Tuesday, February 10, 2009

Eight problems for the mirror neuron theory of action understanding

My critical review of the mirror neuron theory of action understanding is now available in the early access section of the Journal of Cognitive Neuroscience website. The basic conclusion is that there is little or no evidence to support the mirror neuron=action understanding hypothesis and instead there is substantial evidence against it. I would love to hear your thoughts...

16 comments:

David Poeppel said...

It's great to see this out! This is a terrific paper, and we should send it everywhere. Like the New York Times, say ... :-)

Arnie Cox said...

Helpful essay. I'm a music theorist trying to understand the relation between action, perception, imagery, and conceptualization. I wonder if Mahon and Caramazza's idea of motor imagery "coloring" and "enriching" understanding colors our understanding of understanding. I'm imagining something of a continuum of relevant motor experience informing perception, imagery, and conceptualization, which I think is not too far off from your remarks regarding saxophones. But Mahon and Caramazza's characterization (and yours, if I'm reading this fairly) that any motor imagery is added on to an already-abstracted understanding, thus privileging abstract knowledge. Is this consistent with human development, where abstract knowledge follows concrete (motor) experience?

Negri et al. (2007) do not measure brain activity, correct? So the dissociation is in overt mimetic behavior but, leaving mirror neurons aside, this does not necessarily indicate anything about possible covert motor imagery (such as the activation of Stp in your 2003 article).

Thanks for any thoughts you might care to share!

Arnie Cox said...

(Apologies to all for my unedited grammar!)

Greg Hickok said...

Hi Arnie,

Thanks for your comment. I'm not an expert on developmental arguments regarding the relation between abstract and motor experience, but with respect to Mahon and Caramazza's idea as well as my related comments, I don't think it is relevant.

Here is the point I'd like to make. You don't need motor experience to understand actions. We understand flying actions, even though none of us can fly. So we could probably understand walking actions even if we never walked. The question then becomes what does one's personal motor experience with an action add to the concept of that action. One idea, noted by Mahon and Caramazza was that it colors and enriches the core concept. Having experience with walking, upon the observation of walking, we may be able to make several inferences that a non-walker may not such as what the leg muscles may feel like, whether the walker is likely to fall given the terrain, or whatever. None of this is critical, however, to a basic understanding of the action WALK.

Negri et al. do not measure brain activity, you are correct. They demonstrate a double-dissociation in the ability to perform and understand actions, which is strong evidence for the idea that action execution and action understanding do not rely on identical brain circuits.

Arnie Cox said...

Thanks - I think I understand your position better now. I have in mind kinds, degrees, and/or depth of understanding, and accordingly I would be inclined to say that I have a relatively empty understanding of flying compared to what I would have if I had ever flown. I'm also not sure what it is that we understand when we understand of flying.

Is it analogous to saying that I do not need to have the experience of smelling and tasting escargot in order to understand 'escargot'?

I think an important issue here, then, is how one defines 'understanding'. What one person calls a 'core' or 'basic' understanding, someone else might call a 'shell' or 'distal' understanding.

Shaun Gallagher's "How the Body Shapes the Mind" is an interesting source on the ontogenetic development of thought.

Your interpretation of Negri et al. sounds reasonable to me, and I think their finding is fascinating, but still it doesn't indicate that motor imagery did not occur. Has this been tested? It would be interesting if it could be shown that action recognition could occur in the absence of any motor-related imagery - but perhaps that would relate to a different and stronger claim than you intend.

Thanks.

Greg Hickok said...

Arnie,

You've hit on the main issue in the mirror neuron/embodied cognition approach to action semantics: it's all going to boil down to the question, What is a concept?

Our mini-discussion is a perfect example. I say, you can understanding FLY without ever flying, therefore motor involvement in action concepts is just motor priming or at best "augmentation" of the core concept. You say, motor involvement is critical in understanding action concepts, therefore we don't really understand FLY?

