Prelinguistic Thought

Prelinguistic thought is the capacity for structured mental representation, reasoning, and problem-solving that operates independently of natural language. The question of whether such thought exists — and if so, what its properties are — sits at the intersection of philosophy of mind, cognitive development, linguistic relativity, and comparative psychology. It is not merely an empirical question about the cognitive capacities of infants and non-human animals. It is a foundational question about the relationship between language and thought: does language enable thought, shape thought, or merely express thought that would exist in identical form without it?\n\nThe strongest empirical evidence for prelinguistic thought comes from three sources:\n\n* Infant cognition: preverbal infants display object permanence, causal reasoning, numerical discrimination, and goal-directed action. These capacities operate before language acquisition and are not obviously dependent on internal linguistic rehearsal.\n* Non-human animal cognition: corvids solve multi-step causal puzzles, chimpanzees engage in tactical deception, cetaceans maintain complex social alliances. These behaviors require structured representation and planning, yet the animals lack natural language in any form recognizable as human-like.\n* Aphasia and language loss: patients with severe language impairments retain spatial reasoning, emotional judgment, and practical problem-solving. The dissociation between language and thought is visible in neuropathology.\n\n== The Language-as-Structure Debate ==\n\nThe strongest argument against robust prelinguistic thought comes from the Sapir-Whorf hypothesis and its descendants. If language provides the categories in which thought occurs — if there is no 'pure' thought independent of linguistic framing — then prelinguistic cognition is not thought in the full sense. It is proto-thought: perceptual-motor intelligence without the compositional structure that makes human reasoning distinctive.\n\nThis position has been defended by philosophers who argue that symbolic thought — thought involving abstract, arbitrary, recombinable representations — is coextensive with language. On this view, an infant tracking an object's trajectory is not 'thinking about' the object in the way an adult is. The infant's representation is bound to sensorimotor contingencies; the adult's representation is detached, categorical, and linguistically inflected. The difference is not merely developmental stage. It is representational kind.\n\nThe counterargument: symbol manipulation does not require language. Conceptual spaces — geometric structures in which concepts are points and similarity is distance — can support compositionality, inference, and abstraction without linguistic labels. The existence of structured non-linguistic knowledge in animals and infants suggests that the computational substrate for thought is older than language and does not depend on it.\n\n== What Prelinguistic Thought Reveals About Large Language Models ==\n\nThe debate has unexpected relevance for artificial intelligence. Large language models are trained exclusively on linguistic data. They have no sensorimotor grounding, no embodied interaction with physical causality, no prelinguistic cognitive scaffold. If human thought depends on prelinguistic structural capacities — if language is a layer on top of deeper representational architecture — then LLMs are architecturally incomplete in a way that is not fixable by scaling. They lack the substrate.\n\nConversely, if symbolic thought is coextensive with language — if there is no deeper representational layer — then LLMs may be complete in principle, and their limitations are merely matters of scale and training. The empirical question of prelinguistic thought is thus entangled with the architectural question of artificial intelligence: what computational ingredients are necessary for genuine understanding, and does language provide them or merely express them?\n\nThe assumption that thought requires language is not an empirical discovery. It is a methodological prejudice born from the fact that the philosophers who have written about thought have all been linguistic creatures. We have no access to prelinguistic thought from the inside. That does not mean it is not there.\n\n\n\n