Study on the learning of formal languages consisting of natural language sentences and their semantic expressions based on distributional learning

2017 Fiscal Year Final Research Report

• Project/Area Number: 26330013
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Theory of informatics
• Research Institution: Tohoku University (2016-2017); Kyoto University (2014-2015)
• Principal Investigator: Yoshinaka Ryo 東北大学, 情報科学研究科, 准教授 (80466424)
• Co-Investigator(Renkei-kenkyūsha): KANAZAWA Makoto 国立情報学研究所, 情報学プリンシプル研究系, 准教授 (20261886)
• Project Period (FY): 2014-04-01 – 2018-03-31
• Keywords: 文法推論

Outline of Final Research Achievements

In recent years, approaches genericallly called ``distirbutional learning'' towards learning mildly context-sensitive languages have been making many positive results. Our project developped the theory of ``distributional learning'' further and tackled learning even more complex grammar formalisms. We gave a uniform view on those existing learning algorithms for mildly context-sensitive languages and derived a general condition with which a grammar formalism shall be distributionally learnable. We targeted grammar formalisms more complex than mildly context-sensitive grammars, including conjunctive grammars, which may define the intersection of languages, and non-linear lambda grammars, which have copying production rules. Furthermore, we have succeeded in weakening the two representative conditions that make grammars distributionally learnable and showed that even richer classes of languages than those used to be defined are distributionally learnable.

Free Research Field

文法推論，アルゴリズム論，形式言語理論