Knowledge Graph Mediated Translation (KGMT) + APE provides a major jump in quality, even for domains with unusual concepts in quantity or due to domain specificity.

The Knowledge Graph grounds the LLM into deterministic knowledge and reduces the hallucination to zero, or near zero. The Knowledge Graph provides controlled delivery of context to guide the LLM to heed to the context, which is a big booster in long form, narrative, abstract, domain specific and lower resource domain.

It can also differentiate company specific opinion, when it quotes other text and it can include version control based off dates. Knowledge Graphs can imply brand characteristics, as needed.

This Knowledge Graph-based technique has been turned into a formal specification.

This specification was created in collaboration with Edwin Trebels, Founder of LangOptima, and Lead Semantics in their knowledge transformation tool Textdistil. KGMT is available stand-alone and for integration with Translation Management or other RAG Systems.

Automatic Post Editing (APE) and COMET scores are optional add-ons.