With this kind of disagreement about the nature of action concepts, it is very hard (impossible maybe!) to test these ideas empirically. And as we all know, an untestable hypothesis is not scientifically legitimate. So we have a problem here that can only really be solved by deciding ahead of time what is going to count as a concept. If we can't decide, then we don't have much of a game and we might as well spend our efforts on a testible theory -- at least this will save us a lot of scanner time. ;-)

Re: Negri... so your idea is that patients with action execution deficits can nonetheless understand action because they have retained an ability for motor-related imagery. This is an interesting question -- one that has been proposed more or less in the speech literature -- and may be worth a dedicated post. Stay tuned.

Patti said...

Hi Greg,

I greatly enjoyed reading this paper, it was about time a paper like this appeared, especially after 10 years of people explaining everything in terms of mirror neurons as the solution to everything. However, I find that your review of the evidence against the motor theory of speech perception seems rather harsh. You state that the motor theory predicts that damage to the structures supporting production should damage perception and vice versa. And as no compelling evidence to this effect was found, you conclude that the motor theory is not true (in its strong form). But most of the lesion studies involve patiens with unilateral lesions, and it could very well be the case that only bilateral lesions lead to deficits in perception or production. (Note that I'm making the same argument as Rizzolatti & Craighero (2004) that you discuss on page 4.) It's been argued on many occasions that lesion studies cannot provide a conclusive answer for this type of argumentation. Maybe TMS is the answer here? Also, Galantucci et al also argue that, while the strongest claims of the motor theory are obviously false, it is undeniably the case that speech perception and production are linked. (Even though you state this as well at the bottom of page 11 right column). I don't think that there's any researcher in the speech field who still supports the motor theory in its strong form. Anyway, I can see your point and I believe you're right, I just think that it misrepresents the field to assume that anyone still believes in the motor theory as it was presented 50 years ago... (Now I'll read your TiCS paper ;-) )

Greg Hickok said...

Thanks for your comments Patti. There were important enough that I elevated my response to a new post. I'm pleased you enjoyed the paper.

A.P. Saygin said...

Dear Dr Hickok, I spoke to you a couple of times here and there but a long time ago and my work is not focused on language much so you may not remember me.

You cited a paper of mine in your interesting and timely review. This is by no means a well-thought out response to your paper. I am rather selfishly and self-centeredly responding to the bit about my work ☺

I am extremely agnostic about mirror neurons. I think they are cool, but their functional properties are not well-defined, and their roles have been over-interpreted in various domains (e.g., evolution of language, autism, empathy, conceptual representations). Sometimes I get annoyed at the exaggerated claims (and at being lumped as someone who makes such claims since my work does involve action perception) but mainly I am bored by the whole thing. In fact I am "post-bored" by them, if there is such a thing...

Anyway, more specifically, you have cited a paper of mine here (2004 Neuropsychologia). I don't blame you if you didn't read it in entirety – I really do mean it: it's so long, I can barely remember everything in it. I wanted to clarify a few things...

1) I did not understand why you think having tested aphasic patients biases to find IFG involvement in action comprehension. Testing aphasic patients was indeed the purpose of the study (see below). In other studies I did test aphasic and non-aphasic patients (Saygin, 2007, Brain, below).

2) There were reasons I used static actions: The experiment aimed to compare linguistic & non-linguistic processing, following an earlier paper of ours on speech and environmental sound perception. These two set of experiments follow a neuropsych literature going back several decades on the issue of linguistic specificity of aphasia. Particularly, it's been suggested that deficits in pantomime comprehension should be correlated with reading comprehension. (That is not in fact what we found, as you note). In our studies we wanted to maintain contiguity with the older papers. Also since the linguistic stimuli would be written text, I felt realistic action videos with all the added visual information (esp motion) would be even more problematic. I am well aware that static pictures will not tell you the whole story about mirror neurons but we were never intending to do that anyway in that paper. I in fact believe there is no way to have perfect, nicely matched stimuli for comparing linguistic & non-linguistic processing in the visual modality (except perhaps in sign languages) and I have not continued to try to compare nonlinguistic and linguistic perception since.

3) If the experiment does not tap into semantic processing as you suggest, I don't see why there would be a robust semantic distractor effect (in fact as I remember equally strong in the nonlinguistic condition as the linguistic). There was no effect of "affordance-related" distractors. Either my manipulation did not work, or the task was indeed not tapping into the motor/embodied aspects of processing. But I’d think it does tap into semantic processing.

4) The lesion results came out the way they did and we published the paper with as best an interpretation as we could. I was never looking for any evidence for mirror neurons and I try not to make claims specifically about mirror neurons (since I have never tested action production except I collaborated in speech domain the Wilson et al paper). However I think it is fair to say such and such lesion-deficit relationship could be related to action understanding related results in neurophysiology and neuroimaging literature.

5) In another study I have found a relationship between lesions in the same region with biological motion perception. I was again not looking for mirror neurons, hoping to corroborate neuroimaging results that superior temporal regions are critical for the perception of biological motion. This worked but there came out again the inferior frontal lesion focus... (There is also a paper by Tranel et al in 2003)

6) In many ways I agree with your paper. As I wrote in the above paper on this topic: "Having established [this], it will now be important to identify the precise functional roles played by each region in [biological motion perception]". However, while we may disagree on interpretation or strength of interpretation, I am very confident in the data. Consider that I have done multiple experiments with roughly the same group of patients. In 3 of them we found IFG: The above two tasks (pictured actions and point light biological motion perception) and speech production (speech output measures in autobiographical interviews, Borovsky et al 2007 Neuropsychologia). We did not find IFG for environmental sound comprehension, speech comprehension, reading comprehension (action sentences), selective comprehension problems of passive sentences or of the Grodzinsky trace deletion sentences, maybe a few others I am forgetting... In other words, the two times that we found IFG lesions in "receptive" tasks, it was with action-related stimuli. I don't know that this means mirror neurons are involved or if it is impaired motor simulation that underlies deficits, but I think the data show that lesions here are related to action perception or comprehension.

It is good that mirror neuron system receives criticism as well as love. Action understanding (or any complex domain) will not have a simple story. I hope however that we reach a sort of balance in the field where most people don't love or hate mirror neurons (or social cognition etc), but simply view it as an active and evolving area of study.

Thanks for letting me clarify a few things about our work (hope that's OK! I figured it's a blog so you probably welcome comments? Apologies otherwise and please feel free to delete this)

Best regards, Ayse P Saygin

Greg Hickok said...

Hi Ayse,

This is exactly the kind of thing I like to see posted here. Discussion is good. Thanks for posting... Here's some responses to your posts.

First let me say that I'm not arguing with your interpretation of your study, nor am I questioning whether your study was appropriately designed to address the questions you were interested in answering. I'm arguing against a mirror neuron interpretation of your data; e.g., some people have pointed to your paper as evidence for the mirror neuron theory of action understanding. I probably should have clarified this in the paper.

1) I did not understand why you think having tested aphasic patients biases to find IFG involvement in action comprehension. Testing aphasic patients was indeed the purpose of the study (see below). In other studies I did test aphasic and non-aphasic patients (Saygin, 2007, Brain, below).

The point was that by including aphasics your sample was going to include lots of patients with frontal lesions and maybe not as many patients with exclusively parietal lesions. This may have reduced the chance of you finding a correlation between action understanding as you measured it and parietal damage. Your subject selection process was perfectly appropriate for your goals which was a study of aphasia.

