Integrating language models with an organization's knowledge
takes a new kind of conceptual model.
Isagog ontologies are built for this: they guide the reasoning of both language models and the knowledge graph.
Language models can talk about almost anything, but they don't know an organization's data, rules and vocabulary. Connecting them to that knowledge takes a conceptual model: a shared, precise vocabulary stating what is being talked about, how those things are related and what can be inferred from them.
Traditional ontologies were born for databases and logical systems: they file everything into a rigid hierarchy of categories. Natural language works differently: the same words change meaning with context, and the same object is described in different ways depending on what is needed. A model meant to work with language models has to take this into account.
These pages explain how we have rethought our ontologies, what they are for and how we use them. In the explorer you can browse them yourself.
WHERE TO START
Concepts as perspectives
Why a concept is not a category to lock things into, with four examples taken from the models.
Three layers, plus yours
Top level, agents and frames: what each one describes, and how an organization's own ontology is built on them.
One vocabulary, two readers
How the same definitions guide both language models and reasoning over the knowledge graph.
Explore the ontologies
Concepts and relations, with operational definitions, perspectives, domains and ranges.