Logical reasoning does not depend on the brain’s language system

What is it about?

Scientists, linguists, and philosophers have debated for millennia about the relationship between language and reasoning. Is language merely a tool for expressing our thoughts, or is it the very medium in which complex thought takes place? Formal logical reasoning presents a powerful test of this question because language and logic both combine smaller “atomic” elements into structured representations according to codified rules: words into sentences according to the laws of grammar in one case, and propositions into valid inferences according to the laws of logic in the other. Although previous research had suggested that language is separable from several other forms of thought, the role of language in abstract logical reasoning remained surprisingly unclear. We therefore combined brain imaging in healthy adults with behavioral studies of individuals who had profound aphasia to ask whether the brain’s language system is either engaged in, or necessary for, logical inference. The central finding of our work is a striking dissociation. In the brain imaging experiments, the language network responded strongly to language, but showed little or no sensitivity to the additional demands of 3 different logical reasoning tasks (e.g. the language response was more than four times larger than the small response observed during induction). At the same time, the two participants with profound aphasia solved 76-98% of the complex induction problems, and scored 83-87% on difficulty matrix reasoning problems, performance that was typical or above average for their age group. The significance is that sophisticated logical reasoning can remain intact even when the capacity to understand and produce language is profoundly impaired. People often assume that complex thoughts take the form of an internal monologue, based in language. Our findings suggest that, at least for formal logical reasoning in adults, this is not strictly the case. The brain systems that process language are distinct from those that perform logical inferences. Language is extraordinarily useful for communicating ideas and in many cases externalizing our thoughts helps us directly organize or advance them, but the underlying reasoning itself appears to take place in a different representational format.

Featured Image

Photo by Joshua Sortino on Unsplash

Photo by Joshua Sortino on Unsplash

Why is it important?

The people with profound aphasia who made this work possible were extraordinary. In the clinic, many described the extreme difficulty they experienced communicating with people in everyday life, which is an unsurprising but devastating consequence of aphasia. It was deeply moving, then, that many had found ways to preserve a rich understanding and a way of conveying meaning with their closest loved ones, even when ordinary communication with less familiar people had become nearly impossible. It was a beautiful demonstration that when language is lost, the rest of the person, their thoughts, relationships, personality, and capacity for love and connection, remains. This ancient question has also taken on new importance with the rise of AI and computational perspectives on intelligence more broadly. Large language models (LLMs), like ChatGPT and Claude, are trained by predicting text inputs and express their outputs in text, yet they can convincingly emulate some/many forms of human reasoning. If the human brain instead separates language from abstract logical thought, understanding this architectural difference may help explain both the successes and characteristic failures of reasoning in LLMs, and could inform the design of future AI systems. Even though we ultimately discovered that this separation between language and reasoning is a feature of human cognition, that does not mean in principle that this is the only architecture capable of supporting complex thought. Alternatives are possible. Artificial systems, for instance, may achieve similar abilities using very different internal representations and computations. Our findings reveal the particular solution realized in the human brain, but not necessarily the only possible solution. The convergence between the two approaches was also important to us. Brain imaging allowed us to ask whether language areas participate in logical reasoning, while the aphasia cases allowed us to ask the stronger causal question of whether an intact language system is necessary for reasoning.

Perspectives