MODELER Overview
This web utility is designed to formally understand English grammar forms
in terms of an ontology that extends ISO's Topic Map paradigm with
Conceptual Graphs.
By calling this utility properly from your own software, you can enable:
-
Situationally aware dialogs between your software and its users
-
Web apps which automatically track topics for a collaborating group
-
Better recall and precision for your information-retrieval software
-
Accurate mapping of paragraphs into assertions for a rules engine
Once properly set-up, our pure-Java web app assigns topics to English phrases and
clauses passed it in HTTP requests. Linguistically, it is limited to
co-operative speakers, but this is no problem in typical interactive UIs.
The MODELER design arose (in 3 parts)
as a back-end to our own English paragraph analysis software. But with a few tricks
at API levels, many other software modules might also exploit it to retrieve:
- Referent models, found or built for given English phases or clauses.
- Anaphora options based on discourse history, rules and constraints.
- Semantic plausibility ratings that assist in resolving various ambiguities.
- Prototype instances useful as typical examples for classes and sets.
- A chart representing the meaning of any given English paragraph.
To provide such services and data, MODELER maintains an evolving CONTEXT
model holding past and expected topics. Its intelligent agent client
acts as the MVC controller, and uses a web API to request contextual
semantic models under various dynamic views. On their desktops,
each user controls:
- Frame-like storage holding instances of the classes in your
ontology, able to inherit and use your scripted constraints and behavior
rules.
- Metadata source code and documentation, version-synchronized, which
gives each user public specs on all the topics currently in CONTEXT.
- Query facilities using a SQL-like syntax to search for topics cited in
the discourse or expected to relate to whatever is being discussed.
- Scriptable utilities that can generate reports; answer questions;
persist, read or adjust topics; or swap data with modeled J2EE applications.
- Real-time human I/0 that can augment heuristics in a learning mode
that lets domain-specific vocabulary be acquired by reading related texts.
MODELER is driven by its input stream, to which it reacts by using logic that is
fixed for its repertoire of known speech acts. But what happens
then is mostly up to the concepts and scripts loaded in its knowledge base.
You can control these at ontology levels, by configuring CONTEXT with a base of
scripted lexikons holding domain-specific vocabulary
and world knowledge. Organize these topic maps wisely (we can help), and
they will keep MODELER reacting to its inputs with coherence, and possibly even
intelligence.
|
