DiscoTrace wins EMNLP 2026 Best Paper: LLM answers lack humans' community-specific rhetorical diversity

TuhinChakr · x · 2026-10-28

DiscoTrace, by Neha Srikanth, Jordan Boyd-Graber and Rachel Rudinger (University of Maryland), won Best Paper at EMNLP 2026.

Key takeaways:

• The method represents answers as sequences of question-related discourse acts paired with question interpretations, annotated over Rhetorical Structure Theory parses
• Applied to answers from nine communities, it reveals diverse rhetorical preferences; LLMs show no such diversity even when prompted to mimic community guidelines
• LLMs systematically opt for breadth, addressing interpretations human answerers deliberately skip
• The framework can guide building more pragmatic, context-sensitive LLM answerers

First author Srikanth also won an Outstanding Paper at NAACL 2025 and is on the job market.

Original post →

More from Research

• TaskNet uses AI interviews to map tasks across 6 occupations, surfacing overlooked work — Diyi_Yang · 2026-10-28
• A fine-tuned 'cat obsession' control fools every AI welfare criterion — MirrorEthic_Anchor · 2026-10-28
• Sweeping every Qwen3-8B neuron, rated for weirdness by Haiku — Sauers_ · 2026-10-28
• Follow-up repo machine-verifies Hamiltonian cycle results on top of OpenAI/Anthropic ETH refutation claims — xennygrimmato_ · 2026-10-28
• New arXiv paper frames world model representation as reachability, admission, assignment — SoloGen · 2026-10-28
• OpenAI shares internal frontier model math results, Terence Tao responds in meme'd thread — krisspy39 · 2026-10-28

Research channel →