• Search Research Projects
  • Search Researchers
  • How to Use
  1. Back to previous page

Compositional Inference Systems based on Formal Semantics and Natural Language Processing

Research Project

Project/Area Number 20K19868
Research Category

Grant-in-Aid for Early-Career Scientists

Allocation TypeMulti-year Fund
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionThe University of Tokyo (2021-2023)
Institute of Physical and Chemical Research (2020)

Principal Investigator

Yanaka Hitomi  東京大学, 大学院情報理工学系研究科, 准教授 (10854581)

Project Period (FY) 2020-04-01 – 2024-03-31
Project Status Completed (Fiscal Year 2023)
Budget Amount *help
¥3,900,000 (Direct Cost: ¥3,000,000、Indirect Cost: ¥900,000)
Fiscal Year 2023: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2022: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2021: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2020: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Keywords自然言語推論 / 体系性 / 構成性 / 深層学習 / 自然言語処理 / 形式意味論 / 意味解析 / 事前訓練済み言語モデル / 構成性原理 / 人工知能 / 汎化性能 / 単調性 / 推移性
Outline of Research at the Start

文の意味を計算処理可能な形式で表し、文と文との意味的関係を判定する含意関係認識の実現は、計算機による言語理解に向けて解決すべき最重要課題である。近年、深層学習に基づく言語モデルを用いた含意関係認識の研究が盛んに行われているが、文の構成的意味における言語モデルの表現力は明らかではなく、未知のデータに対する頑健性や表現学習の効率性が不透明である。一方で、形式意味論では、文の構成的意味を推論の妥当性から分析する研究が成熟しつつある。そこで本研究では、形式意味論と自然言語処理の融合による構成的言語モデルの実現を目指す。

Outline of Final Research Achievements

Deep learning models have been well studied in natural language processing. However, the extent to which models capture the compositional meaning of sentences is not clear, and their robustness to unseen data is uncertain. In this study, we investigate the extent to which models capture the compositional meaning of sentences, and develop inference systems that consider the compositional meaning of sentences.
Specifically, we constructed English and Japanese benchmarks to analyze whether models capture the compositional meaning of sentences based on the systematicity of inference, and identified the generalization capacity of current deep learning models. Furthermore, we developed inference systems that map sentences into their semantic representations based on compositional semantics and perform inference. Some of the results were accepted by top international conferences and journals, such as ACL and TACL. Our benchmarks and inference systems have been made available for research use.

Academic Significance and Societal Importance of the Research Achievements

本研究成果の中で特定された深層学習のモデルの文の意味における汎化性能の課題は、大規模言語モデルの信頼性に関わり、社会的影響が大きいものである。また、既存の自然言語推論ベンチマークの多くは英語であり、日本語の自然言語推論ベンチマークの枯渇は日本語の言語処理技術の発展に向けて深刻な問題である。本研究成果の一部である日本語の自然言語推論ベンチマークは、いずれも研究利用可能な形で公開しており、大規模言語モデルの基礎解析や日本語の言語処理技術の評価に貢献するものである。

Report

(5 results)
  • 2023 Annual Research Report   Final Research Report ( PDF )
  • 2022 Research-status Report
  • 2021 Research-status Report
  • 2020 Research-status Report
  • Research Products

    (40 results)

All 2024 2023 2022 2021 2020 Other

All Journal Article (8 results) (of which Peer Reviewed: 8 results,  Open Access: 8 results) Presentation (28 results) (of which Int'l Joint Research: 9 results,  Invited: 6 results) Book (1 results) Remarks (3 results)

  • [Journal Article] Do AIs obtain foundations of language? From the viewpoint of inferential systematicity2024

    • Author(s)
      谷中 瞳、峯島 宏次
    • Journal Title

      Cognitive Studies: Bulletin of the Japanese Cognitive Science Society

      Volume: 31 Issue: 1 Pages: 27-45

    • DOI

      10.11225/cs.2023.078

    • ISSN
      1341-7924, 1881-5995
    • Year and Date
      2024-03-01
    • Related Report
      2023 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Medc2l: Compound-Word Analysis and Inference System for Japanese Clinical Texts2023

