I came across this email I sent to David Poeppel in July 1998 that sketches some of the basic ideas that congealed into our dual stream model. You can see the starting point was David's prior work on word deafness and my thinking about how conduction aphasia fits in. It's also clear that Nina Dronker's work on the insula was prominent at the time and in fact I had been corresponding with Nina about some of those details. The comment at the end "Nina says grammar is in the anterior STS" was a reference to a presentation she gave on lesion correlates of sentence comprehension deficits that implicated the ATL. David and I later convinced her to publish the study in our guest edited volume of Cognition in 2004.
Anyway, here's the text of the email:
ok, so the prediction that pure word deaf patients should have paraphasic speech seems to have panned out. If the other prediction that PWD patients who had a left hemi infart first would have been classified as conduction aphasic before the right hemi lesion, then it might be interesting to further test the idea using rTMS. Find a conduction aphasic with the left auditory cortex lesion, interrupt right hemi posterior auditory cortex function using rTMS and we should have an instant PWD patient with musical perception preserved!
Then all we have to do is figure out what the relation is between the posterior superior temporal plane, the anterior insula (a la Dronkers), Broca's area, and the SMG. Damage to any of these areas seems to produce some sort of phonemic-like deficit, whereas damage to more lateral-posterior temporal lobe areas seems to produce more semantic type deficits (e.g., semantic paraphasics in Wernicke's and transcortical sensory aphasia). Ok, I got it: phonemic processing (undefined) is carried out in a network in the post. sup. temporal plane bilaterally, this network interfaces (1) with lexical-conceptual systems in the inferior temporal lobe (and elsewhere) via lateral posterior temporal lobe areas (wernicke's area, angular gyrus), (2) with phonetic-motor articulatory systems in Broca's area via the anterior insula, and (3) with conscious phoneme-based working memory systems in frontal and parietal lobes via the supramarginal gyrus (which is connected with frontal systems, I think). Thus, the anatomical model of language is reduced to interface necessities (might I say, limited to those of virtual conceptual necessity) with: (1) motor articulation systems, (2) conceptual-semantic systems, and (3) known working memory systems. Find a place for grammar and we're home free! Oh, Nina says grammar is in anterior STS, fine we'll go with that.
Sounds like a pheory to me.