面向汉语文本理解的语义计算方法

自然语言理解是人工智能研究的核心内容之一，旨在探索人类自身语言能力和思维活动的本质。要实现对汉语文本的正确理解，汉语语义的表示、分析、计算和推理都是不可缺少且亟待解决的手段。本项目将结合面向汉语文本理解的传统方法以及当前流行的深度机器学习方法，以建立有效的高性能汉语语义计算新方法为总体研究目标，从语义表示、语义计算模型和语义推理三个层次展开研究。期望能够创建出一套新的汉语语义计算的理论框架和实现方法。包括：为解决汉语语义知识资源不足的问题，创建有效的语义知识库自动构建新方法；为解决现有基于符号的语义表示和计算方法的稀疏性问题，创建基于深度学习和分布式表达的语义表示和计算新方法；为解决现有推理算法不能考虑全局信息的问题，将现有的语义逻辑演算单元从词汇级别扩展到句子乃至篇章的级别上，满足更加复杂的文本智能推理的需求。在上述理论研究基础上，实现面向特定领域领域的高水平汉语自动问答系统。

Natural language processing is one of the core research content of artificial intelligence, aiming at exploring the nature of humans' language ability and thinking activity. To understand Chinese texts accurately, the representation, analysis, computing and reasoning of Chinese semantics are essential and urgently need to be addressed means. This project combines traditional methods of Chinese understanding and currently popular deep machine learning methods, aiming to build efficient and high-performance new ways of Chinese semantic computing, and studies on three layers: semantic representation, semantic computing and semantic reasoning. A new set of theory frameworks and implement methods of Chinese semantic computing are expected to be built, including: efficient methods of automatically building semantic knowledge base, to settle the problem of shortage of Chinese semantic knowledge resources; new ways of semantic representation and computing based on deep learning and distributed representation, solving the sparse problem of current symbol-based semantic representation and computing methods; expanding the current unit of semantic logical calculus from words to sentences, and even to discourses, meeting more complex demands of text intelligent reasoning, because current reasoning algorithms do not consider global information. To demonstrate our theories and methods, we will implement an advanced domain specific automatic question-answering system.