An Interdisciplinary Analysis of gender-based discrimination in Translation Technology

Due to its ubiquitous presence in our societies, Machine Translation systems have been subject to heavy scrutiny, in particular, their current tendency to perpetuate (gender) biases. For example, the English sentence ‘my husband is a kindergarten teacher ’is translated by Google Translate as ‘mijn man is kleuterjuf [feminine]’ in Dutch (April 19, 2023).

While there have been attempts to address gender biases in Machine Translation systems, the issue remains largely unsolved and the solutions proposed focus merely on the computational linguistic aspect but do not take into consideration the broader societal and cultural context nor the actual stakeholders.

This project focuses on investigating why and under which circumstances current Machine Translation systems perpetuate gender bias, how an ethical framework can evaluate and tackle it, and whether existing legal frameworks recognize and redress gendered linguistic injustice in this context.

Project objectives

Gender bias in Machine Translation is not just a technical challenge, but a social, ethical, and legal one that stems from language and everyday communication. This project aims to provide an interdisciplinary framework, taking into account linguistic justice, legal compliance, and computational linguistics. Such a framework can be used to offer practical and ethical recommendations to tackle gender biases present in current Machine Translation systems.

Potential impact

This project will provide a more comprehensive understanding of gender bias in the context of translation technology which can lead to the development of innovative, more holistic solutions to gender in Machine Translation.

Aside from the potential to inform the Machine Translation community on principles of linguistic justice which can lead to more inclusive language and the promotion of diversity/equality, the legal evaluation has the potential  to highlight the need for regulations.

The collaboration with and input from stakeholders such as Google and Transgender Network Nederland ensures that the outcomes are relevant and applicable to real world scenarios.

To increase the project’s impact, the findings will be disseminated to a broad audience beyond the academic community (e.g. policymakers, industry, general public) through various communication channels with the aim of promoting the adoption of more ethical and inclusive practices in Machine Translation development.

Duration

The project will run for one year starting from September 2023 onwards.

Multidisciplinary project team

Computational linguist Dr. Eva Vanmassenhove is an Assistant Professor at the Department of Cognitive Science and Artificial Intelligence. Her pioneering work on gender bias earned her invited talks at Google, Amazon and Microsoft and multiple awards. She will bring the technical and linguistic skills to answer why and under which circumstances current Machine Translation systems perpetuate gender bias(es). This will serve as the basis for Dr Seunghyun Song’s work on the framework.

Linguistic justice scholar Dr. Seunghyun Song works as an Assistant Professor at the Department of Philosophy. She will answer the question to what extent an ethical framework grounded in principles of linguistic justice can help to evaluate/tackle gender bias in Machine Translation, investigating the possibilities and limits of 'digital constitutionalism'.

Legal theorist Dr. Hanna Lukkari is an Assistant Professor at the Department of Public Law and Governance. Lukkari will use her expertise in constitutionalism and digitalization to evaluate the project’s findings from a legal perspective and determine whether existing legal frameworks are able to recognize and redress gendered linguistic injustice, using gender bias in Machine Translation as a case study.

The team will be supported by student assistant Sonja Siebeneicher.

External stakeholders, Dr. Jasmijn Bastings on behalf of Google DeepMind and Santi van den Toorn on behalf of Transgender Netwerk Nederland (TNN), will actively shape the project's direction/outcomes. Bastings will offer real-world Machine Translation technology insights, while TNN will provide perspectives on the experiences of transgender individuals and how Machine Translation can better address their needs.

This project is funded by Tilburg University’s Digital Sciences for Society program: