Learning how faces vary: Representing variance across multiple instances of an unfamiliar face.

Abstract

This thesis explores the mechanisms which underlie the improvement to face-identification which arise from familiarity with a face. Experiments investigated whether information is aggregated across multiple encounters with a single face, and assessed the influence of experiencing variance in appearance across these encounters. To answer these questions, pairs of face instances were either correctly presented as a single target identity (1ID condition), or misleadingly presented as two different target identities (2ID condition). Participants completed face-matching tasks with these targets. Initially (Chapter 2) the manipulation did not affect face-matching performance, reflecting an ineffective experimental familiarisation procedure. In sequential face-matching tasks, response bias was more liberal (Chapter 3; photographs), and overall accuracy was higher (Chapter 4; video-clips) in the 1ID condition, although identical pictorial information was available in the 2ID condition. Thus, it appears that identification in the 1ID condition relied on the combined information from both target instances, rather than a single “most-informative” instance. This is consistent with theories suggesting that we aggregate experiences with familiar faces into a single representation of that face (Bruce & Young, 1986). Throughout these experiments, participants experienced greater or lesser amounts of variance in appearance across images of a face. Exposure to more “within-face” variance resulted in improved sensitivity when matching that face (Chapter 5). Greater exposure to within-face variance resulted in a consistently more liberal criterion placement for that specific face (Chapters 3-5), and also for other faces in a block of face-matching trials (Chapter 6). This relationship between variance and criterion held for non-face objects (Chapter 6). In summary, these results confirm that information is combined across multiple experiences with a face, perhaps on a representational level. Moreover, exposure to within-face variance mirrored the effects of familiarity in improving identification performance. These experiments also highlight the importance of analysing criterion in identification tasks, as a means of assessing the variance incorporated into internal face representations. These findings provide theoretical insight into the nature of face representation, and have implications for the design of photo-ID, to optimise real world identification performance.