IssueTracker

We use the cloud translation API to provide machine translations of our website from English into French and Spanish.

A contributor recently flagged the fact that in a list of her co-authors, all of whom are women, the job titles of some were rendered in the masculine form (e.g. two 'co-directors' of the same organisation became a 'co-directrice/co-directora' and a 'co-directeur/co-director' respectively).

The API results appear to be correct for those contributors who have names that are common for women in western Europe and default to the masculine form otherwise, so this appears to be a gender bias with a cultural aspect to it.

Obviously this is concerning to us. In the long term I'm sure you'll want to eliminate this kind of bias. In the meantime, it seems that returning both translations (so that our automated translation pipeline could flag potential ambiguities) is not supported in the API (unlike in web search). This would be a useful feature, especially as there are names that are of ambiguous gender that we'd want to handle properly in future (in the current case, they are not of ambiguous gender).

(NB this is more specific and limited than the existing request for alternative translations with probability scores)