Wednesday, March 07, 2007

Are Faces Special?

modified from Carmel & Bentin (2002) and Gauthier et al. (1998)

An ever-controversial topic in the field of high-level vision and object recognition is the question of whether faces have a privileged status relative to other objects, processed by a special modular region of ventral temporal cortex called the fusiform face area (Kanwisher et al., 1997; Kanwisher & Yove, 2006; McKone et al., 2007), or whether faces are just one example of a stimulus class that requires substantial expertise in order to distinguish between similar exemplars (Gauthier et al., 1999, 2000; Gauthier & Bukach, 2007). A new article in Nature Neuroscience tackles this issue and comes up with a surprising answer.

Thierry and colleagues (2007) recorded event-related potentials (ERPs), which are synchronized brain waves time-locked to the occurrence of particular stimuli or events. In particular, the N170 component is thought to be a highly specific ERP response to faces (as opposed to other objects) that shows a peak at 170 msec after stimulus presentation. This ERP was first reported by Bentin et al. (1996), although other researchers observed a related face-specific response back in the late 80's (Jeffreys, 1989). The N170 is most pronounced at the posterior temporal electrodes and is greater over the right hemisphere than the left.

Bentin et al. (1996)

In the new NN paper, Thierry et al. argue that previous studies of the N170 component did not adequately control for variability across stimulus classes, i.e., face stimuli have been much more similar to each other than the non-face stimuli.

Faces and cars with high and low interstimulus perceptual variance (ISPV).

Thierry et al. (2007)

The actual stimuli used in Experiment 1 are shown below.

From Fig 4a (Thierry et al. 2007)

It turned out that interstimulus perceptual variability alone affected the size of the N170, with low ISPV stimuli (both faces and cars) producing a larger N170 than high ISPV stimuli.

Furthermore, when controlling for ISPV, the face-specific effect for the N170 went away. The face N170 peaked earlier in time than the car N170, but they did not differ in amplitude. However, a face-dominant effect for an earlier component, the P1, emerged (which was an unexpected finding, since it suggests that category-specificity can manifest itself at 100 msec, 70 msec before the N170 component).

Hmm, interesting. Why did this happen? The authors are cautious in their interpretation:
On the other hand, the P1, a peak generally regarded as an index of lower level perceptual processing, was surprisingly unaffected by differences in ISPV, but was sensitive to object category in both Experiments 1 and 2. This should not be interpreted as evidence for absolute category selectivity because only three object categories were tested here. Indeed, faces, cars and butterflies differ in terms of overall composition and various perceptual properties such as outline, contrast, subparts and complexity, which were not manipulated here. It is therefore possible that the P1 still reflects perceptual differences between these objects; that is, that apparent category selectivity in this component is an emergent property arising from low perceptual invariants.
Thierry G, Martin CD, Downing P, Pegna AJ (2007). Controlling for interstimulus perceptual variance abolishes N170 face selectivity. Nature Neuroscience. Published online: 04 March 2007.

Establishing when and how the human brain differentiates between object categories is key to understanding visual cognition. Event-related potential (ERP) investigations have led to the consensus that faces selectively elicit a negative wave peaking 170 ms after presentation, the 'N170'. In such experiments, however, faces are nearly always presented from a full front view, whereas other stimuli are more perceptually variable, leading to uncontrolled interstimulus perceptual variance (ISPV). Here, we compared ERPs elicited by faces, cars and butterflies while—for the first time—controlling ISPV (low or high). Surprisingly, the N170 was sensitive, not to object category, but to ISPV. In addition, we found category effects independent of ISPV 70 ms earlier than has been generally reported. These results demonstrate early ERP category effects in the visual domain, call into question the face selectivity of the N170 and establish ISPV as a critical factor to control in experiments relying on multitrial averaging.

Hypothesized neural generators of the face N170.

Bentin et al. (1996)

- - - - - - - - - - - - - - - - - - - - - - - -
see also Domain specificity vs. Connectionism


Bentin S, Allison T, Puce A, Perez E, McCarthy G (1996). Electrophysiological studies of face perception in humans. J Cog Neurosci. 8:551-565.

Gauthier I, Bukach C. (2007). Should we reject the expertise hypothesis? Cognition 103:322-30.

Gauthier I, Skudlarski P, Gore JC, Anderson AW. (2000). Expertise for cars and birds recruits brain areas involved in face recognition. Nat Neurosci. 3:191-7.

Gauthier I, Tarr MJ, Anderson AW, Skudlarski P, Gore JC. (1999). Activation of the middle fusiform 'face area' increases with expertise in recognizing novel objects. Nat Neurosci. 2:568-73.

Jeffreys DA. (1989). A face responsive potential recorded from the human scalp. Exp Brain Res. 78:193–202.

