DB-MAT [Angelova & Bontcheva 96a,v.Hahn & Angelova 94] aims at application of knowledge based methods in translation aid tools. The central idea is that the user (human translator) receives linguistic and/or domain knowledge support. Domain knowledge explanations are generated in NL from an underlying KB of Conceptual Graphs (CG). All system resources are hidden behind a user-friendly, menu-driven interface where the translator highlights unknown terms, selects a request from a menu and receives NL explanations. Requests for clarification or new explanations are formulated within the context of previous explanations. Consequently, DB-MAT offers a user-directed exploration of the subject matter.
The core is a single, language-neutral KB, related to all lexicons (one per language). CG were chosen due to ( i) support of nested graphs which allows knowledge items with different granularity to be represented and also linked to the lexicon entries for each language; ( ii) well-defined operations. The application of inference rules also supports explanations containing knowledge which is not directly encoded in the KB.
Figure: Conceptual Graph in Graphical Notation