摘要
本文提出了用词语潜在依存关系分析汉语ｂａｓｅＮＰ结构的模型，它有以下的特点：①将依存语法知识融入概率模型中，使得ｂａｓｅＮＰ结构分析在依存语法知识的指导下进行，其性能优于纯粹的概率模型——相邻模型；②词语潜在依存强度的获取算法是基于ＭＤＬ原则的，在模型建造时既考虑数据拟合性，又考虑模型归纳性，其性能优于基于极大似然原则的词语潜在依存强度获取算法；③词语潜在依存强度获取算法在复杂特征集上进行，可以有效地解决参数估计中的数据稀疏问题．实验结果显示，这个模型对于汉语ｂａｓｅＮＰ结构分析是有效的．
The paper puts forward a potential dependency relation based model for Chinese baseNP structure analysis, which has the following characteristics: The dependency grammar is integrated into the statistical model so that the baseNP structure can be analyzed under the supervision of dependency grammar. The performance of the model is superior to that of pure statistical model, adjacency model; The proposed acquisition algorithm of potential dependency strength is based on MDL principle, in which both the data fitness and the generality of the model are considered. The performance of the algorithm is superior to that of the traditional ML based algorithm; The acquisition algorithm is implemented on the basis of complex feature set, so that the data sparseness problem is solved successfully. The experiment shows that the proposed model is suitable for Chinese baseNP structure analysis.
出处
《计算机学报》
EI
CSCD
北大核心
1999年第2期141-146,共6页
Chinese Journal of Computers
基金
国家自然科学基金