Kanwisher N, McDermott J, Chun MM. (1997). The fusiform face area: a module in human extrastriate cortex specialized for face perception. J Neurosci. 17:4302-11.

Kanwisher N, Yovel G. (2006). The fusiform face area: a cortical region specialized for the perception of faces. Philos Trans R Soc Lond B Biol Sci. 361:2109-28.

McKone E, Kanwisher N, Duchaine BC. (2007). Can generic expertise explain special processing for faces? Trends Cog Sci. 11:8-15.

Subscribe to Post Comments [Atom]


At March 07, 2007 3:26 PM, Blogger Dan Dright said...

This is an excellent post. It resonates particularly well with me, as I am finishing up a study right now that essentially throws a whole lot of cold water on a rather well-entrenched idea. More at a later date on that, but as far as this is concerned:

It has never made sense to me that a particular area would be dedicated to such a narrow stimulus, in terms of adaptive advantage. I'd think that a more broadly-tuned sensor would allow for a more robust advantage over the long evolutionary haul.

It "feels" right, in a philosophical, humans are special sense that we ought to have a special face area, but scientifically, not so much.

Thanks. I am off to read the whole paper. Nice choice and treatment.

At March 08, 2007 10:58 AM, Blogger The Neurocritic said...

Thanks. Very true. Like most things, the truth is probably somewhere in the middle. Developmental data on face recognition in newborns are quite compelling in suggesting that "faces are special," e.g.,

Valenza E, Simion F, Cassia VM, Umilta C. Face preference at birth. J Exp Psychol Hum Percept Perform. 1996; 22(4):892-903.

Turati C, Simion F, Milani I, Umilta C. Newborns' preference for faces: what is crucial? Dev Psychol. 2002; 38(6):875-82.

But what this means in terms of the existence and development of a narrowly-tuned, highly specialized neural module for face processing is unclear. Recent studies have examined low-level perceptual features of faces and why infants' visual systems might be attuned to them:

de Heering A, Turati C, Rossion B, Bulf H, Goffaux V, Simion F. Newborns' face recognition is based on spatial frequencies below 0.5 cycles per degree. Cognition. 2007 Jan 17; [Epub ahead of print]

A critical question in Cognitive Science concerns how knowledge of specific domains emerges during development. Here we examined how limitations of the visual system during the first days of life may shape subsequent development of face processing abilities. By manipulating the bands of spatial frequencies of face images, we investigated what is the nature of the visual information that newborn infants rely on to perform face recognition. Newborns were able to extract from a face the visual information lying from 0 to 1 cpd (Experiment 1), but only a narrower 0–0.5 cpd spatial frequency range was successful to accomplish face recognition (Experiment 2). These results provide the first empirical support of a low spatial frequency advantage in individual face recognition at birth and suggest that early in life low-level, non-specific perceptual constraints affect the development of the face processing system.

And this one on gaze direction:

Farroni T, Menon E, Johnson MH. Factors influencing newborns' preference for faces with eye contact. J Exp Child Psychol. 2006; 95(4):298-308.

...Overall, these results further the view that relatively simple perceptual biases in newborns may be an essential foundation for later social-cognitive development.

At March 08, 2007 3:23 PM, Anonymous tirta said...

Two points:

(1) faces are special, as is often argued, because they form a class of visual object that naturally has low ISPV. When the cited study equates cars on this dimension, one can argue that it is the face perception system that is actually doing the car discrimination task, hence N170 dissapears.

(2) neural signatures (N170, FFA) form only part of the story. Faces are also special in terms of the computational mechanisms employed, which result in behavioral signatures such as the inversion effect, holistic processing, part-whole effect, composite effect, and the likes. These aren't directly observable in the brain and its neural signals.

That said, I think the question whether faces are or are not special is not really a fruitful one. Of course faces are special for us social beings, and thus there is no other visual objects comparable to them. But there is no reason to suppose that the face processing mechanisms are not used for other visual discrimination and identification purposes. The more worthy questions are those asking how is it the faces are special -- and the cited study indeed gives one more clue: natural ISPV.

At March 10, 2007 8:30 AM, Blogger The Neurocritic said...

I agree with you that the question of whether faces are or are not special is not really a fruitful one. It does make for lively debates in journals and at conferences, however, and controversy sells.

I do have one point of contention:

These [inversion effect, holistic processing, part-whole effect, composite effect] aren't directly observable in the brain and its neural signals.

Ultimately, they have to occur in the brain, and some of these behavioral effects do have corresponding neural signatures, e.g., these studies on face inversion and N170.

At May 29, 2007 1:54 PM, Anonymous Anonymous said...


Nice website. I have been working on these issues for quite a while actually:

I found your wesbite because we have submitted a reply to the Thierry et al. paper published in Nature Neuroscience and claiming that the larger N170 to faces is abolished when controlling

Basically, the whole study is crap. We proved that tens of studies had actually controlled for this factor and still got a much larger N170 component for faces.

