A Controllable Model of Grounded Response Generation

Authors

  • Zeqiu Wu University of Washington
  • Michel Galley Microsoft Research
  • Chris Brockett Microsoft Research
  • Yizhe Zhang Microsoft Research
  • Xiang Gao Microsoft Research
  • Chris Quirk Microsoft Research
  • Rik Koncel-Kedziorski University of Washington
  • Jianfeng Gao Microsoft Research
  • Hannaneh Hajishirzi University of Washington Allen Institute for AI
  • Mari Ostendorf University of Washington
  • Bill Dolan Microsoft Research

Keywords:

Conversational AI/Dialog Systems

Abstract

Current end-to-end neural conversation models inherently lack the flexibility to impose semantic control in the response generation process, often resulting in uninteresting responses. Attempts to boost informativeness alone come at the expense of factual accuracy, as attested by pretrained language models' propensity to "hallucinate" facts. While this may be mitigated by access to background knowledge, there is scant guarantee of relevance and informativeness in generated responses. We propose a framework that we call controllable grounded response generation (CGRG), in which lexical control phrases are either provided by a user or automatically extracted by a control phrase predictor from dialogue context and grounding knowledge. Quantitative and qualitative results show that, using this framework, a transformer based model with a novel inductive attention mechanism, trained on a conversation-like Reddit dataset, outperforms strong generation baselines.

Downloads

Published

2021-05-18

How to Cite

Wu, Z., Galley, M., Brockett, C., Zhang, Y., Gao, X., Quirk, C., Koncel-Kedziorski, R., Gao, J., Hajishirzi, H., Ostendorf, M., & Dolan, B. (2021). A Controllable Model of Grounded Response Generation. Proceedings of the AAAI Conference on Artificial Intelligence, 35(16), 14085-14093. Retrieved from https://ojs.aaai.org/index.php/AAAI/article/view/17658

Issue

Section

AAAI Technical Track on Speech and Natural Language Processing III