Latent Group Dropout for Multilingual and Multidomain Machine Translation

Anonymous

Paper Link: https://openreview.net/forum?id=iP2cWgCQz3

Paper Type: Long paper (up to eight pages of content + unlimited references and appendices)

Abstract: Multidomain and multilingual machine translation often rely on parameter sharing strategies, where large portions of the network are meant to capture the commonalities of the tasks at hand, while smaller parts are reserved to model the peculiarities of a language or a domain. In adapter-based approaches, these strategies are hardcoded in the network architecture, independent of the similarities between tasks. In this work, we propose a new method to better take advantage of these similarities, using a latent-variable model. We also develop new techniques to train this model end-to-end and report experimental results showing that the learned patterns are both meaningful and yield improved translation performance without any increase of the model size.

Copyright Consent Signature (type Name Or NA If Not Transferrable): Minh-Quang PHAM

Copyright Consent Name And Address: Laboratoire Interdisciplinaire des Sciences du Numérique, Campus universitaire bât 507 Rue du Belvedère F - 91405 Orsay cedex