Knowing When Not to Answer: Cross-Domain and Multi-Turn Generalization of Latent Underspecification Signals

Key points

• Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has been said.1
• We contribute a turn-labeled multi-turn benchmark (423 conversations, 1,661 labeled turn-states) and an evaluation harness with a simulated user who answers clarifying questions, and use them with six datasets and six open-weight LLMs to test how far probes for unanswerability carry.1
• Probes transfer robustly between datasets that share a ground of unanswerability: missing information in math (AUROC 0.77-0.97) and in a passage (SQuAD 2.0<->MuSiQue, 0.77-0.90).1
• Single-turn probes fail zero-shot to detect when a conversation becomes answerable; in-structure probes recover it, but no better than a bag-of-words classifier.1

Sources (1)

• [1]Knowing When Not to Answer: Cross-Domain and Multi-Turn Generalization of Latent Underspecification Signals
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 02:18 PM

  Unanswerability is linearly decodable from hidden states, but it is unclear which of its forms share a representation and whether the signal is useful in dialogue.

  Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has been said.

Extractive summary: sentences quoted from the sources.