基于语义提升HMM的语义标注

语义标注所用标签数目众多，训练数据更为稀疏，用HMM作语义标注面临参数估计不准的难题。不同于传统的解决数据稀疏方法，以《同义词词林》的层次式结构为依据，提出了利用语义层次的提升来改善HMM(hidden Markov model)中参数的估计质量；在算法实现中，采用选择受限策略来解决因语义提升而引起的模型辨别力下降问题。测试表明，在训练数据相对稀疏的情况下，适度调整模型的语义层次可大幅提高语义标注的精度，该方法表现出较好的可塑性。

Semantic tagging using HMM with semantic induction

It is difficult for hidden Markov models to get precise parameter estimation when applied to semantic tagging as the number of semantic tags is large and training data is insufficient. Different from classic method for solving data sparsity problem, a method which takes advantage of the hierarchical structure of Synonymy Thesaurus was presented to improve the quality of HMM parameter estimation by semantic induction. Restrictive selection policy was used to reverse the decline of model discriminability caused by the method. Tests indicate that the method is feasible to semantic tagging and tuning according to training data with size can greatly improve semantic tagging.