Effective Adversarial Regularization for Neural Machine Translation

Sato, Motoki; Suzuki, Jun; Kiyono, Satoshi

doi:10.18653/v1/p19-1020

Cited by 33 publications

(54 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some authors further applied the adversarial training to various NLP tasks, such relation extraction (Wu et al, 2017) , part-of-speech tagging (Yasunaga et al, 2018) and jointly extracting entities and relations (Bekoulis et al, 2018). A recent work (Sato et al, 2019) investigates the effects of AT on neural machine translation. studied the effects of applying AT to different set of variables in MRC tasks.…”

Section: Related Workmentioning

confidence: 99%

Robust Machine Comprehension Models via Adversarial Training

Wang¹,

Bansal²

2018

Proceedings of the 2018 Conference of the North American Chapter Of the Association for Computational Linguistics: Hu

View full text Add to dashboard Cite

It is shown that many published models for the Stanford Question Answering Dataset (Rajpurkar et al., 2016) lack robustness, suffering an over 50% decrease in F1 score during adversarial evaluation based on the AddSent (Jia and Liang, 2017) algorithm. It has also been shown that retraining models on data generated by AddSent has limited effect on their robustness. We propose a novel alternative adversary-generation algorithm, AddSentDiverse, that significantly increases the variance within the adversarial training data by providing effective examples that punish the model for making certain superficial assumptions. Further, in order to improve robustness to AddSent's semantic perturbations (e.g., antonyms), we jointly improve the model's semantic-relationship learning capabilities in addition to our AddSentDiversebased adversarial training data augmentation. With these additions, we show that we can make a state-of-the-art model significantly more robust, achieving a 36.5% increase in F1 score under many different types of adversarial evaluation while maintaining performance on the regular SQuAD task.

show abstract

Section: Related Workmentioning

confidence: 99%

Robust Machine Comprehension Models via Adversarial Training

Wang¹,

Bansal²

2018

Proceedings of the 2018 Conference of the North American Chapter Of the Association for Computational Linguistics: Hu

View full text Add to dashboard Cite

show abstract

“…Due to its formidable search space, this paradigm simply perturbs on a small ratio of token positions and greedy search by brute force among candidates. Note that adversarial example generation is fundamentally different from noised hidden representation in adversarial training (Cheng et al, 2019;Sano et al, 2019), which is not to be concerned in this work.…”

Section: Adversarial Examples In Nlpmentioning

confidence: 99%

A Reinforced Generation of Adversarial Examples for Neural Machine Translation

Zou¹,

Huang²,

Xie³

et al. 2020

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

View full text Add to dashboard Cite

Neural machine translation systems tend to fail on less decent inputs despite its significant efficacy, which may significantly harm the credibility of these systems-fathoming how and when neural-based systems fail in such cases is critical for industrial maintenance. Instead of collecting and analyzing bad cases using limited handcrafted error features, here we investigate this issue by generating adversarial examples via a new paradigm based on reinforcement learning. Our paradigm could expose pitfalls for a given performance metric, e.g., BLEU, and could target any given neural machine translation architecture. We conduct experiments of adversarial attacks on two mainstream neural machine translation architectures, RNN-search, and Transformer. The results show that our method efficiently produces stable attacks with meaning-preserving adversarial examples. We also present a qualitative and quantitative analysis for the preference pattern of the attack, demonstrating its capability of pitfall exposure.

show abstract

“…2019; Motoki Sato, 2019). However, we train the model with only the adversarial samples for the sake of fair comparison with the baselines.…”

Section: Problem Definitionmentioning

confidence: 99%

“…Our method extends the adversarial training framework, which was initially developed in the vision domain (Goodfellow et al, 2014) and has begun to be adopted in the NLP domain recently (Jia and Liang, 2017;Belinkov and Bisk, 2018;Samanta and Mehta, 2017;Miyato et al, 2016;Motoki Sato, 2019;Wang et al, 2019a;Cheng et al, 2019). Miyato et al (2016) adopted the adversarial training framework on text classification by perturbing embedding space with continuous adversarial noise.…”

Section: Related Workmentioning

confidence: 99%

Adversarial Subword Regularization for Robust Neural Machine Translation

Park¹,

Sung²,

Lee³

et al. 2020

Findings of the Association for Computational Linguistics: EMNLP 2020

View full text Add to dashboard Cite

Exposing diverse subword segmentations to neural machine translation (NMT) models often improves the robustness of machine translation as NMT models can experience various subword candidates. However, the diversification of subword segmentations mostly relies on the pre-trained subword language models from which erroneous segmentations of unseen words are less likely to be sampled. In this paper, we present adversarial subword regularization (ADVSR) to study whether gradient signals during training can be a substitute criterion for exposing diverse subword segmentations. We experimentally show that our model-based adversarial samples effectively encourage NMT models to be less sensitive to segmentation errors and improve the performance of NMT models in low-resource and out-domain datasets.

show abstract

Effective Adversarial Regularization for Neural Machine Translation

Cited by 33 publications

References 13 publications

Robust Machine Comprehension Models via Adversarial Training

Robust Machine Comprehension Models via Adversarial Training

A Reinforced Generation of Adversarial Examples for Neural Machine Translation

Adversarial Subword Regularization for Robust Neural Machine Translation

Contact Info

Product

Resources

About