Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Google translations from a language that doesn't specify gender (mathstodon.xyz)
13 points by ColinWright on March 17, 2021 | hide | past | favorite | 12 comments


This is interesting, and I think that from two perspectives

How would I feel if these were all the male pronoun he. I think I would feel odd, but not as odd. Unsure what that says

How did we get here. Is this yet another case of ML picking up on bias in seed data?


That wouldn't be too strange because in many languages including English (though I typically use singular they) the male pronoun is also the ungendered pronoun.

Edit: looked it up to confirm my suspicion; English alai traditionally uses he for the ambiguous pronoun.


> Is this yet another case of ML picking up on bias in seed data?

That was my first thought, that this probably reflects more the views of the original translators than the engineers at Google.


I thought that the Google Translate corpus consisted of largely documents that were translated between multiple languages. In that case there was probably context telling the translator the gender, or they were translated from English with a gender to gender-neutral.

In this case I would expect that barring any other influence these are just whatever gender happened to be the most common for each verb in the input texts.


He cooks.

She reads.

She teaches.

He washes the dishes.

He raises the children.

She is a researcher.

She is an engineer.

He’s a cleaner.

She is working.

I think we should consider the version from our mirror universe to put things in perspective.


While I could definitely see this coming from ML biases, I would like to know the language in question and test some more examples.


In what way does the word "Hungarian" in the first column not answer the first question?

Edit: For the people who've downvoted me ... this is a genuine question. The word "Hungarian" is there in the image, but the person to whom I'm answering says they want to know what language is being discussed. So clearly their question isn't answered by that, and I'd like to know why or how.

So it's a genuine question ... if you've downvoted me, perhaps you could suggest how I should have phrased it.


Maybe he meant he'd like to be able to speak the language in question. Anyway, I guess you could just translate verbs to hungarian and then back, using the genderless pronoun.


> Maybe he meant he'd like to be able to speak the language in question.

Ooh, that's a parsing I never considered.

Thank you.


Hah well you were sorta right, i was being dumb and didn't see that (ironically I was very hungry when making that comment, which may have thrown me off)

But the suggestion of the parent comment to yours afaik is not a great idea. I know that languages can be very complex. To properly understand what is going on, I think it would be most appropriate for someone who knows Hungarian to analyze this. I think it would be especially important to have someone who understands gender roles in Hungarian culture to look at it. They might be able to craft tests that show if this was an issue with the biases (assuming that a bias is proved) from the original Hungarian translators or from Google's overall system.

So yes, in one sense I do now know the language, but in a much more real sense, I do not _know_ the language.


>Maybe he meant he'd like to be able to speak the language in question

Or she.


Random gender assignation?

Oh, no sorry, it's AI.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: