Page 1 of 3 123 LastLast
Results 1 to 10 of 21

Thread: Google Translate and Genderless-to-Gendered Translations

  1. #1
    Curious Conlanger epistemophiliac's Avatar
    Type
    INTP
    Join Date
    Nov 2014
    Location
    Wonderland
    Posts
    133

    Google Translate and Genderless-to-Gendered Translations

    I saw this on my Facebook feed, and I thought it was interesting. I've never really thought about it, but it kind of makes sense when you do.

    Basically, Turkish has no gendered pronouns; the word "o" can mean "he", "she", or "it". But, when Google Translate translates it to English, it automatically assigns it a gender based on the number of times a specific pronoun has occurred in a sequence of text along with the other words being translated.

    So... thoughts? What could be some of the implications of this?

    I might have to test this with other languages; maybe, in other male-vs-female gendered languages, some of the pronouns in the sentences might be translated differently than they were in English. That could show more clearly that some greater cultural forces are potentially at work here and how they differ.

    Pedin i phith in anÝron, a nin ˙-cheniog.
    Kup tar-tor n'zhitlar ik istau - eh ri ken-tor du nash-veh.
    Sa's drisa tonabhiala tomu˝ s'eph de'imal'il, sa se lodhe togal'n.
    B'lir inus vorbilim bi mek inus, hiam gi b'ek ilimlimr.

  2. #2
    Meae Musae Servus Hephaestus's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Location
    Ceti Alpha V
    Posts
    13,402
    INTPx Award Winner
    I think it's strange they went to such work when they could have just done what I just did.
    Most of time, when people ask why something terrible happened, they don't realize they are looking for someone to blame.

    --Meditations on Uncertainty Vol ξ(x)

  3. #3
    just dont think about it mhc's Avatar
    Type
    iNtP
    Join Date
    Feb 2014
    Posts
    751
    Quote Originally Posted by Hephaestus View Post
    I think it's strange they went to such work when they could have just done what I just did.
    I’m guessing they’re treading carefully and want to be able to show they have a solid, formulated logically based algorithm in place if anyone quizzes them re any gender biasing.
    Just look at the blue sky

  4. #4
    Curious Conlanger epistemophiliac's Avatar
    Type
    INTP
    Join Date
    Nov 2014
    Location
    Wonderland
    Posts
    133
    Quote Originally Posted by Hephaestus View Post
    I think it's strange they went to such work when they could have just done what I just did.
    Lol, But that's not really how translation machines work. Generally, statistical natural language processing techniques and neural networks are used to teach a machine - in this case, Google translate - 1) what a correct input looks like (i.e. is the sentence being input grammatical in the selected/detected language) and 2) what a correct output looks like (i.e. is the sentence in the target language grammatical). This is mostly done with statistical analyses of texts and other corpora; so, for example, in English, it just so happens that "he" is more often used by speakers with the phrase "is hard-working", while "she" is more often used by speakers with the phrase "is lazy", and so google translate, when not given a gender in the input language, automatically resorts to the most commonly used pronoun (whether it be 'he', 'she', 'they', or 'it'). It's not google translate, it's pretty much English itself and its speakers that are shaping the algorithms here. And it'd be interesting to see how other gendered languages are shaping the algorithms, lol. (I'm going to test this at some point!)

    I mean, look at the simple Turkish sentence above. The phrase "o bir doktor" literally translates as "he/she/they/it one doctor". You can't just have the translator spit out word-for-word translations. Natural languages are way too messy, complicated, different, and often way too ambiguous for that, so you gotta turn to machine learning and neural networks for that stuff. This way, instead of looking at just one word at a time, the machine can take on entire phrases or chunks of text and, using its 'memory', make a probable guess to what the correct translation is of all the words working together in context. And even then it's not perfect, lol :P

    Edit: Also, many male-vs-female gendered languages don't have a neutral (animate) pronoun to use like how many English speakers use "they". So the speakers/machine automatically have to resort to either 'he' or 'she'. In Polish, for example, "water" is feminine so you generally would say "I'm drinking her" (instead of "I'm drinking it") when referring to some water. And many languages (Polish included) inflect their verbs and/or adjectives depending on the gender of their subject or what they're modifying. So, if you're a female speaking in the past tense in Polish, then you conjugate your verbs differently than a male would. Somewhat similar morphology also exists in French, Arabic, Hebrew, Russian, etc.
    Last edited by epistemophiliac; 12-04-2017 at 11:53 AM.
    Pedin i phith in anÝron, a nin ˙-cheniog.
    Kup tar-tor n'zhitlar ik istau - eh ri ken-tor du nash-veh.
    Sa's drisa tonabhiala tomu˝ s'eph de'imal'il, sa se lodhe togal'n.
    B'lir inus vorbilim bi mek inus, hiam gi b'ek ilimlimr.

