Do we need words to think? For centuries, philosophers and linguists have leaned toward yes, the idea that language is the scaffolding of reasoning itself. A new study from MIT’s McGovern Institute for Brain Research says otherwise: logical reasoning and language live in separate systems in the brain, and one can function perfectly well without the other.

Testing logic without words

Led by MIT associate professor Evelina Fedorenko and postdoc Hope Kean, the research team faced an obvious problem: it’s hard to separate language from thought when your test subjects are, well, using language to think. So they teamed up with neuroscientist Rosemary Varley at University College London, who studies people with acquired language disorders.

Two stroke patients with severe aphasia, meaning their ability to understand and produce language was badly damaged, were given language-free logic puzzles. In one task, they had to spot the hidden rule linking two lists of numbers, such as reversing digits or filtering out values above a threshold, then apply that rule to new examples. In another, they had to complete a matrix of geometric patterns.

The result: both patients solved the puzzles just as well as a control group with no language impairment, communicating their answers through gestures or sketches. Logic, it turns out, doesn’t need a vocabulary.

What brain scans show

The team also scanned the brains of neurotypical volunteers while they worked through similar logic games, plus “if-then” syllogisms (“if the ball is red, it’s big. The ball is red. Is it big?”). The scans mapped each participant’s language network and their “multiple demand network,” a separate system linked to general problem-solving.

The language network stayed quiet during both inductive reasoning (figuring out a hidden rule) and deductive reasoning (checking whether a conclusion follows). Interestingly, the multiple demand network lit up for inductive tasks but not deductive ones — a distinction Kean is now digging into further.

This fits a growing pattern from Fedorenko’s lab: other high-level cognitive tasks, like categorizing objects or reasoning about social situations, also don’t appear to rely on the brain’s language centers.

Why it matters

For people with aphasia, this is more than an academic point. Clinicians have long known that losing language doesn’t mean losing intelligence, people with aphasia can still play chess, solve sudoku, or manage household finances. But it’s easy for others to mistake communication difficulties for a decline in thinking ability. This study adds hard evidence that abstract logical reasoning can stay fully intact even when language is severely impaired.

There’s also a curious angle for AI. Large language models are trained entirely on text and generate text, yet they simulate forms of reasoning fairly convincingly. If the human brain keeps logic and language on separate tracks, comparing that architecture to how language models operate could offer useful clues for building future systems.

Kean calls this a new frontier in mapping the geography of thought, and after these findings, it’s easy to see why.

Original paper: Evidence from formal logical reasoning reveals that the language of thought is not natural language (PNAS, 2026)