Automatic Summarization from Multiple Documents (Extended Abstract)

Since the late 50’s and Luhn [Luh58] the information community has expressed its interest in summarizing texts. The domains of application of such methodologies are countless, ranging from news summarization [WL03, BM05, ROWBG05] to scientific article summarization [TM02] and meeting summarization [NPDP05, ELH03]. Summarization has been defined as a reductive transformation of a given set of texts, usually described as a three-step process: selection of salient portions of text, aggregation of the information over various selected portions, abstraction of this information to a more general level, and finally presentation of the final summary text [MB99, Jon99]. The summarization community nowadays includes a significant number of scientists with increasing interest in the multi-document aspect of summarization. Major issues towards multi-document summarization that have been identified by the researchers include the detection of (possibly query-related) salient information in texts, the preservation of coherence and linguistic quality in summarized text, the control of redundancy in the summaries, and the language independence of summarization methods. Up to date, many summarization systems have been developed and presented, especially within such endeavours as the Document Understanding Conferences (DUC) and Text Analysis Conferences (TAC). The DUC and TAC have strengthened the summarization community and have helped in identifying