Commentary on “The role of domain-general cognitive control in language comprehension” by Fedorenko

Fedorenko (2014) examines “the relationship between high-level language processing and domain-general cognitive control, with a focus on the brain systems that support these cognitive capacities (ital hers).” She addresses two questions –“(i) when (i.e., under what circumstances) the cognitive control mechanisms get engaged during language understanding; and (ii) whether this engagement is necessary for comprehension.” The first part of Fedorenko’s paper presents data that she says shows that (1) a neural “language system” is stable within and across individuals, time, modality of presentation, and language, and that (2) a “multipledemand (MD) system” that is activated by tasks that contrast conditions that vary in difficulty (an executive control system) is anatomically separate from the language system. This forms the backdrop to a far-ranging discussion of the two questions in the second part of the paper. As Fedorenko’s emphasis is on the contribution of neural data to answering these questions, my commentary will focus on the relevance of the BOLD signal data reported in Part 1 to them. I note at this point that there are issues about the results Fedorenko presents in Part 1 of her paper. For instance, the within-subject stability of the activations produced by Federonko’s language localizer across time is documented in only four individuals—higher than the industry standard of zero for the number of participants tested twice for a BOLD signal effect, but not a large number. However, my comments will accept Fedorenko’s results and ask what they show. There are two functional neuroanatomical findings reported in Section 1 of Fedorenko’s paper that could bear on her questions. The first is the finding that tasks that activate the MD/executive system do not activate the language system. I do not think this result is relevant to the questions posed. The crucial question is whether the executive control system needs to activate the comprehension functions localized by Fedorenko’s language localizer when it performs executive functions in the tasks that Fedorenko used to localize it. As far as I can see, it does not. Fedorenko localized the high level language comprehension system by subtracting BOLD signal associated with item recognition in lists of non-words from that associated with recognizing words in sentences (and, separately, by contrasting passive listening to these stimuli), and localized the MD system by varying complexity of arithmetic processing and working memory and with the Stroop task. The executive system does not regulate operations that perform high level language comprehension when it is involved in arithmetic processing and Stroop (whether it does so when it performs WM tasks is a matter of debate; Caplan and Waters, 1999, 2013). In Stroop, for example, the executive functions needed in the interference condition and not in the baseline conditions involve lowering activation of the word derived from print relative to that of the name of the color in order to select a word for production. The aspect of language processing that is regulated is the word production system (Roelofs, 2008), not high level comprehension. The fact that Stroop does not activate areas identified by Federorenko’s language localizer therefore does not imply that executive functions do not control language operations in language comprehension tasks. The second finding that might inform the questions Fedorenko poses is the reverse aspect of the non-overlap—that Fedorenko’s language localizer does not activate the MD system. However, this direction of inference also fails. Power aside, the absence of a difference in BOLD signal in executive areas in the contrast of an experimental and a baseline task does not show that executive functions are not needed in the experimental task; it only shows that they are not used to a greater extent in the experimental than in the baseline task (Caplan, 2009). The absence of an effect of the [sentence-nonword] condition on MD areas is entirely consistent with the MD/executive system performing operations critical to high level comprehension but performing the same operations, or ones that require the same degree of executive control, in the baseline task. Turning to more general issues, there is an important gap in Fedorenko’s discussion—she does not present a framework for understanding the relation between executive control operations and the operations performed by domainspecific processors in the performance of a task. Further consideration of the effects of the language localizer on BOLD signal in MD areas shows why such a model is needed. Fedorenko found that the [nonword-sentence] contrast activated