Lecture Translation Benchmark from Online Educational Videos

• Item type: SIG Technical Reports(1)
• 公開日: 2026-02-28
• タイトル / 言語: ja
• タイトル: Lecture Translation Benchmark from Online Educational Videos
• タイトル / 言語: en
• タイトル: Lecture Translation Benchmark from Online Educational Videos
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 機械翻訳
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: 奈良先端科学技術大学院大学
• 著者所属: 同志社大学
• 著者所属: 奈良先端科学技術大学院大学
• 著者所属: 奈良先端科学技術大学院大学
• 著者所属: 奈良先端科学技術大学院大学
• 著者所属: 同志社大学
• 著者所属: 奈良先端科学技術大学院大学
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Doshisha University
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Doshisha University
• 著者所属(英): en
• 著者所属(英): Nara Institute of Science and Technology
• 著者名: Frederikus,Hudi; Nanakorn,Nina; Justin,Vasselli; Yusuke,Sakai; Hidetaka,Kamigaito; Akihiro,Tamura; Taro,Watanabe
• 著者名(英): Frederikus Hudi; Nanakorn Nina; Justin Vasselli; Yusuke Sakai; Hidetaka Kamigaito; Akihiro Tamura; Taro Watanabe
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: We introduce LecTrans, a large-scale benchmark for evaluating multimodal translation in academic lectures. While existing MT benchmarks primarily focus on text-only or tightly aligned inputs, academic lectures require translation across spoken and visual modalities in long-form, knowledge-intensive settings, making them a challenging and informative testbed for multimodal language understanding. LecTrans is composed of approximately 350 hours of expert-taught online lectures and includes transcripts, presentation slides, and professionally verified translations in seven languages. A 162-minute evaluation subset further includes human-corrected transcripts and domain-term annotations across 17 subjects. We formulate lecture translation as a dual-task evaluation problem that distinguishes between transcription translation and slide translation. This formulation enables modality-aware analysis of translation behavior and error propagation. We provide strong text-only and multimodal baselines and present analysis that reveals challenges in long-context and domain-specific translation. LecTrans serves as a benchmark for assessing the capabilities of multimodal translation systems on academic content.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: We introduce LecTrans, a large-scale benchmark for evaluating multimodal translation in academic lectures. While existing MT benchmarks primarily focus on text-only or tightly aligned inputs, academic lectures require translation across spoken and visual modalities in long-form, knowledge-intensive settings, making them a challenging and informative testbed for multimodal language understanding. LecTrans is composed of approximately 350 hours of expert-taught online lectures and includes transcripts, presentation slides, and professionally verified translations in seven languages. A 162-minute evaluation subset further includes human-corrected transcripts and domain-term annotations across 17 subjects. We formulate lecture translation as a dual-task evaluation problem that distinguishes between transcription translation and slide translation. This formulation enables modality-aware analysis of translation behavior and error propagation. We provide strong text-only and multimodal baselines and present analysis that reveals challenges in long-context and domain-specific translation. LecTrans serves as a benchmark for assessing the capabilities of multimodal translation systems on academic content.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10115061
• 書誌情報: 研究報告自然言語処理（NL）; 巻 2026-NL-267, 号 2, p. 1-8, 発行日 2026-02-28
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8779
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会