摘要
搭配是一种词与词之间的组合关系,搭配的自动提取,是自动句法分析的重要环节,同时也是重要的知识来源。本文在分析搭配性质的基础上提出了一个基于统计的动宾搭配自动识别的算法。我们从经过了人工标注的50万词的训练语料中获取数据,并把所获数据用于自动识别。封闭测试和开放测试的正确率分别是947%和812%。实验结果表明,从训练语料中获取的数据可以比较有效地用于自动识别,本文选取的四项指标也是自动识别比较合适的统计量。
Collocation describes how a word is used in relation to others in a language and extraction of collocations has a wide range of applications in both language learning and natural language processing. Verb-Object collocations is the most important part due to verb pivotal position in a sentence. In this paper, we present an efficient statistical algorithm for extracting verb-object collocations in Chinese in which three statistical features are combined. In our experiments the algorithm obtains precisions of 94.7% and 81.2% respectively for closed test and open test. The error analysis shows that the performance can be improved by applying shallow parsing technique as a preprocessor and adding more data for training.
出处
《语言文字应用》
CSSCI
北大核心
2005年第1期137-143,共7页
Applied Linguistics
关键词
动宾搭配
句法分析
概率
分布
verb-object collocation
syntactic Parsing
probability
distribution
