英语介词短语附着决策的研究

Research on Solution of the Prepositional Phrase Attachment

提出了ID3决策树和WordNet语义知识相结合的方法 ,用此来确定英语介词短语附着对象。首先将介词附着问题抽象成词汇和语义元组的组合问题 ,然后转化为对每个词汇和语义元组属性值的决策。单个介词附着决策的正确率达 85.15%。此外 ,该方法可用于处理多重介词附着的两个显著问题 :数据稀疏和附着歧义更大 ,即提出了递归运用前面的决策结果来解决多重介词附着问题的算法。实验表明 ,该方法取得了较好的消歧结果。

A hybrid method, which combines the decision tree with semantic information, is introduced to solve the problem of the PPA (prepositional phrase attachment). The problem of the PPA is abstracted into the combination of words and semantic class groups, then transferred it into the induction of their values of the attributes. Solution to the PPA with this method reports an accuracy of 85.15% on open test. The method is extended to deal with two related problems of multiple PPA: more sparse data and more severe syntactic ambiguities. In resolving subsequent multiple PPA, an algorithm which solves the two problems through recursive use of the result of the decision obtained from first PP is presented. The experiment shows that the hybrid method has given better results than others.