Unsupervised Pretraining for Neural Machine Translation Using Elastic Weight Consolidation

TL;DR

This paper explores unsupervised pretraining for neural machine translation by initializing with language models and fine-tuning with Elastic Weight Consolidation, achieving faster convergence without needing original data during fine-tuning.

Contribution

It introduces a novel combination of language model initialization and EWC regularization for NMT, improving training speed and data efficiency.

Findings

EWC with decoder achieves BLEU scores comparable to previous methods.

Model converges 2-3 times faster with EWC.

EWC regularization is less effective when tasks are unrelated.

Abstract

This work presents our ongoing research of unsupervised pretraining in neural machine translation (NMT). In our method, we initialize the weights of the encoder and decoder with two language models that are trained with monolingual data and then fine-tune the model on parallel data using Elastic Weight Consolidation (EWC) to avoid forgetting of the original language modeling tasks. We compare the regularization by EWC with the previous work that focuses on regularization by language modeling objectives. The positive result is that using EWC with the decoder achieves BLEU scores similar to the previous work. However, the model converges 2-3 times faster and does not require the original unlabeled training data during the fine-tuning stage. In contrast, the regularization using EWC is less effective if the original and new tasks are not closely related. We show that initializing the bidirectional NMT encoder with a left-to-right language model and forcing the model to remember the original left-to-right language modeling task limits the learning capacity of the encoder for the whole bidirectional context.