Graves et al. have reported a pretty cool study in JoCN examining the cortical regions involved in phonological access in picture naming (Graves, Grabowski, Mehta, & Gordon, 2007, JoCN, 19, 617-631). The goal was to use the word frequency effect (WFE) as a way of identifying phonological word form access. The assumption is that the WFE -- naming times are longer for lower frequency words -- reflects phonological word form access during production, and therefore regions that show greater activity for lower frequency words are those involved in phonological access. Because frequency is highly correlated with other variables such as word length and concept familiarity, the authors also quantified these variables and attempted to factor them out of the WFE analysis. Although I'm not completely convinced that these other variables are completely controlled, the findings are nonetheless interesting and worth paying attention to. Basically, they found that the WFE was associated with three main regions, one in the inferior frontal gyrus, one in the ventral occiptial-temporal cortex, and one in the posterior temporal gyrus. Only the pSTG location however, was specific in its sensitivity to frequency -- the other two regions were, in addition, sensitive to concept familiarity.
The authors conclude that "the left pSTG is specifically involved in accessing lexical phonology" [in picture naming] p. 629, and this seems to be a reasonable conclusion that is consistent with, for example, Indefrey and Levelt's (2004) meta-analysis conclusions.
One question I have, though, is whether this is THE region involved in accessing lexical phonology in production, or whether it is part of a circuit that participates in this process under "high-load" conditions. The region they identified (-51, -37, 20) precisely corresponds to area Spt (see Brad Buchsbaum's comment on the "Where is area Spt" blog entry). We proposed in our NRN paper that the dorsal stream network including area Spt is involved in the production of low frequency words, because sensory-guidance is required during motor sequence programming, whereas higher frequency words may be stored as motor chunks that can simply be activated without Spt mediation. The WFE found in area Spt is consistent with this claim. This view does leave open the question of what circuits are involved in naming of high frequency items. Perhaps STS regions? Any ideas?