Revisiting Low-Resource Neural Machine Translation: A Case Study

TLDR

Neural machine translation performance drops sharply in low‑resource settings, falling below phrase‑based statistical machine translation and requiring large auxiliary data to compete. This study re‑examines those findings, contending that the shortfall stems from insufficient system adaptation to low‑resource contexts. We outline pitfalls in training low‑resource NMT and present recent techniques that form a set of best practices for such settings. Experiments on German–English with varying IWSLT14 data show that an optimized NMT system can outperform PBSMT using far less data, and applying the same techniques to Korean–English yields a 4‑BLEU improvement over prior results.

Abstract

It has been shown that the performance of neural machine translation (NMT) drops starkly in low-resource conditions, underperforming phrase-based statistical machine translation (PBSMT) and requiring large amounts of auxiliary data to achieve competitive results. In this paper, we re-assess the validity of these results, arguing that they are the result of lack of system adaptation to low-resource settings. We discuss some pitfalls to be aware of when training low-resource NMT systems, and recent techniques that have shown to be especially helpful in low-resource settings, resulting in a set of best practices for low-resource NMT. In our experiments on German–English with different amounts of IWSLT14 training data, we show that, without the use of any auxiliary monolingual or multilingual data, an optimized NMT system can outperform PBSMT with far less data than previously claimed. We also apply these techniques to a low-resource Korean–English dataset, surpassing previously reported results by 4 BLEU.