GENERATING CONCISE NATURAL-LANGUAGE SUMMARIES

Summaries typically convey maximal information in minimal space, In th is paper, we describe an approach to summary generation that opportuni stically folds information from multiple facts into a single sentence using concise linguistic constructions. Unlike previous work in genera tion, how information gets added into a summary depends in part on con straints from how the text is worded so far. This approach allows the construction of concise summaries, containing complex sentences that p ack in information, The resulting summary sentences are, in fact, long er than sentences generated by previous systems, We describe two appli cations we have developed using this approach, one of which produces s ummaries of basketball games (STREAK) while the other (PLANDOC) produc es summaries of telephone network planning activity; both systems summ arize input data as opposed to full text. The applications implement o pportunistic summary generation using complementary approaches. STREAK uses revision, creating a draft of essential facts and then using rev ision rules constrained by the draft wording to add in additional fact s as the text allows, PLANDOC uses discourse planning, looking ahead i n its text plan to group together facts which can be expressed concise ly using conjunction and deleting repetitions, In this paper, we descr ibe the problems for summary generation, the two domains, the linguist ic constructions that the systems use to convey information concisely and the textual constraints that determine what information gets inclu ded.