2) There were reasons I used static actions: The experiment aimed to compare linguistic & non-linguistic processing, following an earlier paper of ours on speech and environmental sound perception. These two set of experiments follow a neuropsych literature going back several decades on the issue of linguistic specificity of aphasia. Particularly, it's been suggested that deficits in pantomime comprehension should be correlated with reading comprehension. (That is not in fact what we found, as you note). In our studies we wanted to maintain contiguity with the older papers. Also since the linguistic stimuli would be written text, I felt realistic action videos with all the added visual information (esp motion) would be even more problematic. I am well aware that static pictures will not tell you the whole story about mirror neurons but we were never intending to do that anyway in that paper. I in fact believe there is no way to have perfect, nicely matched stimuli for comparing linguistic & non-linguistic processing in the visual modality (except perhaps in sign languages) and I have not continued to try to compare nonlinguistic and linguistic perception since.

Making comparisons with what has been done in the literature is perfectly reasonable. I just think that trying to depict actions with static images imposes a processing load that just isn't found in pictures of objects or in text, so it is a bit of a confound.

3) If the experiment does not tap into semantic processing as you suggest, I don't see why there would be a robust semantic distractor effect (in fact as I remember equally strong in the nonlinguistic condition as the linguistic). There was no effect of "affordance-related" distractors. Either my manipulation did not work, or the task was indeed not tapping into the motor/embodied aspects of processing. But I’d think it does tap into semantic processing.

My argument was that since performance on the written version of the task did not correlate with frontal lesions, and since written and pictured versions dissociated, damage to frontal regions were not tapping semantic knowledge. That is, if damage to frontal regions disrupted action-semantic *knowledge* then it shouldn't matter how that information was accessed; it should be disrupted across the board. How to explain the semantic distractor effect? Maybe frontal lesions are important for *accessing* semantic knowledge, or for some executive process that attempts to resolve ambiguity -- I'm not sure, but I know it is not a knowledge problem based on the written condition.

4) The lesion results came out the way they did and we published the paper with as best an interpretation as we could. I was never looking for any evidence for mirror neurons and I try not to make claims specifically about mirror neurons (since I have never tested action production except I collaborated in speech domain the Wilson et al paper). However I think it is fair to say such and such lesion-deficit relationship could be related to action understanding related results in neurophysiology and neuroimaging literature.

Again, I think the findings from the written version of the test argues against the view that "action semantics" resides in the left inferior frontal lobe. Otherwise, how do you explain the dissociation?

5) In another study I have found a relationship between lesions in the same region with biological motion perception. I was again not looking for mirror neurons, hoping to corroborate neuroimaging results that superior temporal regions are critical for the perception of biological motion. This worked but there came out again the inferior frontal lesion focus... (There is also a paper by Tranel et al in 2003)

I may have looked at this study but I don't recall any details. The first think I would do is look at the task to see if it might implicate executive or working memory processes that might involve frontal structures. Tell me more about the study...

6) In many ways I agree with your paper. As I wrote in the above paper on this topic: "Having established [this], it will now be important to identify the precise functional roles played by each region in [biological motion perception]". However, while we may disagree on interpretation or strength of interpretation, I am very confident in the data. Consider that I have done multiple experiments with roughly the same group of patients. In 3 of them we found IFG: The above two tasks (pictured actions and point light biological motion perception) and speech production (speech output measures in autobiographical interviews, Borovsky et al 2007 Neuropsychologia). We did not find IFG for environmental sound comprehension, speech comprehension, reading comprehension (action sentences), selective comprehension problems of passive sentences or of the Grodzinsky trace deletion sentences, maybe a few others I am forgetting... In other words, the two times that we found IFG lesions in "receptive" tasks, it was with action-related stimuli. I don't know that this means mirror neurons are involved or if it is impaired motor simulation that underlies deficits, but I think the data show that lesions here are related to action perception or comprehension.

I don't doubt your data. Lesion there are related to action perception/comprehension *as you measured it*. What I'm questioning is whether your measurement reflects the action semantic knowledge or some process that is involved in accessing such knowledge on your tasks. Remember, discrimination of speech sounds (ba-ga) is related to damage in frontal cortex, but this ability double dissociates from processing speech sounds in normal comprehension (as your data seem to indicate).