  5. #5
    Curious Conlanger epistemophiliac's Avatar
    Type
    INTP
    Join Date
    Nov 2014
    Location
    Wonderland
    Posts
    133
    Quote Originally Posted by mhc View Post
    I’m guessing they’re treading carefully and want to be able to show they have a solid, formulated logically based algorithm in place if anyone quizzes them re any gender biasing.
    It's funny, because the person who shared this was complaining about Google Translate showing gender bias and about how the people making these machines are gender biased and thus programing it into the machines... Me: That's not how it worksssssssss.

    The machines are learning from English and how its speakers use it; so does that make English and the majority of its speakers gender biased? Or is it just a statistical fluke that the machine picked up on? I guess a better way to test for the former would be to form an experiment with the same sentences in the picture but without the pronouns, and then asking English speakers to fill in the blank with whatever pronoun comes to mind first; or, to make it more casual so it's naturally elicited, you could replace the pronouns with "someone" and have them write a narrative about that "someone" and see what pronoun the majority of people resort to.

    But then again, maybe our brains work similarly to the machine. Perhaps, in male-vs-female gendered languages, the speakers just subconsciously resort to the most statistically probable pronoun given the context and their prior experiences? hmm.....
    Pedin i phith in anÝron, a nin ˙-cheniog.
    Kup tar-tor n'zhitlar ik istau - eh ri ken-tor du nash-veh.
    Sa's drisa tonabhiala tomu˝ s'eph de'imal'il, sa se lodhe togal'n.
    B'lir inus vorbilim bi mek inus, hiam gi b'ek ilimlimr.

  6. #6
    just dont think about it mhc's Avatar
    Type
    iNtP
    Join Date
    Feb 2014
    Posts
    751
    Quote Originally Posted by epistemophiliac View Post
    It's funny, because the person who shared this was complaining about Google Translate showing gender bias and about how the people making these machines are gender biased and thus programing it into the machines... Me: That's not how it worksssssssss.

    The machines are learning from English and how its speakers use it; so does that make English and the majority of its speakers gender biased? Or is it just a statistical fluke that the machine picked up on? I guess a better way to test for the former would be to form an experiment with the same sentences in the picture but without the pronouns, and then asking English speakers to fill in the blank with whatever pronoun comes to mind first; or, to make it more casual so it's naturally elicited, you could replace the pronouns with "someone" and have them write a narrative about that "someone" and see what pronoun the majority of people resort to.

    But then again, maybe our brains work similarly to the machine. Perhaps, in male-vs-female gendered languages, the speakers just subconsciously resort to the most statistically probable pronoun given the context and their prior experiences? hmm.....
    unconsciously people like to anthropomorphise and tend to label or project onto things their own gender, so perhaps a swing one way or another may indicate which gender narrates more.
    Just look at the blue sky

  7. #7
    Curious Conlanger epistemophiliac's Avatar
    Type
    INTP
    Join Date
    Nov 2014
    Location
    Wonderland
    Posts
    133
    Quote Originally Posted by mhc View Post
    unconsciously people like to anthropomorphise and tend to label or project onto things their own gender, so perhaps a swing one way or another may indicate which gender narrates more.
    Maybe. The way gender arises in language is for redundancy and clarification of the information for the receiver. By categorizing nouns in a systematic way and having verbs and/or adjectives agree with that categorization of the noun, it narrows down the category of things the speaker could be talking about in case the listener has missed some information along the way. It just so happens that some language developed this categorization from gender. In English, however, we no longer have that, and the only remnants of the Old English noun gender system are the pronouns he/him, she/her, and it. Many speakers commonly use 'they' as a singular neutral animate pronoun nowadays or to refer to an unidentified entity (furthering the ambiguity of English pronouns lol; I mean, just look at the pronoun 'you', and now 'they' is becoming just as ambiguous - not fun for translation machines).

    Imagine you're translating from English to Hebrew and you have the simple phrase, "Did you eat?" But Hebrew has a masculine 'you' and a feminine 'you' each with a singular and plural form; that means you're going from one multi-purpose, ambiguous 'you', to a more specific set of four different 'you's that rely on context. lol

    In French, actually, the possessive form of the third person pronouns is the same for both "he" and "she". So my French gf almost always uses the wrong one when she goes to say "his" or "her" in English and it's hilarious. (i.e. "Ma mŔre a fixÚ sa table" can be translated as "My mother fixed her table" or "My mother fixed his table" depending on the context.)
    GF: "Did you see? My mom fixed his table."
    Me: "Huh?? Whose table?"
    GF: "His table!"
    Me: "But who's 'he'?"
    GF: "Not 'he', 'she'. It's my mom's table!"
    Me: *facepalm* "You mean, 'her table'?."
    Pedin i phith in anÝron, a nin ˙-cheniog.
    Kup tar-tor n'zhitlar ik istau - eh ri ken-tor du nash-veh.
    Sa's drisa tonabhiala tomu˝ s'eph de'imal'il, sa se lodhe togal'n.
    B'lir inus vorbilim bi mek inus, hiam gi b'ek ilimlimr.

