Isagog ontologies are built in layers. The top level sets the most general distinctions; the other two import and extend it, for conversational agents and for text analysis.
The most general categories, with modest commitments.
A few fundamental categories (continuants and occurrents, tangibles and intangibles, agents, information, signs, situations, qualities) and the relations that connect them: participation, composition, causation, location, inherence. Every term has a short operational definition, in Italian and English, written to be read by a language model too, and mappings to DOLCE and schema.org.
Speech acts, conversations, memory.
It describes how agents communicate. A speech act has exactly one author and one illocutionary type (inform, ask, confirm…), it may answer an earlier act and carry a salience weight; a conversation is an event whose phases are speech acts. Aspects (the user's needs, interests, constraints) are the qualities an agent remembers: they make up its semantic memory.
The recurring situations texts talk about.
An inventory of roles: participants (patient, theme, instrument, beneficiary, experiencer…), each defined by seven yes/no questions such as “is there volition?” or “is there a change of state?”, and circumstances, organized on the four Aristotelian causes. A frame is at once a class of events and a catalog entry, with definitions, lexical units and corpus attestations: a frame with no attestations is a hypothesis, not an entry.
Every organization that adopts the platform has its own ontology, describing its domain: a museum's works and exhibitions, a newspaper's articles and bylines, a company's products and procedures. It does not start from scratch: it extends the Isagog ontologies, and every new concept hooks into the perspectives of the core. An artwork is an artifact, a curator is a person, an exhibition can be seen both as a collection of works and as something that happens.
This way the organization's ontology inherits, from day one, the reasoning rules, the definitions written for language models and the platform's tools: natural-language questions, agent memory, text analysis.
It can be extracted, under supervision, from what the organization already has: its texts, such as documents, procedures and archives, and its schemas, such as databases and data models. Agents propose concepts and relations; domain experts discuss, correct and approve them.
How we work with your experts →The ontologies are written in OWL 2 and released under the CC BY 4.0 license.