People with Williams syndrome process faces holistically

Abstract

This study compared the performance of 47 adolescents and adults with Williams syndrome to 39 age-matched controls on a face recognition task. Using the whole–part paradigm developed by Tanaka and his colleagues, we found that although performance overall was lower in the participants with Williams syndrome, both groups showed similar patterns of performance across the different conditions. Both groups performed significantly better in the whole-face than in the isolated-part test condition for upright faces, but not for inverted faces. The whole-face advantage only in the upright condition provides strong evidence that people with Williams syndrome encode and recognize faces holistically in the same way as normal controls, suggesting the use of similar underlying neurocognitive mechanisms. These findings contradict earlier reports in the literature that people with Williams syndrome process faces abnormally.

Introduction

Williams syndrome (WMS) is a genetically based neurodevelopmental disorder that has captured the interest of cognitive neuroscientists because of the striking and unusual pattern of cognitive abilities that is characteristic of individuals with this syndrome (e.g. Bellugi et al., 1988, Karmiloff-Smith et al., 1997, Mervis et al., 1999). Relative strengths in auditory short-term memory and language coupled with profound deficits in visual-spatial constructive skills define the cognitive phenotype (Bellugi et al., 1994, Mervis et al., 2000, Morris and Mervis, 1999). Although overall cognitive level is in the moderate range of mental retardation, within the visual-spatial domain face recognition skills appear to be relatively preserved in WMS. On standardized tests including the Benton Test of Facial Recognition and the Rivermead Behavioral Memory Test, people with WMS generally perform within the normal range, and better than mental-age-matched controls (Bellugi et al., 1988, Karmiloff-Smith, 1997, Udwin and Yule, 1991). Despite their strengths in recognizing faces, numerous researchers claim that people with WMS do not process faces normally (Deruelle et al., 1999, Elgar and Campbell, 2001, Gagliardi et al., 2003, Karmiloff-Smith, 1997, Karmiloff-Smith et al., 2002). The goal of this study was to address this apparent paradox regarding face processing skills in WMS: is it possible to be good at recognizing faces using abnormal processing mechanisms?

There is general agreement that our ability to recognize and discriminate faces so easily represents one of the most remarkable human skills and depends on specific cognitive and neural mechanisms (e.g. Farah et al., 1998, Haxby et al., 2002). To a much greater degree than most other objects, faces are encoded holistically rather than by local features (Bradshaw and Wallace, 1971, Farah, 1996). Specifically, faces are encoded in terms of a template-like representation of the whole rather than simply in terms of their component parts. It has been shown that this mechanism typically breaks down when faces are presented in inverted orientation, as inversion is thought to disrupt the expected configuration of the face (Rhodes et al., 1993, Valentine, 1988). Although children's performance on face processing tasks improves with age, there is evidence using inverted faces and other paradigms that even young children process faces holistically (Baenniger, 1994, Carey and Diamond, 1994, Freire and Lee, 2001).

Tanaka and his colleagues introduced the whole–part method, which was specifically designed to operationalize the distinction between holistic and feature-based face processing (Tanaka and Farah, 1993, Tanaka et al., 1998). In the whole–part procedure, a sample upright face stimulus is briefly exposed, followed by two types of recognition tests. In the whole-face test condition, the sample face and a foil face differing from the sample by only one feature (eyes, nose, or mouth) are presented. Alternately, in the isolated-part test condition, one feature from the sample face and the corresponding foil feature are presented. The logic of the whole–part method is that if upright faces are processed holistically, the individual features of a face will be recognized more easily in the context of the whole face in which they were learned than in isolation, but that any holistic processing advantage would not be operative for whole and part test stimuli presented in inverted orientation. Using this method, Tanaka and his colleagues have found that adults and children as young as 6 years old exhibit a whole-face test advantage for upright but not inverted faces. Joseph and Tanaka 2003 successfully adapted the whole–part method to investigate face recognition in children with autism. They found that children with autism showed a holistic processing advantage when face recognition depended on mouths, but were significantly impaired, compared to matched controls, in recognizing the eye region of the face. Their study demonstrated that the whole–part method is especially suited to identifying normal and aberrant face processing mechanisms.

