Interesting that you included a screenshot for en2zh, which shows the quality of the translation is still laughably (or painfully) bad. Throw those sentences to a Chinese person and he wouldn't be even able to make a reasonable guess what it means... Google translate still needs work, A LOT of work..

Yes, you are right... except for "hello" the translations aren't good... even though "how are you" apparently can still be understood, Google's translation being just more formal than "你好吗"... I added your comment to the post.

Heh. I was going to post about this.

I think it's interesting to see that the bots "know" which language you've entered so that they don't try translating things which don't require translation. Of course, if you enter a bit of text with multiple languages, it can trigger multiple bots, producing some weird results when they try to translate the wrong languages. But I guess you're not supposed to do that... :-)

> I think it's interesting to see that the bots "know" which
> language you've entered so that they don't try
> translating things which don't require translation.

Actually sometimes they do. When I'm chatting with en2zh and enter the German "Wie geht es dir?" ("how are you") the bot will try to translate, causing a confusing mash of German and Chinese... (accessibility issues with the Flash-based Google Talk gadget seem to prevent me from copying, but here's a screenshot)

http://blogoscoped.com/files/google-translation-bot-gets-it-wrong.png

One of these days we might be chatting to other tongues "transparently" (there will no visible bot, but an interface displaying everything I originally say in e.g. German to the other person in e.g. Chinese, and vice versa) but only if the translation quality is much, much better...

Chatting in German with the English/Chinese bot probably would confuse it! :-)

<< One of these days we might be chatting to other tongues "transparently" (there will no visible bot, but an interface displaying everything I originally say in e.g. German to the other person in e.g. Chinese, and vice versa) but only if the translation quality is much, much better... >>

That's kind of what SDLChat Translator does when used with AIM. One party has to have both SDLChat Translator and AIM open, but you can follow the whole conversation in just one window. So, I could theoretically use SDLChat Translator to chat to you in German – I would only see and type English and you only see and type German.

[Disclaimer: I work for SDL.]

You're probably too humble to link but I found this via Google:

http://www.chattranslator.com

Erm... more like too ashamed. That site is awful and really doesn't do the technology the justice it deserves!

A neat thing one can do is start a group chat with the two bots for the same language pair (say de2en and en2de) and throw sentences in both languages or mixed sentences at them. If a sentence is only in one language, the right bot will translate it. But if the sentence is mixed, both will give their target language's full translation correctly.
This is very cool, very fun and very handy, but I doubt it will be replacing my dictionaries any time soon.

Should join en2zh and zh2en to one. Judge English and Chinese very easy.
And for Chinese, en2zh and zh2en not good name. named 谷歌翻译家(guge fanyijia, Google Translater) is better , Then use GF(also means Girl Friend, well know word in china) as the name.

(20:39:22) Hong Xiaowan: all of you
(20:39:24) GF: all of you 你们全体

(20:39:33) Hong Xiaowan: 你们全体
(20:39:34) GF: 你们全体 all of you

en2zh

google will be displayed as Google公司

Seriously ... I can't agree more with Jade .. they should have find a sentence that their translator could do better .. not translating 'are you free tomorrow' into 'it's your freedom, how about tomorrow?' ... I would say that after Google has used their own translator rather than that of Systran: http://blogoscoped.com/archive/2007-10-23-n54.html ...

Also, I wonder if these bots can actually understand abbreviations .. like tmr instead of tomorrow ...

Ionut noticed that there's some new (not available in Google Translator main page) translations available with this tool, too:

Bulgarian -> English
Finnish -> English
Hindi -> English
Croatian -> English
Ukrainian -> English
Urdu -> English

http://googlesystem.blogspot.com/2007/12/translation-service-for-google-talk.html

I think whats interesting is that the bot can be driven via a handheld too.. this means I can begin 'talkin' to peep's in their own language.. at least in read mode.. (for now!!)

All that is needed is a txt2speech ..think of blackberry (Google Talk client for BlackBerry.) translating from english to <any lingo> but in speech mode.. that would be a very powerfull connector!!

FI2EN is joke. just a mirror bot (it will send you back whatever you type)

LOL... Croatian bot is very funny. Translations are usually understandable, when they _are_ translated. But roughly 80% of phrases are just left in Croatian. So it is completely unlike a (stereo)typical Croat, who does "something" even if he doesn't quite know how. ;-)

Translation is fundamentally AI subject. Experts had long realized that mathematical/syntax base models won't work. Statistic base models had become mainstream approach for Voice2Text and translation.

I bet this translation bot will sooner or later become a portal of gathering statistics of translation between languages, as Google may find a way to encourage users to refine translations. So, Google continues its saga of collective intelligence.

