Logical Reasoning Survived Severe Aphasia, and Recruited No Language Regions in Healthy Brains
Two patients whose strokes left them severely impaired in both understanding and producing language solved formal logic problems as well as unimpaired controls. In a parallel imaging arm, the brain’s language network stayed inactive during both inductive and deductive reasoning. The finding matters less for what it settles about thought than for what it implies about every cognitive instrument that asks a patient to explain the answer.
Two-arm study. The behavioural arm tested two patients with severe acquired aphasia following stroke, against an unimpaired control group, on language-free reasoning tasks: inferring the hidden rule that transformed one list of numbers into another, then applying it to fresh examples, and selecting the pattern that completed a geometric matrix. The imaging arm scanned neurotypical adults during similar induction tasks plus syllogistic deduction, with separate localiser tasks mapping each participant’s language network and multiple demand network, and task difficulty varied to test load sensitivity. Led by Evelina Fedorenko at MIT’s McGovern Institute with Hope Kean as first author, in collaboration with Rosemary Varley at University College London. Published in PNAS, 8 July 2026.
- Both patients with severe language impairment matched controls as the puzzles grew harder, and conveyed the rules they had inferred through gesture or sketch. Kean frames this as overturning the position that symbolic rule induction requires linguistic capacity.
- In neurotypical adults the language network showed no engagement during either reasoning type: not inductive reasoning, where participants discovered a hidden rule, and not deductive reasoning, where they judged whether a syllogistic conclusion followed.
- The multiple demand network, which many had expected to carry logical reasoning, engaged during induction but appeared not to participate in deduction. The authors flag this as unresolved and under active investigation.
- The result extends earlier work from the same lab reporting that object categorisation and social reasoning also run without recruiting language regions.
Cognitive endpoints in neurology trials are almost entirely language-mediated. Instruments in routine use ask the participant to hear or read an instruction, hold it verbally, and answer in words or writing. Every one of those steps runs through the system these patients had lost.
If reasoning and language are separable, a language-mediated instrument cannot distinguish a patient who cannot reason from a patient who cannot report. The two score identically and the trial records cognitive impairment either way.
That has three practical consequences. Populations with acquired or progressive language impairment get screened out of trials as unable to complete assessments, which removes exactly the patients whose reasoning the instrument mismeasures. Endpoint movement in stroke recovery and aphasia programmes may track recovered expression rather than recovered cognition. And capacity determinations made on verbal assessment carry an assumption this work puts under pressure. Anyone running a cognitive endpoint in a language-impaired population should establish which of the two it is actually measuring.