Tuesday, June 12, 2007

Jealous of the fusiform? I am.

Why do people go on and on so about the fusiform gyrus and face recognition? What's the deal? I think they carry on because ... well, because they can! As it turns out, the neuronal activity in the fusiform face area (FFA; see e.g. Kanwisher N, McDermott J, Chun MM. 1997. The fusiform face area: a module in human extrastriate cortex specialized for face perception. J Neurosci. 17(11):4302-11) really does have a tight link to face recognition processes. Evidently, the holistic aspects of face processing are 'elementary' or 'primitive' (in a computational-representational sense) such that a circumscribed cortical area forms the basis for the functional process. That does not mean that other areas are not involved, but apparently the FFA plays a privileged role. Who would have thunk it - face recognition has a place.

Anyway, I am jealous. I would like for us speech-types to have an area like that, too. The fusiform face area and the parahippocampal place area are practically household names (well ... in pretty nerdy households). What do we have? Why don't we have something like the 'superior speech field' SSF, or the 'middle speech gyrus' MSG? Can we have that please?

Well, it's not so clear that that would make sense. What we have is an increasingly articulated functional and anatomic boxology -- for example in the way Greg and I presented in in our 2007 Nature Reviews Neuroscience paper (see earlier blog entry). I think this does make sense, because 'speech perception' or 'spoken word recognition' are in my view not monolithic; rather, a number of subroutines are necessary parts of successful speech perception. No single area is responsible for a sufficient number of processes to by itself deserve the name 'speech area.' So, although we often continue to search for the speech area, I think this is wrong-headed. We should be radical decompositionalists instead, and identify (cognitively, computationally) all the subroutines implicated in speech processing. And find a neurobiologically plausible implementation (a la Marr) for the computational subroutines.

Of course, I could be wrong about that ... Reasonable people can disagree, so maybe there is an area 'specialized for speech'. I'd certainly like to hear the arguments.


Martin Meyer said...

David, don't you think it should at least considered a progress that the idea of a brain centre for syntax or semantics has meanwhile (at least by the reasonable people you mention) been abandoned? Just a few years ago it was quite fancy to propose that Broca's area accommodates a core 'module' for syntactic operations. I am quite relieved that we have overcome and survided the "age of centrology".

Greg Hickok said...

Don't worry David, lots of guys have fusiform envy. You know, if you really want us to have an AREA, there always the LASTISA (Left Anterior Superior Temporal Intelligible Speech Area; see Scott et al., 2006, JASA) -- it's got nice ring to it, with a bit of Latin flare... Then there's always LASTISA's bedfellow, the ATLSA (Anterior Temporal Lobe Sentence Area), and there are probably a few candidates for the STPA (Superior Temporal Phoneme Area). With the right marketing, we could be rich in AREAS just the like vision people. But as you and Martin point out, I don't think that is what we want. Does anyone really believe that a single area supports the processing of intelligible speech? And I think Martin is right that it is progress that we don't think of say, Broca's area, as the seat of syntax, and instead we are thinking in terms of more distributed networks that support various functions. But then we are left with two problems. (1) what do we make of these focal areas that show up in so many functional imaging experiments? Are these the endpoints of a long chain of computations supporting a given function? Some sort of "convergence zone" in the Damasio sense? Or just an artifact of overzealous thresholding? And (2) how do we develop and test models of these distributed networks? It is so much easier to try to isolate our favorite language function and call the resulting 25 voxels our FLFA (Favorite Language Function Area). And anyway, who wouldn't want an AREA of their own. I think I've got a touch of area envy as well...

David Poeppel said...

Martin, Greg,

I respectully disagree with you both. I don't think we have to abandon what Martin calls "centrology". And, accordingly, I don't think we have to adopt, with reckless abandon, networks.

I think there is no question that *some* form of functional localization is the fact of the matter -- but the question is "what excatly is being done by those neural circuits that are implicated" by the imaging data, neuropsychology, etc. In other words, I see it as a GRANULARITY problem (sorry -- I just love using that word :-) If we had an appropriately granular analysis of what, say, syntax or word recognition actually was, we could hypothesize that certain computational subroutines are in fact localized. This is what I have called in the past "computational organology". I would like to take the good part of Gall's work -- the organology -- and abandon the less-good part -- the phrenology.

I guess what I am arguing for is that we decompose the speech/language/cognitive issues into primitives of a granularity that could in prinicple be mediated by neuronal circuitry. This is sort of what I was advocating in the paper with Dave Embick (Poeppel & Embick 2005, in Anne Cutler's book).

So (a) I am pro Gall; (b) I am pro organology and centers, but at a different level of analysis; (c) I am I guess sort of on board with the network thing insofar as the network parts are constituents that can be localized etc.

Geez, I need to think this through some more ... Ugh.

Greg Hickok said...

As much as I would like to argue with you David (arguing is fun!), I don't think we actually disagree. I think of "networks" as being composed of a set of localized, computationally specialized areas. But what is localized is not as broad as say, "phonological processing" or "speech perception" (whatever that means) but something more specific in terms of granularity (ooh, that is an exciting word to use!). So I guess the problem is not with the concept of organology -- I'm a Gall fan too! -- but with what is organalized (how's that for productive derivational morphology!). So we've dispelled with destructiveness and hope as mental organs (but see the recent mirror neuron lit), now we need to jettison "phonology" and "semantics" and "syntax." It's just a question of granularity.

Unknown said...

david, don 't hate the player, hate the game!

Anonymous said...

You just need a catchy name. Preferably alliterative and with the linguistic or speech area represented by a common non-tecnical word.

Why not make up one of those "match the columns" games and we can play. Column one can be an anatomical part, column two the micro-area of linguistic processing and column three can be words denoting the space "area, section, gyrus" etc.