Dissecting the features and neural mechanisms supporting naturalistic social inference

点击翻译按钮获取中文摘要

Project Summary This first-time fellowship application proposes an integrated training and research plan focused on human social interactions. The applicant has a strong and broad background in computational modeling of behavioral human data, and will now leverage this in her postdoc at Caltech to learn how to model and analyze human interactions from text-based chats and video interactions in behavior and in the brain (using fMRI). Aim 1 will utilize state-of-the-art computer vision and natural language processing methods to annotate the visual, auditory and text features of social interactions. Specific features, such as facial expressions and the semantic content of text, will be extracted from 300 recorded social interactions between healthy adults. These features, in turn, will be used to fit models that predict the social judgments that participants make about one another: how trustworthy, friendly or arrogant do they judge their partner to be? A main goal of this first Aim is to extend human social judgments beyond the typically narrow context used in past studies (e.g., static photos of faces) into the naturalistic, interactive context we actually encounter in the real world. This emphasis will also be important for future applications to psychiatric populations with critical deficits in social cognition (such as autism, outside the scope of the present fellowship). Aim 2 translates the interactions of Aim 1 into neuroimaging, and BOLD-fMRI will be acquired while participants watch previously recorded social interactions. This will reveal the brain regions and networks that track the dynamic features of the stimuli. Neural representations of social inferences will be extracted from multivariate activation patterns using representational similarity analysis, and similarity in neural processing will be compared to similarity in behavioral social inference using inter-subject correlation analysis. All analyses will both apply a whole-brain approach and will query specific social cognition networks. A main goal of Aim 2 is to characterize the neural systems that subserve the dynamic construction of social attributions, information that in future studies can be linked to individual differences and psychiatric illness.