Friday, November 6, 2009

The Pinker Panther Strikes Again - Recording from Broca’s Area

In a recent issue of Science (Vol 326, 445-449), Sahin, Pinker, Cash, Schomer, and Halgren summarize their findings from direct recordings from Broca’s region in patients undergoing presurgical epilepsy evaluation.

In an earlier, 2006, paper in Cortex, Sahin, Pinker, and Halgren reported fMRI data in participants doing one of the garden-variety past-tense tasks used often by Steve Pinker and his students and colleagues (the style of the experiment is something like this “visual cue: Yesterday I was in the park and ___. Target: to walk” -> participant produces “walked”). This 2009 paper is the fancier, intra-cranial recording companion piece (ICE, in their terminology, intracranial electrophysiology :-).

The piece represents something like the ‘harmonic convergence’ between the current enthusiasm for intracranial electrophysiological data (is anyone not doing this?), the long (historical) reach of Pinker’s past-tense-as-psycholinguistic-drosophila philosophy (yes, I remember having to read Pinker and Prince in grad school; and Greg even worked on some of this stuff!), and the growing interest in better cognitive neuroscience of language models.

Peter Hagoort and Pim Levelt provide a perspective in the same issue of Science (Vol 326, 372-373), largely because these data are directly linked to the Levelt production model. The numbers reported by Sahin et al. match nicely with Levelt’s production model (see, e.g. Indefrey & Levelt, 2004, Cognition) -- so the Max-Planck guys are certainly happy.

The centerpiece of the study -- recordings from three patients who also underwent fMRI scanning prior to electrode implantation -- concerns data from electrodes in Broca’s region, perhaps Brodmann’s area 45 (that point is not made with sufficient clarity). They identified in the electrode response three peaks, or rather a tri-phasic response. Across all three patients, there were peaks/valleys at ~200 ms, ~320 ms, and ~450 ms post-target onset. The first, 200 ms, peak was modulated by lexical manipulations (frequency), the second, 320 ms, peak by inflectional demands (grammatical manipulations), the third, 450 ms, peak by articulatory requirements. Based on these observations, they conclude (and this is the title of the article): “Sequential processing of lexical, grammatical, and phonological information within Broca’s area.”

The results are not particularly surprising. When presented with a word, it stands to reason that it has to be accessed/identified before it can be repeated … (Planning a word’s articulation before even seeing it would indeed be pretty novel.) Moreover, if any operation on the input representation is required prior to articulation, it is also not super-surprising that it would be temporally interposed between lexical access and articulatory planning/output generation. What would be the alternative? What is interesting in these data is that there is evidence for these stages in one very small region. Of course, many other regions will also play a role – here, by clinical necessity, only a small region can be investigated. What is not clear is whether the activation observed is functionally critical, i.e. whether the reported triphasic Broca’s region activity is necessary for the execution of these language tasks. If we want to conclude that Broca’s region provides the neuronal substrate for multiple different operations that participate causally in the execution of multiple language tasks – again, is there a credible alternative? – it would help to get a better sense of the role such localized frontal activation plays. In any case, the paper reflects the growing use of intra-cranial data in the study of language (see, e.g., the studies by Boatman, Crone, Knight, etc.)


Greg Hickok said...

I hope they controlled for phonotactic frequency because that could explain much of their findings.

Anonymous said...

Curious if anyone else views this as a glaring confound in the Sahin et al. experiment--

Regarding the interpretation of the contrast between the "read" and both of the "inflect" conditions: there is substantially less context in the read condition, compared to the inflect condition. By "context", I mean the presence of other concepts, and syntactic and phonological structure, in the task (i.e. the stuff before the blank in Fig. 1B)

How can they conclude that the reduced amplitude of the Channel A3-4 (Area 45) peak at 320ms is related to inflection, as opposed to the presence of less context in the read condition?

Dirk Koester said...

Even if the findings are valid, the main question remains, are these components functinoally critical? Or, what does it mean, if the same neural curcuits or brain areas are involved in distinct linguistic processes?

Even though ICE is certainly intriguing, I don't think it is fundamentally superior that rather "boring" scalp potentials (ERPs) as these also work nicely to estimate temporal relations in language production; overt production without written language materials as stimuli: