Discussion Why machine translation has a very long way to go...

Discussion in 'Novel General' started by DZ_Spellcaster, Feb 21, 2020.

Thread Status:
Not open for further replies.
  1. DZ_Spellcaster

    DZ_Spellcaster Well-Known Member

    Joined:
    May 22, 2017
    Messages:
    544
    Likes Received:
    430
    Reading List:
    Link
    AliceShiki, Snowbun, Sam7 and 2 others like this.
  2. imK

    imK Artful Dodger

    Joined:
    Oct 20, 2018
    Messages:
    1,221
    Likes Received:
    2,630
    Reading List:
    Link
    On the plus side it's better than babelfish. Anyone who lived through that horror likely carries the scars.
     
    Snowbun, 0000000 and Geor like this.
  3. Geor

    Geor Well-Known Member

    Joined:
    May 10, 2016
    Messages:
    567
    Likes Received:
    395
    Reading List:
    Link
    my thought, that's nice but without a GOOD proofreader and editor it will still fry your brain to read it ... why I have more then a few sites and books blocked on my list here ... I did love how Google translated names as he, bet it even does that when the person is a she lol seen that before too ...
     
    Araenn and abewan like this.
  4. null

    null Procrastinate Lv MAX

    Joined:
    Oct 20, 2015
    Messages:
    1,088
    Likes Received:
    1,384
    Reading List:
    Link
    That is 2018, Google cloud translation already take leap and bounds in the last few years. Its now good enough that people from law and medical circle using it to translate things.

    Btw, this is not the google translate for free public use. Last time I checked GCT cost 20 USD / 1m words.
     
    Geor, imK, Snowbun and 2 others like this.
  5. abewan

    abewan 〘Nishikigoi Dungeon〙

    Joined:
    Dec 25, 2015
    Messages:
    676
    Likes Received:
    624
    Reading List:
    Link
    Google Translate tends to ignore phrases they don't understand, especially slang. Lots of time happened tbh.
     
  6. 0000000

    0000000 I B SMILING!

    Joined:
    Sep 24, 2016
    Messages:
    5,571
    Likes Received:
    5,494
    Reading List:
    Link
    Because AI still does not have the capability to actually differentiate proverbs and the unique tone set by some writers. Hell, some human translators also fail to take note and require editors and proofreaders to prevent readers' head from melting.
     
    Vilidious, AliceShiki, Geor and 2 others like this.
  7. Kuro_0ni

    Kuro_0ni Cocooned in a Life transition

    Joined:
    Jan 21, 2018
    Messages:
    2,022
    Likes Received:
    4,774
    Reading List:
    Link
    The thing that Machine translation engines, lose out on, whether its scripted or a deep-learning AI. Is the cultural nuance expressed from the Source Language.

    Its like sarcasm, you can't always get it if you read things off a medium.

    Same thing with Creole, Pidgin or foreshortening of words. (What is up? = Wazzup, Whatup. etc)
     
    AliceShiki and imK like this.
  8. Kaminomikan

    Kaminomikan 神のみ感

    Joined:
    Feb 8, 2016
    Messages:
    983
    Likes Received:
    1,071
    Reading List:
    Link
    probably some ppl said the same in some industries in the past, but the truth is, we now have automated assembly lines, we are having auto drive cars... etc, etc...
     
  9. Beer_Kitty

    Beer_Kitty Just a very very drunk kitty that likes beer

    Joined:
    Mar 12, 2017
    Messages:
    3,104
    Likes Received:
    2,304
    Reading List:
    Link
    I still see it as not yet good. Good enough but theres still words it wont translate properly in English and its still japanese in english letters at some point.
     
  10. Vilidious

    Vilidious Well-Known Member

    Joined:
    Jul 28, 2017
    Messages:
    699
    Likes Received:
    892
    Reading List:
    Link
    Well, programming languages are a small subset of natural languages and, by Rice's theorem, cannot decide non-trivial semantics.

    In other words, the meaning of words is beyond machines and must be programmed by humans.

    Ergo, real translation is beyond machines in the first place.
     
  11. AliceShiki

    AliceShiki 『Ms. Tree』『Magical Girl of Love and Justice』

    Joined:
    Apr 27, 2016
    Messages:
    24,647
    Likes Received:
    98,364
    Reading List:
    Link
    But law and medical texts are like... Some of the easiest things to translate in the world?

    I mean, sure, you need to actually know what their lingo means, but if there is one thing that a computer can do well, is storing the meaning of words, so this is certainly trivial for google translate.

    I think the author was talking primarily about text that doesn't deal with objective things specifically, but subjective things.

    Google translate can translate law and medical texts because the only requirement to translate those texts is knowing what the words mean... The problem with Machine Translation is when you try to put it together with stuff that require understanding ideas... Such as when metaphors are used, or rhymes, or neologisms or idioms... Yanno, the kind of thing that doesn't revolve around words alone.

    Machine Translating Software is wonderful for objective texts, but it's pretty useless for stuff like novels.
     
    chan88, Kuro_0ni and Vilidious like this.
  12. Vilidious

    Vilidious Well-Known Member

    Joined:
    Jul 28, 2017
    Messages:
    699
    Likes Received:
    892
    Reading List:
    Link
    On the same vein, all that Artificial Intelligence, and Intelligent and Learning systems is about is making stochastically "good" decisions according to limited pre-programmed input/scenarios... usually also with limited and preprocessed real world training data (and validation sets).

    It's not a silver bullet and it won't replace humans in anything but severely limited cases which don't really require deep/creative thinking in the first place.

    Emotional AIs that love you and interact with you like a real creature would?

    Have fun collecting data and programming a combinatorial explosion worth of constantly changing, non-deterministic, fixed scenarios.

    Not happening with Turing complete automata.
     
Thread Status:
Not open for further replies.