Merleau-Ponty argued that language is not a system of signs that encode pre-existing thoughts but an expressive gesture of the body-subject — an act through which meaning is created, not merely transmitted. The speaker does not first formulate a thought in some pre-linguistic mental medium and then search for corresponding words. The speaker thinks through speaking — through the rhythm, weight, and felt trajectory of language itself, through the motor activity of articulation that carries the thought forward into regions the speaker did not fully anticipate. This analysis strikes at the deepest assumption of both linguistics and the philosophy of mind, and it has direct implications for what AI systems do when they generate text: they arrange tokens according to statistical distribution, without the embodied intentionality that constitutes expression.
The expressive gesture account of language was developed most fully in the chapters on speech in Phenomenology of Perception and in the posthumous fragments published as The Prose of the World. Merleau-Ponty drew on linguists like Ferdinand de Saussure but departed from the structuralist framework by insisting that language's fundamental character is not systemic but performative — an activity of embodied beings engaged in the creation of meaning.
The distinction between empirical and creative speech is central. Empirical speech is the routine use of established expressions — conventional phrases, ready-made sentences. Creative speech is the production of new meaning — expression that says something not said before, that reaches beyond established formulations into territory where thought and language are being forged simultaneously.
For AI analysis, the distinction illuminates what language models do. They produce text that is syntactically correct and semantically coherent — often indistinguishable from human writing on the surface. But the production involves no motor apparatus, no felt rhythm, no embodied intentionality. The text is arranged by probability distribution over tokens, not by the body-subject's expressive engagement with language as a medium.
A 2025 paper in AI and Ethics put the point precisely: algorithmic conceptions of language, 'whether symbolic or statistical,' are 'historically derived from, and dependent upon, a prior field of embodied expression that they cannot fully exhaust.' The models extract patterns from the sedimented outputs of embodied expression. What they cannot extract is the expression itself — the motor, temporal, situated activity of body-subjects engaged in the act of creating meaning.
Merleau-Ponty's analysis of language developed across his career, building on Husserl's phenomenology of expression and engaging critically with Saussurean structuralism. The central chapter on speech in Phenomenology of Perception established the foundations; the posthumously published lecture notes and the unfinished The Prose of the World developed the implications.
The account anticipates much contemporary work in gesture studies, embodied cognition, and the phenomenology of language — fields that have accumulated substantial empirical evidence supporting Merleau-Ponty's claim that speech production involves the motor system at every stage, not merely as an output mechanism but as constitutive of thinking itself.
Language as act, not code. Speech is expressive gesture, not the encoding of pre-linguistic thought in conventional signs.
Thinking through speaking. The speaker does not formulate thought and then express it. The motor activity of speech is itself the thinking.
Empirical vs. creative speech. Routine use of established expressions versus the creation of genuinely new meaning through language.
Embodied temporality. Expression unfolds in time through the body's motor engagement with language as a medium.
Diagnostic for AI. Language models produce text through statistical distribution, without the embodied intentionality that constitutes expression.