Recitation-Augmented Language Models

TL;DR

RECITE introduces a novel approach where language models recite relevant knowledge from their own memory before answering, significantly improving accuracy in knowledge-intensive NLP tasks without external retrieval.

Contribution

The paper presents RECITE, a new paradigm that enhances LLMs by reciting relevant passages internally before answering, achieving state-of-the-art results in closed-book question answering.

Findings

RECITE improves performance on multiple CBQA tasks.

Effective across various pre-trained models.

Achieves state-of-the-art results in knowledge-intensive tasks.

Abstract

We propose a new paradigm to help Large Language Models (LLMs) generate more accurate factual knowledge without retrieving from an external corpus, called RECITation-augmented gEneration (RECITE). Different from retrieval-augmented language models that retrieve relevant documents before generating the outputs, given an input, RECITE first recites one or several relevant passages from LLMs' own memory via sampling, and then produces the final answers. We show that RECITE is a powerful paradigm for knowledge-intensive NLP tasks. Specifically, we show that by utilizing recitation as the intermediate step, a recite-and-answer scheme can achieve new state-of-the-art performance in various closed-book question answering (CBQA) tasks. In experiments, we verify the effectiveness of \method~on four pre-trained models (PaLM, UL2, OPT, and Codex) and three CBQA tasks (Natural Questions, TriviaQA, and HotpotQA). Our code is available at "https://github.com/Edward-Sun/RECITE".