Multi-Sentence Resampling: A Simple Approach to Alleviate Dataset Length Bias and Beam-Search Degradation

Ivan Provilkov, Andrey Malinin


Abstract
Neural Machine Translation (NMT) is known to suffer from a beam-search problem: after a certain point, increasing beam size causes an overall drop in translation quality. This effect is especially pronounced for long sentences. While much work was done analyzing this phenomenon, primarily for autoregressive NMT models, there is still no consensus on its underlying cause. In this work, we analyze errors that cause major quality degradation with large beams in NMT and Automatic Speech Recognition (ASR). We show that a factor that strongly contributes to the quality degradation with large beams is dataset length-bias - NMT datasets are strongly biased towards short sentences. To mitigate this issue, we propose a new data augmentation technique – Multi-Sentence Resampling (MSR). This technique extends the training examples by concatenating several sentences from the original dataset to make a long training example. We demonstrate that MSR significantly reduces degradation with growing beam size and improves final translation quality on the IWSTL15 En-Vi, IWSTL17 En-Fr, and WMT14 En-De datasets.
Anthology ID:
2021.emnlp-main.677
Volume:
Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing
Month:
November
Year:
2021
Address:
Online and Punta Cana, Dominican Republic
Editors:
Marie-Francine Moens, Xuanjing Huang, Lucia Specia, Scott Wen-tau Yih
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
8612–8621
Language:
URL:
https://aclanthology.org/2021.emnlp-main.677
DOI:
10.18653/v1/2021.emnlp-main.677
Bibkey:
Cite (ACL):
Ivan Provilkov and Andrey Malinin. 2021. Multi-Sentence Resampling: A Simple Approach to Alleviate Dataset Length Bias and Beam-Search Degradation. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8612–8621, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
Multi-Sentence Resampling: A Simple Approach to Alleviate Dataset Length Bias and Beam-Search Degradation (Provilkov & Malinin, EMNLP 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.emnlp-main.677.pdf
Video:
 https://aclanthology.org/2021.emnlp-main.677.mp4
Code
 yandex-research/msr
Data
LibriSpeech