Mar 252014

Rezlescu, Barton, Pitcher & Duchaine  (RBPD) recently published an article in PNAS, presenting evidence that two patients with acquired prosopagnosia were able to become “Greeble experts”. RBPD argue that this refutes a fundamental prediction of the expertise hypothesis.

Although it is always interesting to hear about what specific patient can and cannot do, I would argue we cannot learn much that is compelling from this evidence. I was asked to comment on the article by Andreas von Bubnoff, a journalist for Nature News (see his coverage), but since my comments will likely be greatly abbreviated, I thought I would elaborate here.

Not an actual prediction.
RBPD argue that a “fundamental prediction” of the expertise hypothesis is that people with prosopagnosia should also be impaired at acquiring expertise for other objects. Although at first glance this may sound right, it is incorrect. The expertise account proposes that the reason we observe specialization for faces in behavior and in the brain is due to a specific kind of experience with faces. Or, to state it more specifically: if you have to individuate non-face objects that are visually similar and you use the same strategy most people use to learn faces, then the same kind of specialization may occur for non-face objects.

The qualifications are important, because expertise is not ONE thing. Expertise is just becoming really good in a given domain that was difficult to start with. My colleagues and I have published a considerable amount of work in which we show that how you become an expert constrains the kind of expert you become (e.g., McGugin et al., 2011; Wong et al., 2012; Wong et al., 2009a). Clearly one can become an expert in non-face domains, like cars, and recruit the face area (McGugin et al., 2012; in press; Xu, 2005) and in other domains, like print or musical notation, and engage other brain areas (e.g., Wong et al., 2009b; Wong et al., 2012). How you become an expert is likely to depend on a number of factors, including the information available in the stimuli, the task you are trying to perform, whatever biases you may bring to the situation, and what strategies are available to you.

This last point about strategies is really important to interpret RBPD’s new article. They used a procedure that was designed to elicit a specific strategy—holistic processing— in normal subjects, but their argument depends on the assumption that this is the only strategy one could use. We may be responsible for this misconception in not being sufficiently clear in our early work. However, our work with prosopagnosic patient LR (Bukach et al., 2012), who also became a Greeble expert like the patients in RBPD’s article, is very clear on this point. Because each Greeble has unique parts, there is no principled reason one has to use holistic processing in a Greeble training study. The fact that normal subjects seem to adopt a holistic processing strategy does not speak to what patients would do if integrating across parts is not easy for them. By testing controls and patient LR’s processing of Greebles that did not have unique parts after training, we were able to show that similar performance can come from different strategies. This was not verified by RBPD about their patients, so we cannot know whether their processing of Greebles is holistic and normal, just that their performance individuating Greebles is as good as controls.

Suggesting that Greebles and faces are somehow equated.
The critical evidence in RBPD’s article is that the patients can learn Greebles and not faces. Thus, whether the two tasks are equated is important. In their abstract, they called the face procedure “matched” to the Greeble task. But their description makes it clear that they selected faces to be highly similar, whereas they did not do the same for Greebles. They used 20 Greebles during training, 4 Greebles from each family (apparently 2 from each gender, defined by all parts pointing up or down), which means that each Greeble in the experiment only had a single similar foil (another Greeble with the same body shape and direction of parts). The original Greeble set used in the first Greeble training study (Gauthier & Tarr, 1997) and in the work with patient LR (Bukach et al., 2012) used a more difficult set with 30 Greebles (2 similar foils per object). The faces RBPD used are more homogenous than the Greebles on any metric we can think of (this is obvious from their figure).

Imagine that the deficit in prosopagnosia has to do with a difficulty integrating across parts of an object. Integrating across parts will be particularly useful when the local featural information is not very diagnostic. Nobody predicts that RBPD’s patients should have difficulty processing the overall body shape of these objects or the direction of their parts. Thus, the Greeble task requires RBPD’s patients to learn to distinguish each Greeble from a single foil, and this can be done by looking at a single part, say the top right appendage (because all Greebles have unique parts). In contrast, the Face task could be performed based on single parts but each face has 19 other similar foils). These two tasks are not matched in difficulty. This is apparent when you consider that the control subjects, who have had a lifetime of experience with faces and no experience with Greebles, did equally well at learning greebles as they did faces.

These properties should not be taken as limitations of the Greebles or the Greeble training procedure. These were designed on purpose to be easier than faces, as a much simplified reduced “world” in which a limited amount of training might be able to produce holistic processing and FFA activity. The training itself is a manipulation and was not designed as a test of any kind. Being able to learn the Greebles was never supposed to be a test of whether someone is using face-like processing (although how a patient may learn them might be telling, Bukach et al., 2012).


Bukach, C.M., Gauthier, I., Tarr, M.J., Kadlec, H., Barth, S., Ryan, E., Turpin, J.  & Bub, D. (2012). Does acquisition of expertise in prosopagnosia rule of out a domain-general deficit?, Neuropsychologia, 50(2): 289-304.

Gauthier, I., & Tarr, M.J. (1997). Becoming a “Greeble” expert: Exploring mechanisms for face recognition, Vision Research, 37(12), 1673-1682.

McGugin, R.W., Van Gulick, A.E., Tamber-Rosenau, B.J., Ross, D.A. & Gauthier, I. (in press). Expertise effects in face selective areas are robust to clutter and diverted attention but not to competition. Cerebral Cortex.

McGugin, R.W., Gatenby, Gore, J.C.,Gauthier, I. (2012). High-resolution imaging of expertise reveals reliable object selectivity in the FFA related to perceptual performance. Proceedings of the National Academy of Sciences, 109(42), 17063-17068.

McGugin, R.W., Tanaka, J.W., Lebrecht, S., Tarr, M.J., & Gauthier, I. (2011). Race-Specific perceptual discrimination improvement following short individuation training with faces. Cognitive Science, 35(2):330-47.

Wong, A. C.-N., Palmeri, T. J., Gauthier, I. (2009a). Conditions for face-like expertise with objects: Becoming a Ziggerin expert – but which type? Psychological Science. 20(9), 1108-17.

Wong, A.C.-N., Jobard, G., James, K.H., James, T.W., & Gauthier, I. (2009b). Expertise with characters in alphabetic and non-alphabetic writing systems engage overlapping occipito-temporal areas. Cognitive Neuropsychology, 26(1): 111-127.

Wong, Y.K., Folstein, J.R., & Gauthier, I. (2012). The nature of experience determines object representations in the visual system. Journal of Experimental Psychology: General, 141(4):662-98.

Wong, Y.K. & Gauthier, I. (2012). Music-reading expertise alters visual spatial resolution for musical notation, Psychonomic Bulletin and Review, 19(4):594-600.

Xu, Y. (2005). Revisiting the role of the fusiform face area in visual expertise.Cerebral Cortex15(8), 1234-1242.