Automode: Choosing the Best Machine Translation Supplier Based on Source Text

Abstract

This study presents Automode, a framework consisting of two quality estimators which, within the wide-range of options available nowadays, aims to automatically select the MT supplier which is more likely to provide a good translation to a source text in terms of fluency and mitigation of gender bias. Among six analysed MT suppliers, results show that a simple machine learning regression can learn how to highly score the supplier which is more likely to generate a fluent translation to a source text, while reducing the translation cost compared to the best performing single supplier in some scenarios. However, for gender-bias we noticed that an open-source supplier is the one which generates the less biased translations regarding gender.

Tasks

Reproductions