{"links":{},"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:02000995","sets":["1164:4179:1740452116224:1740452168372"]},"path":["1740452168372"],"owner":"80578","recid":"2000995","title":["大規模言語モデルによる対話型依存構造解析"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2025-03-01"},"_buckets":{"deposit":"22dec2a9-e7fd-46d6-a26d-0c907f5e3aed"},"_deposit":{"id":"2000995","pid":{"type":"depid","value":"2000995","revision_id":0},"owners":[80578],"status":"published","created_by":80578},"item_title":"大規模言語モデルによる対話型依存構造解析","author_link":[],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"大規模言語モデルによる対話型依存構造解析","subitem_title_language":"ja"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"大規模言語モデル","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2025-03-01","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"株式会社リクルートMegagon Labs, Tokyo"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/2000995/files/IPSJ-NL25263017.pdf","label":"IPSJ-NL25263017.pdf"},"date":[{"dateType":"Available","dateValue":"2027-03-01"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-NL25263017.pdf","filesize":[{"value":"1.2 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"23"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"2bc57b66-31ae-44da-8616-e77ebf14bc3d","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2025 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"松田,寛"}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10115061","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8779","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"近年，構文解析分野でも大規模言語モデル(LLM)の応用が進んでいる．本論文では，事後学習により対話応答性能が強化されたLLMに対して，入力文とその依存構造解析結果の関係を指示-応答のプロンプト形式で表現した教師データでFine-tuningを行い，指示プロンプトにより依存構造解析タスクを実行する対話型依存構造解析を提案する．提案手法は，先行研究で用いられてきた括弧の対応関係による構文木表現は用いず，単語テーブル上で依存先の単語インデックスを参照する形で依存構造を表現することで，可読性を向上すると同時に，交差を含めた依存構造を表現可能にしている．提案手法の解析精度をUniversal Dependenciesの英語と日本語のデータセットで評価した結果，英語においては提案手法が従来のパイプライン型の解析手法の精度を上回り，日本語においては同程度の精度となることを確認したが，ベースモデルの学習時に評価用データセットが混入している可能性を考慮する必要がある．同時に，提案手法の出力は非巡回・シングルルートなど一般的な構文木の制約を満たさない場合がごく稀にあること，ハルシネーションにより系列が部分的に変化し入力と出力と整合しない場合があることを確認した．提案手法は一般的なLLMホスティングサービス上で容易に実現できるため，従来のパイプライン型の解析処理に比べて開発負荷を大きく減らすことが可能であり，今後の利用の広まりが期待される．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"7","bibliographic_titles":[{"bibliographic_title":"研究報告自然言語処理（NL）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2025-03-01","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"17","bibliographicVolumeNumber":"2025-NL-263"}]},"relation_version_is_last":true,"weko_creator_id":"80578"},"updated":"2025-02-25T05:35:33.253869+00:00","created":"2025-02-25T05:35:29.282272+00:00","id":2000995}