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

Natural Language Interface Technology to Support Complex Tasks

Research Project

Project/Area Number 21H03502
Research Category

Grant-in-Aid for Scientific Research (B)

Allocation TypeSingle-year Grants
Section一般
Review Section Basic Section 61030:Intelligent informatics-related
Research InstitutionNational Institute of Informatics

Principal Investigator

Aizawa Akiko  国立情報学研究所, コンテンツ科学研究系, 教授 (90222447)

Project Period (FY) 2021-04-01 – 2024-03-31
Project Status Completed (Fiscal Year 2023)
Budget Amount *help
¥17,290,000 (Direct Cost: ¥13,300,000、Indirect Cost: ¥3,990,000)
Fiscal Year 2023: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000)
Fiscal Year 2022: ¥6,110,000 (Direct Cost: ¥4,700,000、Indirect Cost: ¥1,410,000)
Fiscal Year 2021: ¥6,630,000 (Direct Cost: ¥5,100,000、Indirect Cost: ¥1,530,000)
Keywords言語モデル / 多段階推論 / 対話システム / 意味解析 / 基盤化 / 視覚言語
Outline of Research at the Start

本研究では、複雑な作業を言語でコンピュータに指示するための対話技術の実現を目指す。ここでの「複雑な作業」とは、事前登録した関数を呼び出すだけでは解決できず、複数の関数の呼び出しや詳細な条件設定などが必要となる処理である。このような複雑な作業は、単発の自然な発話では指示が困難であることから、本研究では対話に焦点をあてて、基盤化による共通理解モデルの構築や意味解析技術の研究に取り組む。

Outline of Final Research Achievements

In this study, we addressed natural language processing for instructing computers to perform complex tasks in natural language. Complex tasks” in this study are those processes that cannot be solved by solely using single-question answering, but require multi-step reasoning and conversational interactions. In the research period, we worked on multi-step machine reading comprehension, which combines information from multiple source documents to answer a single question; semantic parsing, which converts natural language sentences into SQL queries or executable programs; and mutual grounding, which identifies entities commonly referred to by multiple interlocutors in their utterances. We have constructed datasets for the construction and evaluation of language models, and demonstrated their usefulness through analytical evaluation of the models.

Academic Significance and Societal Importance of the Research Achievements

生成型の言語モデルの登場と急速な大規模化によって、言語モデルの推論能力は劇的に向上している。このような言語モデルは、さまざまな場面における人間の作業支援ツールとして大きな期待が寄せられる一方で、実際の環境において必要となる複雑な指示の理解やコマンド生成などの能力については、さらなる改良が求められている。また、複数の情報ソースに分散する情報をさがしあてて、質問への正確な回答を生成する処理は、ハルシネーション対策としても重要である。本研究では、複雑な推論タスクに関する複数のデータセットを構築して公開しており、今後の研究開発における言語モデルの訓練・評価に貢献するものである。

Report

(4 results)
  • 2023 Annual Research Report   Final Research Report ( PDF )
  • 2022 Annual Research Report
  • 2021 Annual Research Report
  • Research Products

    (15 results)

All 2023 2022 2021

All Journal Article (12 results) (of which Peer Reviewed: 11 results,  Open Access: 11 results) Presentation (3 results)

  • [Journal Article] Predicting Numerals in Text Using Nearest Neighbor Language Models2023

    • Author(s)
      Taku Sakamoto, Akiko Aizawa
    • Journal Title

      Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL)

      Volume: - Pages: 4795-4809

    • DOI

      10.18653/v1/2023.findings-acl.295

    • Related Report
      2023 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Analyzing the Effectiveness of the Underlying Reasoning Tasks in Multi-hop Question Answering2023

    • Author(s)
      Ho Xanh、Duong Nguyen Anh-Khoa、Sugawara Saku、Aizawa Akiko
    • Journal Title

      Findings of the Association for Computational Linguistics: EACL 2023

      Volume: 1 Pages: 1163-1180

    • DOI

      10.18653/v1/2023.findings-eacl.87

    • Related Report
      2023 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] How Well Do Multi-hop Reading Comprehension Models Understand Date Information?2022

    • Author(s)
      Xanh Ho, Saku Sugawara, Akiko Aizawa
    • Journal Title

      Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing (Short Papers)

