Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation
This work addresses the challenge of trust in natural language outputs for users of foundation models, though it is incremental as it builds on existing uncertainty estimation methods.
The paper tackles the problem of measuring uncertainty in large language models for tasks like question answering by addressing semantic equivalence, where different sentences can have the same meaning. It introduces semantic entropy, an unsupervised method that incorporates linguistic invariances, and shows it is more predictive of model accuracy than baselines in ablation studies.
We introduce a method to measure uncertainty in large language models. For tasks like question answering, it is essential to know when we can trust the natural language outputs of foundation models. We show that measuring uncertainty in natural language is challenging because of "semantic equivalence" -- different sentences can mean the same thing. To overcome these challenges we introduce semantic entropy -- an entropy which incorporates linguistic invariances created by shared meanings. Our method is unsupervised, uses only a single model, and requires no modifications to off-the-shelf language models. In comprehensive ablation studies we show that the semantic entropy is more predictive of model accuracy on question answering data sets than comparable baselines.