Google adds instant Japanese translation to Translate's iPhone camera-based Word Lens

This is really cool! Japanese is a very complex language to read and write. The Kanji characters can have different readings depending on how they are used or combined while you can look up a character doing so in even a semi-efficient manner requires a basic knowledge of the characters, meaning translation is all but impossible for someone who doesn't know the language. Automating this via optical recognition is a pretty impressive feat

Must dig out Word Lens, and check the language packs. Non-roman languages are something I might find it more useful for. I've only had to use it a couple of times in Germany or France, and the problem then was network connectivity when travelling.

Because, 'I gotta pen…i gotta apple' translates to: phuq you English, we're never going to learn you.

Sukebe7 said:

Because, 'I gotta pen…i gotta apple' translates to: phuq you English, we're never going to learn you.

Nice of you to barge in. 

MplsP said:

This is really cool! Japanese is a very complex language to read and write. The Kanji characters can have different readings depending on how they are used or combined while you can look up a character doing so in even a semi-efficient manner requires a basic knowledge of the characters, meaning translation is all but impossible for someone who doesn't know the language. Automating this via optical recognition is a pretty impressive feat

It's probably the easiest language for a computer to translate however because the nature of the language, and if AI has been used to drive it, people still use google translate to line-by-line read Japanese games.

- Kanji - Japanese Names, Nouns and Verbs that have direct meanings. Some Kanji have multiple meanings, which is why context matters.
- Katakana - Foreign Names and sounds
- Hiragana - Everything but Foreign Names

So if it detects Kanakana, it knows to try and and sound out what the word is. Most Japanese sentences are a mixture of Kanji and Hiragana, and the younger the reader targeted, the less Kanji there will be. Try using this to read print shoujo/shonen manga, I bet it would get enough right.

Getting the nuances of the language however, AI is unlikely to get. This is why translating something like English or Korean tends to be a crapshoot, because of words like "their,there,they're" , "your,you're" and "its,it's" because these sound the same, and people use the wrong words all the time (especially online) so without knowing what that word sounds like in the context of the sentence the AI has to read the entire sentence to determine a meaning. With Kanji, we know Kanji isn't going to be used for onomatopoeia, and can be translated typically two-characters at a time.

(I've amateur translated stuff before, and while Japanese IS hard for a human to learn, English is in fact harder because of the tendency to use idioms and metaphors. So an AI is more likely to be mislead in English than it is with Japanese. However the one thing I did find in Japanese that neither Google or Bing could figure out, are phrases that roughly mean "I will do my best today" with an obscure dialect.)