The interlingual text language created for this project is a variation of the one used in DIOGENES and currently being used in the Mikrokosmos project. In general, it is a hierarchical arrangement of propositions linked together into propositional clusters (which in turn can be linked together) by rhetorical relations. The propositional representation relies on an ontology, or language-independent world model. Instances of ontological concepts become elements of text meaning representation - either free-standing entities (frames) or values of other entities' properties (slot fillers). Statements of the event's aspectual and modal properties are also included. Each concept used in the proposition has its place in the ontological hierarchy. Discourse and pragmatics-related information is represented through speaker attitudes to text meaning elements (e.g. expectation, evaluative, etc.) and general stylistic features.
In analysis, using the world knowledge contained in the ontology is extremely important in disambiguation. For instance, knowing that a ball is used in soccer helps disambiguate the sentence ``He had a ball'' in a soccer context. In generation, world knowledge is used to a much smaller extent. The ontology essentially provides hooks on which lexical items can be hung. There are cases, however, where a concept's ancestry is important; for instance, knowing whether a particular concept is animate may influence the choice of verb. The ontology is also important in generating definite descriptions.
The speaker's intent in
communication can also be specified, although, at present, none of our
rules make use of such information.
Our experiment assumes a hierarchical backbone, but beyond that would be able to handle a large variety of semantic information. Anything can be attached to one of the nodes, and rules created to handle its occurrences. We are very interested in issues of IL representation and look forward to applying advances in that area to this paradigm.