Comments on: Google Translation http://faganm.com/blog/2005/10/10/744/ The personal weblog of Michael Fagan Fri, 21 Nov 2008 20:14:21 +0000 http://wordpress.org/?v=1.5 by: kats http://faganm.com/blog/2005/10/10/744/#comment-61 Mon, 10 Oct 2005 17:03:53 -0700 http://faganm.com/blog/2005/10/10/744/#comment-61 <p>At the google infosession last week, one of the things they talked about was their translation software.. the guy there mentioned that they&#8217;ve been able to throw so much data at their software&#8217;s learning algorithm that it now beats out all other translation software. Apparently there&#8217;s some translation contest that&#8217;s held every year and they won in all the categories. So yeah, it might be that they now consider their translation software da bomb! and ready to be rolled out.</p> <p>Also, your DeWitt link is broken (extra quote mark from the looks of it). </p> At the google infosession last week, one of the things they talked about was their translation software.. the guy there mentioned that they’ve been able to throw so much data at their software’s learning algorithm that it now beats out all other translation software. Apparently there’s some translation contest that’s held every year and they won in all the categories. So yeah, it might be that they now consider their translation software da bomb! and ready to be rolled out.

Also, your DeWitt link is broken (extra quote mark from the looks of it).

]]> by: mfagan http://faganm.com/blog/2005/10/10/744/#comment-62 Tue, 11 Oct 2005 09:10:21 -0700 http://faganm.com/blog/2005/10/10/744/#comment-62 <p>it&#8217;s true, they won in all categories. although the scores are self-reported. see <a href='http://www.nist.gov/speech/tests/mt/mt05eval_official_results_release_20050801_v3.html' rel='nofollow'>http://www.nist.gov/speech/tests/mt/mt05eval_official_results_release_20050801_v3.html</a></p> <p>even if this is a rollout of their own translation software, it&#8217;s not complete. I don&#8217;t see Arabic on their translation page at all&#8230; although I haven&#8217;t tried hacking the urls yet </p> it’s true, they won in all categories. although the scores are self-reported. see http://www.nist.gov/speech/tests/mt/mt05eval_official_results_release_20050801_v3.html

even if this is a rollout of their own translation software, it’s not complete. I don’t see Arabic on their translation page at all… although I haven’t tried hacking the urls yet

]]> by: Mattt http://faganm.com/blog/2005/10/10/744/#comment-67 Sat, 15 Oct 2005 23:16:52 -0700 http://faganm.com/blog/2005/10/10/744/#comment-67 <p>Arabic? That sounds like terrorist talk to me, Mike. Are you against Freedom? Why don&#8217;t you and John Kerry take your communist loving translation software and go find some other country to ruin? </p> Arabic? That sounds like terrorist talk to me, Mike. Are you against Freedom? Why don’t you and John Kerry take your communist loving translation software and go find some other country to ruin?

]]> by: Fred http://faganm.com/blog/2005/10/10/744/#comment-69 Sun, 16 Oct 2005 21:58:37 -0700 http://faganm.com/blog/2005/10/10/744/#comment-69 <p>It looks like the Google Version of Japanese-English Translation is fooled by variety of Japanese Text that works fine with</p> <p>Excel.co.jp (Fujitsu Ltd OEM)<br /> Yahoo.co.jp/honyahku (systran)</p> <p>Any clue what is going on? Gievn that MT is very different than translation who do you suggest is working with google on this?</p> <p>Is it possible that they have Japanese Subconractor (maybe someone like Fujitsu that has a 20% leading share of the Japanese Market) ? </p> It looks like the Google Version of Japanese-English Translation is fooled by variety of Japanese Text that works fine with

Excel.co.jp (Fujitsu Ltd OEM)
Yahoo.co.jp/honyahku (systran)

Any clue what is going on? Gievn that MT is very different than translation who do you suggest is working with google on this?

Is it possible that they have Japanese Subconractor (maybe someone like Fujitsu that has a 20% leading share of the Japanese Market) ?

]]> by: mfagan http://faganm.com/blog/2005/10/10/744/#comment-70 Sun, 16 Oct 2005 22:34:02 -0700 http://faganm.com/blog/2005/10/10/744/#comment-70 <p>I figure google is/has moved all language translation to their own software, but of course I could be wrong</p> <p>thanks for the info though </p> I figure google is/has moved all language translation to their own software, but of course I could be wrong

thanks for the info though

]]> by: mara http://faganm.com/blog/2005/10/10/744/#comment-82 Sat, 12 Nov 2005 13:15:44 -0800 http://faganm.com/blog/2005/10/10/744/#comment-82 <p>fiecare om are dreptul la viata<br /> toti oamenii sunt egali in drepturi<br /> libertatea este atunci cand poti face orice lucru care nu ii deranjezi pe ceilalti </p> fiecare om are dreptul la viata
toti oamenii sunt egali in drepturi
libertatea este atunci cand poti face orice lucru care nu ii deranjezi pe ceilalti

]]> by: wolfgang http://faganm.com/blog/2005/10/10/744/#comment-91 Thu, 24 Nov 2005 14:22:56 -0800 http://faganm.com/blog/2005/10/10/744/#comment-91 <p>Google hired one of the most famous MT experts, Franz-Josef Och (he was at RWTH Aachen before). They are working on statistical machine translation SMT. One of Google&#8217;s strong point in it is that they have massive computer power. For SMT to work, you need a lot of parallel text (source and translation) to train the engine. Note that the NIST score probably favours SMT in comparison to rule based systems like Systran (RBMT). </p> Google hired one of the most famous MT experts, Franz-Josef Och (he was at RWTH Aachen before). They are working on statistical machine translation SMT. One of Google’s strong point in it is that they have massive computer power. For SMT to work, you need a lot of parallel text (source and translation) to train the engine. Note that the NIST score probably favours SMT in comparison to rule based systems like Systran (RBMT).

]]> by: mfagan http://faganm.com/blog/2005/10/10/744/#comment-92 Thu, 24 Nov 2005 14:27:54 -0800 http://faganm.com/blog/2005/10/10/744/#comment-92 <p>Interesting. The whole idea of using anything other than humans to compare the quality of translations seems flawed to me.. </p> Interesting. The whole idea of using anything other than humans to compare the quality of translations seems flawed to me..

]]> by: Jaak Henno http://faganm.com/blog/2005/10/10/744/#comment-144 Mon, 12 Dec 2005 14:02:23 -0800 http://faganm.com/blog/2005/10/10/744/#comment-144 <p>A rather classical test of MT systems is double translation, e.g. english-german-back to english; with Google website this is very easy to implement, two-way translation is provided for all languages. In the following are results of copy-paste two-way translations from a sentence from the same page <a href='http://www.google.com/language_tools' rel='nofollow'>http://www.google.com/language_tools</a> (only European languages were used) - can you tell what was the original sentence?</p> <p>english-language1-english:<br /> Adjust the homepage Google, the announcements and the keys to announcement in your preselected language over our preference side.</p> <p>english-language2-english:<br /> Rules the homepage of Google, the messages and the keys to exposure in your language selected via our page of preferences. </p> <p>english-language3-english:<br /> Fix the homepage of Google, the messages, and the bellboys to the exhibition in its language selected via our page of the preferences. </p> <p>So - &#8220;Adjust the homepage Google, the messages and the bellboys to the exhibition in its language selected via our page of preferences&#8221;. Clear? </p> A rather classical test of MT systems is double translation, e.g. english-german-back to english; with Google website this is very easy to implement, two-way translation is provided for all languages. In the following are results of copy-paste two-way translations from a sentence from the same page http://www.google.com/language_tools (only European languages were used) - can you tell what was the original sentence?

english-language1-english:
Adjust the homepage Google, the announcements and the keys to announcement in your preselected language over our preference side.

english-language2-english:
Rules the homepage of Google, the messages and the keys to exposure in your language selected via our page of preferences.

english-language3-english:
Fix the homepage of Google, the messages, and the bellboys to the exhibition in its language selected via our page of the preferences.

So - “Adjust the homepage Google, the messages and the bellboys to the exhibition in its language selected via our page of preferences”. Clear?

]]> by: nunzio http://faganm.com/blog/2005/10/10/744/#comment-153 Mon, 26 Dec 2005 16:25:08 -0800 http://faganm.com/blog/2005/10/10/744/#comment-153 <p>Google Sprachtool – Test 1</p> <p>Testsatz: „Es war so schön mit dir. Wann werden wir uns wiedersehen?“</p> <p>Die Übersetzung: “It was with you when we will so beautifully see itself again?“</p> <p>Die Gegenprobe: „Es war mit Ihnen, wenn wir so schön uns wieder sehen?“</p> <p>8 mistakes in one sentence, wow! </p> Google Sprachtool – Test 1

Testsatz: „Es war so schön mit dir. Wann werden wir uns wiedersehen?“

Die Übersetzung: “It was with you when we will so beautifully see itself again?“

Die Gegenprobe: „Es war mit Ihnen, wenn wir so schön uns wieder sehen?“

8 mistakes in one sentence, wow!

]]>