Isagog Ontologies
REASONING WITH ONTOLOGIES

Two readers,
one vocabulary.

An Isagog ontology has two readers: language models, which read the words, and the knowledge graph, which applies the rules. The same concept guides both, which is why their answers can be compared and checked.

In language models

Operational definitions state in a few lines what each term means, in the language required. We give models a compact view of the ontology, not the whole of it: they know which concepts they may talk about and in what sense, with a smaller context.

When an agent extracts facts from a conversation, it can only use concepts from the catalog: anything that does not fit is discarded. In the same way, text analysis only recognizes the expected frames, with their roles. The chosen perspective steers interpretation, instead of leaving it to the model's imagination.

In the knowledge graph

On the graph, the same concepts are axioms applied by an inference engine. Whoever brings an event about is, by definition, an agent; an event with at least two coparticipants is recognized as reciprocal; if a document yields a sign referring to something, the document is about that thing.

Natural-language questions become SPARQL queries over the ontology's vocabulary, and consistency checks flag data that violate the axioms. Every answer can be traced back to the data and rules it comes from.

Words steer the model, rules constrain the graph: the same ontology holds the two together.