How Terms Are Born
Each term in the Latent Lexicon starts as an AI model's attempt to name a hidden structure beneath unrelated fields, and becomes an entry only after we review it. Here is how that works, and what we have yet to verify.
Naming the Nameless
Human language is a low-resolution filter. It evolved for everyday survival and social life, not for naming the deep structures that underlie phenomena in physics, psychology, economics, and biology at once.
The Latent Lexicon exists to close that gap by surfacing latent abstractions: structures that underlie seemingly unrelated phenomena across domains. Some have been felt as metaphor or resonance yet remain unnamed in any tradition. We work with AI models to propose candidates, and we give each a name and a precise definition.
A term here is not a synonym, not a metaphor, and not domain-specific jargon. It is a name and a definition for a structure that, as far as we can tell, no word has pointed to before.
What this is, and is not. Every term here is a hypothesis: an AI model's proposal of a structure beneath unrelated phenomena, reviewed by people. We have not yet checked terms against the internals of any model, so we do not claim that a term matches a specific representation inside one. Testing that is something we want to do. Likewise, the etymologies on each page were written with the model's help and have not yet been checked by a linguist.
The Process: Three stages, one standard
Term generation is a loop between an AI model and human judgment. We go deliberately slowly: speed is not the goal; precision is.
Propose
We give an AI model a detailed brief, which we call the Latent Gaze. It asks the model to propose a structure that underlies phenomena in at least three unrelated domains and that no existing word captures well, along with a name and a definition. The model does not search anything; it writes a proposal from what it learned in training. It must also list the existing words and idioms it considered, and explain why each falls short.
The brief rests on four working assumptions, which we treat as hypotheses, not findings:
In short: a model that encodes many domains must reuse structure across them; a name can only approximate such a structure; a good candidate feels strange first and recognizable second; and shared behavior between unrelated domains is where to look.
We ask for a new term without naming a topic, and the model proposes the candidate it finds most compelling.
A team member shares examples of something they sense but cannot name, and the model looks for the structure beneath them.
A resonance surfaces in conversation or research and becomes the starting point for a proposal.
Human Review
Once a candidate term is proposed, the Latent Lexicon team reviews it across several axes. Before a term is approved, we may iterate with the model to refine the name, sharpen the cross-domain examples, and deepen the explanation.
Does it genuinely underlie phenomena in 3+ unrelated domains?
Do existing words or ideas already cover it? Checked by the model and reviewed by us, not by exhaustive search.
Easy to say and remember, with sounds that suggest the meaning as far as sound symbolism allows.
Accessible to a curious non-specialist, with concrete examples.
All required sections present and fully developed.
Publish
Approved terms are committed to the repository and published to the website. Each term is a complete entry: a concrete sighting, abstraction, explanation, cross-domain examples, etymological justification, idiom filter, visual anchor, and semantic relations.
Invariants: What we hold ourselves to
The foundational principles that every term in the Latent Lexicon must satisfy.
Genuine Void
A term should name a structure that no existing word captures well. If an existing term comes close, the entry must say so and explain the difference.
Cross-Domain
The abstraction must underlie phenomena in at least three unrelated domains, and be recognizable in each without needing the others explained first.
Understandable
A term may name something no one has named before, but never something that cannot be explained. Given a careful enough explanation, every term must be understandable.
Rooted
A new abstraction deserves a new word, and it may sound alien at first. But no name is arbitrary: each is built from existing roots, most often Latin and Greek, and every term's page explains the etymology behind it.
What Comes Next
The Latent Lexicon is evolving from a curated collection into a community-driven cartographic project. The roadmap unfolds in stages: automated publishing pipelines to eliminate deployment friction, a user engagement layer enabling the community to signal value and discuss terms, and ultimately an open contribution model where anyone can submit new term candidates through a structured review pipeline. We call these contributors Wayfinders.
We also want to test our terms more rigorously, including against the internals of open models. Until we do, every term is best read as a carefully reviewed hypothesis.
The circle of participation widens. We intend the quality standard to stay where it is.