Correcting Length Bias in Neural Machine Translation

Proceedings of the Third Conference on Machine Translation (WMT), Volume 1: Research Papers, pages 212–223 Belgium, Brussels, October 31 - Novermber 1, 2018. c⃝2018 Association for Computational Linguistics https://doi.org/10.18653/v1/W18-64022 Correcting Length Bias in Neural Machine Translation Kenton Murray and David Chiang Department of Computer Science and Engineering University of Notre Dame {kmurray4,dchiang}@nd.edu Abstract We study two problems in neural machine translation (NMT). First, in beam search, whereas a wider beam should in principle help translation, it often hurts NMT. Second, NMT has a tendency to produce translations that are too short. Here, we argue that these prob- lems are closely related and both rooted in la- bel bias. We show that correcting the brevity problem almost eliminates the beam problem; we compare some commonly-used methods for doing this, finding that a simple per-word reward works well; and we introduce a simple and quick way to tune this reward using the perceptron algorithm. 1 Introduction Although highly successful, neural machine trans- lation (NMT) systems continue to be plagued by a number of problems. We focus on two here: the beam problem and the brevity problem. First, machine translation systems rely on heuristics to search through the intractably large space of possible translations. Most commonly, beam search is used during the decoding process. Traditional statistical machine translation systems often rely on large beams to find good translations. However, in neural machine translation, increas- ing the beam size has been shown to degrade per- formance. This is the last of the six challenges identified by Koehn and Knowles (2017). The second problem, noted by several authors, is that NMT tends to generate translations that are too short. Jean et al. (2015) and Koehn and Knowles address this by dividing translation scores by their length, inspired by work on au- dio chords (Boulanger-Lewandowski et al., 2013). A similar method is also used by Google’s pro- duction system (Wu et al., 2016). A third sim- ple method used by various authors (Och and Ney, 2002; He et al., 2016; Neubig, 2016) is a tunable reward added for each output word. Huang et al. (2017) and Yang et al. (2018) propose variations of this reward that enable better guarantees during search. In this paper, we argue that these two problems are related (as hinted at by Koehn and Knowles) and that both stem from label bias, an undesirable property of models that generate sentences word by word instead of all at once. The typical solution is to introduce a sentence- level correction to the model. We show that mak- ing such a correction almost completely eliminates the beam problem. We compare two commonly- used corrections, length normalization and a word reward, and show that the word reward is slightly better. Finally, instead of tuning the word reward using grid search, we introduce a way to learn it using a perceptron-like tuning method. We show that the optimal value is sensitive both to task and beam size, implying that it is important to tune for every model …

Content truncated to 500 words