Monday, October 27, 2008

"The battle for Broca’s region" -- lost again

There is a new paper in the journal Trends in Cognitive Sciences that, once again, examines the role of Broca's area and language processing.

The battle for Broca’s region, by Yosef Grodzinsky and Andrea Santi, summarizes four positions about the role of Broca's area and concludes -- who would have thunk it? -- that the 'syntactic movement account' is the best account to date.

Grodzinsky and Santi distinguish between four positions: an "action perception" model (advocated, for example, by Arbib and Rizzolatti), a "working memory" model (Caplan), a "syntactic complexity" model (Goodglass, Friederici), and a "syntactic movement" model (supported by the authors). I think one can quibble about the attributions, but by and large this is more or less of a fair characterization of various positions. The former two are of the "general" variety; the latter two are language specific. The authors examine these positions in light of data from deficit-lesion correlation and neuroimaging evidence, basically from fMRI. They argue, reasonably, that a single model account is likely to underspecified. That being said, they conclude that the recent evidence is most consistent with a "syntactic movement" model Broca's area.

I have rather mixed feelings about this brief review/perspective piece. On the one hand, it is perfectly reasonable for Yosef to work hard on supporting the view that he is fought hard for for a long time. Indeed, attempting to identify a particular kind of computation that's executed in a chunk of brain tissue seems like a sensible goal. On the other hand, I do think it's really time to go further now, and I wish that these authors might lead the way on a more biologically sophisticated perspective.

The fact that their view is too simple is something they state repeatedly. "Importantly, Broca’s region might well be multi-functional." And: "Indeed, Broca’s region might be multifunctional." And so on. Well, yes, then let's actually entertain that...

The fact that we have to make careful distinctions between areas 44, 45, 47, and the frontal operculum is now well established. Yosef has supported important progress in this area, and Friederici and her colleagues as well as Amunts and her colleagues have provided impressive evidence for functionally relevant subdivisions. Moreover, even for a single piece of tissue a la Brodmann, the probability is very a very high that more than one operation is executed. Obviously ... Look, take Brodmann area 17 (primary visual cortex, striate cortex). Beyond subdivisions into ocular dominance columns, orientation pinwheels, and -- obviously -- six differentiated layers of cortical tissue, there are further functionally critical subdivisions into cytochrome oxidase blobs, etc. We are perfectly comfortable attributing multiple functions to local pieces of tissue in the visual system. Yet we persist in trying to find surprisingly monolithic interpretations of the chunk of brain as extensive as Broca's region. Now admittedly we don't have the necessary cell biological analysis of this part of the brain; nevertheless, isn't it time we come up with some more nuanced hypotheses about what gets calculated in these various different parts of the frontal lobe?

Inquiring minds want to know. I'm pretty frustrated with the state-of-the-art in this area of research. Please, somebody, figure this piece of brain out!

Y GRODZINSKY, A SANTI (2008). The battle for Broca’s region Trends in Cognitive Sciences DOI: 10.1016/j.tics.2008.09.001


Greg Hickok said...

Was there any mention of the possible role of Broca's region in speech production in that paper? Or is that idea completely out of the picture? I suppose you could argue this position given that a lesion restricted to Broca's area doesn't cause Broca's aphasia nor any significant deficit in speech production. Come to think of it, I bet a lesion restricted to Broca's area wouldn't cause asyntactic comprehension, working memory deficit, or action understanding difficulty.

It's actually a very puzzling chunk of tissue: it activates for almost anything but when damaged doesn't seem to result in any permanent deficit!

RE: multi-functionality, I fully agree, David. Even though I am guilty of it myself, I think it is time to stop talking about "Broca's area" as if it were a single functional unit. Same goes for the "planum temporale" b.t.w.

How about we figure out Broca's area(s) right here on Talking Brains? Group project!

Tom said...

To start with, someone's going to have to decide what counts as Broca's area. Or would it be better to abandon that label completely in favour of talking about pars opercularis, triangularis, orbitalis etc? If we're arguing from functional imaging data, we're probably on shakey ground with BA 44/45/47.

David Poeppel said...

my colleague al braun from the NIH still has the best Broca's area line: "any experiment, if properly designed, will activate Broca's area." (personal communication :-)

tom's point is well taken. is the label helpful, at this point, other than in the context of historiography? Broca's region encompasses so much heterogeneous cortical tissue that it's unclear to whether it makes any sense at all ...

amunts and colleagues have nice data supporting certain anatomic subdivisions, but there too the work is guided by the 'originary presupposition' that it can be captured by a few labels.

You know how in France the Bourbaki Circle was a group of folks that published together under that name in mathematics? We as an intellectual collective could publish a Broca's area agenda for research under some name. Like greg suggests, a group project.

Tom said...

