Poster
Linguistic Calibration of Language Models
Neil Band · Xuechen Li · Tengyu Ma · Tatsunori Hashimoto
Language models (LMs) may lead their users to make suboptimal downstream decisions when they confidently hallucinate. This issue can be mitigated by having the LM verbally convey the probability that a claim is correct, but existing models cannot produce text with calibrated confidence statements. Through the lens of decision-making, we formalize linguistic calibration: an LM is linguistically calibrated if its generations enable its users to make calibrated probabilistic predictions. This definition enables a training framework where a supervised finetuning step bootstraps an LM to emit long-form generations with confidence statements such as “I estimate a 30% chance of...” or “I am certain that...”, followed by a reinforcement learning step which rewards generations that enable a user to provide calibrated answers to related questions. We linguistically calibrate Llama 2 7B and find in automated and human evaluations of long-form generations that it is significantly more calibrated than strong finetuned factuality baselines with comparable accuracy, including under distribution shifton question-answering and person biography generation. Our results demonstrate that long-form generations may be calibrated end-to-end by shifting objectives from the space of text to the those of downstream predictions.
Live content is unavailable. Log in and register to view live content