Tuesday, September 22, 2009

Mirrors in the Brain -- Comments on Rizzolatti & Sinigaglia, 2008

Apparently I'm obsessed with mirror neurons because I can't seem to stop reading what people say about them. Now I'm reading Rizzolatti & Sinigaglia's 2008 book, Mirrors in the Brain, translated from the original Italian by Frances Anderson and published by Oxford.

I'm only about halfway through so far but already I find the book both useful in terms of its summary of the functional anatomy of the macaque motor system and frustratingly sloppy in terms of its theoretical logic.

Let me provide one example of the latter. At the outset of the book the authors describe the functional properties motor neurons (not necessarily mirror neurons) in macaque area F5. They argue that F5 motor cells
code motor acts (i.e., goal-directed movement) and not individual movements (p. 23).

As evidence they note that
...many F5 neurons discharge when the monkey performs a motor act, for example when it grasps a piece of food, irrespective of whether it uses its right or left hand or even its mouth ... [and] a particular movement that activates a neuron during a specific motor act does not do so during other seemingly related acts; for example, bending the index finger triggers a neuron when grasping, but not when scratching. (p. 23)

They conclude,
Therefore the activity of these neurons cannot be adequately described in terms of pure movement, but taking the efficacy of the motor act as the fundamental criterion of classification they can be subdivided into specific categories, of which the most common are 'grasping-with-the-hand-and-the-mouth', 'grasping-with-the-hand', 'holding', 'tearing', 'manipulating', and so on. (p. 23)

So the claim is that F5 cells are coding something higher-level that is defined by the goal, the "efficacy", of movement.

Clearly, F5 cells are coding something that is at least one-step removed from specific movements (e.g., finger flexion), but the leap from this observation to the idea that it is coding categories or goals such as 'tearing' is suspect. Perhaps these complex movements are being coded -- separately for the mouth and hand, for tearing in one manner versus another, etc. -- by the population of cells in F5 rather than in individual cells. In other words, the fact that a single cell responds to grasping with the hand and grasping with the mouth doesn't necessarily mean that it is coding an abstract concept of grasping.

But we don't need to argue with Rizzolatti and Sinigaglia on this theoretical point because they argue against their own view rather convincingly (although unwittingly) on empirical grounds. Specifically, in contrast to the claim that F5 cells code goal-directed actions, they give examples of how these cells code specific, albeit complex, movements.

Most F5 neurons ... also code the shape the hand has to adopt to execute the act in question... (p. 25)

This strikes me as a rather specific individual movement that, for example, would not apply to the same "act" executed by the mouth. More pointedly though, in their discussion of mirror neurons in F5, Rizzolatti and Sinigaglia make a big deal of cells that show a strict relation between the observed and executed act. They provide a striking example:

...the monkey observes the experimenter twisting a raisin in his hands, anti-clockwise and clockwise, as if to break it in two: the neuron discharges for one direction only. (p. 82)

So here is a case where two movements have the same goal (breaking the raisin in two), but the F5 cell only fires in response to one of the movements. Apparently this cell is coding movements not goals.

Has anyone else read Rizzolatti and Sinigaglia's book? Any thoughts?


James Kilner said...

Dear Greg,

Firstly I have to confess that I have not read this book. However, it would appear although you may have made at least two errors in your interpretation of this data.

The first is a logical error. The fact that SOME neurons in area X "cannot be adequately described in terms of pure movement" should not be interpreted as ALL neurons in area X "cannot be adequately described in terms of pure movement". As I understand it the discovery of the "grasping" type of F5 mirror neurons demonstrates that some of these cells are responsive to some "high-order" features of the movement and not just the movement not that all of them are. In the Gallese et al. 1996 Brain paper they give some numbers on F5 mirror neurons that an be described as stictly congruent and broadly congruent.

The second error is the fact that you have interpreted the goal of the second set of data you present as "breaking the raisin in two". What if this assumption of yours was not right. If I said there were two goals - to twist clockwise and to twist anticlockwise - then the all the data you have selected to report in this blog would be logically consistent.

Sometimes I think that there is a real desire for some people to reduce what are likely to be complicated coding patterns in F5 during both execution and observation to some easy to label yet poorly defined function like goals. To repeat slightly what I have said earlier, the fact that some neurons can be categorised into action goals does not mean that all neurons can be categorised like this.

Greg Hickok said...

Hi James,

You are correct that just because some neurons code a specific feature doesn't mean that all of them do and it is true that there is a distribution of response properties in F5 with some being more specific than others. I guess the point I was trying to make (perhaps not very effectively) is that Rizzolatti and company are reasoning from the properties of a subset of cells in the region and not others. They note that some cells don't respond to specific actions and code grasping (for example) of any kind. Therefore it must be the abstract goal of the action that is being coded.

I just flipped that argument around and pointed to the population of cells that respond to specific movements and (using their own logic) inferred that the function of F5 is to code complex movements not goal oriented acts.

So you caught my faulty logic. The question is, why don't people catch Rizzolatti's?! Suppose F5 really is just coding (via *populations* of cells) complex movements such as grasp-with-hand-using-precision-grip and not more abstract notions such as 'grasp' or 'tear'. Just because some cells respond to grasping with hand AND grasping with mouth doesn't mean that the population isn't coding something more specific.

Regarding your second point, the interpretation of the twisting motion was implied by R&S (re-read the quote); it was not my interpretation. Further, a problem with your suggested interpretation is that it is not goal-oriented: what object-oriented goal does twisting clockwise accomplish that is distinct from twisting anti-clockwise?

But the details don't matter as the point I'm trying to make is more general. I'm not necessarily arguing that F5 neurons aren't coding goals, I'm pointing out that existing discussions are very one-sided and fail to consider other possibilities. If we get some alternative hypotheses on the table, then maybe they can be tested empirically.