BERT, mBERT, or BiBERT? A Study on Contextualized Embeddings for Neural Machine Translation
Haoran Xu, Benjamin Van Durme, Kenton Murray
Abstract
The success of bidirectional encoders using masked language models, such as BERT, on numerous natural language processing tasks has prompted researchers to attempt to incorporate these pre-trained models into neural machine translation (NMT) systems. However, proposed methods for incorporating pretrained models are non-trivial and mainly focus on BERT, which lacks a comparison of the impact that other pre-trained models may have on translation performance. In this paper, we demonstrate that simply using the output (contextualized embeddings) of a tailored and suitable bilingual pre-trained language model (dubbed BIBERT) as the input of the NMT encoder achieves state-of-the-art translation performance. Moreover, we also propose a stochastic layer selection approach and a concept of dual-directional translation model to ensure the sufficient utilization of contextualized embeddings. In the case of without using back translation, our best models achieve BLEU scores of 30.45 for EnDe and 38.61 for DeEn on the IWSLT'14 dataset, and 31.26 for EnDe and 34.94 for DeEn on the WMT'14 dataset, which exceeds all published numbers 12 .