A post or two ago I got waylaid on my way to mentioning some important papers by David Poeppel and Dave Embick (P&E). They have been writing about Broca’s Problem; how to fit linguistic work on FL/UG with neuro work on brains. Several of their papers (e.g. here and here) review the state of the art. For the neuro illiterate (e.g. me) this is a useful quick review of what “we” know about how brains embody language. The general impression I get (though please read these for yourself) is that we know relatively little. Or to be more exact, we know a little about one kind of problem, the “where” problem (“Where are certain capacities localized?” i.e. what Poeppel (here) calls this the map problem), but almost nothing about (what I, and P&E, take to be the more important issue) the “how” problem (“How do brains realize these capacities?”, i.e. what Poeppel calls the mapping problem). Let me expand a bit.
P&E describe two hurdles to successfully integrating linguistic work with neuro-research. They dub these the Granularity Mismatch Problem (GMP) and the Ontological Incommensurability Problem (OIP).
GMP names a fact that becomes obvious when a linguist (e.g. me) sits in on neuro-lab meetings (which I often do to make myself feel good about work and data in syntax) and listens to discussions about localizing syntax or semantics or whatever. The units being mapped in the brain are, from a syntactician’s perspective, very big. Syntax for me is not one thing but a richly structured system with many kinds of objects (nouns, verbs, phrases, chains, complements, adjuncts, antecedents, bound pronouns etc.), many kinds of relations (binding, agreement, case marking, subcategorization etc.) and subserved by (possibly many) different kinds of operations (e.g. Merge, AGREE, Copy, Delete etc.), while in the neuro literature it is one undifferentiated thing, usually found on the left in conspicuous proximity to Broca’s area or BA 44/45 (sounds like the Arizona redoubt where the government keeps captured aliens doesn’t it?). Similarly, semantics as studied in the neuro literature hardly ever worries about compositionality or scope or binding but more about the interpretation of words and the semantic fields they are part of, something that linguistic semantics says almost nothing about (what's the meaning of ‘life’? Well, LIFE’ or lx.LIFE(x), helpful huh?). P&E identify several problems that arise from the fact that “the distinctions made in the neurological study of language are coarse in comparison with the distinction made by linguists (7).” One that they highlight concerns the (incorrect, in their view) identification of Broca’s area with syntactic computation. I’ve mentioned another review reaching the same conclusion (here).
P&E make two important points: (i) at a coarse level of description many apparently different things light up Broca’s area (ii) that this should not be surprising for the mistaken identification of Broca’s area with “syntax” rests on identifying syntax as a “simplex unstructured computation” looking for a “single undifferentiated cortical region” to call home. Of course syntax is not simple and Broca’s area is not undifferentiated. What P&E believe to be true is that “one or perhaps several of the computational subroutines that are essential for syntactic processing are computed in the Inferior Temporal Gyrus (IFG). But these are not ‘syntax’ per se -- they are computational sub-components of syntax (9).” The game should be to identify these suboperations and see how they fit.
This is music to my ears for from a minimalist perspective, we should see many of the basic processes operative in the syntax also implicated in other kinds of cognitive manipulations. In other words, getting the grain right leads us to expect that Broca’s area should light up quite often. Or as P&E put it:
The natural assumption is that the differently structured cortical areas are specialized for performing different types of computations, and that some of these computations are necessary for language but also for other cognitive functions (10).
P&E also address the OIP and suggest that ‘computation’ is where linguistics and neurology might meet to unify into a serious neurolinguistics. They note that this first requires dumping “terms like ‘psychologically real’ or ‘neurologically real.’” Why? Because, these notions mislead. How? They “imply that there is some other type of reality to linguistic computations beyond being computed in the brain.” (Yes! Yes! Yes!) Linguistic analyses, they insist, are proposals about the “computations/representations that are computed in the minds/brains of speakers.” So understood the envisioned job of neurolinguistics is to identify how “these computations are implemented” in neural hardware (12). In other words, what linguistics study and what neuro-linguists study is one and the same object described in different ways. The aim is to unify these different descriptions and this requires dispensing with the silly distinction between true and psychologically/neurally real.
Poeppel suggests that right “place” for this unification is at the level of the ‘circuit.’ Circuits can be described functionally (this circuit adds two quantities, this one takes the difference between them) as well as physiologically (two and-gates plus a not-gate) and anatomically (current flows here when open, there when closed). This is congenial to certain minimalist ways of talking (e.g. here) wherein the prize goes to s/he who finds the right basic circuits (and their wiring) that underlie the complex operations of the grammar (e.g. binding, phrase structure, movement). At any rate, maybe ontology and granularity can happily join hands at the computational circuit.
The above just touches lightly on the content of these papers. Poeppel also elaborates on the useful distinction between the map vs mapping problem. The latter is closely related to the OIP issues discussed in P&E. with additional instructive bells and whistles. Let me end with a sample quote to whet your appetites for more (and to show that I stole the ‘circuit’ talk from David P) (p.35):
In a typical cognitive neuroscience study …participants will engage in some…task…while their brain activity is monitored. The analyses show that some area or areas are selectively modulated, and it is then argued that activation of a given area underpins, say, phonological processing, or lexical access or syntax. …[I]t is fair to say that the canonical results –very much at the center of current research- are correlational…but we have no explanation, no sense of which properties of neuronal circuits that we understand account for the execution of function. How to proceed? …[W]e [should] decompose the cognitive tasks under investigation into computational primitives that can be related to local brain structure and function, in a sense instrumentalizing the computational theory of mind…more aggressively.
And let us all say: Amen!