SumREN: Summarizing Reported Speech about Events in News

A primary objective of news articles is to establish the factual record for an event, frequently achieved by conveying both details specified event (i.e., 5 Ws; Who, What, Where, When and Why regarding event) how people reacted it reported statements). However, existing work on summarization almost exclusively focuses details. In this work, we propose novel task summarizing reactions different speakers, as expressed their statements, a given event. To end, create new multi-document benchmark, SumREN, comprising 745 summaries statements from various public figures obtained 633 discussing 132 events. We automatic silver-training data generation approach our task, which helps smaller models like BART achieve GPT-3 level performance task. Finally, introduce pipeline-based framework speech, empirically show generate that are more abstractive than baseline query-focused approaches.