LLMs struggle to improve dialogue parsing via input rewriting

Researchers explored input rewriting techniques using Large Language Models (LLMs) to improve dialogue discourse parsing, particularly in scenarios where supervised clarification is unavailable. Their findings indicate that parser-agnostic rewriting often introduces more errors than corrections, and even parser-aware methods with GRPO show limitations in consistently improving parsing accuracy. The study highlights the need for a rewritability prediction capability to determine if an utterance can be repaired before intervention, which is crucial for optimizing frozen discourse parsers and enhancing agentic pipelines. AI

IMPACT Highlights limitations in current LLM capabilities for complex NLP tasks, suggesting new research directions for agentic systems.

RANK_REASON Academic paper detailing a novel approach to dialogue discourse parsing. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• dialogue discourse parsing
• GRPO
• LLMs
• natural language processing
• Segmented Discourse Representation Theory

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →