First post here because I've got a very specific question. To give some background: I'm currently learning Japanese and I learn mostly through cross referencing jisho.org, Tae Kim's grammar guide and Google Translate (GT herein). I was doing some studying and realized again that I'm a "top-down" learner. I also teach English in Japan and this made me think of one of my students. So I tried to think of how to say "I learn from the top down." I came up with what I thought it would be, and checked with GT by translating "I learn downwards from the top." My sentence matched GT:


When I reverse translated, it gave me "I learn from the top down." - as desired. Fine and dandy, but I wanted to say it politely, so I changed the ぶ at the end to びます and retranslated 「僕は上から下に学びます。」To my surprise, GT spit out:

"I will learn from the top down." (Without the asterisks of course)

Now, from my understanding there's no discrimination between future and present tense in Japanese (provided a lack of context), so the polite and plain forms shouldn't translate any differently. Is there a specific reason why GT would do this? Or is this just the intensity of translation presenting itself as a flaw within Google's algorithms?

Thanks for the long read, Erik

  • 2
    Google translate knows a lot less about Japanese grammar than you do. It demonstrates that massive data pattern-matching can produce something not-wrong often enough to be (very) useful. But utterly utterly useless for learning Japanese. – Brian Chandler Jun 2 '17 at 8:28
  • Eh I disagree, GT is great for learning because it can still help you learn a lot of nuances that you wouldn't otherwise be able to learn from a dictionary or something similar. You can go to JSE or a linguist to ask 20 questions but I often see questions not satisfactorily answered on JSE and a linguist wouldn't have time for you. For example GT showed me the nuances of the kanji 的, which has incredible resolving power as far as ambiguity is concerned. It can do things like turn (some, afaik) no-adjectives into na-adjectives, and more that I don't have enough space left to put here (lol) – Erik Jun 2 '17 at 9:34
  • It's honestly a terrible tool for the English–Japanese language pair. If you use it, you'll spend lots of time chasing your tail, as in this example. – snailboat Jun 2 '17 at 15:44
  • Sure, if you're trying to get a concrete understanding of a specific notion, but as far as getting a general idea of something goes it's pretty good, then you can look elsewhere for further clarification. It's not like I spent hours on this example, I was playing around for 10 sec to get a general idea of something then if I really wanted to I'd clarify with native speakers. – Erik Jun 2 '17 at 19:46
  • One way of eliminating the ambiguity has more to do with expression, in this case. Here, you're stating the action that happens, but in English, this phrase also has the effect of being used as a description about you. In Japanese, this works in the rudimentary sense, but you could also try the equivalent phrasing of "I am someone who learns from the top down," instead. This would be understood more unequivocally, if the translation is correct. – psosuna Sep 14 '18 at 23:17

Whether 学ぶ or 学びます translate as learn or will learn is dependent entirely on context. Google Translate does not know the context of your sentence so they can only make a guess based off known translations of similar sentences.

Google uses machine learning and a J-E corpus (collection of matching Japanese and English sentences) to power their translation. The neural network that they use to translate the sentences is built from this corpus.

This is an oversimplification, but it's possible that there is a similar sentence to yours in the corpus which has a corresponding English sentence in the future tense. Thus, the resulting neural network will translate yours to the future tense in English because it considers it to have a higher chance of being correct.

  • Awesome, thanks for the input. I had figured this was the case, as I know a decent bit about machine learning, but I figured I'd check to see if I was missing something before I go spraying words at Japanese folks. – Erik Jun 2 '17 at 5:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.