So their claim is wrong. We also show that the reasons why the did not replicate previous effects is because they did not use the right methodology (electrodes, stimulus control).

Worse, these guys did not control at all for interstimulus variance in their own study !

there is a need to be careful with publications in journals like that: they are loking for controversial findings to make a lot of noise and get these papers cited.

The paper will soon be in the trash, just wait for a couple of weeks and check my website.

another comment on this:

These [inversion effect, holistic processing, part-whole effect, composite effect] aren't directly observable in the brain and its neural signals.

Wrong: we recently showed a nice composite face illusion (holistic processing) in the occipito-temporal cortex (Schiltz & Rossion, Neuroimage).

Inversion effects, as you say, have been reported in these areas as well as on the N170.

thanks for this website, good luck with it.


At May 29, 2007 6:44 PM, Blogger The Neurocritic said...


Thanks for your comments. I'd heard through the grapevine that 2-3 labs planned to submit letters to the editor of Nature Neurosci to complain about the paper. It was remiss of me to not include some of your objections in my original post (e.g., about the stimuli), but I'm not an expert.

I did talk to a researcher who was offended by this paragraph:

These results are difficult to relate to the intracranial recording literature. Face-selective responses from the inferior temporal lobe have been reported 200 ms after stimulus onset in individuals with intracortical electrode implants, although other reports indicate that face-related activity occurs as early as 50 ms after stimulus presentation. Comparisons between ERPs and intracranial recordings can only be tentative, as cortical activity in pharmacoresistant epileptic individuals may be affected by cognitive impairment following repeated seizures, intake of anticonvulsant medication or functional reorganization subsequent to the presence of epileptic foci.
[Thierry et al., refs removed]

I look forward to reading your reply to Thierry et al. and will mention it in my blog.

At October 19, 2007 9:41 AM, Anonymous Anonymous said...


A few months ago, you asked me to keep you updated on this issue. We have now in press (Neuroimage) a full paper deconstructing the claim of Thierry et al. that the N170 would not be larger to faces than other visual stimuli :

Rossion, B. & Jacques, C. (in press). Does physical interstimulus variance account for early electrophysiological face sensitive responses in the human brain? Ten lessons on the N170. NeuroImage.

You can get it here :

The short reply that the editors of Nature Neuroscience allowed us to write was not enough, and gave the opportunity to Thierry and colleagues to throw more confusion on this issue (as also acknowledged by the comment on this website). So we thought it deserved a full commentary, deconstructing their paper, and trying to take the positives from this unfortunate publication (i.e. what can we learn about this for N170 research, suggest some kinds of guidelines, clarify a number of theoretical and methodological points).

In a nutshell, in our paper, we :

- Explain clearly the nature of their claim, why it’s ill defined and not to be confused with the real debate about the NATURE of the larger N170 to faces.
- Explain why their claim was not really plausible for reasons that are related to EEG/ERP analysis : an increase of intertrial variance should have delayed/smeared the N170, which is due to a fixed increase of power time-locked to the stimulus onset.
- Show that Thierry et al. were wrong with respect to previous studies not controlling for the factor they mention. In fact, ironically again, one of the few studies that suffered from such limitations, is their only published study before this one.
- Explain clearly why they failed to find a N170 effect in their study (we replicate this « finding » with the wrong electrodes considered)
- Show that they did not control for the factor that was supposed to be controlled and for which they were blaming other studies.
- Provide an account for their « ISPV » finding, which merely reflects a comparison of high-quality images to low quality image sets.
- Discuss why the N170 face effect is not related to low-level visual factors, whereas the earlier P1 effect (emphasized by Thierry et al.) is likely to be related to such factors.
- Emphasize that the larger N170 to faces is an important phenomenon for researchers to understand the time-course of face processing, and that this effect is in line with a large body of data from other sources.

I hope this will set the record straight. The paper of Thierry et al. was accepted in NN it seems precisely because, done by novices in this area ( it was very controversial and would make a lot of noise (i.e. citations). I believe it is not only intellectually dishonest, but reveal a dangerous trend in some « high impact factors » journals to publish papers first and foremost because they appear novel, catchy or controversial rather than on their scientific credibility.

Perhaps at some point you will be able to reformulate the question as « why are faces special » ?

All the best

Bruno Rossion

At October 19, 2007 10:03 AM, Blogger The Neurocritic said...

Hi Bruno,

Thanks for the informative update on your latest N170 paper and the complete burial of the Thierry et al. findings published in Nature Neuroscience. I agree with you about the disturbing trend for high profile journals to publish headline-grabbing manuscripts, no matter their scientific quality (sad to say).

I will take the liberty to repost your comments in a new entry.

The Neurocritic


Post a Comment

Links to this post:

Create a Link

<< Home

eXTReMe Tracker