More at
http://webandlife.blogspot.com/2007/12/google-talk-translation-bots.html

it surprised me that people are so picky about the translation quality. computerized translation is a hard topic and google has made its progress in the realm and it's good. imagine how you comminicate with someone who speaks a totoally different language, you don't speak sentences; you don't expect the other side can follow you in a single try. and that is the attitude we should have toward translation bots. actually it is far better than nothing.

> actually it is far better than nothing.

J, I think to judge whether it's "better than nothing" you would have to do a test. Speak to someone Chinese for instance who honestly doesn't speak a word of English, and you two use nothing but the en2zh and zh2en bots. Will you be able to communicate about the stuff you want to? Or will the conversaton go like this...

A: "Have you seen the new Stallone movie?"
B: "Why do you call my mother an ox?"
A: "No no... I mean the Stallone movie!"
B: "Stop calling my mother an ox!"

it's a very useful and interesting function with great potential. it's just funny that they pick suck a bad example. usually even if the translation is not perfect, at least it makes some sense.
i wish it'll have pronunciation in the future.

to Philipp Lenssen

your example works fine:
me:Have you seen the new Stallone movie?
en2zh:你见过新的史泰龙电影吗？

> Or will the conversaton go like this...

I think if that happened, you'd quickly realise you needed to completely reword your sentences – just like when speaking to foreigners if you're on holiday.

And no, I don't mean just type in capitals and leave spaces between each letter to simulate speaking louder and more slowly... ;-)

to Hong Xiaowan

don't agree with you. the original naming system is fine. it's logical and easy to remember and use. sometimes people only need 1way translation. and I don't think chinese will only use en/zh translation, more chinese are learning more languages nowadays. your method have no extensibility. now I'm using zh2en and en2jp together to translate zh to jp. obviously in the future such new bots will be added. and 谷歌翻译家？这名字太学究，就像“谷歌”一样会被人骂。就叫机器人挺好，实事求是

> your example works fine:

Again, I think to see if this survives reality one would need to get together two people of two different languages, each not speaking a word of the other tongue (but both tongues supported by Google language bots). A neutral observer who speaks both languages can then judge the quality. Judging just a single sentence is not enough, IMO..

> I think if that happened, you'd quickly realise you
> needed to completely reword your sentences –
> just like when speaking to foreigners if you're on
> holiday..

I think there's 1 very important difference: if you talk to a foreigner, you will hear their accent is completely broken, hence you can make assumptions on the quality of the rest of their understanding. But with an automated translation, especially with the statistical approach Google seems to take, the quality of the sentence may be completely fine, so the other person may forget they are seeing a very rough translation.

For instance, the English "How are you" was translated correctly into Chinese but to a much more formal way of saying it... so now the other person may think "He's using such a formal way of talking to me, he must feel distant to me and not be friendly." And that may be the exact reverse of what you have wanted to express with a casual "how are you".

Granted, both parties *know* there's a chat bot in the room, but are you always able to tell when the bot screws up, and when the bot correctly translates something? I think only a real test would show how much all of this affects everyone in a real chat.

Anybody knows a Chinese who doesn't speak English who has time for a test on Google chat?

I want a he2en and en2he. as far as i know ther is none – correct?

meng zh-jp will be better than zh2en and en2jp. In fact, Japanese and Chinese similar. Easy to get better translation result.

In fact, I did not like any tools for translation. Try to understood the meaning and try to spoke in my language.

best one for me so far

me: they don't give a damn of anything
en2de: Sie nicht geben, der alles scheißegal

me: they don't give a damn of anything
en2zh: 他们不给妈什么

me: 他们不给妈什么
zh2en: What they do not give mother

good discussion going on here i see.. to Philip: I'd volunteer my dad or grandma except I'm nowhere near them now, ;o), kinda hard to find a completely non-English speaking Chinese who regularly surfs (and use googletalk!) I can short-circuit my English speaking sector for a while, if you wanna test. =D. just let me know.

Yes Jade, if you know someone please contact me at infoblogoscoped.com so we can work out the details... it could be an interesting experiment...

or what about German to Chinese? That one I can volunteer myself. I'm not sure if English is the most developed among all the languages google included yet...

The problem, there is no de2zh or zh2de bot available apparently... so German <-> Chinese won't work with Google yet (and using an intermediate language like English in-between the two also doesn't seem to be a good solution)...

Hmmm, I see. An easy (but tedious) solution might be just open two conversations, one with the bot and one with the real person, and pasting back and forth..

^ I thought that was the idea of it, no?