Elsevier

Cognition

Volume 103, Issue 1, April 2007, Pages 34-79
Cognition

No face-like processing for objects-of-expertise in three behavioural tasks

https://doi.org/10.1016/j.cognition.2006.02.008Get rights and content

Abstract

In the debate between expertise and domain-specific explanations of “special” processing for faces, a common belief is that behavioural studies support the expertise hypothesis. The present article refutes this view, via a combination of new data and review. We tested dog experts with confirmed good individuation of exemplars of their breed-of-expertise. In all experiments, standard results were confirmed for faces. However, dog experts showed no face-like processing for dogs on three behavioural tasks (inversion; the composite paradigm; and sensitivity to contrast reversal). The lack of holistic/configural processing, indicated in the first two of these tests, is shown by review to be consistent rather than inconsistent with previous studies of objects-of-expertise.

Introduction

Recognising faces is something people are very good at. It is well established that the process used to recognise another person’s face differs from the process used to recognise nonface objects, at least when subjects have no specific expertise in the relevant object domain (e.g., Farah, 1996, Grill-Spector et al., 2004, Moscovitch et al., 1997, Tanaka and Farah, 1993, Yin, 1969). Two main theories have emerged to explain this difference.

The domain-specificity hypothesis (e.g., Kanwisher, 2000, McKone and Kanwisher, 2005, Rhodes et al., 2004, Yin, 1969) suggests that the ‘special’ processing used for faces occurs only for faces. This hypothesis does not, per se, propose a mechanism for the origin of the special processing. It is possible, however, that special processing for faces has an innate component (de Haan et al., 2002, Morton and Johnson, 1991) and/or that it is necessary to obtain appropriate face experience at a particular time in development (e.g., a sensitive/critical period during infancy for the development of normal face processing, Le Grand et al., 2001, Le Grand et al., 2003, Le Grand et al., 2004).

In contrast, the expertise hypothesis (e.g., Carey, 1992, Diamond and Carey, 1986, Gauthier and Tarr, 1997, Meadows, 1974) suggests that ‘special’ processing for faces is a potentially generic ability that arises for faces because of substantial experience in individual-level discrimination; this predicts that the special processing can also arise for any other object class through the same mechanism (e.g., in expert dog show judges looking at dogs from their breed-of-expertise). An important assumption of the expertise hypothesis is that the period of life when this experience is obtained is irrelevant: object expertise can be developed entirely as an adult, and the predictor of processing style is merely the amount of practice. This assumption was made explicit by the original proposers of the hypothesis (e.g., Carey, 1992, Diamond and Carey, 1986), and has remained implicit in subsequent research, which tests for face-like processing in subjects who have in many cases obtained their expertise as adults or teenagers (e.g., Gauthier et al., 2000, Gauthier and Tarr, 1997, Grill-Spector et al., 2004, Xu, 2005).

In terms of empirical tests of the different hypotheses, key neuropsychological results support domain specificity. In novices (i.e., subjects with no particular expertise in the object domain tested), there have been cases where face recognition is spared but object recognition is damaged (e.g., Moscovitch et al., 1997), and vice versa (e.g., De Renzi, 1986, McNeil and Warrington, 1993), showing a double dissociation between face and object recognition. In experts, this has been extended to a double dissociation between face processing and processing of objects-of-expertise. That is, individual patients have shown intact face processing with impaired recognition of objects-of-expertise (toy soldiers; Moscovitch et al., 1997), and also impaired face processing with intact processing of objects-of-expertise (brass instruments, Dixon, Desmarais, Gojmerac, Schweizer, & Bub, 2002; sheep, McNeil & Warrington, 1993). It is also possible for an object agnosic and prosopagnosic patient to become an expert with greebles despite not having learned expertise with faces, either after several years of exposure post-injury (Behrmann, Marotta, Gauthier, Tarr, & McKeeff, 2005), or after a lifetime of exposure in a developmental case (Duchaine, Dingle, Butterworth, & Nakayama, 2004). These results, particularly the double dissociation, should not be possible according to the expertise hypothesis, which would have predicted that the normal vs. impaired status of expert object processing should track that for face recognition, not dissociate from it.

However, studies using functional magnetic resonance imaging (fMRI) have produced less clear-cut results. In novices, the fusiform face area (FFA) shows a BOLD response approximately 2–3 times as strong to individual faces as to individual members of other object classes (Kanwisher et al., 1997, McCarthy et al., 1997, Tong et al., 2000). In testing expertise effects, two results in the literature show no increase in activation in the FFA for objects in experts compared to novices (cars, Grill-Spector et al., 2004, Gauthier et al., 2000). Other findings, however, have indicated higher FFA activation in experts than novices (birds, Gauthier et al., 2000, Xu, 2005, Rhodes et al., 2004), even if the level remains lower than that normally obtained for faces.

How should these results be understood? One interpretation is that, despite the neuropsychological evidence, the fMRI results should be taken as supporting the expertise hypothesis. An alternative view is that the small BOLD increase in fMRI studies represents something other than the brain carrying out the same type of computations for faces and objects-of-expertise. There are at least two reasonable candidates for an alternative explanation (e.g., Xu, 2005). One is that the increased signal is due to greater attention to the objects by the experts than the novices: the FFA produces some response to objects even in novices (e.g., Haxby et al., 2001), and attention is known to increase BOLD response (Wojciulik, Kanwisher, & Driver, 1998). According to this explanation, the attempts in most studies to equate level of attention to the objects between experts and novices, such as using one-back-matching tasks to force processing at the individual-exemplar level, are not fully successful. A second important consideration is that fMRI has limited spatial resolution, and the measured FFA may be combining the outputs of smaller areas with differing functions (as demonstrated in Schwarzlose, Baker, & Kanwisher, 2005). According to this interpretation, objects-of-expertise could appear to be activating the FFA when, in fact, greater resolution would reveal closely located but separate areas for faces and for objects-of-expertise.

The present study addresses whether face-like computational processing emerges for objects-of-expertise. Given the difficulties of interpreting the fMRI data, this is a critical question in evaluating the expertise hypothesis.

The nature of the computations the brain performs with faces and objects-of-expertise is most cogently addressed by behavioural experiments. Currently, the common position in the literature is a presumption that behavioural studies support the expertise hypothesis. In the present article, we refute this position. We do so partly based on new results from three experiments conducted in dog experts, all of which show no evidence of face-like processing of dogs. Equally important, we include a careful review of previous data. This review shows that previous findings on standard tasks do not, in fact, support the expertise hypothesis.

Theoretically, the primary issue we are concerned with is whether experts develop the configural/holistic style of processing known to be ‘special’ to upright faces, or whether they continue to use the part-based processing style used for inverted faces and objects in novices. The exact definition of the terms configural and part-based are a matter of ongoing debate (e.g., see Maurer et al., 2002, Peterson and Rhodes, 2003, Tanaka and Sengco, 1997). Here, we take configural (or holistic) processing to mean strong integration of information from across multiple regions of the face at once, which occurs at a perceptual level, and includes processing of detailed spatial relational information between face features (e.g., nose-mouth distance) or perhaps between lower-order ‘features’ (e.g., distance from corner of eye to edge of nostril). We take part-based processing to mean independent local processing of individual features, where features are defined by obvious contour boundaries in the face.

Empirically, a number of standard tasks have been associated with configural processing. The details of these tasks, and their exact theoretical relationship to configural processing, are discussed in later sections of this article. Briefly, the phenomena we consider are: the disproportionately large inversion effect on memory for faces (Yin, 1969; see Experiment 1 for both our results and the review of previous findings), the Tanaka and Farah (1993) part-whole effect (see introduction to Experiment 2 for review), and the Young, Hellawell, and Hay (1987) composite effect (see Experiment 2 for review and new results).

We also consider the sensitivity of face and object processing to contrast reversal (i.e., making the face look like a photographic negative; Experiment 3). We do so because of evidence that, in nonexperts at least, shape-from-shading information is particularly important in identifying faces (Galper, 1970, Kemp et al., 1990, Subramaniam and Biederman, 1997). In addition, a previous study (Gauthier, Williams, Tarr, & Tanaka, 1998) has considered contrast reversal relevant to testing the expertise hypothesis.

According to the expertise hypothesis, face-like processing is most likely to emerge for objects when (a) the processing requirements of the task are the same for faces and objects, (b) the subjects are sufficiently expert in the object domain, and (c) the object class is matched to faces in important ways. In our own experiments, we satisfied the first of these criteria by using tasks that require discrimination of individual exemplars for both faces (e.g., Mary vs. Jane) and the selected object class (e.g., Dog 1 vs. Dog 2).

In judging what comprises sufficient expertise, we note that previous studies have used both experiment-trained subjects with approximately 8–10 h of training on individual members of the object class (the greeble studies, e.g., Gauthier & Tarr, 1997), and real world experts with years of experience (e.g., Diamond & Carey, 1986). Clearly, testing real world experts provides the best chance of identifying expertise effects. It is also important to provide experimental evidence that the expert subjects are, indeed, good at recognising their objects-of-expertise.

