Can Retriever-Augmented Language Models Reason? The Blame Game Between the Retriever and the Language Model

TL;DR

This paper critically evaluates retriever-augmented language models' reasoning capabilities, revealing limitations in retrieval quality and reasoning strength, and highlights the potential and challenges of multihop retrieve-and-read approaches.

Contribution

It provides a comprehensive analysis of popular retriever-augmented models, identifying key weaknesses and potential directions for improving reasoning in such systems.

Findings

Retriever similarity metrics are insufficient for complete reasoning.

Language models show limited reasoning even with necessary statements.

Performance drops significantly with imperfect retrieval, e.g., Flan-T5 drops 28.6%.

Abstract

Augmenting pretrained language models with retrievers has shown promise in effectively solving common NLP problems, such as language modeling and question answering. In this paper, we evaluate the strengths and weaknesses of popular retriever-augmented language models, namely kNN-LM, REALM, DPR + FiD, Contriever + ATLAS, and Contriever + Flan-T5, in reasoning over retrieved statements across different tasks. Our findings indicate that the simple similarity metric employed by retrievers is insufficient for retrieving all the necessary statements for reasoning. Additionally, the language models do not exhibit strong reasoning even when provided with only the required statements. Furthermore, when combined with imperfect retrievers, the performance of the language models becomes even worse, e.g., Flan-T5's performance drops by 28.6% when retrieving 5 statements using Contriever. While larger language models improve performance, there is still a substantial room for enhancement. Our further analysis indicates that multihop retrieve-and-read is promising for large language models like GPT-3.5, but does not generalize to other language models like Flan-T5-xxl.