Two people who had lost nearly all ability to understand or produce language sat down with a set of number puzzles. They had to work out the hidden rule that turned one list into another, then apply it to new examples. They performed as well as people whose language was intact, and they communicated the rules they had worked out through gestures or a sketch.
That result, published in the Proceedings of the National Academy of Sciences, is a pointed piece of evidence in a long argument about whether humans need words to think. The MIT team behind it says the answer is no.
Taking Language Out of the Equation
The question is genuinely hard to study, because researchers normally use language to explain a task and collect an answer. To get around that, the group led by Evelina Fedorenko at MIT's McGovern Institute for Brain Research worked with Rosemary Varley, a neuroscientist at University College London who studies acquired language disorders.
The two patients had suffered strokes damaging language-processing regions, leaving them with severe impairment in both comprehension and production. The researchers designed logic games that required no language at all. One asked participants to infer the rule connecting two lists of numbers, such as reversing the digits or removing values above a threshold, and then apply it. Another presented a set of geometric patterns and asked participants to identify the one completing the matrix.
The puzzles got progressively harder. The patients kept up.
What the Scanner Showed
The behavioral half of the study establishes that language is not required. The imaging half addresses whether the language system gets used anyway.
Neurotypical participants came to MIT for MRI scans while completing logic games similar to the patients' tasks, as well as syllogistic problems using if-then statements. They also performed tasks that mapped, individually for each person, the location of their language-processing regions and their multiple demand network, a distributed system that supports complex problem solving.
The language network stayed out of it. Scans showed it was not engaged during inductive reasoning, when participants worked out hidden rules, or during deductive reasoning, when they judged whether a conclusion followed from premises. Researchers varied puzzle difficulty and compared inferring a rule against merely applying a given one, and the separation held.
One result surprised the team. The multiple demand network, which many scientists expected to support logical reasoning, was active during inductive reasoning but did not appear to participate in deductive reasoning. Deduction instead drew on regions distinct from both networks, a thread that first author Hope Kean, a postdoctoral researcher in Fedorenko's lab, is following in ongoing work.
The findings extend earlier results from the same lab showing that object categorization and social reasoning also proceed without the language system.
Why This Matters for People with Aphasia
About 2 million people in the United States live with aphasia, according to figures cited by the NIDCD from the National Aphasia Association. Stroke is the leading cause. Speech-language pathologists have long understood that losing language does not mean losing intelligence. People with aphasia manage household finances, play chess, and finish sudoku puzzles. The problem is that everyone else routinely reads difficulty speaking as difficulty thinking.
In a statement from MIT, Fedorenko said the work adds to a growing body of research establishing that even severely aphasic people can preserve their capacity for abstract logical thought. Linguistic difficulty, whether from aphasia, stuttering, or speaking English as a second language, is "not indicative of how smart or capable someone is," she said.
That is the practical takeaway for families and clinicians, and it is not a small one. Assumptions about cognitive capacity shape how much autonomy a person retains after a stroke, how they are addressed, and which decisions they are allowed to participate in.
The Sample Size Problem
The clinical evidence rests on two patients. That number needs to sit in plain view.
Aphasia severe enough for this design is uncommon, and patients who can complete extended nonverbal testing are even rarer, so small sample sizes are typical in this literature. But two people cannot support claims about aphasia in general, about milder forms, or about individual variation in how reasoning survives brain damage. The MIT team does not make those claims.
What the lesion approach does offer is something imaging cannot. If a brain system is severely damaged and a function remains, the argument that the system is required for that function weakens considerably. Pairing that with scans showing the language network sitting idle in healthy brains during the same tasks is what makes the combined case stronger than either half alone.
The researchers also note an implication outside medicine. Large language models are trained entirely on text and produce text, yet they can convincingly simulate certain kinds of human reasoning. If the human brain keeps language and abstract logic in separate systems, the difference between those architectures may be worth studying rather than glossing over. The lab has documented the language-versus-thought distinction across multiple studies over more than a decade.
Key Questions Answered
What did the study find?
People with severe language impairment from stroke solved logic puzzles as well as unimpaired controls, and brain scans showed the language network was not engaged during reasoning in healthy adults.
How many patients were tested?
Two. Both had strokes damaging language-processing regions, leaving severe impairment in understanding and producing language. The imaging portion used a separate group of neurotypical participants.
How do you test reasoning without language?
The researchers built nonverbal puzzles using numbers and geometric patterns. Participants inferred hidden rules and applied them, sometimes communicating answers through gestures or drawings.
What was the unexpected result?
The multiple demand network, long suspected of handling logical reasoning, was engaged during inductive reasoning but appeared uninvolved in deductive reasoning.
Why does this matter for aphasia?
It supports what specialists already know, that losing language does not mean losing intelligence, and gives clinicians and families evidence against treating communication difficulty as impaired thinking.
Does this apply to all forms of aphasia?
No. The clinical sample was very small and included severe acquired impairment, so the results cannot be generalized to milder forms or used to predict individual recovery.
