How Google Translate works

Open Forum
21

That was a deliberate (if not overly creative) calque from the Russian “Тогда я слегка свихнулся” (Then I went a little crazy).

Looks like my attempt to imitate the meaninglessness of the machine translation of that phrase into Russian was successful.

22

Aha or ага (or фрф on a qwerty keyboard)! I shall now be able to sleep so much better. Thank you and good night!

23

Night-night.

xoxo

24

Google Translate cannot understand what a “tomcat” says. I imagine that It does not have any knowledge of grammatical rules.

  1. 俺は、猫だ。 —> I is a cat. （×）
  2. 俺は猫だ。 —> I was a cat. （×）
  3. おれは、猫である。 —> Me is a cat. （×）
  4. おれは猫である。 —> Is my cat. （×）
  5. おれは、猫だった。 —> Me was a cat. （×）

From 1) to 4), the correct answers should be the same, that is ,“I am a cat.”
5) should be “I was a cat.”

25

Errata:
“It does not have” should read “it does not have.”

26

There are a lot of grammatical rules that can be acquired unbeknown to us by reading a lot and by listening a lot. But that is not the case when GT is searching through a huge database relying on some kind of statistical analysis. It cannot learn what we humans can learn, I suppose.

27

I am tempted to think that GT is rather like LingQ. Previous machine translation methods focus on teaching the computer the grammar rules. It is a complex and difficult task. GT doesn’t use grammatical rules. Lots of content is fed to the computer to let it work out its own rules. It’s kind of like how we LingQ users forgo formal grammar learning and just trust that our brain will be able to sort out how the target languge works based on a massive amount input. It’s not an exact analogy, but the similarities are interesting neverthelss.

28

I think that there are two questions to be asked.

  1. Can grammatical rules be found out only by statistical analysis?
  2. Can the algorithms that were given to the computer system change themselves?
29
  1. By analysis, yes, but not only by statistical analysis.
  2. If the main algorithm allows for changing other algorithms it governs, or even itself, then yes.
30

astamoore-san,

I don’t know if computer software can develop (or learn) like us humans.

31

No, not exactly like humans, of course, but a great deal of research has been done in the area of artificial intelligence.

Now, don’t get me wrong, I don’t believe for a second that machine translation will ever come anywhere close in accuracy or naturalness of human translators. But it certainly can be improved.

32

*to human translators (typo)

33

"It certainly can be improved. "
I agree with you on that. Thank you for your replies, astamoore-san.

34

Google never fails to impress me. So /resourceful/…