It is good that mirror neuron system receives criticism as well as love. Action understanding (or any complex domain) will not have a simple story. I hope however that we reach a sort of balance in the field where most people don't love or hate mirror neurons (or social cognition etc), but simply view it as an active and evolving area of study.

I agree. I have no issue with mirror neurons, and the hypothesis that they support action understanding is interesting as well as conceivable. What bothers me is the belief in a theory that has no basis in the evidence.

A.P. Saygin said...

Thanks for your response and comments!

I don't know how to include your reply here so I hope if this is not very unstructured.

Our data are in conflict with the asymbolia view of aphasia, and also with the idea that there are amodal representations for action or action knowledge specifically in the IFG. We suggested this IFG region is linked to deficits in the (visual) processing of action information in patients.

Any neuropsych study is limited by power issues related to lesion distributions. Indeed it is something we discuss often and there are ideas and methods being used or in development (e.g. Husain and Coslett have some papers on this). But I don't think possible sampling bias based on aphasia caused the IFG finding since we do find posterior lesion sites linked to other measures -- indeed in the same experiment we find parietal for the reading task. We also published (Borovsky et al 2007 Fig 1) a map of lesion distribution and you can see that we have pretty good power outside of frontal cortex including parietal. I’d love to have had a more uniform sample - in fact, the most obvious limitation of my work actually is there is little or no info from the right hemisphere! Alas, it’s difficult to do patient work so it’s not always possible to have an ideal sample. These were the patients I (was lucky enough to have) had access to at the time.

Re: Static pictures. I felt it would be more of a confound to have rich moving stimuli and plain old boring text. There are many ways in which this comparison simply was not possible to un-confound. I ran this experiment based on the neuropsych history, and also to go with our paper in the auditory modality but I gave up on the endeavour of comparing linguistic and nonlinguistic of visual stimuli as I decided the stimuli would have one confound or another... However in retrospect, since I am now working primarily in vision, it is interesting to have used still stimuli (my current work is on what the roles of motion cues are in body perception)

I don't follow why the lesion site would reflect executive processing or working memory since it's not in those "classic" executive areas, and the task was the same across domains, which dissociated. Anyway I think we have a more general issue here that has to do with the semantics of "semantics" :-) I am afraid my understanding of these issues is rather superficial. I have not tried to (and don't know how to) dissociate between action semantics and access to action semantics (at least with the data and methods at hand).

The action/mirror literature was very active at the time but had very little work from lesion studies. I this it was reasonable to discuss our data in relation to other data from action perception. It seems to me obvious that some kind of amodal representation of action semantics is not located to this region and I don't know if anyone has cited my work to claim it does but if so they would be wrong. We have not found some sort of absolute deficit, an action-agnosia... But simply that deficits in (at least visual) processing of action stimuli are linked to IFG lesions. This is important to note because activation in an imaging study has little ability to make a causal link between brain and bheavior and neuropsych and TMS are thus very much complementary to these.

And yes, of course whatever we reveal about action processing is “as we measured it”. But isn’t this true of any experiment? For example the study did not even have non-action stimuli, it was designed to fit in a different theoretical framework. It is in my view impossible to conclude something satisfactory about action semantics from this one study (even though we all try and make some conclusions about data in discussion sections – I’d personally rather just write what I did and what happened, but that’s not the culture…) But you make incremental progress over time so for instance the same area showing up in the biomotion study indicates the finding not specific to static stimuli or form-based visual stimuli, or the object matching task… And so on, over time, we learn more and adapt our theories.

I still think the semantic distracter effect (across modalities and domains, always significant but without any interaction with linguistic vs non-linguistic domain) is indicative that this task taps into something semantic, or at least there is interference based on semantics. We did discuss this a little bit in the paper (in the light of some very useful comments from one referee) but again I am afraid I have to plead either agnosticism or ignorance when it comes to what you call "action knowledge". I don't indeed know how to focus on specifically that.

