Applying Reinforcement learning (RL) following maximum likelihood estimation (MLE) pre-training is a versatile method for enhancing neural machine translation (NMT) performance. However, recent work has argued that the gains produced by RL for NMT are mostly due to promoting tokens that have already received a fairly high probability in pre-training. We hypothesize that the large action space is a main obstacle to RL’s effectiveness in MT, and conduct two sets of experiments that lend support to our hypothesis. First, we find that reducing the size of the vocabulary improves RL’s effectiveness. Second, we find that effectively reducing the dimension of the action space without changing the vocabulary also yields notable improvement as evaluated by BLEU, semantic similarity, and human evaluation. Indeed, by initializing the network’s final fully connected layer (that maps the network’s internal dimension to the vocabulary dimension), with a layer that generalizes over similar actions, we obtain a substantial improvement in RL performance: 1.5 BLEU points on average.
|Original language||American English|
|Number of pages||13|
|Journal||Proceedings - International Conference on Computational Linguistics, COLING|
|State||Published - 2022|
|Event||29th International Conference on Computational Linguistics, COLING 2022 - Gyeongju, Korea, Republic of|
Duration: 12 Oct 2022 → 17 Oct 2022
Bibliographical noteFunding Information:
This work was supported by the Israel Science Foundation (grant no. 2424/21), and by the Applied Research in Academia Program of the Israel Innovation Authority.
© 2022 Proceedings - International Conference on Computational Linguistics, COLING. All rights reserved.