软件学报
軟件學報
연건학보
JOURNAL OF SOFTWARE
2014年
11期
2528-2555
,共28页
洪宇%严为绒%车婷婷%梁颖红%姚建民%朱巧明%周国栋
洪宇%嚴為絨%車婷婷%樑穎紅%姚建民%硃巧明%週國棟
홍우%엄위융%차정정%량영홍%요건민%주교명%주국동
篇章%隐式篇章关系%关系检测%平行线索%篇章关系消歧
篇章%隱式篇章關繫%關繫檢測%平行線索%篇章關繫消歧
편장%은식편장관계%관계검측%평행선색%편장관계소기
discourse%implicit discourse relation%relation detection%parallel cue%discourse relation disambiguation
篇章是论元经过语义关联和结构化组织形成的自然语言文体.篇章分析研究的核心任务之一是解释论元的语义关系,其中,显式关系因具有直观线索而易于检测,目前检测精度高达90%;相对而言,隐式关系因缺乏直观线索而难于检测,目前精度仅约40%.针对这一问题,基于一种“论元平行则关系平行”的假设,并利用显式篇章关系易于检测的特点,通过平行论元的识别与平行关系的消歧,实现了一种显式关系平行推理隐式关系的隐式篇章关系检测方法.利用标准宾州篇章关系树库(Penn discourse TreeBank,简称PDTB)对这一检测方法进行评测,结果显示,精确率提升达17.26%.
篇章是論元經過語義關聯和結構化組織形成的自然語言文體.篇章分析研究的覈心任務之一是解釋論元的語義關繫,其中,顯式關繫因具有直觀線索而易于檢測,目前檢測精度高達90%;相對而言,隱式關繫因缺乏直觀線索而難于檢測,目前精度僅約40%.針對這一問題,基于一種“論元平行則關繫平行”的假設,併利用顯式篇章關繫易于檢測的特點,通過平行論元的識彆與平行關繫的消歧,實現瞭一種顯式關繫平行推理隱式關繫的隱式篇章關繫檢測方法.利用標準賓州篇章關繫樹庫(Penn discourse TreeBank,簡稱PDTB)對這一檢測方法進行評測,結果顯示,精確率提升達17.26%.
편장시론원경과어의관련화결구화조직형성적자연어언문체.편장분석연구적핵심임무지일시해석론원적어의관계,기중,현식관계인구유직관선색이역우검측,목전검측정도고체90%;상대이언,은식관계인결핍직관선색이난우검측,목전정도부약40%.침대저일문제,기우일충“론원평행칙관계평행”적가설,병이용현식편장관계역우검측적특점,통과평행론원적식별여평행관계적소기,실현료일충현식관계평행추리은식관계적은식편장관계검측방법.이용표준빈주편장관계수고(Penn discourse TreeBank,간칭PDTB)대저일검측방법진행평측,결과현시,정학솔제승체17.26%.
Discourse is a literary form that consists of semantically-related and well-structured arguments. One of the key tasks of discourse analysis is to resolve semantic relationship between arguments. Explicit relation is easy to detect with an accuracy of nearly 90% because of its direct cues. In contrast, implicit relation is difficult to detect with only an accuracy of nearly 40% since it has no direct cues. To solve the problem, paper proposes a hypothesis that parallel arguments normally have the consistent semantic relations. Based on the hypothesis and by utilizing the characteristics that explicit relation is easy to detect, the paper implements a method of implicit relation detection which uses explicit relation to infer implicit relation among parallel arguments. We evaluate the method on the standard penn discourse Treebank (PDTB). The experimental results show an improvement of 17.26%.