      Volume: - Pages: 470-479

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Debiasing Masks: A New Framework for Shortcut Mitigation in NLU2022

    • Author(s)
      Johannes Mario Meissner, Saku Sugawara, Akiko Aizawa
    • Journal Title

      Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing

      Volume: - Pages: 7607-7613

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Measuring Text-to-SQL Semantic Parsing Model on the Question Generalizability2022

    • Author(s)
      Thanakrit Julavanich and Akiko Aizawa
    • Journal Title

      Proceedings of the 6th International Conference on Natural Language Processing and Information Retrieval (NLPIR 2022)

      Volume: -

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed
  • [Journal Article] Do BERTs Learn to Use Browser User Interface? Exploring Multi-Step Tasks with Unified Vision-and-Language BERTs2022

    • Author(s)
      Taichi Iki and Akiko Aizawa
    • Journal Title

      arXiv: 2203.07828

      Volume: -

    • Related Report
      2022 Annual Research Report
    • Open Access
  • [Journal Article] Penalizing Confident Predictions on Largely Perturbed Inputs Does Not Improve Out-of-Distribution Generalization in Question Answering2022

    • Author(s)
      Kazutoshi Shinoda, Saku Sugawara, and Akiko Aizawa
    • Journal Title

      The AAAI-23 Workshop on Knowledge Augmented Methods for NLP (KnowledgeNLP)

      Volume: -

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Which Shortcut Solution Do Question Answering Models Prefer to Learn?2022

    • Author(s)
      Kazutoshi Shinoda, Saku Sugawara, and Akiko Aizawa
    • Journal Title

      Proceedings of the 37th AAAI conference on Artificial Intelligence (AAAI)

      Volume: -

    • Related Report
      2022 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Effect of Visual Extensions on Natural Language Understanding in Vision-and-Language Models.2021

    • Author(s)
      Taichi Iki, Akiko Aizawa
    • Journal Title

      The 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP 2021)

      Volume: - Pages: 2189-2196

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Maintaining Common Ground in Dynamic Environments.2021

    • Author(s)
      Takuma Udagawa, Akiko Aizawa
    • Journal Title

      Transactions of the Association for Computational Linguistics

      Volume: 9 Pages: 995-1011

    • DOI

      10.1162/tacl_a_00409

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Can Question Generation Debias Question Answering Models? A Case Study on Question-Context Lexical Overlap.2021

    • Author(s)
      Kazutoshi Shinoda, Saku Sugawara, Akiko Aizawa
    • Journal Title

      The 3rd Workshop on Machine Reading for Question Answering (MRQA), at the 2021 conference on Empirical Methods in Natural Language Processing (EMNLP)

      Volume: - Pages: 63-72

    • DOI

      10.18653/v1/2021.mrqa-1.6

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Journal Article] Predicting Numerals in Natural Language Text Using a Language Model Considering the Quantitative Aspects of Numerals.2021

    • Author(s)
      Taku Sakamoto, Akiko Aizawa
    • Journal Title

      The Second Workshop on Knowledge Extraction and Integration for Deep Learning Architectures (DeeLIO), collocated with NAACL 2021

      Volume: - Pages: 140-150

    • DOI

      10.18653/v1/2021.deelio-1.14

    • Related Report
      2021 Annual Research Report
    • Peer Reviewed / Open Access
  • [Presentation] k近傍言語モデルを用いたテキスト中の数字の予測2022

    • Author(s)
      阪本 拓功, 相澤 彰子
    • Organizer
      人工知能学会 第36回全国大会 (JSAI2022)
    • Related Report
      2022 Annual Research Report
  • [Presentation] 抽出型質問応答における相対位置バイアスの除去2022

    • Author(s)
      篠田一聡, 菅原朔, 相澤彰子
    • Organizer
      言語処理学会第28回年次大会(NLP2022)
    • Related Report
      2021 Annual Research Report
  • [Presentation] 言語モデルに対するトークンのキャンセルアウト手法の比較2022

    • Author(s)
      鈴木淳平, 菅原朔, 相澤彰子
    • Organizer
      言語処理学会第28回年次大会(NLP2022)
    • Related Report
      2021 Annual Research Report

URL: 

Published: 2021-04-28   Modified: 2025-01-30  

Information User Guide FAQ News Terms of Use Attribution of KAKENHI

Powered by NII kakenhi