These findings move beyond prior literature on one-shot impressions and the magnitude of impression changes by comprehensively characterizing the informational drivers and their unique roles for how trait impressions dynamically unfold in naturalistic contexts.

  • Open Access

    Stereotypes as Byproducts of Hierarchical Bayesian Inference

    by Junsong Lu, Chujun Lin

    Stereotypes are ubiquitous even when inaccurate. Why do they arise readily and persist? We propose a novel computational account linking person perception and group stereotypes: stereotypes emerge as byproducts of hierarchical Bayesian inference (HBI), which aims to maximize the accuracy of person perception. While efficient for individual-level perception, HBI can facilitate biased group stereotypes. We tested and supported four stereotype-related predictions derived from HBI using five simulations and seven empirical samples (n = 1,273). Consistent with HBI, stereotypes not only emerged when perceivers were asymmetrically exposed to different groups (Study 1) but also when exposure was sufficient, because HBI enables faster acquisition of group-level impressions (Study 2). Established group beliefs persisted because counterevidence was attributed to idiosyncrasies (Study 3) or to an individual's secondary group membership (Study 4). Together, these findings suggest HBI as a unifying account of stereotypes, advancing a generalizable computational framework for social cognition.

  • Open Access

    Facial and body actions speak louder than context when inferring facial expressions of emotions in naturalistic images

    by Ruoying Zheng, Kevin Liu, Jiabin Zou, Chujun Lin

    Inferring others’ emotions is crucial for survival. Much work focused on facial expressions; others argue that rich body and contextual cues also shape emotion inferences from faces, often based on experimentally combined face images and contextual cues. However, people may use a different strategy to infer emotions in real-world contexts where rich cues naturally co-occur. Here, we had US representative participants (N = 475) view everyday images (N = 1,125) and infer the targets’ facial expressions of 8 emotions. We modeled how rich cues in the images quantified by computational algorithms and US representative participants (N = 1,093) predicted inferences of facial expressions of emotions. We found significant predictions by not only facial cues (e.g., action units, facial structure) but also body cues (e.g., posture, high-level actions) and contexts (e.g., environmental valence, gist). However, these predictions were based on different mechanisms: Facial and body actions carried unique information that was unexplainable by other co-occurring cues, whereas other cues primarily carried information that was shared. Given the differential mechanisms, we investigated whether integrating merely cues that carried unique information was sufficient for predicting human emotion inferences. We built artificial neural networks that allowed for process-level manipulations: by lesioning artificial neurons, we found that neurons that were crucial for human-like inferences integrated rich cues asymmetrically beyond facial and body actions. Together, these findings suggest that naturalistic inferences of facial expressions of emotions may be driven by facial and body actions that are contextualized by other co-occurring cues to support a coherent interpretation.

  • Open Access

    A Small-World Mind: Towards a General Computational Principle of Social Cognition across Contexts

    by Junsong Lu, Chujun Lin

    People describe social inferences using hundreds of words. How do people mentally represent these inferences? A long-standing tradition suggests that social inferences are underpinned by a small number of dimensions (e.g., warmth, competence). Another long-standing perspective argues they are more complex and require many dimensions to capture – an idea that has regained support by recent work using naturalistic designs. Motivated by this debate, we propose a network theory explaining how social inferences are mentally structured and generated in response to social information. Specifically, we propose that mental representations of social inferences form a small-world network with two principles: growth – the network expands by adding inference concepts, and preferential attachment – new concepts preferentially connect to well-connected ones. This representation is scalable while remaining cognitively parsimonious. We discuss the plausibility of this account based on existing social, developmental, and neurological findings. Our network theory accounts for classic phenomena—halo effects and psychological dimensions—as emergent network dynamics. Specifically, in small-world networks, nodes (social inferences) are connected with only a few steps from one another, so initial activation in a small set of nodes from constrained stimuli propagates broadly, producing synchronous covariation that mimics halo effects or low-dimensional structures. Dimensionality of activation gradually increases with input complexity. This network account reveals that established psychological phenomena may emerge from computations performed on mental representations in response to specific contextual inputs. By reconciling seemingly contradictory findings, our theory advances understanding of how the mind efficiently organizes vast social knowledge while adapting to environmental complexity.

  • Open Access

    Inferring corruptible politicians from faces applies only if they are White

    by Yufei Mao, Chujun Lin

    First impressions from faces shape important real-world outcomes. We investigated to what extent and by which mechanism political corruption was associated with facial impressions across politicians’ gender, race, and perceivers’ culture. Across four preregistered studies (N = 1,352), we found that Caucasian male and female politicians who looked corruptible were more likely to have been convicted of corruption. We replicated this finding across perceivers from U.S. representative samples, Europe, and Asia (Studies 1-2). This association was absent for African American politicians, buffered by the Teddy Bear effect (Study 3). We surveyed participants working in industries that are vulnerable to corruption (e.g., finance) or combat corruption (e.g., legal). We found that corruptible-looking politicians made participants want to investigate them more but not bribe them more (Study 4). Together, these findings reveal a new mechanism underlying facial bias in justice and caution against the potential under-scrutiny of genuinely corrupt but trustworthy-looking politicians.

  • Open Access

    Beyond trait summaries: stereotypes as multimodal mental representations of groups across perceivers, situations, and intersecting categories

    by Yilin Wang, Chujun Lin

    Stereotypes are often studied as trait summaries of groups. Yet in everyday life, people observe and represent others concretely through multiple modalities, including how others look, act, think, and feel, which are only partially captured by traits. In two pre-registered studies with U.S. representative samples (N = 1,721), we investigated multimodal stereotypes by letting participants freely describe spontaneous mental images of 9 groups across 6 situations and 30 intersections. We found both idiosyncratic perceptions and consistent cultural stereotypes. Multimodal stereotypes changed across contexts, but some elements were more rigid than others. Intersecting groups shaped multimodal stereotypes to different degrees; two exceptions were “Asian” and “unattractive”, which dominated the other intersecting groups. Across 30 intersectional groups, 27 supported the additive mechanism, and 3 showed interactive patterns. Together, these findings show that investigating stereotypes as multimodal representations is key to understanding how different elements of stereotypes change across perceivers, situations, and intersections.

  • Open Access

    Not what you were born with: people prioritize controllable cues when forming first impressions in naturalistic contexts

    by Ruoying Zheng, Chujun Lin

    First impressions are ubiquitous and consequential. Prior work identified a rich set of cues shaping impressions by studying one cue at a time (e.g., facial structure, body shape, hair color, environmental valence). Here we ask whether integrating rich cues simultaneously offers new insights into how people make snap judgments in naturalistic contexts. This poses a technical challenge—characterizing and manipulating rich cues in naturalistic stimuli. We tackled this challenge using novel computational methods. Across three large-scale, pre-registered studies across U.S. and Chinese representative participants (N = 3,734), we found that, with rich cues, people relied on only a small subset to make snap judgments. These cues were often controllable by the targets (e.g., action, clothing, rather than facial structure). We replicated these findings in both the U.S. and China, across seven judgments, and in novel stimuli that the models were not trained on. A general mechanism underlies these rich cues and impressions: people form impressions using cues that i) carried unique information beyond other cues, which makes these cues uniquely informative, or ii) shared information with other cues, which makes these cues reliably reinforcing judgments. By computationally manipulating one cue at a time in naturalistic images, we showed that changing just one of these controllable cues was sufficient to change others’ impressions in naturalistic contexts. Together, our findings advance the mechanism of snap judgments in the real world: the human mind may leverage the structure of co-occurring cues that it learns in daily life to prioritize a small subset of cues for impressions.

  • Open Access

    Trait judgments of faces using a task without word labels

    by Yue Xu, Umit Keles, Chujun Lin, Ralph Adolphs

    People reliably attribute psychological traits to faces. Prior research typically elicits these judgments using specific words provided by the experimenter (e.g., ratings on “competent,” “trustworthy”), raising questions about completeness (tasks use a limited set of words) and generalizability (e.g., across languages). In two experiments (Experiment 1 with predominantly White faces; Experiment 2 with White, Black, and Asian faces) we asked over 3,000 participants to spatially arrange unfamiliar faces based on perceived trait similarity, without providing any word labels. Experiment 1 yielded five dimensions we interpret as approachability, gender-specific warmth, competence, morality and youthfulness. Experiment 2 also produced five dimensions, which showed prominent effects of stereotypes associated with race. Comparisons with four other studies that used word labels argue that trait judgments are characterized by 2-5 dimensions, that approachability/warmth emerges as a universal evaluative factor, and that race and gender stereotypes emerge depending on the set of face stimuli used.

  • Open Access

    Network models reveal high-dimensional social inferences in naturalistic settings beyond latent construct models

    by Junsong Lu, Chujun Lin

    Long-standing research suggests that social inferences are captured by a few latent dimensions (e.g., warmth and competence). Others argue that social inferences are more complex but lack sufficient empirical support. Here, we conducted two pre-registered studies to test the high-dimensional properties of social inferences. To maximize generalizability, we computationally sampled diverse naturalistic videos and recruited U.S. representative participants (Study 1, N = 1598). Participants freely described people in videos using their own words. Cross-validation identified 25 latent dimensions which explained only 15% of the variance in the data. Alternatively, a sparse network model representing the unique correlations between inferences better represented the data. The network models informed the dynamics of naturalistic inferences, revealing how different inferences co-occurred and how they unfolded over time from concrete to abstract (Study 1). The network models also indicated cultural differences in how one inference was related to another between samples (Study 2, Asian N = 651, European N = 792). Together, these findings show that the high-dimensional network approach provides an alternative model for understanding the mental representation of social inferences in naturalistic contexts, which provides new insights into the dynamics and diversities of social inferences beyond the static, universal structure found with traditional low-dimensional latent-construct approaches.