Facial expressions of sentence comprehension

Abstract

Understanding facial expressions allows access to one's intentional and affective states. Using the findings in psychology and neuroscience, in which physical behaviors of the face are linked to emotional states, this paper aims to study sentence comprehension shown by facial expressions. In our experiments, participants took part in a roughly 30-minute computer mediated task, where they were asked to answer either 'true' or 'false' to knowledge-based questions, then immediately given feedback of 'correct' or 'incorrect'. Their faces, which were recorded during the task using the Kinect v2 device, are later used to identify the level of comprehension shown by their expressions. To achieve this, the SVM and Random Forest classifiers with facial appearance information extracted using a spatiotemporal local descriptor, named LPQ-TOP, are employed. Results of online sentence comprehension show that facial dynamics are promising to help understand cognitive states of the mind.