The Cortical Dynamics of Intelligible Speech

This is the title of a new paper in J. Neuroscience by Alexander Leff and company (Jennifer Crinion, Karl Friston, and Cathy Price among others) at the Wellcome Trust Centre, University College London. The report is beautifully straightforward and fills an important gap in our understanding of the pathways that support the processing of meaningful speech.

They set out to test two competing hypotheses regarding information flow in the temporal and frontal lobes during the processing of intelligible speech. One hypothesis, put forward by Sophie Scott and Richard Wise, suggests that the pathway for intelligible speech projects anteriorly into the temporal lobe from primary auditory cortex. The other hypothesis, recently promoted by us (Hickok & Poeppel, 2000, 2004, 2007), but by no means unique to us (it is a rather conventional view), holds that the posterior STS is an important projection target for acoustic speech information on its way to being comprehended.

Leff, et al. used fMRI to identify a network of brain regions active during the perception of intelligible speech, which was defined as regions that responded more to word pairs than to time reversed versions of word pairs. Here is a summary map of the regions activated by this contrast:


They didn't see much bilateral activation (must be something in the London water because we have just finished a similar experiment and see TONS of activation on both sides -- more on this in the future), but that's not the point of the paper. Notice that there are foci of activation in the posterior as well as anterior STS, and an inferior frontal area as well that falls within BA47, outside of Broca's region.

They then used dynamic causal modeling and Baysian parameter estimation to determine the model of information flow among these three nodes that best fit their data. Of 216 models tested -- all possible combinations of input (squares with arrows) and interactions between ROIs (dotted lines), diagram on left -- the winning model (right diagram) had sensory input entering the network only via pSTS and projecting in separate pathways to aSTS on the one hand and IFG on the other.


In other words, information flow is not exclusively anterior from primary auditory cortex, nor is it flowing in parallel from A1 to aSTS and pSTS, but rather projects first posteriorly and then anteriorly within the temporal lobe; i.e., the ventral stream runs through the pSTS.

In proposing an exclusively anterior-going pathway from primary auditory cortex, Scott and Wise were particularly persuaded by three observations. (i) monkey data suggested anterior projections from the auditory core, (ii) their own imaging data suggested an anterior focus of activity for intelligible versus unintelligible speech, and (iii) semantic dementia involves word level semantic deficits and has anterior temporal degeneration as a hallmark feature. Their proposal was quite reasonable in light of these facts, but it just didn't seem to pan out: (i) monkey data is useful as a guide, but may not generalize to humans especially when language systems are involved, (ii) subsequent experiments looking at intelligible speech, such as the present one, clearly identified posterior activation foci, and (iii) it seems that the deficit in semantic dementia is to some extent supramodal, i.e., may be well beyond the linguistic computations that appear to be supported by the pSTS, and lesion (stroke) evidence implicates posterior temporal regions in word-level semantic deficits.

To be fair, we didn't completely predict the findings of the Leff, et al. study either. Specifically, we posited no direct projection from pSTS to aSTS, and discussed the function of the anterior temporal region in the context of grammatical type processes only. Neither did we discuss a direct influence of pSTS on the IFG (BA47) within the ventral stream. (Notice that this link does not, presumably, reflect the dorsal stream, which involves more posterior portions of the IFG and should not be a dominant node in network supporting language comprehension.)

Know that we know a bit more about the nature of information flow in this network, it's time to try to figure out exactly what these different regions might be doing. Our suggestion regarding the posterior STS is that it supports phonological processing of some sort. This still makes sense I think. But what is the anterior STS doing?

References

A. P. Leff, T. M. Schofield, K. E. Stephan, J. T. Crinion, K. J. Friston, C. J. Price (2008). The Cortical Dynamics of Intelligible Speech Journal of Neuroscience, 28 (49), 13209-13215 DOI: 10.1523/JNEUROSCI.2903-08.2008

Sophie K. Scott, C. Catrin Blank, Stuart Rosen and Richard J. S. Wise (2000) Identification of a pathway for intelligible speech in the left temporal lobe. Brain, Vol. 123, No. 12, 2400-2406