Representations for face recognition: The 53rd Bartlett Lecture

Models of human face recognition rely on the notion of representation, but rarely describe this in detail. Here, I will argue that our conception of face representations is often 'essentialist' - assuming that there is some fixed set of values that captures a particular person's face. However, this conception is inadequate for the purpose of familiar face recognition, and I will suggest that representations instead need to incorporate the statistical properties of our exposure to all the faces we know, including variability and sampling. I will review findings from empirical and simulation research suggesting that the idiosyncratic properties of each perceiver result in a unique set of representations, which can be difficult to understand using traditional experimental approaches. Methodological diversity seems to offer the best route for understanding face recognition - a problem that remains stubbornly unsolved.