Four year-olds use norm-based coding for face identity

Abstract

Norm-based coding, in which faces are coded as deviations from an average face, is an efficient way of coding visual patterns that share a common structure and must be distinguished by subtle variations that define individuals. Adults and school-aged children use norm-based coding for face identity but it is not yet known if pre-school aged children also use norm-based coding. We reasoned that the transition to school could be critical in developing a norm-based system because school places new demands on children’s face identification skills and substantially increases experience with faces. Consistent with this view, face identification performance improves steeply between ages 4 and 7. We used face identity aftereffects to test whether norm-based coding emerges between these ages. We found that 4 year-old children, like adults, showed larger face identity aftereffects for adaptors far from the average than for adaptors closer to the average, consistent with use of norm-based coding. We conclude that experience prior to age 4 is sufficient to develop a norm-based face-space and that failure to use norm-based coding cannot explain 4 year-old children’s poor face identification skills.

Introduction

Faces convey a wealth of information that we use to guide our social interactions. As adults we swiftly extract information about identity, gender, ethnicity, age and emotional state from faces. Face identification, in particular, requires exquisite sensitivity to very subtle differences between highly similar visual patterns. Norm-based coding represents an efficient and elegant solution to this problem of representing visual patterns that share a common structure and must be distinguished by subtle variations that define individuals. A norm-based system represents what is distinctive about each face by coding how faces deviate from a perceptual norm or prototype (see Fig. 1). This system may be more efficient than one that codes a complete structural description, most elements of which are shared by all faces and therefore redundant. Moreover, face norms are updated by experience, fine-tuning our face perception coding mechanisms to our diet of faces (e.g., Rhodes, Jeffery, Watson, Clifford, & Nakayama, 2003). A variety of behavioral and neurophysiological evidence supports use of norm-based coding for facial identity in adults (see Rhodes & Leopold, 2011, for a review). The functional importance of adaptive norm-based coding is suggested by evidence that stronger adaptive norm-based coding of faces is associated with better face identification performance (Dennett, McKone, Edwards, & Susilo, 2012) and that groups with known face perception difficulties show reduced adaptive norm-based coding of facial identity (Congenital Prosopagnosia – Palermo, Rivolta, Wilson, & Jeffery, 2011; Autism Spectrum Disorders – Pellicano, Jeffery, Burr, & Rhodes, 2007).

Preschool aged children perform poorly on face identification tasks, relative to older children and adults (Bruce et al., 2000) but the source of their poor performance is controversial (Crookes and McKone, 2009, McKone et al., 2012). Early research suggested that a key mechanism of face perception, holistic coding (the integration of information across a face, including features and the spatial relations between them) was either absent or immature in young children (Carey et al., 1980, Diamond and Carey, 1977). However, more recent studies have established that holistic coding is present by age three (Macchi Cassia, Picozzi, Kuefner, Bricolo, & Turati, 2009) and is mature by five (see McKone et al., 2012, Pellicano and Rhodes, 2003). Therefore poor face identification performance in young children cannot be attributed to immaturities in holistic coding. Here we ask whether failure to use norm-based coding, another key mechanism of face perception, could explain young children’s poor performance.

Norm-based coding of facial identity has been demonstrated in adults and 7–9 year-old children using face adaptation techniques (e.g., Jeffery et al., 2011, Leopold et al., 2001, Rhodes and Jeffery, 2006, Robbins et al., 2007, Webster and MacLin, 1999). Adaptation (exposure) to faces biases the appearance of subsequently viewed faces so that they look wider after seeing narrow faces, more male after seeing female faces, and so on (e.g., Webster, Kaping, Mizokami, & Duhamel, 2004). In a norm-based account these “aftereffects” are argued to reflect opponent coding, with pairs of neural channels tuned to above- and below-average values on each dimension in face space and the norm signaled by balanced activity in both pools. The further the adaptor is from the norm the larger the effect of adaptation on an average test face because more extreme adaptors produce greater activation, and greater subsequent suppression, in the preferred channel (Rhodes et al., 2005, Robbins et al., 2007). Alternative models of face coding that do not posit a norm, e.g., exemplar coding, neurally instantiated by multichannel coding, predict a different pattern. Multichannel models predict that small increases in the distance of the adaptor from the average face will increase aftereffects but thereafter increasing adaptor distance will reduce aftereffects because these extreme adaptors will have less effect on channels that code the average face (Clifford et al., 2000, Dickinson et al., 2010). The pattern of aftereffects predicted by norm-based coding has been found for adults (Robbins et al., 2007) and 7 year-old children (Jeffery et al., 2011).

Here we tested whether norm-based coding for facial identity emerges earlier in development, by measuring 4-year-olds’ face identity aftereffects for near and far adaptors. Four year-olds are the youngest age group who do not yet to attend school but can complete adult-like adaptation tasks. The transition from preschool to primary (grade) school typically results in a substantial and sudden increase the number of individuals that children need to distinguish and remember. This sudden increase in experience may drive critical changes in face-space organization, consistent with the assumption that face-space is built via experience with faces (Johnston and Ellis, 1995, Valentine, 1991) and proposals that changing demands on face perception skills during development prompt reorganization of the face-perception system (Scherf & Scott, 2012). Certainly performance appears to improve more steeply between these ages than during later childhood (Bruce et al., 2000, Ellis, 1992, Kinnunen et al., 2012), consistent with a qualitative shift in processing between 4 and 7 years of age. While 4 year-old children possess at least a rudimentary face-space that produces distinctiveness effects (e.g., McKone & Boyer, 2006) and atypicality biases (Tanaka, Meixner, & Kantner, 2011) it has not been established that their face-space is norm-based. It remains possible that young children represent faces in an exemplar manner (neurally instantiated by multichannel coding) until sufficient demands are placed on their face-recognition system to prompt reorganization of face-space to be norm-based. Increased exposure to faces at school may also be crucial in developing a representation of the average face that is sufficiently good to function as a norm, or in developing sufficiently good representations of the dimensions on which to code individual faces as deviating from this norm.

We sought evidence of norm-based coding of face identity in 4 year-olds using the face identity aftereffect in which adaptation to an individual face (e.g. Ted) biases perception of an identity neutral average face so that it resembles an individual with characteristics opposite the adapting face (e.g., antiTed, see Fig. 1). Identity aftereffects directly tap face identification processes (Rhodes, Evangelista, & Jeffery, 2009) and have not previously been demonstrated in children. We asked whether children’s identity aftereffects would be larger for more extreme versus less extreme adaptors, consistent with norm-based coding.