A paper I co-authored^[1] has recently been published in the journal of Information Processing and Management, on the topic of the gender–career implicit bias.

Implicit biases are those which are unconscious, and result in a mismatch between thought and reality. The gender–career implicit bias is the increased unconscious association between maleness and professionalism or certain professions, as distinct from femaleness. A classic example is that resumés with female-coded names will receive lower ratings for employability than identical documents with male-coded names attached.

As well as being present in people’s psychology, the gender–career bias is reflected in the statistics of language use, which in turn can reinforce those biases. In other words, maybe male names are more frequently used in contexts of discussing professions than female names are, and in turn this can induce or reinforce that association in people exposed to that linguistic imbalance. This is also important because the statistics of languages are currently being crystallised and amplified by — among other things — large-language-model “AI” systems. So understanding how these biases relate to the statistics of language.

In this paper, we investigate how the gender–career bias is reflected in language, and how different statistical models of language capture and exhibit these biases.

If you have trouble finding a copy of the paper, I have a copy on my website.