Task Calibration: Calibrating Large Language Models on Inference Tasks

TL;DR

This paper introduces task calibration (TC), a method to improve large language models' reasoning on inference tasks by reducing reliance on spurious correlations, leading to better zero-shot and few-shot performance.

Contribution

The paper proposes a novel task calibration approach inspired by mutual information that enhances LLM reasoning by balancing premise and hypothesis considerations.

Findings

TC significantly improves zero-shot performance on 13 inference tasks.

TC is effective in few-shot and diverse natural language understanding tasks.

TC is robust to different prompt templates and compatible with other calibration methods.

Abstract

Large language models (LLMs) have exhibited impressive zero-shot performance on inference tasks. However, LLMs may suffer from spurious correlations between input texts and output labels, which limits LLMs' ability to reason based purely on general language understanding. In other words, LLMs may make predictions primarily based on premise or hypothesis, rather than both components. To address this problem that may lead to unexpected performance degradation, we propose task calibration (TC), a zero-shot and inference-only calibration method inspired by mutual information which recovers LLM performance through task reformulation. TC encourages LLMs to reason based on both premise and hypothesis, while mitigating the models' over-reliance on individual premise or hypothesis for inference. Experimental results show that TC achieves a substantial improvement on 13 inference tasks in the zero-shot setup. We further validate the effectiveness of TC in few-shot setups and various natural language understanding tasks. Further analysis indicates that TC is also robust to prompt templates and has the potential to be integrated with other calibration methods.