Narrowing the Knowledge Evaluation Gap: Open-Domain Question Answering with Multi-Granularity Answers

TL;DR

This paper introduces GRANOLA QA, an evaluation framework for open-domain question answering that considers multi-granularity answers, and proposes DRAG, a decoding method that improves answer accuracy by aligning response granularity with model uncertainty.

Contribution

The paper presents GRANOLA QA and GRANOLA-EQ datasets, and introduces DRAG, a decoding algorithm that enhances answer accuracy by considering answer granularity and model uncertainty.

Findings

DRAG improves accuracy by nearly 20 points on GRANOLA-EQ.

Standard decoding often produces overly specific answers that are incorrect.

Multi-granularity evaluation reveals more knowledge in language models than standard methods.

Abstract

Factual questions typically can be answered correctly at different levels of granularity. For example, both ``August 4, 1961'' and ``1961'' are correct answers to the question ``When was Barack Obama born?''. Standard question answering (QA) evaluation protocols, however, do not explicitly take this into account and compare a predicted answer against answers of a single granularity level. In this work, we propose GRANOLA QA, a novel evaluation setting where a predicted answer is evaluated in terms of accuracy and informativeness against a set of multi-granularity answers. We present a simple methodology for enriching existing datasets with multi-granularity answers, and create GRANOLA-EQ, a multi-granularity version of the EntityQuestions dataset. We evaluate a range of decoding methods on GRANOLA-EQ, including a new algorithm, called Decoding with Response Aggregation (DRAG), that is geared towards aligning the response granularity with the model's uncertainty. Our experiments show that large language models with standard decoding tend to generate specific answers, which are often incorrect. In contrast, when evaluated on multi-granularity answers, DRAG yields a nearly 20 point increase in accuracy on average, which further increases for rare entities. Overall, this reveals that standard evaluation and decoding schemes may significantly underestimate the knowledge encapsulated in LMs.