As noted above, people with WMS perform within the normal range of performance for age on standardized tests of face recognition, classification, and memory (Bellugi et al., 1992, Bellugi et al., 1994, Karmiloff-Smith, 1997, Udwin and Yule, 1991). However, it has been suggested by a number of researchers that they do not process faces holistically. For example, in their recent review of the literature on face recognition in developmental disorders, Elgar and Campbell argue, “relatively good performance on face discrimination tasks in Williams syndrome is achieved via a piecemeal route, as in individuals with autism” (Elgar & Campbell, 2001, p. 709). And Karmiloff-Smith concludes that “Williams syndrome face processing is not intact and it develops differently” (Karmiloff-Smith, 1997, p. 522).

These claims of aberrant face processing in WMS are based on two studies. Karmiloff-Smith (1997) gave a group of ten adolescents and adults with WMS and ten age-matched controls the Benton test. After the test was administered in the standard way, she then presented some of the test faces in inverted orientation, and questioned the participants about how they remembered the faces. In a follow-up experiment with the same participants, Karmiloff-Smith administered a computerized face processing test battery developed by Campbell and her colleagues for children aged 4–10 years. Although no data or statistics were presented, Karmiloff-Smith reported that the participants with WMS were less disturbed than controls when faces were presented in inverted orientation and that they pointed to features when asked how they remembered a face. Karmiloff-Smith also reported that item analyses for the second experiment suggested that the participants with WMS were at chance on test items that required configural processing. It is difficult to evaluate this study since few methodological details were presented, and the informal nature of the procedures used and analyses conducted do not allow one to objectively evaluate the conclusions. Karmiloff-Smith (1997) acknowledged the preliminary nature of her study, which included a very small number of participants with WMS who varied widely in both age and IQ.

A second experimental study on face recognition in WMS was conducted by Deruelle et al. (1999; Experiment 2). They compared 12 participants with WMS between the ages of 7 and 23 to two control groups, one matched on age and the other matched on mental age, on a same–different discrimination task in which faces and houses were presented in upright or inverted orientation. All the groups made relatively more errors on the inverted faces than on the houses, and for the control groups this pattern was confirmed by a significant interaction between stimulus type and orientation. This interaction did not reach statistical significance for the WMS participants. However, the data for the WMS group (Deruelle et al., 1999, Table 2, p. 287) show that the mean difference in errors on the inverted faces compared to upright faces (1.23) was higher than on inverted houses (0.43), and this difference on faces was actually higher for the WMS group than the age-matched controls (0.83). Thus, the data obtained from this study were in the direction predicted if the participants were using a holistic processing mechanism. The lack of a statistically significant effect is not surprising given the small number of participants, who varied widely in age and ability level. It is doubtful that there was sufficient statistical power to detect a significant effect from this group of participants. Nevertheless, the authors conclude from this experiment that their participants with WMS “are incapable of encoding faces in terms of configurational information and encode both upright and inverted faces through local characteristics” (Deruelle et al., 1999, p. 288). Taken together, neither of these studies on face recognition in WMS provide compelling support for the view that people with WMS process faces in an aberrant or atypical way.

In light of the evidence from standardized data, which suggests that face processing is a relative strength in WMS, we were surprised by the claims that in people with WMS face recognition depends on abnormal mechanisms that process featural information (Elgar and Campbell, 2001, Gagliardi et al., 2003, Karmiloff-Smith et al., 2002). Our study was designed to explore the question of how people with WMS recognize faces using a more rigorous and sensitive methodology. We address the methodological weaknesses of earlier work by directly testing whether people with WMS process faces relying primarily on holistic or featural mechanisms. We included a relatively large group of adolescents and adults with WMS, and used the whole–part method developed by Tanaka and his colleagues (Tanaka & Farah, 1993). In this paradigm, accuracy differences between the whole- and part-face conditions can be unambiguously interpreted as evidence for the use of holistic or featural processing mechanisms in face recognition. We used the same stimuli and similar procedures as those developed by Joseph and Tanaka 2003 because they had been used successfully to investigate face processing in individuals with autism, a different developmental disorder with which WMS has been compared (cf. Elgar and Campbell, 2001, Karmiloff-Smith et al., 1995).