篇章中指代消解研究综述

篇章中指代消解研究综述 Intra-Document Anaphora Resolution:A Survey 周炫余 ZHOU Xuanyu 1 first-author 刘娟 LIU Juan 1 卢笑 LU Xiao 2 武汉大学计算机学院 武汉大学计算机学院 School of Computer,Wuhan University School of Computer,Wuhan University 湖南大学电气与信息工程学院 湖南大学电气与信息工程学院 College of Electrical and Information Engineering,Hunan University College of Electrical and Information Engineering,Hunan University 指代消解的基本任务是识别篇章中对现实世界同一实体的不同表达.由于消解不同实体表达是自然语言理解的关键因素之一,因此许多学者在该领域做了大量的研究工作.本文将指代消解领域的经典算法综述为两大类:①基于启发式语言学规则的指代消解方法;②基于数据驱动的指代消解方法.在过去的几十年里,由于标注语料MUC,ACE和OntoNotes的出现,指代消解系统从基于启发性规则的消解方法转向基于数据驱动的消解方法.然而最近两年出现一个比较有趣的现象,只要能获取足够多的表达信息,基于启发性规则的方法也能取得很好的消解效果,基于启发性规则的方法又有兴起的趋势.随着指代消解研究的发展,它在其余自然语言处理技术中得到了广泛的应用,例如:机器翻译、信息抽取和自动文摘等领域.计算模型不精确、篇章知识匮乏以及缺少标注语料是限制指代消解发展的主要因素.未来,候选消解项的识别与抽取、跨文本指代消解、多语言的指代消解和各种深层篇章知识的联合利用将成为指代消解发展趋势. The task of anaphora resolution is to determine linguistic expressions that refer to the same real-world entity in a text.Because of its significance for robust natural language understanding,much work has been done by researchers in the area of anaphora resolution in the past decades.We present a literature survey of the anaphora resolution methods,which are roughly divided into two categories:heuristic-based approaches and data driven approaches. In the past decades,the focus of computational anaphora resolution has exhibited a shift from heuristic-based approachs to data driven approaches,which can be partly attributed to the advent of the public availability of annotated anaphora corpora（MUC,ACE and OntoNotes）.However,in the past two years some researchers have found an interesting phenomenon that such a rich representation is available,even the heuristic-based approach can achieve stateof-the-art performance.The trend of heuristic-based approaches is developing again.With the development of anaphora resolution,it is widely used in machine translation,information extraction,automatic summarization and so on. The accurate computational models,plenty of corpus and abundant of texture knowledge are the main reasons restricting the development of anaphora resolution.The development trade is to identify and extract anaphoricity,cross-text, multi-language and association of deep texture knowledge for anaphora resolution. 自然语言处理 指代消解 共指消解 机器学习 数据挖掘 natural language processing anaphora resolution coreference resolution machine learning data mining 10.14188/j.1671-8836.2014.01.003 TP391.1 国家自然科学基金（61272274）资助项目 2014-01-01 2021-04-01 1 60