Generating Diverse Translation from Model Distribution with Dropout
This addresses the problem of limited translation variety for NMT users, but it is incremental as it builds on existing dropout and Bayesian methods.
The paper tackled the lack of diversity in neural machine translation by generating diverse translations through Bayesian modeling and dropout sampling, achieving a better trade-off between diversity and accuracy in Chinese-English and English-German tasks.
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.