Revisiting the Weaknesses of Reinforcement Learning for Neural Machine Translation

Revisiting the Weaknesses of Reinforcement Learning for Neural Machine Translation
复制标题

重新审视神经机器翻译强化学习的弱点

DOI:
10.18653/v1/2021.naacl-main.133
复制
发表时间:
2021
期刊:
ArXiv
影响因子:
--
通讯作者:
Julia Kreutzer
Julia Kreutzer
中科院分区:
--
文献类型:
--
作者:
Samuel Kiegeland;Julia Kreutzer

文献摘要

参考文献

被引文献

相似文献

策略梯度算法在 NLP 中得到了广泛采用,但最近受到批评,人们怀疑它们是否适合 NMT。乔申等人。 (2020)发现了多个弱点,并怀疑他们的成功是由产出分布的形状而不是奖励决定的。在本文中,我们重新审视这些主张并在更广泛的配置下研究它们。我们对域内和跨域适应的实验揭示了探索和奖励缩放的重要性,并为这些主张提供了经验反证。
Policy gradient algorithms have found wide adoption in NLP, but have recently become subject to criticism, doubting their suitability for NMT. Choshen et al. (2020) identify multiple weaknesses and suspect that their success is determined by the shape of output distributions rather than the reward. In this paper, we revisit these claims and study them under a wider range of configurations. Our experiments on in-domain and cross-domain adaptation reveal the importance of exploration and reward scaling, and provide empirical counter-evidence to these claims.
DOI: 10.18653/v1/p17-1138
发表时间: 2017-04
期刊: --
影响因子: --
作者:
Julia Kreutzer;Artem Sokolov;S. Riezler
通讯作者: Julia Kreutzer;Artem Sokolov;S. Riezler
DOI: 10.18653/v1/d19-3019
发表时间: 2019-07
期刊: ArXiv
影响因子: --
作者:
Julia Kreutzer;Jasmijn Bastings;S. Riezler
通讯作者: Julia Kreutzer;Jasmijn Bastings;S. Riezler