---

ACL2025読み会@名大

---

概要

• 会場: 名古屋大学工学部IB電子情報館中棟1階 IB015講義室
• 日時: 2025年09月26日(金)
• 形式: 1人18分(発表15分、質疑3分)

---

プログラム

• 9:30-9:35: Opening [大鹿, 矢野 slide]
• 9:35-10:30: Session1: LLM:応用
  • Evaluating Language Models as Synthetic Data Generators [大鹿, slide]
  • Completing A Systematic Review in Hours instead of Months with Interactive AI Agents [徳久, slide]
  • Faithful and Robust LLM-Driven Theorem Proving for NLI Explanations [加藤, slide]
• 10:40-11:35: Session2: データセット、ベンチマーク
  • ConLoan: A Contrastive Multilingual Dataset for Evaluating Loanwords [田中, slide]
  • RAGEval: Scenario Specific RAG Evaluation Dataset Generation Framework [小島, slide]
  • Can Knowledge Graphs Make Large Language Models More Trustworthy? An Empirical Study Over Open-ended Question Answering [黄, slide]
• 11:35-12:35: Lunch Break
• 12:35-13:30: Session3: モデル理解
  • The Impact of Token Granularity on the Predictive Power of Language Model Surprisal [笹野, slide]
  • A Theory of Response Sampling in LLMs: Part Descriptive and Part Prescriptive [井田, slide]
  • Understanding the Dark Side of LLMs’ Intrinsic Self-Correction [張, slide]
• 13:40-14:35: Session4: 埋め込み表現
  • On the Acquisition of Shared Grammatical Representations in Bilingual Language Models [山口, slide]
  • Length-Induced Embedding Collapse in PLM-based Models [矢野, slide]
  • Mapping 1,000+ Language Models via the Log-Likelihood Vector [木下, slide]
• 14:35-14:55: Short Break
• 14:55-15:31: Session5: 文書理解
  • The Noisy Path from Source to Citation: Measuring How Scholars Engage with Past Research [伊藤, slide]
  • Hierarchical Document Refinement for Long-context Retrieval-augmented Generation [韓, slide]
• 15:40-16:16: Session6: LLM:推論
  • Large Language and Reasoning Models are Shallow Disjunctive Reasoners [大崎, slide]
  • Can Language Models Reason about Individualistic Human Values and Preferences? [銭本, slide]
• 16:16-16:40: Short Break
• 16:40-17:35: Session7: モデル効率化
  • 500xCompressor: Generalized Prompt Compression for Large Language Models [井上, slide]
  • Byte Latent Transformer: Patches Scale Better Than Tokens [佐藤, slide]
  • Unifying Continuous and Discrete Text Diffusion with Non-simultaneous Diffusion Processes [金児, slide]
• 17:35-17:40: Closing

---

発表者

• 笹野遼平, 韓毅, 大鹿雅史, 矢野千紘, 佐藤蒼馬, 田中邦朋, 張有馳 (笹野研), 銭本友樹 (東中研), 加藤芳秀, 伊藤滉一郎, 黄晟昊 (松原研), 井田龍希, 井上純大, 金児一矢, 大崎稜司, 木下貴人, 小島亮 (豊田工大), 徳久良子 (愛工大), 山口真 (静大)

---

過去の読み会

• ACL2024読み会@名大
• ACL2023読み会@名大
• ACL2019読み会@名大
• ACL2018読み会@名大

---