0.7 Lexical creativity: extending a language

Human language makes theory extension unusually concrete. Speakers continually introduce new expressions and new senses for old expressions. Some disappear; others acquire stable grammatical roles, meanings, and relations to the rest of a language. The Oxford English Dictionary records this continuing process in quarterly updates that add words and senses and revise existing entries from historical evidence [ Oxford University Press , 2026 ] . Lexicographical inclusion is recognition of an innovation that has acquired sufficient evidence of use, not the act that originally created it.

This exposes an important difference between producing an unfamiliar string and changing a language. A conventional language model normally has a fixed tokenizer and a fixed set of learned parameters. Subword tokenization allows it to assemble known pieces into rare or previously unseen surface forms [ Sennrich et al. , 2016 ] . It can therefore spell a plausible word that did not occur in its training corpus. But the generation alone does not add a new token to the model, assign the expression a persistent grammatical and semantic type, or revise a shared lexicon for later speakers.

It is useful to distinguish three achievements.

Novel string generation.

Produce a sequence not previously observed. This may be combinational novelty over a fixed token vocabulary.

Conceptual innovation.

Use an expression to mark a useful distinction, operation, object, or social practice that the prior vocabulary did not mark adequately.

Lexical accommodation.

Modify a persistent language so that the new word or sense becomes a reusable compositional resource for future speakers.

The final achievement can be represented as a declaration extension

Here the extended sketch does not merely contain another character string. It records a new lexical generator—or a new sense of an existing one—together with relevant syntactic behavior, semantic relations, and conventions of use. Restriction along \(J_{\mathrm{lex}}\) forgets the innovation and recovers the earlier lexicon. Transport in the other direction asks how old utterances and distinctions survive after the extension.

Human linguistic communities perform proposal, exploration, and admission at different scales. An individual or group coins an expression. Speakers test it across contexts, alter its meaning, and sometimes derive related forms. Sustained public use supplies evidence that the expression has become part of a shared language. In Piagetian terms, a novel string may be assimilated by existing grammar, whereas a genuinely new conceptual distinction may require accommodation of the lexicon. In Boden’s terms, the episode can begin as combination, become exploratory as uses proliferate, and become transformational when it changes what speakers can concisely express and reason about.

. Do not identify linguistic creativity with out-of-vocabulary generation. The synthetic target is a persistent extension of the language: a new expression or sense with a type, compositional role, relations to prior expressions, and an admission history.

This example sharpens the book’s contrast with a compression-centered account. Predictive compression may induce a useful latent representation of an emerging word. Synthetic creativity asks for the additional transition by which that distinction becomes an explicit, transportable component of the language in which later thought is conducted.