AGent: A Novel Pipeline for Automatically Creating Unanswerable Questions

TL;DR

This paper introduces AGent, an automated pipeline for generating unanswerable questions to improve extractive question answering models, reducing manual annotation efforts and maintaining high data quality.

Contribution

AGent is a novel automated method for creating unanswerable questions, enhancing dataset generation for EQA without manual annotation.

Findings

Created unanswerable question sets with low error rates

Models trained on these sets perform comparably to those trained on SQuAD 2.0

Demonstrated effectiveness on multiple EQA benchmarks

Abstract

The development of large high-quality datasets and high-performing models have led to significant advancements in the domain of Extractive Question Answering (EQA). This progress has sparked considerable interest in exploring unanswerable questions within the EQA domain. Training EQA models with unanswerable questions helps them avoid extracting misleading or incorrect answers for queries that lack valid responses. However, manually annotating unanswerable questions is labor-intensive. To address this, we propose AGent, a novel pipeline that automatically creates new unanswerable questions by re-matching a question with a context that lacks the necessary information for a correct answer. In this paper, we demonstrate the usefulness of this AGent pipeline by creating two sets of unanswerable questions from answerable questions in SQuAD and HotpotQA. These created question sets exhibit low error rates. Additionally, models fine-tuned on these questions show comparable performance with those fine-tuned on the SQuAD 2.0 dataset on multiple EQA benchmarks.