Reproduce “Referential ambiguity and clarification requests: comparing human and LLM behaviour” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chris Madge, Matthew Purver, Massimo Poesio
A comparison of human and Large Language Model behavior in situated, task-oriented dialogue reveals that LLMs, especially with activated reasoning capabilities, are more inclined to ask targeted clarification questions for referential ambiguities than human participants. The research indicates that LLMs perform similarly to humans in identifying task-based uncertainty, but diverge in their handling of linguistic ambiguities.
Key insights
- Humans rarely issue clarification requests for referential linguistic ambiguities, often preferring presumptive actions or using such ambiguities for conversational signals.
- Large Language Models, particularly when equipped with or prompted for reasoning, demonstrate a capacity to proactively clarify referential ambiguities, a behavior distinct from human patterns.
- An LLM's ability to generate relevant clarification questions for literal ambiguities is directly linked to its reasoning capabilities, whether inherent to the model or induced by Chain of Thought prompting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.