Answer-first documentation
Frequently Asked Questions
Concise technical answers about the architecture, embeddings, multilingual alignment, and semantic registry.
What is Embedded Semantics?
Embedded Semantics is a multilingual semantic representation system that maps expressions to stable, registry-backed concept identities.
Is an embedding vector the source of truth?
No. Embeddings rank likely concepts. Stable concept identifiers, definitions, evidence, and versioned registry records are authoritative.
Does the system require every language to use the same words?
No. Different expressions can resolve to the same concept while preserving language, locale, register, and unresolved semantic residue as separate metadata.
Why use hard negatives?
Hard negatives are closely related but incorrect meanings. They teach an embedder to distinguish conceptual neighbors instead of grouping everything from the same topic together.
How does the system handle uncertainty?
Candidate scores and score margins are calibrated so the resolver can abstain when evidence is insufficient instead of forcing a concept assignment.