Movement / multimodal context
Gesture, rhythm, visual attention, action schemas, and multimodal signals can frame meaning before a sentence exists.
Technology
Semantic Isomorphism Technology
A plain-English overview of semantic isomorphism, language residue, hybrid retrieval, the four-layer Neurokinetic AI model, and the normalize-embed-neutralize-resolve-render pipeline.
Accessible definition
Semantic isomorphism means preserving the role of meaning while its surface changes.
A phrase can become a normalized string, an embedding vector, a concept record, a compact message, or a human-readable answer. Semantic isomorphism is the engineering discipline of keeping the important structure intact across those moves.
For Neurokinetic AI, the semantic layer does not ask downstream systems to trust a string match or a raw vector coordinate. It asks them to inspect a concept identity, its aliases, its language-residue profile, its provenance, and the policy used to resolve it.
Implementation boundary
The public WordPress site exposes the product narrative, route structure, metadata, and client-side Alignment Lab. This repository does not expose a hosted embedding model, registry API, credentials, crawler, vector database, or production contact backend; those operational tasks are tracked in deployment documentation.
Four-layer model
Each layer carries a different kind of evidence.
Text policy / Unicode normalization
Encoding, normalization, grapheme boundaries, bidi behavior, identifiers, and confusables need declared policies.
Concept interlingua / neutral routing
Embeddings discover semantic neighborhoods; neutralization reduces language identity residue; the registry establishes identity and evidence.
Protocol surfaces / UAI-1 and IOTA-1
Structured handoff surfaces carry confidence, provenance, integrity, and render targets across AI-to-AI boundaries.
Pipeline
Five stages keep comparison, identity, and rendering separate.
Normalize
Decode and compare safely without losing the original surface expression.
Embed
Map the expression into dense, sparse, or late-interaction neighborhoods for candidate discovery.
Neutralize
Reduce source-language residue and extract side channels such as tone, urgency, locale, and authority.
Resolve
Bind the expression to a versioned concept ID with confidence, evidence, and abstention thresholds.
Render
Generate the target language, symbol sequence, API record, UAI-style envelope, or review artifact.
Retrieval theory
The vector is a search instrument, not the final meaning.
New source research for this pass emphasized a practical distinction: embeddings are excellent at finding semantic neighborhoods, but they can also leak language identity, over-favor English, blur rare entities, and flatten side channels. Neurokinetic AI treats retrieval as evidence gathering before registry resolution.
Semantic neighborhood
Multilingual encoders locate related phrases across language boundaries and give the resolver a candidate set.
Exact-signal protection
Lexical matching protects acronyms, protocol IDs, product names, citations, and terms that dense vectors can wash out.
Residue control
INLP-style or SVD-style projection can be used as a design pattern for reducing language identity while preserving semantic topology.
Registry authority
The concept registry decides whether the evidence is strong enough to assign a stable ID or return a review state.