Towards Constructing Sports News from Live Text Commentary

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 1361–1371, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Towards Constructing Sports News from Live Text Commentary Jianmin Zhang Jin-ge Yao Xiaojun Wan Institute of Computer Science and Technology, Peking University, Beijing 100871, China Key Laboratory of Computational Linguistic (Peking University), MOE, China {zhangjianmin2015, yaojinge, wanxiaojun}@pku.edu.cn Abstract In this paper, we investigate the possibil- ity to automatically generate sports news from live text commentary scripts. As a preliminary study, we treat this task as a special kind of document summarization based on sentence extraction. We for- mulate the task in a supervised learning to rank framework, utilizing both tradi- tional sentence features for generic docu- ment summarization and novelly designed task-specific features. To tackle the prob- lem of local redundancy, we also propose a probabilistic sentence selection algorithm. Experiments on our collected data from football live commentary scripts and cor- responding sports news demonstrate the feasibility of this task. Evaluation results show that our methods are indeed appro- priate for this task, outperforming several baseline methods in different aspects. 1 Introduction There are a huge number of sports games played each day. It is demanding and challenging to write corresponding news reports instantly after various games. Meanwhile, live text commentary services are available on the web and becoming increas- ingly popular for sports fans who do not have ac- cess to live video streams due to copyright reasons. Some people may also prefer live texts on portable devices. The emergence of live texts has produced huge amount of text commentary data. To the best of our knowledge, there exists few studies about utilizing this rich data source. Manually written sports news for match report usually share the same information and vocabulary as live texts for the corresponding sports game. Sports news and commentary texts can be treated as two different sources of descriptions for the same sports events. It is tempting to investigate whether we can utilize the huge amount of live texts to automatically construct sports news, typ- ically in a form of match report. Building such a system will largely relax the burden of sports news editors, making them free from repetitive tedious efforts for writing while producing sports news more efficiently. In this work, we study the possibility to con- struct sports news in the form of match reports from given live text commentary scripts. As a con- crete example we collect live text data and corre- sponding news reports for football (called soccer more often in the United States) games and con- duct our study thereby. However, our methods and discussions made in this paper can be trivially adapted to other types of sports games as well. As a preliminary study, we treat this task as a special kind of document summarization: ex- tracting sentences from live texts to form a match report as generated news. However, generating sports news from live texts is still challenging due to some unique properties …

Content truncated to 500 words