基于Penn Discourse TreeBank(简称PDTB)语料中的隐式篇章关系类型,提出一种无指导的识别方法.主要依据显式与隐式平行论元对之间的映射关系实现显式到隐式关系的推理,即利用显式论元对的篇章关系,推理与之平行的隐式论元对的篇章关系.推理过程主要包括:基于大规模语料库以及基于搜索引擎的平行语料挖掘,平行语料中显式连接词映射到篇章关系过程的消歧.与传统基于监督学习的分类方法相比,基于统计的无指导方法在隐式篇章关系推理的性能上获得显著提升,识别精确率提高了近15.6%.此外,在分析比较各研究方法的同时,指出目前隐式篇章关系推理研究所面临的主要困难与挑战.%We present an unsupervised method based implicit discourse relation recognizer in the Penn Discourse Treebank (PDTB). Our recognizer performs an implicit relation inference by well using Explicit-to-Implicit relation mapping among parallel argument pairs. The inference process contains; parallel resource mining based on a large scaled corpus and search engine, sense disambiguation occurred in discourse relation mapping. Compared with the typical method of supervised relation classification, our results on the PDTB yield a significant 15.6% improvement. Finally, we point out the main difficulty and challenge in inferring implicit discourse relations by analyzing the current related researches.
展开▼