Evaluating BERT-based Rewards for Question Generation with Reinforcement Learning

Research output: Chapter in Book/Conference proceedings/Edited volumeConference contributionScientificpeer-review

Abstract

Question generation systems aim to generate natural language questions that are relevant to a given piece of text, and can usually be answered by just considering this text. Prior works have identified a range of shortcomings (including semantic drift and exposure bias) and thus have turned to the reinforcement learning paradigm to improve the effectiveness of question generation. As part of it, different reward functions have been proposed. As typically these reward functions have been empirically investigated in different experimental settings (different datasets, models and parameters) we lack a common framework to fairly compare them. In this paper, we first categorize existing rewards systematically. We then provide such a fair empirical evaluation of different reward functions (including three we propose here for QG) in a common framework. We find rewards that model answerability to be the most effective.

Original languageEnglish
Title of host publicationICTIR 2021 - Proceedings of the 2021 ACM SIGIR International Conference on Theory of Information Retrieval
PublisherAssociation for Computing Machinery (ACM)
Pages261-270
Number of pages10
ISBN (Electronic)9781450386111
DOIs
Publication statusPublished - 2021
Event11th ACM SIGIR International Conference on Theory of Information Retrieval, ICTIR 2021 - Virtual, Online, Canada
Duration: 11 Jul 202111 Jul 2021

Publication series

NameICTIR 2021 - Proceedings of the 2021 ACM SIGIR International Conference on Theory of Information Retrieval

Conference

Conference11th ACM SIGIR International Conference on Theory of Information Retrieval, ICTIR 2021
CountryCanada
CityVirtual, Online
Period11/07/2111/07/21

Keywords

  • question generation
  • reinforcement learning
  • reward functions

Fingerprint

Dive into the research topics of 'Evaluating BERT-based Rewards for Question Generation with Reinforcement Learning'. Together they form a unique fingerprint.

Cite this