Gender Bias and its Impact
Gender bias are stereotypical beliefs about people on the basis of their gender that influence how they are perceived and/or treated in society. For example, men are expected to be more active and independent, whereas women are expected to be more passive and communal. People with non-binary gender identities are often excluded entirely.
Why is Gender Bias so Harmful?
There are two broad categories of harm when talking about the impact of gender bias: allocational harm and representational harm. Both of them have an impact on people’s power and status in society, with certain groups receiving more power or opportunities than others.
Allocational harm means that certain individuals are less likely to receive support or have access to opportunities, tools and resources because of the ‘group’ they belong to. A classic example would be women with the same qualifications as men being passed up for promotion. Non-binary or trans people might not receive appropriate medical care when incorrect assumptions about their bodies are made, and refusing to affirm a person’s gender identity can cause psychological harm.
Representational harm is often more subtle. It reinforces stereotypes and diminishes or even ignores certain groups in society. Representational harm can manifest through underrepresentation or even lack of representation. Think about image searches only returning pictures of white men when looking for the word ‘professor’, for example, or non-binary people being excluded entirely.
This project focuses on identifying cases of representational harm in machine translation output. The ubiquitous embedding of MT in web applications and social networks such as Facebook or Twitter adds to this problem, as users may not even be aware that they are reading a (biased) translation due to the otherwise fluent MT output. This way, gender stereotypes and misrepresentation are unknowingly reinforced.
