Enhancing Language Model Factuality via Activation-Based Confidence Calibration and Guided Decoding

TL;DR

This paper introduces ActCab, an activation-based calibration method, and CoDec, a confidence-guided decoding strategy, to improve language model factuality and calibration efficiency, leading to more reliable and truthful responses.

Contribution

The paper presents a novel activation-based calibration method and a confidence-guided decoding strategy that significantly enhance language model factuality and calibration performance.

Findings

ActCab reduces expected calibration error by up to 39%.

CoDec improves factuality on challenging QA datasets.

Calibration signals help generate more truthful answers.

Abstract

Calibrating language models (LMs) aligns their generation confidence with the actual likelihood of answer correctness, which can inform users about LMs' reliability and mitigate hallucinated content. However, prior calibration methods, such as self-consistency-based and logit-based approaches, are either limited in inference-time efficiency or fall short of providing informative signals. Moreover, simply filtering out low-confidence responses reduces the LM's helpfulness when the answers are correct. Therefore, effectively using calibration techniques to enhance an LM's factuality remains an unsolved challenge. In this paper, we first propose an activation-based calibration method, ActCab, which trains a linear layer on top of the LM's last-layer activations that can better capture the representations of knowledge. Built on top of ActCab, we further propose CoDec, a confidence-guided decoding strategy to elicit truthful answers with high confidence from LMs. By evaluating on five popular QA benchmarks, ActCab achieves superior calibration performance than all competitive baselines, e.g., by reducing the average expected calibration error (ECE) score by up to 39%. Further experiments on CoDec show consistent improvements in several LMs' factuality on challenging QA datasets, such as TruthfulQA, highlighting the value of confidence signals in enhancing factuality.