  8. #8
    TJ TeresaJ's Avatar
    Type
    INTP
    Join Date
    Apr 2014
    Location
    mosquito-infested hell
    Posts
    3,230
    My reaction is the same as @mhc, especially since they threw a "he-she" in there. It looks like they tweaked their algorithm to recognize when something would be assigned a gender in English and then randomly assign that gender according to pre-set proportions (e.g. 42% male 42% female 6% he-she, whatever). I think if they were smart they would avoid basing their algorithm on usage of the phrase, because you specifically don't want to wind up in the situation where he is always a doctor and she is always a nurse, not if you're trying to be PC, which I'm sure Google is.

  9. #9
    Meae Musae Servus Hephaestus's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Location
    Ceti Alpha V
    Posts
    13,402
    INTPx Award Winner
    Quote Originally Posted by epistemophiliac View Post
    Lol, But that's not really how translation machines work. Generally, statistical natural language processing techniques and neural networks are used to teach a machine - in this case, Google translate - 1) what a correct input looks like (i.e. is the sentence being input grammatical in the selected/detected language) and 2) what a correct output looks like (i.e. is the sentence in the target language grammatical). This is mostly done with statistical analyses of texts and other corpora; so, for example, in English, it just so happens that "he" is more often used by speakers with the phrase "is hard-working", while "she" is more often used by speakers with the phrase "is lazy", and so google translate, when not given a gender in the input language, automatically resorts to the most commonly used pronoun (whether it be 'he', 'she', 'they', or 'it'). It's not google translate, it's pretty much English itself and its speakers that are shaping the algorithms here. And it'd be interesting to see how other gendered languages are shaping the algorithms, lol. (I'm going to test this at some point!)

    I mean, look at the simple Turkish sentence above. The phrase "o bir doktor" literally translates as "he/she/they/it one doctor". You can't just have the translator spit out word-for-word translations. Natural languages are way too messy, complicated, different, and often way too ambiguous for that, so you gotta turn to machine learning and neural networks for that stuff. This way, instead of looking at just one word at a time, the machine can take on entire phrases or chunks of text and, using its 'memory', make a probable guess to what the correct translation is of all the words working together in context. And even then it's not perfect, lol :P

    Edit: Also, many male-vs-female gendered languages don't have a neutral (animate) pronoun to use like how many English speakers use "they". So the speakers/machine automatically have to resort to either 'he' or 'she'. In Polish, for example, "water" is feminine so you generally would say "I'm drinking her" (instead of "I'm drinking it") when referring to some water. And many languages (Polish included) inflect their verbs and/or adjectives depending on the gender of their subject or what they're modifying. So, if you're a female speaking in the past tense in Polish, then you conjugate your verbs differently than a male would. Somewhat similar morphology also exists in French, Arabic, Hebrew, Russian, etc.
    I know how natural language processing works/machine learning works in broad strokes. I'm n-gram aware. Computational linguistics used to be my target. That approach makes sense as a starting point when translating from genderless pronouns to gendered pronouns in languages that lack equivalent genderless pronouns, but if you're going from a language using gender neutral pronouns to a language that has gender neutral pronouns available, doesn't it make sense to take advantage of that rather than do something you know will produce wrong outputs?
    Most of time, when people ask why something terrible happened, they don't realize they are looking for someone to blame.

    --Meditations on Uncertainty Vol ξ(x)

  10. #10
    know nothing pensive_pilgrim's Avatar
    Type
    INTP
    Join Date
    Dec 2013
    Posts
    4,585
    Quote Originally Posted by Hephaestus View Post
    I know how natural language processing works/machine learning works in broad strokes. I'm n-gram aware. Computational linguistics used to be my target. That approach makes sense as a starting point when translating from genderless pronouns to gendered pronouns in languages that lack equivalent genderless pronouns, but if you're going from a language using gender neutral pronouns to a language that has gender neutral pronouns available, doesn't it make sense to take advantage of that rather than do something you know will produce wrong outputs?
    No because singular 'they' is an abomination and I will fight it with my last breath.

Similar Threads

  1. Replies: 10
    Last Post: 04-07-2017, 02:56 PM
  2. Google failed to answer my question. Help
    By stigmatica in forum The Pub
    Replies: 4
    Last Post: 09-25-2016, 11:46 PM
  3. Your INTPx Google Image Footprint
    By OrionzRevenge in forum The Pub
    Replies: 5
    Last Post: 06-25-2016, 02:59 AM
  4. Ask INTPx Google autofill results
    By NedLudd in forum The Playground
    Replies: 16
    Last Post: 06-09-2016, 06:10 PM
  5. Google Glass
    By Osito Polar in forum Math, Science & Tech
    Replies: 11
    Last Post: 04-30-2014, 03:22 AM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •