Tuesday, May 6, 2008

A proof that mirror neurons are not the basis of action understanding?

A couple of comments on the Gallese et al. 1996 paper (Brain, 119:593-609) and Rizzolatti, Fogassi, & Gallese 2001 review paper (Nature Reviews Neuroscience, 2: 661-670.

The Gallese et al. paper is an empirical report, and as such, a good one.  Have a look if you want to know about the functional properties of mirror neurons.  One interesting tidbit: five different hand actions were assessed, grasping, placing, manipulating, hand interaction, and holding.  Most cells, by far, are partial to grasping.  Of 92 cells sampled, 30 were selective for grasping (next highest selective cell count was placing cells & manipulating cells, 7 cells each), and 39 responded to grasping plus some other action.  So grasping is represented in 75% of sampled cells.  This probably just means monkeys grasp a lot, but maybe it's worth noting.

There is some commentary in the discussion section that's worth examining because a clear definition of action understanding is provided.  "By this term [understanding], we mean the capacity to recognize that an individual is performing an action, to differentiate this action from others analogous to it, and to use this information in order to act appropriately" (p. 606). 

This is a strong position.  If we flip the statement around, it implies that without MNs the animal cannot "recognize that an individual is performing an action" or "differentiate this action from others analogous to it" or "use this information in order to act appropriately".

This view seems to rule out the possibility of observational learning.  Now, I'm way outside my area of expertise, so I'm going to resist stepping too far out on a limb.  I am aware however, that observational learning occurs in many species.  E.g., I saw a TV science show once where an octopus learned to get into a jar by watching another octopus open the jar.  (I TOLD you I was outside my area of expertise!)  Anyway, the point is, making a claim that rules out observational learning seems to a lay person like me, the wrong theoretical move to make (see below).

We looked that Rizzolatti, Fogassi, & Gallese (RFG) review paper because it was continually cited in the Rizzolatti & Craighero paper after statements regarding other mechanisms for action understanding.  This is important, because if there is another way to break into the system, you might get out of some of the circularity problems we noted previously with respect to imitation and action understanding, as well as this related problem with observational learning.  I was hoping we'd get a bit of clarification on the question of these other mechanisms, and we did. 

RFG are admirably strong in their claims regarding the importance of MNs in action understanding.  They're certainly not hedging any claims: "We understand action because the motor representation of that action is activated in our brain" (p. 661).

So what about those pesky STS cells that also respond to the perception of action?  Well, RFG provide a nice summary of these cells.  STS cells respond not only to simple actions, but also appear to combine action-related bits of information.  For example, the firing of some action responsive cells in STS is modulated by eye gaze information: the cell fires only if the actor is looking at the target of the action, not if the actor is looking away.

RFG summarize, "The properties of these neurons show that the visual analysis of action reaches a surprising level of complexity in the STSa" (p. 666).  A level that seems to be much more sophisticated than that found in F5, I would interject.  RFG continue, "But the existence of these neurons and, more generally, of neurons that bind different types of visual features of an observed action, is not a sufficient condition for action understanding per se" (p. 666).  I completely agree.  You can't make logically necessary inferences about function on the basis correlated neural activity.  But the same is true of mirror neurons.

I think, though, that the point RFG are trying to make here is that unlike in pure sensory perception (in their view), the "semantics" of an action is somehow inherent in the action itself.  Again, evidence from aphasia leads me to question this position.  How is it that there are patients who can perfectly reproduce an action (repeat speech verbatim) yet fail to understand the meaning?  But let's move on.

So how does STS acquire these complex response properties?  To the best I can tell, RFG hold that the motor system endows the STS with its action perception abilities.  

"We argue that the sensory binding of different actions found in the STSa is derived from the development of motor synergistic actions.  Efferent copies of these actions activate specific sensory targets for a better control of action. Subsequently, this association is used in understanding the actions of others" (p. 666).

So it sounds like the STS action recognition system fully inherits its action recognition/understanding abilities from the motor system.  How it inherits "a surprising level of complexity" from a much simpler system is not clear.  But RFG's position is clear: action understanding is a function of the motor system.  Action understanding can be achieved via other systems, such as the STS, but this is only because the motor system has endowed such systems with its action understanding ability via previous association.

Correct me if I'm wrong, but from this collection of assumptions, namely, >

1.  "We understand action because the motor representation of that action is activated in our brain." RFG p. 661.

2.  "We cannot claim that this is the only mechanism through which actions done by others may be understood." RC p. 172.
3.  These other mechanism acquire their action understanding properties via associations with the motor system. (My paraphrase of above quotes from RFG p. 666.)

it follows that an action cannot be understood unless it has already been executed by the perceiver.  From this it follows that observational learning should be impossible in creatures that use their mirror neurons to understand actions.  I don't know about monkeys, but I believe (without knowledge of the literature) that humans are good at observational learning.

If all of these assumptions are correct, we've just proven that the mirror neuron theory of action understanding is wrong.   

So let's discuss the problems with my assumptions.  Anyone have any ideas?


Brad Buchsbaum said...

With regard to number 3:

"These other mechanism acquire their action understanding properties via associations with the motor system. (My paraphrase of above quotes from RFG p. 666.)"

Can these other mechanisms that "acquire" their action understanding mechanism operate even in the absence of a working mirror neuron system?

If this is the case, then it implies that even if action understanding survived a bilateral lesion to the mirror system, the system would, at least partially, "live on" in the sensory areas to whom it had imbued its knowledge.

