Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The Language of Thought is not Language: Evidence from Formal Induction
Poster D69 in Poster Session D, Saturday, October 26, 10:30 am - 12:00 pm, Great Hall 4
Hope Kean1, Alexander Fung*1, Paris Jaggers*2, Joshua Rule3, Yael Benn4, Joshua Tenenbaum1, Steve Piantadosi3, Rosemary Varley2, Evelina Fedorenko1; 1MIT, 2UCL, 3UC Berkeley, 4Manchester Metropolitan University
The human capacity for inductive reasoning lies at the heart of our ability to learn and think about the world around us, as well as engage in formal thought (e.g., mathematical reasoning). The format of the mental representations that mediate thought remains debated: thoughts can be expressed symbolically, schematically, diagrammatically, in mathematical expressions, but can also be cast into a linguistic format. According to one popular hypothesis about the representational format of thoughts—the “language of thought” (LOT) hypothesis (Fodor, 1975)—thoughts are built out of smaller atomic pieces, just like programs are built out of a small collection of operations, or sentences are built out of words. Some have further explicitly argued that the language of thought is natural language (Chomsky, 1993, 1995; Davidson, 1967, 1975; Dennett, 1991, 1996, 2017; for earlier claims, see Wittgenstein, 1921). We here challenge this claim using a combination of neuroimaging and behavioral investigations of patients with aphasia. In both studies, we use a formal rule induction paradigm (Rule et al., 2024). Participants are presented with an input number list and told what the output list looks like (e.g., 1 4 7 → 7 4 1); their task is to infer the underlying transformation rule. They can then test their hypothesis on a new input list, until they guess the correct rule. The rules can involve a combination of mathematical, list, and structural operations, and require some formal representation (in e.g. first-order logic or lambda calculus) to solve. In Experiment 1, we measured healthy adults’ brain responses using fMRI while they performed the rule induction task; in the control condition, participants were told what the rule was and simply had to apply it to the input list. The response in the language brain areas (Fedorenko et al., 2024) during the critical rule induction condition were low, close to the low-level baseline. Instead, another system—the Multiple Demand system (Duncan et al., 2020)—showed robust responses to the rule induction task, and stronger responses during the induction, compared to the application, condition. Thus, the language areas in healthy adults are not engaged much during formal rule induction. Next, in Experiment 2, we asked whether the rule induction task can be performed without the language system by testing two individuals with global aphasia. These participants have sustained massive damage to the Perisylvian language cortex and display severe language difficulties in both comprehension and production. In spite of this severe linguistic impairment, one participant (age=50) performed better than a control group of n=40 age-matched participants on the rule induction task, and the other (age=78) performed similarly to the controls. Moreover, both participants with aphasia were able to nonverbally communicate (with symbols, numbers, and gestures) a subset of the rules which they had inferred. Overall, our results demonstrate that the left-lateralized fronto-temporal language network is not significantly engaged in and not necessary for the induction of formal rules, thus falsifying the variant of the LOT hypothesis, whereby natural language is the medium of thinking and reasoning.
Topic Areas: Disorders: Acquired, Syntax and Combinatorial Semantics