English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Multilingual Neural Machine Translation (MNMT) enables one system totranslate sentences from multiple source languages to multiple targetlanguages, greatly reducing deployment costs compared with conventionalbilingual systems. The MNMT training benefit, however, is often limited tomany-to-one directions. The model suffers from poor performance in one-to-manyand many-to-many with zero-shot setup. To address this issue, this paperdiscusses how to practically build MNMT systems that serve arbitrary X-Ytranslation directions while leveraging multilinguality with a two-stagetraining strategy of pretraining and finetuning. Experimenting with the WMT'21multilingual translation task, we demonstrate that our systems outperform theconventional baselines of direct bilingual models and pivot translation modelsfor most directions, averagely giving +6.0 and +4.1 BLEU, without the need forarchitecture change or extra data collection. Moreover, we also examine ourproposed approach in an extremely large-scale data setting to accommodatepractical deployment scenarios.

Building Multilingual Machine Translation Systems That Serve Arbitrary X-Y Translations