基于深层学习的汉语句法语义分析研究

项目名称： 基于深层学习的汉语句法语义分析研究

项目编号： No.61273318

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 常宝宝

作者单位： 北京大学

项目金额： 80万元

中文摘要： 句法语义分析的主要任务是：对于输入的自然语言句子，运用计算模型和算法得到句子的句法结构并给句子成分赋以语义角色。句法语义分析是机器翻译、信息提取、自然语言人机接口等应用系统的核心部件，准确高效的句法语义分析对这些应用系统的成功研发有着不可估量的作用。目前句法语义分析在方法上还存在多方面的缺陷，如主要依赖线性化的机器学习技术、过度依赖人工特征工程和有限的标注数据、缺乏有效的联合训练手段，限制了句法语义分析技术的性能。本项目拟将深层学习机制用于汉语句法语义分析，针对目前句法语义分析研究表现出的问题进行探索并期望取得进展。项目除对句法语义分析的基础方法进行探索外，其成果也将能直接支持汉语信息处理应用系统的开发和研究，具有重要的理论意义和应用价值。

中文关键词： 依存句法分析；语义角色标记；汉语分词；深度学习；

英文摘要： Syntactic and Semantic parsing is aimed to generate syntactic structure of natural language sentences and assign proper semantic roles to their constituents.Since syntactic and semantic parser is key component of many natural language appliactions, such as Machine Translation, Information Retrieval, Natural Language User Interface etc., high performance syntactic and semantic paring with satisficatory accuracy is essential to research and development of successful natural language applications. However, state-of-art syntactic and semantic parsing is still error-prone. As we believe, the possible reasons of such low performance might include: (1)limitations in learning capacity of linear machine learning techniques dominating the field, (2)the over-reliance on manual feature engineering, (3)the over-reliance on limited annotated data and (4)the lack of means of joint training. In this project, we propose to use deep learning strategy in Chinese syntactic and semantic parsing, which we beilieve to be a possible way to get away from the current plight of the field or at least to alleviate the problem facing the field. With the adequate preliminary preparation we have already conducted and the well-designed research program, we believe that improvement could be expected.The project focuses on fundamental issues in C

英文关键词： Dependency Parsing；Semantic Role Labeling；Chinese Segmentation；Deep Learning；