In terms of the type of expert, we wished to select experts who naturally discriminate their objects-of-expertise as individuals (Diamond & Carey, 1986). Not all experts do this; for example, bird experts usually discriminate birds at the species level, and car experts at the level of the model and year. Recent authors have suggested that the predictions of the expertise hypothesis should also apply to these experts who employ only subordinate-level categorisation (e.g., Gauthier et al., 2000, Xu, 2005). However, we chose experts who identify their objects-of-expertise at the individual level because these provide the closest match to the type of expertise people have with faces (and incidentally can provide the strongest evidence against the expertise hypothesis if null results are obtained).

In choosing an object class that is ‘matched’ to faces,1 Diamond and Carey (1986) suggested that two conditions are particularly important. First, the objects must share the same first order configuration; for faces, this comprises two eyes above a nose above a mouth. Second, individual exemplars of the class must differ only in second-order ways from this common configuration; faces, for example, differ in the exact distance between the eyes, or exact nose shape.

To these minimum criteria, our own experiments added the following properties. We wanted the objects to be natural rather than artificial, with individual members of the class differing because of genetic variability. Members of artificial object classes (e.g., greebles) tend to vary on only a small number of dimensions, and/or have differences located in a small number of discrete locations. Natural objects are like faces, however, in that individual exemplars differ on many dimensions, with differences spread over the extent of the whole object. Finally, to examine contrast reversal effects, we wanted objects for which, like faces, part boundaries are often gradual, and shape-from-shading information is potentially useful.

These criteria are met by relatively few classes of objects, most obviously by animals (e.g., horses, cats, and dogs). For practical reasons, we chose to test labrador dogs. Labradors are one of the most popular dog breeds in Australia, making it feasible to obtain experts. Labradors are also short-haired, avoiding the problem of clipping differences between individuals. Unlike dogs such as beagles, they have no strong colour boundaries, avoiding the problem of artificially encouraging a part-based processing strategy. Dog photographs are available in a standard ‘show dog’ pose (side on, usually with tail out), avoiding problems of substantial image differences between pictures of different individuals (a problem with cats); importantly dog-show judges and breeders are very familiar with pictures in this standard view (i.e., the side-on view we selected matches the particular expertise of the subjects). Finally, the breed standard for Labradors in Australia requires not only reference to local features of the dog (e.g., well-arched toes), but also to more global properties (e.g., strongly built, short coupled, very active, broad and deep through chest and ribs, broad and strong over loins and hindquarters), and our experts reported that they ‘look at the whole dog’ when judging quality, rather than merely checking a list of local features.

Our experiments were designed to test whether labrador experts demonstrated face-like processing for their objects-of-expertise, on three behavioural tasks (inversion, composite task, and contrast reversal). Two of these tasks also provided psychophysical data relevant to demonstrating expertise in our subjects. We now extract and summarise these data; full Methods for the tasks are described in Experiments 1 and 3.

Our subjects were 15 labrador show judges, breeders and trainers (aged 41–76 years). They had 5–42 years of experience with labradors, with a mean of 23.1 years. Twelve of the subjects had greater than 10 years of experience. This compares very favourably with previous studies testing real-world experts (e.g., Diamond and Carey, 1986, Gauthier et al., 2000, Rhodes et al., 2004, Tanaka and Curran, 2001, Tanaka and Taylor, 1991, Tanaka and Gauthier, 1997). We also derived an estimate of total ‘lifetime’ number of labradors seen, using a time line questionnaire method to assist retrieval from different stages of life (as developed and validated in the drug use field; Anglin, Hser, & Chou, 1993). Scores ranged between 43 individual labradors (in a guide dog trainer who had regular contact with each of these dogs over a year or more) and approximately 11,635 individual labradors; the median was 2716 and the mean was 4553.

During testing, both anecdotal evidence and more formal evidence of expertise emerged. In an example of the former, subjects could sometimes name the breeder of a particular dog. Experts also commonly distinguished the dogs’ country of origin. We had included Australian, British and American dogs, with experts finding the local dogs most attractive (indeed, a common comment was “Those American dogs are ugly!”).

