Natural Language Generation
(NLG) systems require rich knowledge resources. In particular,
in a multilingual environment
the underlying domain knowledge should be represented in a language
independent formalism and structured in language neutral units
[Hovy & Nirenburg 92].
Our experience within the DB-MAT project
has proved that the acquisition of such knowledge sources is an
extremely difficult task [Angelova & Bontcheva 97]. The acquired knowledge
should represent explicitly various domain relations together with
the conditions when they hold. In order to enable the
generation of tailored multilingual explanations from a common
knowledge source, we aim at the development of a
task-dependent knowledge acquisition (KA) approach which accounts for
complex ontological phenomena.
In this paper we discuss some refinements of domain ontologies and their possible use for generation of NL explanations. In section 2 we mention quite briefly the knowledge resources in some current NLP systems. As far as our suggestions concern user perspectives, a related approach from NLG is summarised. Section 3 describes our system as a knowledge-based NLP approach relying on a KB of domain knowledge rather than on ontological representations. Section 4 presents the acquisition of detailed ontological IS-A relations. An overview of the existing DB-MAT generator is given in Section 5. Section 6 discusses how our refined type hierarchy might improve the generation of domain knowledge explanations. An alternative source of improvement for the explanations is outlined in Section 7, while Section 8 contains the conclusion.