Can Retriever-Augmented Language Models Reason? The Blame Game Between the Retriever and the Language Model
This work addresses the reasoning limitations of retriever-augmented language models for NLP researchers and practitioners, highlighting critical bottlenecks in current methods.
The paper evaluated retriever-augmented language models on reasoning tasks, finding that retrievers often fail to fetch necessary statements and language models struggle with reasoning even when given correct statements, with performance dropping significantly when combined with imperfect retrievers, such as a 28.6% decrease for Flan-T5 with Contriever.
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.