FACT: Examining the Effectiveness of Iterative Context Rewriting for Multi-fact Retrieval

TL;DR

This paper introduces FACT, an iterative context rewriting method that improves large language models' ability to retrieve multiple facts simultaneously, addressing the 'lost-in-the-middle' phenomenon and enhancing multi-fact retrieval performance.

Contribution

The paper proposes a novel iterative retrieval approach, FACT, that refines context through multiple rounds, significantly improving multi-fact retrieval in LLMs compared to single-pass methods.

Findings

FACT improves multi-fact retrieval accuracy

The method reduces the 'lost-in-the-middle' phenomenon

Performance gains are task-dependent, less in general QA

Abstract

Large Language Models (LLMs) are proficient at retrieving single facts from extended contexts, yet they struggle with tasks requiring the simultaneous retrieval of multiple facts, especially during generation. This paper identifies a novel "lost-in-the-middle" phenomenon, where LLMs progressively lose track of critical information throughout the generation process, resulting in incomplete or inaccurate retrieval. To address this challenge, we introduce Find All Crucial Texts (FACT), an iterative retrieval method that refines context through successive rounds of rewriting. This approach enables models to capture essential facts incrementally, which are often overlooked in single-pass retrieval. Experiments demonstrate that FACT substantially enhances multi-fact retrieval performance across various tasks, though improvements are less notable in general-purpose QA scenarios. Our findings shed light on the limitations of LLMs in multi-fact retrieval and underscore the need for more resilient long-context retrieval strategies.