Meaning Beyond Truth Conditions: Evaluating Discourse Level Understanding via Anaphora Accessibility
Xiaomeng Zhu, Zhenghao Zhou, Simon Charlow, Robert Frank
摘要
We present a hierarchy of natural language understanding abilities and argue for the importance of moving beyond assessments of understanding at the lexical and sentence levels to the discourse level. We propose the task of anaphora accessibility as a diagnostic for assessing discourse understanding, and to this end, present an evaluation dataset inspired by theoretical research in dynamic semantics. We evaluate human and LLM performance on our dataset and find that LLMs and humans align on some tasks and diverge on others. Such divergence can be explained by LLMs' reliance on specific lexical items during language comprehension, in contrast to human sensitivity to structural abstractions. Dataset and code: .
问 Lune
尽管问,回答会引用原文。
Lune 会读完全文和相关的 Paper,每个回答都附上所依据的原文段落。
它引用的 Paper5
- Prompting is not a substitute for probability measurements in large language modelsJennifer Hu, Roger LevyEMNLP 2023 · 31 次引用
- TaxoLLaMA: WordNet-based Model for Solving Multiple Lexical Semantic TasksViktor Moskvoretskii, Ekaterina Neminova, Alina Lobanova, Alexander Panchenko 等ACL 2024 · 11 次引用
- Entity Tracking in Language ModelsNajoung Kim, Sebastian SchusterACL 2023 · 9 次引用
- Are Natural Language Inference Models IMPPRESsive? Learning IMPlicature and PRESuppositionPaloma Jeretic, Alex Warstadt, Suvrat Bhooshan, Adina WilliamsACL 2020 · 2 次引用
- Implicit Representations of Meaning in Neural Language ModelsBelinda Z. Li, Maxwell I. Nye, Jacob AndreasACL 2021
相关 Paper
- PUB: A Pragmatics Understanding Benchmark for Assessing LLMs' Pragmatics CapabilitiesSettaluri Lakshmi Sravanthi, Meet Doshi, Pavan Tankala, V. Rudra Murthy 等ACL Findings 2024 · 24 次引用
- LIEDER: Linguistically-Informed Evaluation for Discourse Entity RecognitionXiaomeng Zhu, Robert FrankACL 2024
- What Did You Refer to? Evaluating Co-References in DialogueWeinan Zhang, Yue Zhang, Hanlin Tang, Zhengyu Zhao 等ACL Findings 2021 · 2 次引用
- Probing LLMs for Multilingual Discourse Generalization Through a Unified Label SetFlorian Eichin, Yang Janet Liu, Barbara Plank, Michael A. HedderichACL 2025
- Threads of Subtlety: Detecting Machine-Generated Texts Through Discourse MotifsZae Myung Kim, Kwang Hee Lee, Preston Zhu, Vipul Raheja 等ACL 2024 · 1 次引用