In terms of formal testing, only certain conditions from our experiments are relevant to assessing expertise. These conditions are those in which the dog stimuli were presented in their usual form – that is, upright, complete dogs, in normal contrast – as this is the appearance with which experts would have had the opportunity to develop expertise. Two experiments contained relevant conditions. In our recognition memory task, subjects learned one set of dogs, then later discriminated between the studied and the unstudied dog in a series of pairs (see Experiment 1). Table 1 shows percent correct for upright faces and dogs. Experts with more than 10 years’ experience (N = 12) were 11% more accurate for dogs than were control novices matched for age, sex and education (t (11) = 1.97, p < .04, for 1-tailed test of experts better than novices) and this did not reflect a speed-accuracy tradeoff as experts tended to be faster than age matched novices (see Table 4). In contrast, there was no difference between experts and matched novices for faces (1% difference, t < 1). More powerful data (more trials per subject) came from a same-different identity task with simultaneous presentation, where two images were briefly presented side-by-side showing either the same dog or two different dogs (see Experiment 3). Table 2 shows percent correct for upright faces and dogs (in normal contrast). In this case, data for experts and novices cannot be compared directly, because the older experts (age 41–72) were given a longer stimulus presentation time than the young adult novices (age 18–30 years), and because age-sex-education matched novices were not tested in this experiment. However, performance for faces and dogs can be compared within each subject group, because the same presentation time was used for both stimulus classes. The results show that experts (N = 15) were as good with dogs as they were with faces (0% difference, t < 1), while young adult novices were 11% worse with dogs than faces, t (19) = 4.87, p < .001; there was also a highly significant interaction between stimulus class and expertise, F (1,33) = 24.08, p < .001.Our finding that experts were as good with dogs as with faces in simultaneous matching is particularly strong: this argues that 5–42 years of experience has produced a level of perceptual expertise with dogs that is equivalent to that obtained with rather more years of experience with faces. (The same result was not obtained in recognition memory, but this task combines perceptual ability with memory factors.)

Overall, there is considerable evidence that the experts tested in our experiments were sufficiently expert to provide a valid test of the expertise hypothesis. According to the hypothesis, these particular subjects would clearly be predicted to show evidence of face-like processing for labradors.

Section snippets

Experiment 1 – Inversion effects on recognition memory

In object novices, all stimuli are remembered better upright than inverted, but this inversion effect is much larger for faces (usually 15–25%) than for objects (usually 0–8%; see Table 3). As pointed out by Valentine (1988), the disproportionate inversion effect for faces does not per se indicate configural processing for upright faces. In principle, the better performance with upright faces could arise from better part-based processing. For faces, the evidence showing that the large inversion

Experiment 2 – Configural processing and the composite effect

Experiment 1 failed to show face-like processing for objects-of-expertise on a recognition memory task. Note again that, even had it done so, this would not demonstrate directly that configural processing had occurred for objects of expertise: improvements with experience for upright could come from other sources (e.g., improved part-based processing).

In demonstrating configural processing for faces, two paradigms have become standard (e.g., see review by Maurer et al., 2002). In Tanaka and

Experiment 3 – Contrast reversal

The third experiment tested face and dog recognition for the effects of contrast reversal, that is, swapping the luminance values of a picture so that it looks like a photographic negative. Contrast reversal has been shown to have a very adverse effect on face recognition in both identification and matching tasks (Bruce and Langton, 1994, Galper, 1970, Johnston et al., 1992, Kemp et al., 1990, Kemp et al., 1996, Phillips, 1972). However, for objects, in novices, the one published study found no

General discussion

In dog experts, our results showed no evidence of face-like processing for objects-of-expertise on three behavioural tasks. This included only weak and nonsignificant sensitivity to inversion in recognition memory (Experiment 1) and no sensitivity to inversion in simultaneous matching (Experiment 3), no Young et al. (1987) composite effect for dogs even when upright (Experiment 2), and no sensitivity to contrast reversal (Experiment 3). Further, experts showed no increases relative to novices,

Conclusion

In agreement with neuropsychological evidence of a double dissociation in cortical location between faces and objects-of-expertise, behavioural studies investigating style of computational processing do not support the predictions of the expertise hypothesis. Investigations of configural/holistic processing are, instead, supportive of true domain-specificity for faces. The focus of future research should be on why this arises; that is, the extent to which ‘special’ processing for faces is based

Acknowledgements

Research supported by Australian Research Council grants DP0208630 and DP0450636 to EM. We are very grateful to the dog experts for agreeing to participate, and for their assistance in contacting additional experts. We thank Mary Conheady of the Victoria Labrador Retrievers Association for her assistance in finding pictures of dogs, and Jacqui Brewer for assistance in locating and testing age-sex-education matched controls in Experiment 1.

