Verify-and-Edit: A Knowledge-Enhanced Chain-of-Thought Framework

TL;DR

This paper introduces Verify-and-Edit, a framework that enhances factual accuracy in chain-of-thought reasoning by post-editing reasoning chains with external knowledge, improving performance on knowledge-intensive tasks.

Contribution

It presents a novel Verify-and-Edit framework that refines reasoning chains for factual correctness using external knowledge, built upon GPT-3.

Findings

Improved accuracy in open-domain question-answering tasks

Enhanced factual correctness of reasoning chains

Framework applicable to knowledge-intensive NLP tasks

Abstract

As large language models (LLMs) have become the norm in NLP, demonstrating good performance in generation and reasoning tasks, one of its most fatal disadvantages is the lack of factual correctness. Generating unfactual texts not only leads to lower performances but also degrades the trust and validity of their applications. Chain-of-Thought (CoT) prompting improves trust and model performance on complex reasoning tasks by generating interpretable reasoning chains, but still suffers from factuality concerns in knowledge-intensive tasks. In this paper, we propose the Verify-and-Edit framework for CoT prompting, which seeks to increase prediction factuality by post-editing reasoning chains according to external knowledge. Building on top of GPT-3, our framework lead to accuracy improvements in multiple open-domain question-answering tasks.