Google translation: a question for Evan
|
|
Thread rating:  |
Athel Cornish-Bowden - 23 Mar 2008 18:09 GMT Mark Liberman has an interesting poost today on LanguageLog
http://itre.cis.upenn.edu/~myl/languagelog/archives/005485.html#more
about how Google translation can translate "made in Austria" (from German, these words being as I've written them) as "made in USA", with similar oddities for translation between other pairs of languages.
Remembering Evan's illuminating discussion the other day of why it would not be as easy as one would guess for searching to pay attention to punctuation in search strings, I wonder if he has any idea about why this might arise (and why it might be difficult to avoid).
 Signature athel (BrE)
Jitze - 24 Mar 2008 01:28 GMT >Mark Liberman has an interesting poost today on LanguageLog > [quoted text clipped - 8 lines] >to punctuation in search strings, I wonder if he has any idea about why >this might arise (and why it might be difficult to avoid). I asked it to translate (Dutch-to-English) the utterance "Made in Holland" (which is of course English) and it tranlsated it to "Made in USA". For consistency of weird effects, it should have translated it to "Made in England" because that at least would acknowlege that I asked for Dutch-to-English rather than Dutch-to-American.
Like many translators, it seems that if it can't understand (recognize) a word(s) then it just moves them over to the translated stream untouched. So the "Made in" part presumably got moved over as unrecognized in the source language (Dutch) but why Holland became USA I can't explain...
It is somewhat reminiscent of when I was giving a paper at an international congress once, and it was being simultaneously translated into French. The translator who was usualy trailing me by a sentence or so in real time, mistranslated some technical jargon and I heard it. So without thinking, I corrected what she had said - in French. The translator however was working as an automaton and translated my correction into English without skipping a beat. While her brain translated it as if on autopilot, it did not grok the actual meaning of what was being said - i.e. that I was addressing the translator. She seemed to be completely in the mode of "take whatever you hear and translate (create output) in whatever is the opposite language to the input".
It took us about a minute to get back into the swing of things.
Jitze
Evan Kirshenbaum - 25 Mar 2008 00:01 GMT > Mark Liberman has an interesting poost today on LanguageLog > [quoted text clipped - 10 lines] > idea about why this might arise (and why it might be difficult to > avoid). That's just bizarre. I note that "Made in Austria" comes out "Made in Austria" for me going from German to English, although Jitze's "Made in Holland" comes out "Made in USA" when going from Dutch to English".
 Signature Evan Kirshenbaum +------------------------------------ HP Laboratories |English grammar is not taught in 1501 Page Mill Road, 1U, MS 1141 |primary or secondary schools in the Palo Alto, CA 94304 |United States. Sometimes some |mythology is taught under that kirshenbaum@hpl.hp.com |rubric, but luckily it's usually (650)857-7572 |ignored, except by the credulous. | John Lawler http://www.kirshenbaum.net/
Athel Cornish-Bowden - 25 Mar 2008 12:58 GMT >> Mark Liberman has an interesting poost today on LanguageLog >> [quoted text clipped - 14 lines] > Austria" for me going from German to English, although Jitze's "Made > in Holland" comes out "Made in USA" when going from Dutch to English". At the time I sent my original message I hadn't done any tests of my own, but now I have, and I find that I am quite _unable_ to reprduce the errors reported by Mark Libermann and others: every combination I try gives either the correct result or reproduces the original unchanged. Yet you and others report various peculiarities. It's hard to believe it's a question of operating system (I'm on MacOS 10.4.11) because the translation is presumably being done on a server that cares nothing about the client's operating system.
 Signature athel
Peter Moylan - 25 Mar 2008 14:21 GMT >> That's just bizarre. I note that "Made in Austria" comes out "Made in >> Austria" for me going from German to English, although Jitze's "Made [quoted text clipped - 8 lines] > translation is presumably being done on a server that cares nothing > about the client's operating system. Google has many different servers, so it's possible to get different results in different countries. At least, that's the case when doing a web search, so it might also be so for the translator.
 Signature Peter Moylan, Newcastle, NSW, Australia. http://www.pmoylan.org For an e-mail address, see my web page.
Athel Cornish-Bowden - 25 Mar 2008 19:16 GMT >>> That's just bizarre. I note that "Made in Austria" comes out "Made in >>> Austria" for me going from German to English, although Jitze's "Made [quoted text clipped - 12 lines] > results in different countries. At least, that's the case when doing a > web search, so it might also be so for the translator. I tried to avoid that possibility by using the exact same server that one of the commenters at LanguageLog gave as a URL. However, I realize that often doesn't work with Google as they like to send you to the server they consider appropriate, which is not necessarily the one you asked for.
