-
Notifications
You must be signed in to change notification settings - Fork 5
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Remove gender bias (metaissue) #29
Comments
The following PR is a proof-of-concept that addresses part of the non-binary pronoun issue in English: apertium/apertium-eng#37 . |
An example from the "gisting pair" sme-nob: Northern Sámi uses the same pronoun son (gen/acc su) for all genders[0], while Norwegian has to choose between hun/han/hen ( In sme→nob, the first version pretty much just used han, then we got some naïve t2x anaphora resolution choosing between han/hun based on proper nouns which was wrong 30% of the time, and when people started using it on news articles, Giellatekno would be contacted by people who were upset because the translator used the wrong pronoun. So we changed it into h_n to signal that we don't know any better, later changed into hun/han which, though a bit verbose, people seemed to like better. (Unlike in Swedish, we can't really use hen for "we don't know"; if you use hen in Norwegian for a named individual it signals that they explicitly prefer that term over han/hun.) It's not a good solution for quality pairs used for post-editing, but then we don't have any of those going into English anyway … [0] so now my Twitter bio says son/su (if you want to refer to me in the third person, you have to speak Sámi) |
Oh, another item for your list:
|
This paper has some thoughts about this. |
On top of the actual changes to the output of the translator, something we've done at Softcatalà for the neural version of the eng-cat is to try to detect words that in English do not have associated gender, but they do in Catalan, is to show a "warning" in the UI. Probably, this information could be sent through the pipeline using blanks, and then optionally show them in the UI, or on the command line. Do you think that's a good idea, and something doable? |
@xavivars I like this solution, it would also be cool to have a switch (like with Unhammer's thing) to prefer m/f or neutral terms. |
The implementation that comes to mind for what @xavivars describes is to have some way of specifying in the bidix that a particular word-bound blank should be attached to a particular entry. |
For languages like German at least the gender selection is pretty much required for all human nouns or doer nouns at least, not just jobs. I initially just translated the female versions to feminine derivations in deu-fin but that can get quite silly, so it would be nice to have a solution like with user preferences and probably defaulting to neutral forms. How would this be defined, in bidix per entry? |
This conversation mostly seems to be relevant to issue #31. |
I agree, but this causes problems if we try to choose collective forms which are singular instead of plural. For example, in Spanish to avoid the words "
|
Gender bias permeates Apertium language modules and translation pairs. The following are some examples:
<m><pl>
instead of<mf><pl>
;This likely just grazes the surface. I'd like to use this issue as a place to gather general observations, carry out general discussion, and link to individual issues that address these problems in specific language modules and translation pairs.
The text was updated successfully, but these errors were encountered: