A recent report in Frontiers (link to it here) by my good friends Oiwi Parker Jones (first author) and Cathy Price (senior author) challenges the claim that area Spt is a sensorimotor integration area for vocal tract actions. Their attack comes from multiple fronts, both fMRI and lesion data. On the fMRI side they sought to determine whether Spt was more active during repetition tasks, particularly for pseudowords which demand sensory-to-motor translation, compared to two auditory naming tasks. One involved listening to animal sounds and naming the animal and the other involved listening to someone humming and naming the gender of the hummer. These tasks did not involve direct translation between a heard auditory code and a motor code and so shouldn't activate Spt as vigorously as the repetition tasks, they argued. The lesion data involved 8 patients with auditory repetition deficits were studied and their lesions mapped to identify the anatomical source of the problem. We have shown previously (Buchsbaum et al. 2011) that area Spt, as mapped using fMRI in 100+ participants, overlaps the lesion distribution of conduction aphasia (N=14).
What Parker Jones et al. found was that "No brain areas, including Spt, were more activated by auditory repetition of pseudowords compared to sound naming." They further found that the lesions associated with repetition deficits involved the arcuate fasciculus, not necessarily Spt.
They conclude, "the results were most consistent with Spt responding to bottom up or top down auditory processing, independent of the demands on auditory-to-motor integration... [and] most consistent with the neurological tradition that emphasizes the importance of the arcuate fasciculus in the non-semantic integration of auditory and motor speech processing." Back to 1960s, er 1870s, we go.
There are two main problems with this study that undermine their conclusions regarding Spt.
1. The used overt speech production. We have always used covert production to study Spt. Why? Because Spt responds to acoustic stimulation including the sound of one's own voice. Using overt feedback makes it very hard to separate sensory from motor-related activity in Spt. Case in point: primary auditory cortex showed exactly the same response profile as did Spt in the Parker Jones study: more activity during animal sound naming than pseudoword repetition (the animal sounds were longer). Therefore, this study measured primarily the acoustic response properties of Spt, not its sensorimotor properties.
2. The idea that Spt isn't particularly involved in naming (whether it's pictures, words, animal sounds, or genders) is incorrect. To be fair, before I fully came to grips with what Spt is doing computationally, I mostly thought of it as performing an auditory-to-motor coordinate transform. Pseudoword repetition would seem to be the most direct task to tap into this process. But I have always believed that Spt plays a role in producing words under any speech production condition. And more recently, I have argued (see here and here) that Spt is part of the speech motor planning process, essentially running not only in the auditory-to-motor direction, but also doing a motor-to-sensory forward prediction to detect motor planning errors, i.e., motor plan-to-auditory target mismatches, prior to speech generation. This, I argue, occurs for every speech act, including animal sound naming. So the manipulation that Parker Jones et al. used does not pit auditory-motor integration against no auditory-auditory motor integration. Instead it pits two auditory-motor integration tasks against one another. At best, we might be able to learn something about Spt if the two tasks differ in their integration load.
With this in mind let's look a little closer at the pseudoword versus animal sound naming results. I think it is a reasonable assumption that pseudoword repetition places greater loads on the Spt circuit than animal naming. First, there is no other route, through which the task can be achieved. Second, pseudowords are by definition lower frequency than real words and therefore more likely to induce motor planning errors and more likely to require extra error detection and correction work. Third, conduction aphasics, which according to my model have Spt damage, in general have more difficulty with non-word repetition. So why in the Parker Jones study does animal sound naming activate Spt more than pseudoword repetition? It doesn't, in contrast to what they claim, if we factor out the acoustic response (see problem #1). Here are some numbers, generated by eyeball from their graph:
Auditory cortex activity to animal sound naming: 6.9
Auditory cortex activity to pseudo word repetition: 4.0
*p-word repetition to animal sound naming ratio: .58
Spt activity to animal sound naming: 3.7
Spt activity to pseudo word repetition: 2.9
*p-word repetition to animal sound naming ratio: .78
The relative activation in the pseudo word compared to sound naming condition increases by 20% in Spt versus auditory cortex. In relative terms then (i.e., with the baseline auditory activity factored out), Spt activates more to pseudo words than animal sound naming. This fact is obscured in absolute terms because the acoustically-driven activation dominates the response profile.
What about the lesion data. I have no reason to question their findings, other than on the basis of the small sample size. I can report informally that we have also been collecting data on repetition ability in a much larger lesion sample and Spt (and surrounds) is robustly implicated even when white matter involvement is factored out. We are working on the manuscript now.
In short, Spt's demise as a sensorimotor interface is greatly exaggerated. Which is not to say that I believe we completely understand what Spt is doing. In fact I think it is a bit more nuanced and interesting. I would show you the data if only I could convince an NIH panel to fund the research.