Mind, Matter, and Meaning

Millikan

Ruth Millikan (b. 1933) spent most of her career at the University of Connecticut working on problems most of her contemporaries considered either solved or intractable, which is usually a sign that they are neither. Her Language, Thought, and Other Biological Categories (1984), among the most original and technically demanding works in modern philosophy of mind, argues that representation and meaning are biological phenomena, grounded in natural selection and the proper functions selection has assigned to cognitive mechanisms. This is teleosemantics, and Millikan founded it.

Major Works

  • Language, Thought, and Other Biological Categories: New Foundations for Realism (1984) — the foundational statement of teleosemantics
  • “Biosemantics” (Journal of Philosophy, 1989) — the canonical short statement; the paper most frequently cited
  • White Queen Psychology and Other Essays for Alice (1993) — collected essays developing and defending the teleosemantic framework
  • On Clear and Confused Ideas (2000) — application of teleosemantics to concepts and the architecture of thought
  • Beyond Concepts: Unicepts, Language, and Natural Information (2017) — later work on mental categories

Key Positions

Proper functions. Millikan’s foundational concept: a mechanism has a proper function if it was selected for (by evolution, by individual learning, or by cultural transmission) because it performed that function in the history of the lineage. A heart’s proper function is to pump blood, not to make sounds, even if it makes sounds reliably. An artifact like a thermostat gets its function derivatively, from the purposes of its designers. The proper function of a cognitive mechanism is what it was selected to do, not what it merely does as a matter of fact.

Biosemantics and representational content. A representation receives its content from the proper functions of the producer-consumer organization in which it operates. The familiar frog case shows both the attraction and the difficulty. Consumer systems may privilege content such as frog food, moving prey, or fly over the proximal description small moving dark spot, because their historically selected use answers to what the signal helps the animal do. The exact grain does not follow from selection alone. Misrepresentation becomes possible once a correspondence rule fixes conditions under which a signal can count as accurate even when the consumer fails, or false even when the response happens to succeed.

The consumer-system grounding of content. A crucial feature of Millikan’s account: content is fixed not by the producer of a representation but by its consumer. This means that what a representation is about is determined by what systems down the processing chain were selected to do with it. The content is the world-feature that, under normal conditions, would make the consumer’s response appropriate. This shifts the locus of semantic grounding from the signal to the system that exploits it.

Pushmi-pullyu representations. In addition to purely descriptive representations (which say how things are) and purely directive ones (which say what to do), Millikan identifies a third class: representations that are simultaneously descriptive and directive. A bee’s dance represents both where the nectar is (descriptive) and where to go (directive). Many perceptual-action states have this dual character. This framework illuminates the action-guidance role of perception in ways that more traditional content theories miss.

My View

Millikan supplies the principal teleosemantic framework behind the book’s naturalization of content. Producer-consumer history, proper function, and a correspondence rule can explain how a state acquires accuracy conditions without an interpreter assigning them at the moment of use. The framework supports a qualified reference theory by showing how physical organization can become answerable to a world.

That route does not make every selected or trained mechanism a source of original content. A mechanism carries original content to the extent that its history and producer-consumer use fix accuracy conditions not exhausted by an interpreting practice. Mallory 2026 Teleosemantics for Neural Word Embeddings.md|Fintan Mallory's word2vec analysis supplies a serious account of content about actual word-types and their context distributions. Whether that content is original at this limited grain remains unsettled. The vocabulary, corpus, task, and loss descend from public linguistic practice, but that ancestry alone does not establish semantic dependence. No further proprietor relation settles the issue.

Mechanism content also settles no larger attribution by itself. Integration, correction, memory, planning, and control bear on whether a whole system understands; perceptual mode, attention, and poising bear on phenomenal character and state consciousness. Millikan helps answer what a state represents and how it can misrepresent. The remaining questions require additional arguments from externalism, system organization, and the identity claim.

See Teleosemantics, Reference, and Burge.