| The present research addresses two fundamental questions regarding human reasoning: (a) what are the neural correlates of deductive reasoning; and (b) what is the role of language in deductive inference? In the past ten years both issues have been tackled by only a handful of studies. Results, however, have been inconsistent, possibly because of the variety of tasks used and suboptimal experimental design. We address these questions by adopting a cognitive load methodology comparing simple inferences to linguistically matched complex ones. We first construct four pairs of logical forms and show that, within each pair, the complex one is systematically more difficult to evaluate. We then employ these logical forms in two fMRI experiments contrasting metabolic response to simple versus complex deductions. In addition, to isolate brain regions sensitive to logical form from regions sensitive to lexical contents we presented the same arguments multiple times, varying their lexical content. In both experiments the complexity comparison uncovered a distributed network of regions that is disjoint from areas traditionally thought of as supporting linguistic processing and regions engaged by simple reading of the arguments. Further, while a set of regions appeared consistently engaged by deductive complexity, regardless of contents, other regions appeared to be selectively recruited by the lexicon used. A control experiment ruled out the possibility that the results are epiphenomenal to faster or prolonged reading for complex versus simple arguments. The distinction between regions recruited by logical structures and regions recruited by lexicon is further assessed with a novel multi-voxel pattern analysis approach. Finally, we examine inferences of semantic equivalence of a more "linguistic" nature, and detected no activity in the regions we propose to be crucial for deductive reasoning. These findings suggest that deductive inference relies on a distributed network of brain regions that is (i) language-independent and (ii) plausibly organized into three functionally distinct components. Namely, (a) content-independent "core" regions, embodying logical operations, (b) content-independent "support" regions, extracting and maintaining the logical form, and (c) content-dependent "support" regions that bind logical variables to specific lexicon. |
