The problem that Carlo mentions is even more serious in the Japanese example. If you could read Japanese, you'd see how very little the word-by-word translation had in common with the sentence translation. Take the word "If" which is translated as ??. Conditionals in Japanese are done by inflecting the verb. There's no equivalent of "If". In fact the only word level translations that make it into the sentence level translations are the words for "love", "cannot", and "mark".

If we could do word segmentation and stemming in languages other than english, it would be fun to see what percent of words at the word-by-word translation appeared in the sentence level translation. It might be a useful way to measure how different the grammar of the language is from English.

I really like this way of annotating text though. With some work, it might be useful for language learning.

Interesting idea, there is one slight problem: google translate doesn't work word by word, so if a word means something in context, the API call on the whole sentence will return something different for the whole thing. Look at how be is translated to être first, and est in the context of the sentence.

One can do only so much with just a few lines of code. There is another "tag" in parsing google translate results that can show word definitions with their respective weights. But it would be overkill to do so. I only intended to create a 5-minute code to show one cool, but frail, usage.

Of course I didn't mean to say this is not really cool. I especially liked your MakeBoxes code I think it might be very useful to expose to end users in some form, maybe it should be a way to typeset (tree) Graphs or just general expressions, maybe to show how evaluation works.

Thales, Thank you for an interesting post. I've enjoyed your example from English to Japanese. However, it does not work well from Japanese to English in my circumstance. Adding a bit to your GoogleTranslate[], it worked well.