论文部分内容阅读
为提高口语对话系统中语言理解的稳健性,提出了一种基于最大后验统计框架的两级搜索的理解算法。第一级用概念捆绑达到提取句中关键成分并剔除某些干扰成分的目的;第二级采用改进的基于树扩展的稳健句法分析搜索最佳理解结果,同时引入用户意图推断和句子特征短语两方面的信息对搜索空间进行约束,进一步提高了理解的稳健性和实时率。实验表明,该算法应用于火车信息查询领域,在0.22倍实时下,能得到13.6%的句意理解错误率和25.4%的概念理解错误率,相对基线系统分别为降低了23.2%和9.3%。
In order to improve the robustness of language comprehension in colloquial dialogue systems, a two-level search comprehension algorithm based on the largest posterior statistical framework is proposed. The first level uses the concepts to bind the key components of the extracted sentences and eliminate some of the interference components. The second level uses the improved robust syntax analysis based on tree expansion to search for the best comprehension results. At the same time, the user intention inference and sentence feature phrases are introduced into two The information on the search space constraints, to further improve the understanding of the robustness and real-time rate. Experiments show that the proposed algorithm can be applied to train information inquiry. At 0.22 times of real-time, the algorithm can get 13.6% of sentence understanding error rate and 25.4% of concept understanding error rate, which is reduced by 23.2% and 9.3% respectively compared with baseline system.