RI-Small: Unsupervised Learning of Meaning

RI-Small：无监督意义学习

• 批准号: 0811974
• 负责人: Daniel Jurafsky
• 金额: $ 45万
• 依托单位: Stanford University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-09-01 至 2012-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0811974&HistoricalAwards=false
• 关键词: RI; Small; Unsupervised; Learning; Meaning

The goal of this research to to develop a new generation of robust, highlyeffective unsupervised and semi-supervised models of meaningextraction. Our key methological insights include the use ofglobal methods of inference to simultaneously consider many linguisticaspects of the task, the use of rich typed lexical dependencies,and the semi-supervised use of structured data from the web, suchas dictionaries, thesauruses, encyclopedias, and so on. We areextracting meaning at three levels: word meaning, propositionalmeaning, and conceptual meaning. At the word level, we are learninglexical relations like hyponymy (leptin is-a hormone), synonymy,and others both from raw text and from structured sources like on-linedictionaries. At the propositional level, we are learningpredicate-argument structure using a global unsupervised clusteringmodel as well as developing semi-supervised methods of learning semantic frameextractors. At a larger structural level we are inducing scriptsand structured narrative relations between verbs. These rich modelsof meaning will have a broader impact by providing a critical steptowards the creation of systems with true language understandingcapabilities. The results could thus impact the creation ofnatural-language applications in every field, from educational ortutorial applications, to information extraction tasks like legaldiscovery, to conversational agents. All deliverables of thisproject will be available on the web: WordNet expansions, inducedframes and scripts, our temporal event classifier, and semanticrole, frame, and script inducers.

本研究的目标是开发新一代鲁棒、高效的无监督和半监督意义提取模型。 我们的主要方法论见解包括使用全局推理方法来同时考虑任务的许多语言方面，使用丰富的类型化词汇依赖，以及半监督地使用来自网络的结构化数据，如字典，同义词，词典等。我们在三个层面上提取意义：词义，命题意义和概念意义。 在单词层面上，我们学习词汇关系，如上下义关系（瘦素是一种激素），同义关系，以及其他来自原始文本和结构化资源（如在线词典）的词汇关系。在命题层面，我们正在使用全局无监督聚类模型学习谓词-论元结构，并开发学习语义框架提取器的半监督方法。 在一个更大的结构层面上，我们正在归纳脚本和动词之间的结构化叙事关系。 这些丰富的意义模型将产生更广泛的影响，为创造具有真正语言理解能力的系统迈出关键一步。 因此，这些结果可能会影响每个领域的自然语言应用程序的创建，从教育或教程应用程序，到信息提取任务，如法律发现，再到会话代理。 该项目的所有交付成果都将在网络上提供：WordNet扩展，诱导框架和脚本，我们的时间事件分类器，以及语义角色，框架和脚本诱导器。