    • Author(s)
      Ishida Mana、Yanaka Hitomi、Bekki Daisuke
    • Journal Title

      Journal of Natural Language Processing

      Volume: 30 Issue: 3 Pages: 935-958

    • DOI

      10.5715/jnlp.30.935

    • ISSN
      1340-7619, 2185-8314
    • Related Report
      2023 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Compositional Evaluation on Japanese Textual Entailment and Similarity2022

    • Author(s)
      Hitomi Yanaka, Koji Mineshima
    • Journal Title

      Transactions of the Association for Computational Linguistics

      Volume: 10 Pages: 1266-1284

    • DOI

      10.1162/tacl_a_00518

    • Related Report
      2022 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] Compositional Semantics and Inference System for Temporal Order based on Japanese CCG2022

    • Author(s)
      Sugimoto Tomoki、Yanaka Hitomi
    • Journal Title

      Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop

      Volume: - Pages: 104-114

    • DOI

      10.18653/v1/2022.acl-srw.10

    • Related Report
      2022 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] SyGNS: A Systematic Generalization Testbed Based on Natural Language Semantics2021

    • Author(s)
      Hitomi Yanaka, Koji Mineshima, and Kentaro Inui
    • Journal Title

      Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021

      Volume: - Pages: 103-119

    • DOI

      10.18653/v1/2021.findings-acl.10

    • Related Report
      2021 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] Do Grammatical Error Correction Models Realize Grammatical Generalization?2021

    • Author(s)
      Mita Masato、Yanaka Hitomi
    • Journal Title

      Findings of the Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (ACL-IJCNLP2021)

      Volume: - Pages: 4554-4561

    • DOI

      10.18653/v1/2021.findings-acl.399

    • Related Report
      2021 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] Exploring Transitivity in Neural NLI Models through Veridicality2021

    • Author(s)
      Yanaka Hitomi、Mineshima Koji、Inui Kentaro
    • Journal Title

      Proceedings of the 16th conference of the European Chapter of the Association for Computational Linguistics (EACL2021)

      Volume: 1 Pages: 920-934

    • Related Report
      2020 Research-status Report
    • Peer Reviewed / Open Access
  • [Journal Article] Do Neural Models Learn Systematicity of Monotonicity Inference in Natural Language?2020

    • Author(s)
      Yanaka Hitomi、Mineshima Koji、Bekki Daisuke、Inui Kentaro
    • Journal Title

      Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL2020)

      Volume: 1 Pages: 6105-6117

    • DOI

      10.18653/v1/2020.acl-main.543

    • NAID

      130007956038

    • Related Report
      2020 Research-status Report
    • Peer Reviewed / Open Access
  • [Presentation] Reforging: A Method for Constructing a Linguistically Valid Japanese CCG Treebank2024

    • Author(s)
      Asa Tomita, Hitomi Yanaka, Daisuke Bekki
    • Organizer
      Proceedings of the EACL2024 Student Research Workshop
    • Related Report
      2023 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 文字系列情報による性能への影響からニューラルモデルが有する言語的な傾向を見出せるか2024

    • Author(s)
      黒澤友哉, 谷中瞳
    • Organizer
      言語処理学会第30回年次大会
    • Related Report
      2023 Annual Research Report
  • [Presentation] 言語学的に妥当な日本語CCGツリーバンクの構築と評価2024

    • Author(s)
      富田朝, 谷中瞳, 戸次大介
    • Organizer
      言語処理学会第30回年次大会
    • Related Report
      2023 Annual Research Report
  • [Presentation] Jamp: Controlled Japanese Temporal Inference Dataset for Evaluating Generalization Capacity of Language Models2023