The - "any experiment, if properly designed, will activate Broca's area" - gag reminded me of this paper published by Russell Poldrack a little while ago. He applies Bayesian logic to clearly demonstrate the dangers of making a reverse inference from activation to cognition - i.e. a notion such as, 'if Broca's area is activated by our task, there must be a language-related component to the task'. He argues that the usefulness of such an inference is limited by the selectivity of the region. With a 20mm cube of dorsal left IFG as his seed region, he searches the BrainMap database for contrasts that activate/do not activate this region. Poldrack finds "positive, but relatively for the idea that that activation in [this region] implies engagement of language function".

Reading the paper again, in the context of thinking about 'figuring out Broca's area', it strikes me that;

1) This subregion of 'Broca's Area' (roughly BA44) is remarkably unselective, even when such a crude distinction as "Language Task"/"Non-language Task" is used. Expanding the region of interest to encompass BA44/45/47 will, of course, make matters worse. Any monolithic interpretation of 'the function of Broca's area' is doomed.

2) fMRI may well offer good spatial resolution, but because of the signal/noise problem we have in neuroimaging, it only offers poor 'cognitive resolution'. Firstly, we can't investigate behaviour in the scanner in anything like as thorough or subtle a manner as we do outside the scanner. Most in-scanner tasks are neccesarily crude as subtle behavioural manipulations are likely to get lost in the noise. Maybe the 'cognitive ontologies' approach proposed by Poldrack is a good away around this? I'm not sure how well it would work in practice, because;

(i) asking subjects to perform an explict task over and over again in the scanner is probably not going to give us a clear view of how cognitive processes are embodied under real world conditions. We can guess at which processes are likely to be engaged by the task, but we have little idea about how these processes will interact with each other over the course of the experiment. Is the same thing happening at the first and the fiftieth repetition?

And (ii) it's probably a mistake to try to examine how the brain works by focusing on single regions of tissue at a time. Once we start to talk in terms of activated networks rather than activated regions, we increase the complexity of the problem by a frightening degree.

daniel kislyuk said...

Precisely! What gets calculated there and why it is calculated there and not elsewhere. These questions are truly interesting.

From this point of view it is sad that the authors push forward a strictly modular account of syntactic processing. In such a paradigm syntactic processes in the area in principle can not be connected to & derived from other calculations happening in this region. In the original paper used in the review (Santi & Grodzinski, 2007) the authors take less extreme, and thus IMHO much more productive, position saying that Brocas' area supports working memory that is specifically involved in syntactic movement operation.

I'm also a bit suspicious about the arguments that are used to draw a border btw working memory and syntactic movement accounts. Authors compare the processing of reflexive binding (The cat touched itself) vs. filler-gap dependencies (The cat that the dog chased -- was black). Only the latter uses syntactic movement (SM) while both tax working memory (WM). This assumption was employed in Santi & Grodzinski 2007 study, where the filler-gap and antecendent-pronoun distances were parametrically varied. They found that the increase of distance only in filler-gap, but not in reflexive binding constructions had correlated with the increase of activation in the Broca's area. From this the authors conclude that working memory that resides in the Broca's area subserves specifically the syntactic movement but not general sentence processing. This contrast seems a bit strange for such a conclusion. I would say that working memory (WM) model also does _not_ predict the increased activation of the WM-related areas with growing antecedent-pronoun distance in the reflexive binding condition.

Where from can increased activation of the working memory-related area arise? From a)retention of information for certain period and b)retrieval of information. Let's consider syntactic movement and reflexive binding from these points of view.
The authors note (S&G, 2007) that reflexive binding uses the ordinary word order and the sentence in principle can end without any reflexive pronoun. Cf. "The cat touched itself" and "The cat touched the mouse". Therefore, when we perceive such a sentence we process it in a regular way, not especially trying to retain anything and shipping off the chuncks of information to the next stages of processing in due course. If the reflexive pronoun appears then one single retrieval operation is to be performed independently of the distance between the antecedent and the pronoun. Therefore - no dependance of WM ares activation on the distance.

With gap construction situation is totally different. They require active retention which is evident pretty early in the sentence. "Kate loves the woman who the mailman and the mother pinched". And from this point on the filler is to be actively retained and possibly fit into the first available position. Animal studies suggest high neuronal activity of memory-related neurons during the retention period. No wonder increased period of filler retention evokes higher activity in working-memory related Broca's area!

So, we end up comparing the linguistic operations one of which explicitly calls for high WM involvement during the good part of a sentence, while in the other the need of retrieval becomes evident only in the very end of the sentence. This ground seems pretty unstable to conclude that syntactic movement is of special importance for Broca's area.

Greg & David, pretty often you post some quite critical reviews. Do you think that it would be a nice idea to actually tell the authors that their opuses are being discussed? It will take google scholar a while to start index the citations from here.

Santi and Grodzinski, 2007. Working memory and syntax interact in Broca's area. NeuroImage, 37(1), p8-17