Helpful Google

« previous post | next post »

The marvels of modern natural language processing:

Michael Glazer, who sent in the example, wonders whether Google Translate has overdosed on old Boris and Natasha segments from Rocky and Bullwinkle:


But it seems that the Google speech synthesis systems are not in on the fun, because if I accept Helpful Google's suggestion that I might mean "I vud be grateful if jou vould čonfirm rečeipt of this email so that I čan be sure that is has reačed jou", and then use the synthesize button, what come out sounds less like Boris Badenov and more like a bad reconstruction of proto-indo-european:



11 Comments »

  1. mg said,

    July 17, 2017 @ 6:38 pm

    Thanks – I needed that!

  2. Smut Clyde said,

    July 17, 2017 @ 10:46 pm

    I would be gureitofuru if you would confirm reshīto of this email so zatto I kyan be shuā it has reached you.

  3. Rebecca said,

    July 17, 2017 @ 11:28 pm

    Kind of a classic garbage-in-garbage-out scenario — pick Croatian as your source language but type in English and nonsense it is.

    [(myl) Google actually has excellent language-recognition software, so a more helpful (but less amusing) suggestion would be something like "Um, this seems to be English, not Croatian — do you really want me to pretend that it's Croatian"]

  4. Keith said,

    July 18, 2017 @ 2:51 am

    I wanted to make an apple pie, but had no apples. I had onions, though, so I used those. The pie was somewhat strange.

  5. Sandra Hammond said,

    July 18, 2017 @ 8:47 am

    Seems timely, considering Don Jr. invited them to his offices at Trump tower. (Oh, that wasn't Boris and Natasha? You sure?)

    [(myl) My thoughts exactly. I was going to post pictures of Boris and Natasha juxtaposed with pictures of Rinat Akhmetshin and Natalia Veselnitskaya, but I had to leave for the airport. With a hat and darker moustaches, Mr. Akhmetshin would be an excellent match for Boris. And Ms. Veelsnitskaya certainly looks like someone who "raises tarantulas and is the National Chairman of the Society to Restore The Real Meaning of Halloween." Oh wait, that's the other Natasha.]

  6. rpsms said,

    July 18, 2017 @ 10:33 am

    I hear google is making self-driving cars. What could go wrong?!

  7. Smut Clyde said,

    July 18, 2017 @ 3:08 pm

    Works for Belorusian too.

  8. Daniel Barkalow said,

    July 18, 2017 @ 4:07 pm

    That looks like it's trusting that you're writing Croatian, but thinks you're having trouble with orthography and character set. Then it runs afoul of the fact that it's good at answering "which of these options is most likely", but terrible at answering "are any of these options at all plausible", and good/bad about completely ignoring anything that's not an allowed option.

  9. phspaelti said,

    July 18, 2017 @ 10:02 pm

    RE: @Smut Clyde' comment

    I was wondering if you got this result by trying the same with Japanese. There is of course a big difference. With Japanese there is no "Did you mean" output. Instead with Japanese it just doggedly treats the English as Japanese. The output you get below the input is the transcription, not a joke. If the English word is recognized as a loanword then it gives the "Japanese transcription":

    that –> zatto
    and –> ando
    etc.

  10. phspaelti said,

    July 18, 2017 @ 10:09 pm

    Meanwhile, if i try @Smut Clyde's output as an input (for Japanese) I do get a "Did you mean" suggestion:

    Did you mean: I would be gureitofuru if you would confirm reshīto of this email so zlato I kyan be shuā it has reached you.

    Helpfully the "zlato" is highlighted!

  11. David Marjanović said,

    July 19, 2017 @ 4:57 am

    Has nobody noticed that the input says woud instead of would?

RSS feed for comments on this post · TrackBack URI

Leave a Comment