The computational substrates of holistic face processing in cognitive control

Abstract

Upright faces are considered to be processed more holistically than inverted faces in the face inversion and face composite as well as parts-whole paradigms. However, little is known about whether the holistic face processing is affected by face orientation in the conflict tasks, which computations characterize cognitive control mechanisms of upright and inverted faces in these conflict tasks, and whether the holistic cognitive control mechanism is manifested differentially in determining the direction of arrow and gaze. Here, we address these questions by asking participants to perform a novel arrow-gaze Stroop task including arrow judgment task and gaze judgment task. We show that the congruency effect is ostensibly not affected by the upright or inverted faces in our modified Stroop task. Consistent with previous findings, we find that the magnitude of the congruency effect is larger for gaze than arrow judgment. By conducting a systematic hierarchical Bayesian estimation of three conflict drift-diffusion models (DDMs) – the diffusion model for conflict tasks (DMC), the dual-stage two-phase (DSTP) model and the shrinking spotlight (SSP) model, our study provides the first model-based assessment of holistic face processing-related cognitive control mechanisms. Fits of the models to data reveal a superiority of SSP over DMC and DSTP. The SSP accounts for our findings that 1) holistic face processing only affects the underlying factors in the arrow task instead of ostensible performance, however, part-based processing affects both levels in the gaze task; 2) both shared and dissociable attentional effects exist between eye gaze and arrow stimuli. These results provide valuable evidence of computational substrates in holistic face processing of cognitive control and in social & non-social attentional mechanisms.

Beyond boundaries: conflict processing mechanisms in mental imagery across disparate locations

“…To mitigate the influence of noise variability in model selection, we performed five independent estimations of the log-marginal likelihood for each model for each individual. Wang & Nie, 2022;Wang & Nie, 2023). The log-Bayes factors were computed for each person through the following two steps: (1) identifying the superior model by finding the one with the highest log-marginal likelihood out of the five approximations, and (2) Calculating the log-Bayes factor by taking the difference between the minimum logmarginal likelihood of the superior model and the maximum log-marginal likelihood of the inferior model, reflecting the lowest difference of the two models.…”

“…Their findings supported the hypothesis that conflict can indeed arise even in the absence of stimulus conflict. In their study, Wang and Nie (2022) combined the arrow-gaze Stroop task (Wang & Nie, 2023) with the mental rotation task, creating what they termed the arrow-gaze mental-rotation Stroop task. In their first experiment, the human face was presented either in an upright position or rotated by 90°, allowing the eyes on the face to appear to look left or right in a head-centered frame.…”