Semantic Grounding

Semantic grounding is the problem of how symbols, words, or representations acquire their referential connection to the world — how a word like 'tree' comes to refer to actual trees rather than merely to other words or internal states. It is the bridge between the formal combinatorics of a symbol system and the empirical reality it purports to describe.

The problem is most acute in artificial intelligence. A large language model can generate fluent discourse about trees without ever having seen, touched, or climbed one. Its representations are grounded in statistical patterns over text, not in causal interaction with the world. Whether this constitutes genuine semantic grounding — or merely a sophisticated form of referential opacity in which symbols refer only to other symbols — is one of the central debates in contemporary AI.

Theories of grounding range from embodied cognition (meaning emerges from sensorimotor interaction) to teleosemantics (meaning is grounded in evolutionary function). The systems-theoretic view holds that grounding is not a binary property but a scalar one: a representation is more or less grounded depending on the richness of its causal and inferential coupling with the environment it represents.