    • Author(s)
      Tomoki Sugimoto, Yasumasa Onoe, Hitomi Yanaka
    • Organizer
      Proceedings of the ACL2023 Student Research Workshop (SRW)
    • Related Report
      2023 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 多言語DRS意味解析における文字系列情報の性能分析2023

    • Author(s)
      黒澤友哉, 谷中瞳
    • Organizer
      人工知能学会第37回全国大会
    • Related Report
      2023 Annual Research Report
  • [Presentation] 言語学的に妥当なCCGツリーバンク構築の試み2023

    • Author(s)
      富田朝, 谷中瞳, 戸次大介
    • Organizer
      人工知能学会第37回全国大会
    • Related Report
      2023 Annual Research Report
  • [Presentation] Is Japanese CCGBank empirically correct? A case study of passive and causative constructions2023

    • Author(s)
      Daisuke Bekki, Hitomi Yanaka
    • Organizer
      Proceedings of the 21st International Workshop on Treebanks and Linguistic Theories
    • Related Report
      2023 Annual Research Report
    • Int'l Joint Research
  • [Presentation] 日本語CCGBankは言語学的に妥当か2023

    • Author(s)
      戸次大介, 谷中瞳
    • Organizer
      言語処理学会第29回年次大会
    • Related Report
      2023 Annual Research Report
  • [Presentation] 時間的順序を考慮した日本語論理推論システムの構築2022

    • Author(s)
      杉本智紀, 谷中瞳
    • Organizer
      人工知能学会第36回全国大会
    • Related Report
      2022 Research-status Report
  • [Presentation] 共起確率を用いた構文木の自動選択による推論システムの改善2022

    • Author(s)
      井上裕太, 谷中瞳
    • Organizer
      NLP若手の会第17回シンポジウム
    • Related Report
      2022 Research-status Report
  • [Presentation] サプライザルを利用した日本語の流暢性フィルタリングの試み2022

    • Author(s)
      田村鴻希, 土井惟成, 西田直人, Junjie Chen, 谷中瞳
    • Organizer
      言語処理学会第29回年次大会
    • Related Report
      2022 Research-status Report
  • [Presentation] 説明可能な検索ベースの文書分類手法の提案2022

    • Author(s)
      中井優, 中野雄介, 徳永優也, 上田亮, 谷中瞳
    • Organizer
      言語処理学会第29回年次大会
    • Related Report
      2022 Research-status Report
  • [Presentation] Revisiting the Systematicity Argument through Analyzing Deep Neural Networks2022

    • Author(s)
      Hitomi Yanaka
    • Organizer
      Aspects of Logic Study, celebrating World Logic Day (WLD2022)
    • Related Report
      2022 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] Exploring the Generalization Ability of Neural Models through Natural Language Inference2022

    • Author(s)
      Hitomi Yanaka
    • Organizer
      the 6th International Workshop on Symbolic-Neural Learning (SNL2022)
    • Related Report
      2022 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] 論理に基づく推論システムの再訪2022

    • Author(s)
      谷中瞳
    • Organizer
      情報処理学会第253回自然言語処理研究会
    • Related Report
      2022 Research-status Report
    • Invited
  • [Presentation] JSICK:構成的推論・類似度データセットSICK日本語版の紹介2022

    • Author(s)
      谷中瞳
    • Organizer
      NLPコロキウム
    • Related Report
      2022 Research-status Report
    • Invited
  • [Presentation] 形式言語学の知見を活用した自然言語推論2022

    • Author(s)
      谷中瞳
    • Organizer
      第31回ステアラボ人工知能セミナー
    • Related Report
      2022 Research-status Report
    • Invited
  • [Presentation] Logical Inference System for Temporal Order based on Japanese CCG2022

    • Author(s)
      Tomoki Sugimoto, Hitomi Yanaka
    • Organizer
      the ESSLLI2022 Workshop Natural Logic Meets Machine Learning (NALOMA2022)
    • Related Report
      2022 Research-status Report
    • Int'l Joint Research
  • [Presentation] JSICK: 日本語構成的推論・類似度データセットの構築2021

