Member-only story
Letting Language Models Hallucinate Languages
One conlanger’s quest to construct functional languages for their fictional worlds.
Constructed languages almost always begin the same way: a burst of invention, a small core vocabulary, a handful of phonological choices that feel right in the mouth, and some sketch of grammar that looks clean on paper. Then reality arrives. The first time you try to actually use the language — translate a paragraph, write dialogue, express a subtle tense, coin technical terms — you discover what linguists already know: a language is not a list of rules, it’s an ecology. It lives in edge cases.
A human conlanger can build a strong skeleton and still find themselves stuck when the system needs to behave in places they didn’t design. What happens with borrowed words? How do proper names behave? Which irregularities feel natural rather than arbitrary? How do you avoid a language that seems functional in isolated sentences and falls apart the moment it’s asked to carry real load? Finishing a language is time intensive because completeness isn’t just coverage; it’s consistency under variation.
Natural languages solve this problem in the opposite way from conlangs. They rarely aim for elegance. They evolve. They accumulate repairs. They acquire redundancy and ambiguity, then…