References (75)

  • Q.C. Vuong et al.

    The role of surface pigmentation for recognition revealed by contrast reversal in faces and Greebles

    Vision Research

    (2005)
  • D.D. Wheeler

    Processes in word recognition

    Cognitive Psychology

    (1970)
  • Y. Xu et al.

    The M170 is selective for faces, not for expertise

    Neuropsychologia

    (2005)
  • A.D. Anglin et al.

    Reliability and validity of retrospective behavioural self-report by narcotics addicts

    Evaluation Review

    (1993)
  • M. Behrmann et al.

    Behavioral change and its neural correlates in visual agnosia after expertise training

    Journal of Cognitive Neuroscience

    (2005)
  • I. Biederman

    Recognition-by-components: A theory of human image understanding

    Psychological Review

    (1987)
  • V. Bruce et al.

    The use of pigmentation and shading information in recognizing the sex and identities of faces

    Perception

    (1994)
  • R. Bruyer et al.

    Expertise in person recognition

    Bulletin of the Psychonomic Society

    (1992)
  • Carey, S. (1992). Becoming a face expert. Philosophical Transcripts of the Royal Society London: B 335, 95–102;...
  • S. Carey et al.

    Are faces perceived as configurations more by adults than by children?

    Visual Cognition

    (1994)
  • J. Cohen et al.

    PsyScope: An interactive graphic system for designing and controlling experiments in the psychology laboratory using Macintosh computers

    Behavior Research Methods, Instruments and Computers

    (1993)
  • M. de Haan et al.

    Developing a brain specialized for face perception: A converging methods approach

    Developmental Psychobiology

    (2002)
  • E. De Renzi

    Current issues on prosopagnosia

  • R. Diamond et al.

    Why faces are and are not special: An effect of expertise

    Journal of Experimental Psychology: General

    (1986)
  • M.J. Dixon et al.

    The role of premorbid expertise on object identification in a patient with category-specific visual agnosia

    Cognitive Neuropsychology

    (2002)
  • N. Donnelly et al.

    The mental representations of faces and houses: Issues concerning parts and wholes

    Visual Cognition

    (1999)
  • P.S. Fastenau et al.

    Age differences in retrieval: Further support for the resource-reduction hypothesis

    Psychology and Aging

    (1996)
  • R.E. Galper

    Recognition of faces in photographic negative

    Psychonomic Science

    (1970)
  • I. Gauthier et al.

    Perceptual interference supports a non-modular account of face processing

    Nature Neuroscience

    (2003)
  • I. Gauthier et al.

    Expertise for cars and birds recruits brain areas involved in face recognition

    Nature Neuroscience

    (2000)
  • I. Gauthier et al.

    Unravelling mechanisms for expert object recognition: Bridging brain activity and behavior

    Journal of Experimental Psychology: Human Perception Performance

    (2002)
  • K. Grill-Spector et al.

    Visual recognition: As soon as you know it is there, you know what it is

    Psychological Science

    (2005)
  • K. Grill-Spector et al.

    The fusiform face area subserves face perception, not generic within-category identification

    Nature Neuroscience

    (2004)
  • J.V. Haxby et al.

    Distributed and overlapping representations of faces and objects in ventral temporal cortex

    Science

    (2001)
  • C. Hertzog et al.

    Metacognition in adulthood and old age

  • G.J. Hole

    Configurational factors in the perception of unfamiliar faces

    Perception

    (1994)
  • G.J. Hole et al.

    Evidence for holistic processing of faces viewed as photographic negatives

    Perception

    (1999)
  • Cited by (265)

    • Face specific inversion effects provide evidence for two subtypes of developmental prosopagnosia

      2022, Neuropsychologia
      Citation Excerpt :

      Participants were instructed to respond as quickly and as accurately as possible. Precise details of stimulus creation can be found in Robbins and McKone (2007), but in brief, the top and bottom halves of the stimuli were either aligned so that the edges joined up relatively neatly (aligned condition); or the bottom half of the stimulus was offset by approximately one quarter the width of the stimulus (misaligned condition). The original set of stimuli contained 30 same-identity pairings and 180 different-identity pairings each for dogs and faces.

    • “Super-recognizers” and the Legal System

      2024, The Impact of Technology on the Criminal Justice System: A Psychological Overview
    View all citing articles on Scopus

    This manuscript was accepted under the editorship of Jacques Mehler.

    View full text