首页> 外国专利> SEMANTIC LABEL APPLICATION MODEL LEARNING DEVICE, SEMANTIC LABEL APPLICATION DEVICE, SEMANTIC LABEL APPLICATION MODEL LEARNING METHOD AND PROGRAM

SEMANTIC LABEL APPLICATION MODEL LEARNING DEVICE, SEMANTIC LABEL APPLICATION DEVICE, SEMANTIC LABEL APPLICATION MODEL LEARNING METHOD AND PROGRAM

机译:语义标签应用模型学习设备,语义标签应用设备,语义标签应用模型学习方法和程序

摘要

PROBLEM TO BE SOLVED: To apply an appropriate semantic label for a functional expression of a predicate that becomes an interrogative expression or an assertive expression depending on the flow of conversation.SOLUTION: Learning utterance pair creation unit 22 creates an utterance pair for learning on the basis of a correct answer corpus in which correct answer labels of a functional expression and a response expression are applied to respective morpheme analysis results. An initial value of weight is set in a parameter table for each of a plurality of kinds of identities using, as identity, a row of identities which are a semantic label of a functional expression that appears at the end of the former utterance of the utterance pair and a semantic label of a response expression that appears at the beginning of the latter utterance of the utterance pair. A parameter table creation unit 24 establishes a lattice including all candidate semantic labels for each morpheme using morphological information of the utterance pair and a functional expression dictionary 30, and searches a maximum likelihood path from the lattice structure on the basis of the weight for each identity in the parameter table. The weight in the parameter table is learned so that the maximum likelihood path becomes a semantic label string of correct answers.
机译:解决的问题:为谓词的功能性表达式应用适当的语义标签,该谓词的功能性表达式根据对话的流程而变成疑问句或主张性表达式。解决方案:学习话语对创建单元22创建用于在谓词上学习的话语对。正确答案语料库的基础,其中将功能表达式和响应表达式的正确答案标签应用于相应的词素分析结果。以一排身份作为身份,在参数表中为多种身份中的每一种设置权重的初始值,该行身份是出现在前一发音之前的功能表达式的语义标签对和响应表达的语义标签,该表达出现在言语对的后一个言语的开头。参数表创建单元24使用话语对的形态信息和功能表达式字典30来建立包括针对每个语素的所有候选语义标签的格,并且基于权重为每个标识从格结构中搜索最大似然路径。在参数表中。学习参数表中的权重,以便最大似然路径成为正确答案的语义标签字符串。

著录项

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号