The Ubiqus English-Inuktitut System for WMT20

TL;DR

This paper presents Ubiqus' multilingual Transformer-based system for English-Inuktitut translation in WMT20, addressing challenges of low-resource and agglutinative language complexities through joint training and careful data handling.

Contribution

It introduces a multilingual Transformer approach trained on multiple agglutinative languages specifically for English-Inuktitut translation, tackling low-resource and linguistic challenges.

Findings

Effective multilingual training improves translation quality.

Handling language-specific features enhances system robustness.

Addressed low-resource data limitations successfully.

Abstract

This paper describes Ubiqus' submission to the WMT20 English-Inuktitut shared news translation task. Our main system, and only submission, is based on a multilingual approach, jointly training a Transformer model on several agglutinative languages. The English-Inuktitut translation task is challenging at every step, from data selection, preparation and tokenization to quality evaluation down the line. Difficulties emerge both because of the peculiarities of the Inuktitut language as well as the low-resource context.