Some people have thought that I should be disappointed or at least surprised by the lack of correlation between the domains. I indeed was not. I did not think the match between the domains in the visual modality was satisfactorily divergent selectively or even primarily on the dimension of linguistic vs. not, as I explained above, so I did not expect to find what we found in the auditory modality (in fact I expect I will never find correlations like that in patient data :-)). How do I interpret the dissociation? I will cheat and paste what I wrote in the paper all that time ago: "Finally, note that an embodied cognition view is not at odds with the lack of correlation between domains ... A strong asymbolia view would expect such an outcome... even though task and stimulus level factors were controlled for across the two modalities, there were other varying factors between the two domains. ... non-linguistic action comprehension system could be overlaid very early in development on the body’s own motor, sensory and proprioceptive representational systems, whereas reading, being a later-acquired skill, would be overlaid on a more distributed linguistic and conceptual network. If the systems are acquired and related skills are honed at such different stages in development, the resulting brain networks subserving processing in the two domains will also be rather different ... In contrast, in a very similar study we conducted in the auditory modality, where the linguistic and nonlinguistic stimuli are both perceptually similar and are acquired at similar stages in development (paper just now coming out), we did find tightly correlated deficits in aphasic patients’ performance, along with shared lesion sites." Sounds reasonable to me today that perceptual and developmental timecourse differences may have led to the outcome.

As for the biomotion study with the IFG finding (as well as superior temporal/inf parietal), here is the paper: http://brain.oxfordjournals.org/cgi/content/full/130/9/2452

I am not defending a specific theory here or more generally in my work. Theories do and should change over time. Most people I know who work in this field agree we still have to figure out a lot re the functional properties of these brain areas but perhaps this is not everyone.

Best, Ayse

Greg Hickok said...

Hi Ayse,
I'll respond to some of your points one at a time for clarity. You said,

"I don't follow why the lesion site would reflect executive processing or working memory since it's not in those "classic" executive areas, and the task was the same across domains, which dissociated."

The lesion focus is in Broca's area right? I would agree that this is not a classic executive area, but it has been implicated working memory both for speech and gesture (e.g., sign language). So, suppose that the pictured action task is harder than the text task (probably true right?). And suppose harder tasks require more working memory. Broca's area then becomes correlated with deficits on the picture task but not the text task. This would be a working memory effect, not an action processing effect.

Anonymous said...

Nope, it's the other way around - sorry perhaps I was not clear as I wrote it all very rapidly between meetings. It's the reading task that is harder not the action task.

Anyway it can't be a working memory effect also because in the biomotion task the stimuli plays on until a response is recorded.

I even ran a working memory task with the same stimuli with fMRi (plus the same biomtion stimuli) and those modulations are in clearly distinct areas.

No hurry to respond at all! Perhaps we can talk after you have had a chance to read the papers.

best, Ayse

Geoffrey said...

I felt much excitement when I saw that there was a paper challenging the assumptions surrounding mirror neurons, anyone else?

Anonymous said...

Greg, can you clarify what you mean to say we "understand flying actions" even though we cannot fly? Do you mean intellectually/mechanically? I am not sure I personally do understand the subjective element of flying, that is, what it feels like (in other words, I cannot run a complete 'mental simulation' of what it is like to fly, in the same way I can when I see someone walk or talk or display emotion etc.)

I would therefore be interested to know if any research has been done in relation to the action of mirror neurons (I am assuming here that they do exist) when an individual observes actions from an 'agent' (a necessary condition for mirror-neuron activation, according to Ramachandran) that they themselves would be incapable of carrying out?

Greg Hickok said...

The fact that we have a word in the language, to fly, indicates that we have knowledge of the concept. That's what I mean. Yes, there have been studies of watching actions that we can reproduce (speech) versus those we cannot (barking). I reviewed this in my "Eight Problems" paper.