Grounding text and knowledge base by using distributed representation and their compositions

2017 Fiscal Year Final Research Report

• Project/Area Number: 15H05318
• Research Category: Grant-in-Aid for Young Scientists (A)
• Allocation Type: Single-year Grants
• Research Field: Intelligent informatics
• Research Institution: Tokyo Institute of Technology (2017); Tohoku University (2015-2016)
• Principal Investigator: Okazaki Naoaki 東京工業大学, 情報理工学院, 教授 (50601118)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 自然言語処理 / 人工知能 / 深層学習 / 分散表現 / 知識ベース

Outline of Final Research Achievements

In order to realize computers that can understand and infer on natural languages, a mechanism for collecting and utilizing the commonsense knowledge is essential. We need address three research topics, acquisition of a large-scale knowledge base, grounding a text with the knowledge base, and an inference method on the grounded text. In this project, we built corpora where a text is grounded on instances of a knowledge base. In order to associate relational patterns into a knowledge base, we proposed a novel method for composing a vector of a relation pattern from its constituent words and for disambiguating the sense of a relation pattern. We demonstrated the effectiveness of these studies not only by the experiments on individual tasks but also by applying downstream tasks such as question answering and stance detection.

Free Research Field

自然言語処理