Information Fusion in the Context of Multi-Document Summarization

Information Fusion in the Context of Multi-Document Summarization Regina BarzUay and Kathleen R. McKeown Dept. of Computer Science Columbia University New York, NY 10027, USA Michael Elhadad Dept. of Computer Science Ben-Gurion University Beer-Sheva, Israel Abstract We present a method to automatically generate a concise summary by identifying and synthe- sizing similar elements across related text from a set of multiple documents. Our approach is unique in its usage of language generation to reformulate the wording of the summary. 1 Introduction Information overload has created an acute need for summarization. Typically, the same infor- mation is described by many different online documents. Hence, summaries that synthesize common information across documents and em- phasize the differences would significantly help readers. Such a summary would be beneficial, for example, to a user who follows a single event through several newswires. In this paper, we present research on the automatic fusion of simi- lar information across multiple documents using language generation to produce a concise sum- mary. We propose a method for summarizing a spe- cific type of input: news articles presenting dif- ferent descriptions of the same event. Hundreds of news stories on the same event are produced daily by news agencies. Repeated information about the event is a good indicator of its impor- tancy to the event, and can be used for summary generation. Most research on single document summa- rization, particularly for domain independent tasks, uses sentence extraction to produce a summary (Lin and Hovy, 1997; Marcu, 1997; Salton et al., 1991). In the case of multi- document summarization of articles about the same event, the original articles can include both similar and contradictory information. Extracting all similar sentences would produce a verbose and repetitive summary, while ex- tracting some similar sentences could produce a summary biased towards some sources. Instead, we move beyond sentence extraction, using a comparison of extracted similar sen- tences to select the phrases that should be in- cluded in the summary and sentence generation to reformulate them as new text. Our work is part of a full summarization system (McK- eown et al., 1999), which extracts sets of simi- lax sentences, themes (Eskin et al., 1999), in the first stage for input to the components described here. Our model for multi-document summariza- tion represents a number of departures from traditional language generation. Typically, lan- guage generation systems have access to a full semantic representation of the domain. A con- tent planner selects and orders propositions from an underlying knowledge base to form text content. A sentence planner determines how to combine propositions into a single sentence, and a sentence generator realizes each set of com- bined propositions as a sentence, mapping from concepts to words and building syntactic struc- ture. Our approach differs in the following ways: Content planning operates over full sentences, producing sentence frag- ments. Thus, content planning straddles the border between interpretation and gen- eration. We preprocess the similar sen- tences using an existing shallow parser (Collins, 1996) and a mapping to …

Content truncated to 500 words