Hurdles to Progress in Long-form Question Answering

TL;DR

This paper highlights fundamental challenges in long-form question answering, including evaluation issues and dataset flaws, despite achieving state-of-the-art results on the ELI5 dataset.

Contribution

The paper identifies key problems in LFQA evaluation and datasets, and proposes solutions to improve future research directions.

Findings

State-of-the-art performance on ELI5 achieved with a new system.

Answers are often not grounded in retrieved documents.

Significant overlap between training and validation sets in ELI5.

Abstract

The task of long-form question answering (LFQA) involves retrieving documents relevant to a given question and using them to generate a paragraph-length answer. While many models have recently been proposed for LFQA, we show in this paper that the task formulation raises fundamental challenges regarding evaluation and dataset creation that currently preclude meaningful modeling progress. To demonstrate these challenges, we first design a new system that relies on sparse attention and contrastive retriever learning to achieve state-of-the-art performance on the ELI5 LFQA dataset. While our system tops the public leaderboard, a detailed analysis reveals several troubling trends: (1) our system's generated answers are not actually grounded in the documents that it retrieves; (2) ELI5 contains significant train / validation overlap, as at least 81% of ELI5 validation questions occur in paraphrased form in the training set; (3) ROUGE-L is not an informative metric of generated answer quality and can be easily gamed; and (4) human evaluations used for other text generation tasks are unreliable for LFQA. We offer suggestions to mitigate each of these issues, which we hope will lead to more rigorous LFQA research and meaningful progress in the future.