On the Weaknesses of Reinforcement Learning for Neural Machine Translation.

Leshem Choshen, Lior Fox, Zohar Aizenbud, Omri Abend

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


Reinforcement learning (RL) is frequently used to increase performance in text generation tasks, including machine translation (MT), notably through the use of Minimum Risk Training (MRT) and Generative Adversarial Networks (GAN).
However, little is known about what and how these methods learn in the context of MT.
We prove that one of the most common RL methods for MT does not optimize the expected reward, as well as show that other methods take an infeasibly long time to converge.
In fact, our results suggest that RL practices in MT are likely to improve performance only where the pre-trained parameters are already close to yielding the correct translation.
Our findings further suggest that observed gains may be due to effects unrelated to the training signal, concretely, changes in the shape of the distribution curve.
Original languageEnglish
Title of host publicationICLR 2020
Subtitle of host publicationInternational Conference on Learning Representations
Number of pages14
StatePublished - 2020
EventInternational Conference on Learning Representations, ICLR 2020
- Virtual event
Duration: 26 Apr 20201 May 2020


ConferenceInternational Conference on Learning Representations, ICLR 2020
Abbreviated titleICLR 2020
Internet address


  • Reinforcement learning
  • Minimum risk training
  • Machine Translation


Dive into the research topics of 'On the Weaknesses of Reinforcement Learning for Neural Machine Translation.'. Together they form a unique fingerprint.

Cite this