Monday, May 05, 2008

Google Translation tool: (My First Look)

Just saw today google translation tool, nice to see effort they have done for hindi-to-english and english-to-hindi
playing with it just 5 min gave me following results:

While doing from Hindi-to-english
i/p: तुम कौन हो -> What you have
expected result: who are you
i/p: मे घर जा रहा हु -> Hu is in the home ->
expected result: I am going home

While doing from english-to-Hindi
i/p: who are you -> आप जो कर रहे हैं
expected result: तुम कौन हो
i/p: go away -> दूर
expected result:दूर जा

so i found its accuracy something around 20-30% for sentence. It is good for single word.

while speaking about translation from english-hindi and hindi-english is really very tough job, since
1. For single english word there are lots of hindi words available so its really difficult for tool to understand, It need manual editing for improving accuracy.

2. Meaning of words keeps on changing according to statement while changing from english to hindi

3. Structure of sentence, if you observe carefully, 'is' 'was' comes in the middle of english sentence same translated things comes in the last part of hindi sentence.

4. Determining sex of sentence is a tedious task
Statement "I am going" this statement have two forms depending upon who actually speaking it
male "मे जा रहा हु" Female : "मे जा रही हु"
This can be achieve only if we know who is actually speaking it from previous statements else we need manual editing.

5. Most Important thing it need strong dictionary in the background for all these comparisons and should have all possibles words for particular word.

Good to see google tried it, some more work will definitely improve it lot!!!


Anonymous said...

Looks like you're not a native Hindi {speak,write}r.

> i/p: मे घर जा रहा हु -> Hu is in the home ->

should be:

i/p: मैं घर जा रहा हूं -> I am going home.

Link to the translation of above.

Alex said...

This is a great post, but I think you're being too hard on the service.

1) "तुम कौन हो?" works - I guess it's trying to treat it as a clause from within a larger sentence without the punctuation. The answer you got is still odd.
2) You left out the anusvara, मैं घर जा रहा हूँ works.
3) "Who are you?" gives - rudely but correctly - "तू कौन है ?"
4) "Go away!" gave "दूर पर जाएँ !", while "Go away." gave "जाएँ दूर है ." which is just weird.

For some reason it insists on putting spaces between the final word and the punctuation.

None of the reasons you cite are things which would make Hindi any harder than any other language. In fact gender in Hindi is much easier to recognize than in many languages, as long as the machine is aware of the ergativity. (i.e. Pravin ne kitab kharidi.) Of course no machine could know from the text "I went home," whether the speaker is male or female.

As you rightly point out, it will take a much larger body of work for analysis for the tool to get better at idiomatic and contextual translations. But all in all, it doesn't seem any quirkier than machine translation for other languages. Try it with bigger bodies of text. (I just noticed you can now click on "Translate this page" for Hindi pages in Google results:

Pravin Satpute said...

thanks ashish for pointing mistake
normally i dont type hindi, so done this mistake
sorry for this!!!!!!!!
please let me know if any more worng thing

Anonymous said...

Your blog keeps getting better and better! Your older articles are not as good as newer ones you have a lot more creativity and originality now keep it up!

Anonymous said...

I will not concur on it. I over precise post. Expressly the appellation attracted me to be familiar with the unscathed story.

Anonymous said...

Nice dispatch and this enter helped me alot in my college assignement. Thanks you on your information.

Anonymous said...

Nice brief and this enter helped me alot in my college assignement. Gratefulness you seeking your information.