This again raises the interesting question about how to define a "motor" and a "sensory" region: if the motor system can shape the sensory system (and vice versa) then arguing about what belongs to the MNS and what does not may be moot. Indeed, primary auditory cortex might well be part of the "motor" system, so defined, if its behavior is in some way influenced over the course of learning/development by the motor system.

Greg Hickok said...

They do seem to admit that "sensory" systems, such as that in the STS, can be used in the understanding of actions: "Efferent copies of these actions activate specific sensory targets for a better control of action. Subsequently, this association is used in understanding the actions of others." RFG, p. 666.

Is this STS system sufficient for action understanding? They don't say. My guess is that they want to say no, but they are forced to say yes, to some extent.

Their position gives them a convenient but vacuous "out" whenever someone (like me, for example) points out that damage to motor systems doesn't destroy "understanding." This "out" is vacuous because if motor knowledge can take up residence outside the motor system you can no longer falsify the claim.

I suppose what you'd have to do to test the MN prediction that in development, disruption of the motor speech system will prevent normal receptive speech acquisition. Interestingly, this seems to be false, as Lenneberg pointed out decades ago in the context of cases of childhood dysarthria: "Children may acquire a complete understanding of language without ever having been able to produce intelligible words." (Lenneberg, E.H., 1967, Biological Foundations of Language. New York: John Wiley & Sons. p. 65; see also: Lenneberg, 1962, Understanding language without ability to speak: a case report. J. Abnorm. Soc. Psychol., 65:419-25).

So here's my counter-theory to theirs: All of action understanding is achieved via sensory systems. To the extent that the motor system behaves as if it understands actions -- activity correlates with perception, etc. -- this is simply a function of the sensory action understanding system transferring its knowledge to the motor system via association.

This accounts for mirror neuron response properties (they are acquired via sensory-motor associations), that pesky data from severe Broca's aphasia, the developmental observations of Lenneberg, the "sophisticated" action perception response properties of STS neurons, and observational learning.

Zarinah Agnew said...

I agree with the comment about sensory associations. There is relatively little evidence that the human mirror system has a functional role in the human brain; the patient studies are the best evidence we have (patients with execution deficits sometimes have corresponding perceptual deficits).

In fact a recent study has shown that you can artificially manipulate these sensorimotor associations. Catmur et al., (Current Biology, 2007) subjected healthy subjects to a training phase during which they had to execute a simple finger movement whilst observing a different finger movement. Before training, observing the original finger movement elicited a facilitated motor evoked potential (MEP) in the same muscle involved in executing that action. So this is pretty standard news (Fadiga et al., 1995, etc etc). After training however they found that observing the second, different finger movement, elicited increased MEPs from the original muscle involved in a different action.

In short they created an associated between an executed movement and a different observed movement. Wow. This is pretty exciting I think.

So they propose that this is potential mechanism for development of the human mirror system. More to the point in my mind however, it demonstrates that these associations can be formed without a specific function. During normal development we all perform actions which are accompanied by vision of that action. We observed our own actions and the actions of others; when you sit and eat dinner, you are executing an action and observing the same actions, both by yourself and by others. So it's possible that this is a) how the mirror system developed and b) that it may be involved but is not essential for action perception, recognition or understanding at all. It may in fact be a consequence of meaningless associations formed during motor development.

(PS As far as I know inactivation of the part of area F5 which contains mirror neurons, resulted in motor slowing but ultimately action execution was not eliminated (Fogassi et al., Brain 2001). It would be interesting to see if these animals are capable of action discrimination/ recognition/ prediction (as per Umilta et al., 2001))

Ludmila said...

We are independent researchers studying mirror neurons and have recently come to conclusion that the results of the experiments conducted by Iacoboni et al (2005) do not support the notion that mirror neurons code intention. If you are interested, we would like send you work entitled "The Link Between Mirror Neurons and Intention (Iacoboni et al 2005) Is Subject To Further Investigation." Abstract attached.

Ludmila Vucolova, USA
Peter Boroditch, Russia

This work will analyze the findings of Iacoboni et al., 2005 wherein is stated “the human mirror neuron system does not simply provide an action recognition mechanism, but also constitutes a neural system for coding the intentions of others” and will suggest that the findings are not supported by the results of the experiments.
The analysis of Iacoboni's findings demonstrates that the alleged greater activity in mirror neuron areas in the inferior frontal cortex in the Intention cleaning condition did not show sufficient additional activity to support the principal finding.
The analysis demonstrates that alleged greater activity in mirror neuron areas in the inferior frontal cortex in the Intention drinking condition is not valid since the comparison between Action, Intention drinking and Intention cleaning conditions was not based on the premises of the hypothetical model (identical grasping actions), thus, leading to an improper reading of the data and interpretation of the findings.
The high response in the Intention drinking condition versus Action and Intention cleaning is attributed instead to differences in the types of grasping actions, implemented by Iacoboni and observed by participants. These differences affect the level of mirror neuron response. The high level reported in the Intention drinking is due to the instantaneous recognition of two actions, “grasping” and “bringing to the mouth.” These are found to be two consecutive and adjoining links within the drinking action. The lack of additional response in Intention cleaning is due to the recognition of only the grasping action by viewing the Intention cleaning clip. Consequently, the need to resort to the understanding of intention to account for the high response (in the Intention drinking condition) can be eliminated.