Generating Gender Alternatives in Machine Translation

TL;DR

This paper addresses gender bias in machine translation by generating all grammatically correct gendered alternatives, providing datasets and benchmarks, and introducing a semi-supervised method that integrates with existing models without extra overhead.

Contribution

It introduces a semi-supervised approach for generating gender alternatives in MT, along with open datasets and benchmarks for five language pairs.

Findings

Effective generation of gender alternatives without extra inference cost

Open source datasets and benchmarks for gender translation

Seamless integration with standard MT models

Abstract

Machine translation (MT) systems often translate terms with ambiguous gender (e.g., English term "the nurse") into the gendered form that is most prevalent in the systems' training data (e.g., "enfermera", the Spanish term for a female nurse). This often reflects and perpetuates harmful stereotypes present in society. With MT user interfaces in mind that allow for resolving gender ambiguity in a frictionless manner, we study the problem of generating all grammatically correct gendered translation alternatives. We open source train and test datasets for five language pairs and establish benchmarks for this task. Our key technical contribution is a novel semi-supervised solution for generating alternatives that integrates seamlessly with standard MT models and maintains high performance without requiring additional components or increasing inference overhead.