On the Effectiveness of Offline RL for Dialogue Response Generation

Paloma Sodhi, Felix Wu, Ethan R. Elenberg, Kilian Q Weinberger, Ryan Mcdonald
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:32088-32104, 2023.

Abstract

A common training technique for language models is teacher forcing (TF). TF attempts to match human language exactly, even though identical meanings can be expressed in different ways. This motivates use of sequence-level objectives for dialogue response generation. In this paper, we study the efficacy of various offline reinforcement learning (RL) methods to maximize such objectives. We present a comprehensive evaluation across multiple datasets, models, and metrics. Offline RL shows a clear performance improvement over teacher forcing while not inducing training instability or sacrificing practical training budgets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-sodhi23a, title = {On the Effectiveness of Offline {RL} for Dialogue Response Generation}, author = {Sodhi, Paloma and Wu, Felix and Elenberg, Ethan R. and Weinberger, Kilian Q and Mcdonald, Ryan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {32088--32104}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/sodhi23a/sodhi23a.pdf}, url = {https://proceedings.mlr.press/v202/sodhi23a.html}, abstract = {A common training technique for language models is teacher forcing (TF). TF attempts to match human language exactly, even though identical meanings can be expressed in different ways. This motivates use of sequence-level objectives for dialogue response generation. In this paper, we study the efficacy of various offline reinforcement learning (RL) methods to maximize such objectives. We present a comprehensive evaluation across multiple datasets, models, and metrics. Offline RL shows a clear performance improvement over teacher forcing while not inducing training instability or sacrificing practical training budgets.} }
Endnote
%0 Conference Paper %T On the Effectiveness of Offline RL for Dialogue Response Generation %A Paloma Sodhi %A Felix Wu %A Ethan R. Elenberg %A Kilian Q Weinberger %A Ryan Mcdonald %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-sodhi23a %I PMLR %P 32088--32104 %U https://proceedings.mlr.press/v202/sodhi23a.html %V 202 %X A common training technique for language models is teacher forcing (TF). TF attempts to match human language exactly, even though identical meanings can be expressed in different ways. This motivates use of sequence-level objectives for dialogue response generation. In this paper, we study the efficacy of various offline reinforcement learning (RL) methods to maximize such objectives. We present a comprehensive evaluation across multiple datasets, models, and metrics. Offline RL shows a clear performance improvement over teacher forcing while not inducing training instability or sacrificing practical training budgets.
APA
Sodhi, P., Wu, F., Elenberg, E.R., Weinberger, K.Q. & Mcdonald, R.. (2023). On the Effectiveness of Offline RL for Dialogue Response Generation. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:32088-32104 Available from https://proceedings.mlr.press/v202/sodhi23a.html.

Related Material