A comparative study of face processing using scrambled faces

Abstract

It is a widespread assumption that all primate species process faces in the same way because the species are closely related and they engage in similar social interactions. However, this approach ignores potentially interesting and informative differences that may exist between species. This paper describes a comparative study of holistic face processing. Twelve subjects (six chimpanzees Pan troglodytes and six rhesus monkeys Macaca mulatta) were trained to discriminate whole faces (faces with features in their canonical position) and feature-scrambled faces in two separate conditions. We found that both species tended to match the global configuration of features over local features, providing strong evidence of global precedence. In addition, we show that both species were better able to generalize from a learned configuration to an entirely novel configuration when they were first trained to match feature-scrambled faces compared to when they were trained with whole faces. This result implies that the subjects were able to access local information easier when facial features were presented in a scrambled configuration and is consistent with a holistic processing hypothesis. Interestingly, these data also suggest that, while holistic processing in chimpanzees is tuned to own-species faces, monkeys have a more general approach towards all faces. Thus, while these data confirm that both chimpanzees and rhesus monkeys process faces holistically, they also indicate that there are differences between the species that warrant further investigation.

Figure 1

A schematic representation of the experimental procedure. Subjects would be trained to discriminate between either whole (training condition 1) or scrambled (training condition 2) faces before they completed corresponding test phase. For the purposes of illustration, in each of these samples the same appears at the top of the visual display with the target in the bottom left corner and the distractor in the bottom right corner.

Figure 2

(a) The interaction between congruency and training condition (whole faces, scrambled faces); (b) the interaction between congruency and subject species (chimpanzees, rhesus monkeys); (c) the interaction between congruency and the species of stimulus (chimpanzee faces, monkey faces, human faces) with mean matching performance on the y axis); and (d) mean difference in congruency (congruent trials − incongruent trials) as a function of subject species and stimulus species. Error bars = ±1 SEM.

Figure 3

(a) The main effect of training type on congruency trials; (b) the main effect of training type on new configuration trials; (c) the main effect of stimulus species; and (d) the mean difference in training condition (scrambled faces − whole faces) for each subject species (chimpanzees and monkeys) separated by stimulus species. Error bars = ±1 SEM.