    • Author(s)
      谷中瞳, 峯島宏次
    • Organizer
      人工知能学会第35回全国大会
    • Related Report
      2021 Research-status Report
  • [Presentation] ワークショップ:哲学の応用と社会実装 ─ヘイトスピーチをめぐる 文理共創研究の可能性と課題─2021

    • Author(s)
      荒井ひろみ, 和泉悠, 朱喜哲, 辻大介, 仲宗根勝仁, 谷中瞳
    • Organizer
      応用哲学会第十三回年次大会
    • Related Report
      2021 Research-status Report
  • [Presentation] Abusive Tweets in Japanese during the COVID-19 pandemic2021

    • Author(s)
      Yu Izumi, Hiromi Arai, Hitomi Yanaka, Katsuhito Nakasone, Heechul Ju
    • Organizer
      the 3rd International Workshop HATE SPEECH IN ASIA AND EUROPE Pandemic, Fear, and Hate
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research
  • [Presentation] Do Neural Models Learn Transitivity of Veridical Inference?2021

    • Author(s)
      Hitomi Yanaka, Koji Mineshima, Kentaro Inui
    • Organizer
      the IWCS2021 Workshop Natural Logic Meets Machine Learning
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research
  • [Presentation] Revisiting the Systematicity Argument through Analyzing Deep Neural Networks2021

    • Author(s)
      Hitomi Yanaka
    • Organizer
      Aspects of Logic Study, celebrating World Logic Day (WLD2022)
    • Related Report
      2021 Research-status Report
    • Int'l Joint Research / Invited
  • [Presentation] ニューラルネットが学習する意味表現は体系性を持つか2021

    • Author(s)
      谷中瞳, 峯島宏次, 乾 健太郎
    • Organizer
      言語処理学会第27回年次大会
    • Related Report
      2020 Research-status Report
  • [Presentation] 文法誤り訂正モデルは訂正に必要な文法を学習しているか2021

    • Author(s)
      三田雅人, 谷中瞳
    • Organizer
      言語処理学会第27回年次大会
    • Related Report
      2020 Research-status Report
  • [Presentation] ニューラルネットワークによる自然言語推論の可能性2020

    • Author(s)
      谷中瞳, 峯島宏次, 戸次大介, 乾 健太郎
    • Organizer
      人工知能学会第34回全国大会
    • Related Report
      2020 Research-status Report
  • [Presentation] ニューラルネットは自然言語推論の体系性を学習するか2020

    • Author(s)
      谷中瞳, 峯島宏次, 戸次大介, 乾 健太郎
    • Organizer
      言語処理学会第26回年次大会
    • Related Report
      2020 Research-status Report
  • [Presentation] 文法誤り訂正モデルは文法知識を汎化しているか2020

    • Author(s)
      三田雅人, 谷中瞳
    • Organizer
      NLP若手の会第15回シンポジウム
    • Related Report
      2020 Research-status Report
  • [Book] AIはレイシズムと戦えるのか―自然言語処理分野におけるヘイトスピーチ自動検出研究の現状と課題2021

    • Author(s)
      和泉 悠, 仲宗根勝仁, 朱喜哲, 谷中 瞳, 荒井ひろみ
    • Total Pages
      17
    • Publisher
      岩波書店
    • Related Report
      2021 Research-status Report
  • [Remarks] JAMP: Japanese Temporal Inference Dataset

    • URL

      https://github.com/ynklab/Jamp

    • Related Report
      2023 Annual Research Report
  • [Remarks] JSICK

    • URL

      https://github.com/verypluming/JSICK

    • Related Report
      2022 Research-status Report
  • [Remarks] ccgtemp

    • URL

      https://github.com/ynklab/ccgtemp

    • Related Report
      2022 Research-status Report

URL: 

Published: 2020-04-28   Modified: 2025-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi