Development of Japanese CCG parser "lightblue"

2022 Fiscal Year Final Research Report

• Project/Area Number: 18H03284
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Single-year Grants
• Section: 一般
• Review Section: Basic Section 61030:Intelligent informatics-related
• Research Institution: Ochanomizu University
• Principal Investigator: Bekki Daisuke お茶の水女子大学, 基幹研究院, 教授 (90431783)
• Project Period (FY): 2018-04-01 – 2022-03-31
• Keywords: 計算言語学 / 組合せ範疇文法 / 統語解析 / 深層ニューラルネットワーク

Outline of Final Research Achievements

The aim of this research was to develop a CCG syntactic parser based on a hybrid method of theoretical linguistics and deep neural networks. We successfully formulated and implemented RNN-CCG, a system that converts the grammatical theory of RNNG from CFG to CCG. We also studied automatic theorem prover with Dependent-Type Semantics (DTS) and successfully implemented a proof search algorithm by Haskell. In parallel, we also conducted research to evaluate the performance of natural language inference tasks with large-scale language models and showed their limitations, and empirical research in theoretical linguistics using dependent tye semantics, such as comparative constructions, Weak Crossover and the Proviso problem.

Free Research Field

計算言語学

Academic Significance and Societal Importance of the Research Achievements

本研究では、理論言語学と機械学習を融合させた深い意味解析を行う手法や、自然言語理解の自動化を可能にするCCGパーザと依存型意味論(DTS)の研究など、自然言語処理における重要な課題に取り組んできました。これらの研究成果は、査読付き国際学会や国内学会の発表を通じて学術的に評価され、また企業向けのセミナーやメディア掲載などを通じて社会に還元されました。今後、自然言語処理のさらなる進展に向けて、理論言語学と深層学習のハイブリッドアプローチを進めることが求められます。