Interlingua for Bilingual Statistical Reports

This paper describes the interlingua used in a text generation system applied to the domain of bilingual statistical reports (LFS; [Iordanskaja et al. 92]). This interlingua is a sentential conceptual interlingua ([Polguère 91]) that is transformed directly into English and French semantic representations ([Mel’cuk 81]) via declarative rules stored in an interlingual dictionary. However, this domain contains problematic cases illustrating that a direct transformation of the interlingua into semantic representations may not always be appropriate. Namely, some bilingual examples from the corpus show important differences at the level of the conceptual content or at the level of thematic structure. As an initial solution to these problems, I propose to consider the stylistic preferences associated with each target language in order to paraphrase (adjust) the interlingua for these languages.