Generating Diverse Translation from Model Distribution with Dropout

TL;DR

This paper introduces a Bayesian dropout-based method for neural machine translation that generates diverse translations by sampling from a distribution of models, improving diversity without sacrificing accuracy.

Contribution

It proposes a novel approach using concrete dropout and variational inference to derive multiple models for diverse translation generation in NMT.

Findings

Enhanced diversity in translation outputs

Maintained or improved translation accuracy

Effective trade-off between diversity and quality

Abstract

Despite the improvement of translation quality, neural machine translation (NMT) often suffers from the lack of diversity in its generation. In this paper, we propose to generate diverse translations by deriving a large number of possible models with Bayesian modelling and sampling models from them for inference. The possible models are obtained by applying concrete dropout to the NMT model and each of them has specific confidence for its prediction, which corresponds to a posterior model distribution under specific training data in the principle of Bayesian modeling. With variational inference, the posterior model distribution can be approximated with a variational distribution, from which the final models for inference are sampled. We conducted experiments on Chinese-English and English-German translation tasks and the results shows that our method makes a better trade-off between diversity and accuracy.