期刊名称：Association for Computing Machinery New Zealand Bulletin
印刷版ISSN：1176-9998
出版年度：2006
卷号：2
期号：1
出版社：New Zealand Chapter
摘要：This paper proposes a Tree-based Predicate-Argument Recognition Algorithm (PARA) that identifies boundaries for all predicates and arguments in a sentence, along with a few additional syntactic features such as grammatical functions, voice, and so forth, useful for other Natural Language Processing (NLP) applications such as Semantic Role Labeling. Evaluation shows the algorithm, without training, is capable of automatically converting syntactic tree-based structures to their corresponding flat structures and achieves F 1: 80.72 for boundary recognition on the WSJ23 test dataset and F 1: 79.87 on the Brown Corpus test dataset from the CoNLL-2005 Shared Task. The execution time for each sentence is 0.005 second, which is much faster than state-of-the-art Machine Learning approaches, suggesting that PARA would be more useful for some practical NLP applications such as real-time Machine Translation.
关键词：Semantic role labeling ; Predicate-Argument Recognition ; Boundary Recognition