 Signature athel
Jitze - 25 Mar 2008 23:08 GMT >I tried to avoid that possibility by using the exact same server that >one of the commenters at LanguageLog gave as a URL. Any given URL does not necessarily map always to the same actual server. Particularly with "large" environments, the URL takes you first to a front-end switcher who then passes on the connection/data stream to some other server in a farm that actualy generates the response to the request. This switching may be done on the basis of the client's location, or the nature of the request, or merely an effort at load balancing.
Jitze
Peter Duncanson (BrE) - 26 Mar 2008 00:36 GMT >>I tried to avoid that possibility by using the exact same server that >>one of the commenters at LanguageLog gave as a URL. [quoted text clipped - 6 lines] >may be done on the basis of the client's location, or the nature >of the request, or merely an effort at load balancing. http://en.wikipedia.org/wiki/Google_platform
 Signature Peter Duncanson, UK (in alt.usage.english)
TsuiDF - 30 Mar 2008 21:39 GMT Using Google in Belgium (a country for which German is one of the official languages), I can get it to come up with a correct translation for 'product of Germany' but for 'made in Germany' I just get 'made in Germany'. Same thing with 'Austria' and with 'made in France' or 'made in Belgium'. But if I go the other way, I can get it to correctly translate 'product of Belgium' from French into English; whereas 'made in Belgium' comes out (for rather obvious reasons) 'done in Belgium'.
Harumph.
cheers, Stephanie in Brussels
Peter Moylan - 25 Mar 2008 14:05 GMT > That's just bizarre. I note that "Made in Austria" comes out "Made in > Austria" for me going from German to English, although Jitze's "Made > in Holland" comes out "Made in USA" when going from Dutch to English". Another peculiarity: if you translate "Made in Holland" from English to almost any European language (including Greek, but not including Russian) the result is "Made in Holland".
If you omit the "in Holland", you get a correct translation for "Made".
 Signature Peter Moylan, Newcastle, NSW, Australia. http://www.pmoylan.org For an e-mail address, see my web page.
Mark Brader - 30 Mar 2008 00:49 GMT Athel Cornish-Bowden:
> Mark Liberman has an interesting poost today on LanguageLog > [quoted text clipped - 3 lines] > German, these words being as I've written them) as "made in USA", with > similar oddities for translation between other pairs of languages. Wow.
I ran into a curious one last night. Apparently someone on a TV singing contest had decided to replace the song lyric
Voulez-vous coucher avec moi ce soir? [1]
with the less racy
Voulez-vous chanter avec moi ce soir? [2]
and a viewer commented on this in alt.fan.cecil-adams (getting the spelling and punctuation slightly wrong, but this is irrelevant to what follows; I've tried all the variations).
A non-French-speaking reader of the group then ran the two sentences through Google translation to see what they meant -- and found that for the first one, Google *left it untranslated*, as if censoring it.
I was curious enough to try to reproduce this, and *I couldn't* -- for me it translated both sentences. It took some experimentation to find out why -- I hadn't bothered using the shift key when retyping the sentences. It turns out that Google Language Tools will happily translate sentence 1 into English *if you start the sentence with a lower-case letter*, but not if you capitalize it correctly. But sentence 2 works either way.
Just now I tried substituting "foutre" for "coucher", and also "me foutre" for "coucher avec moi", and it was happy to translate those forms. So I don't think censorship is really what's going on. But... good grief!
 Signature Mark Brader | "We may take pride in observing that there is Toronto | not a single film showing in London today which msb@vex.net | deals with one of the burning issues of the day." | -- Lord Tyrell, British film censors' chief, 1937
My text in this article is in the public domain.
Jitze - 30 Mar 2008 07:41 GMT >I ran into a curious one last night. Apparently someone on a TV singing >contest had decided to replace the song lyric [quoted text clipped - 24 lines] >those forms. So I don't think censorship is really what's going on. >But... good grief! I managed to reproduce these results exactly. Sentence [1] fails with a captial "V" at the beginning, but is translated quite happily with a lower-case "v".
Replacing the sentiment with "foutre avec moi" or "me foutre" worked either way and was case insensitive.
Now *that's* weird
Jitze
J. J. Lodder - 30 Mar 2008 09:22 GMT > >I ran into a curious one last night. Apparently someone on a TV singing > >contest had decided to replace the song lyric [quoted text clipped - 33 lines] > > Now *that's* weird Probably a case sensitive dictionary of fixed expressions,
Jan
Peter Moylan - 30 Mar 2008 11:43 GMT >>> I ran into a curious one last night. Apparently someone on a TV singing >>> contest had decided to replace the song lyric [quoted text clipped - 34 lines] > > Probably a case sensitive dictionary of fixed expressions, Possibly even a list of song titles. There's a certain logic to not translating song titles.
Is there a song called "Made in USA"?
By the way, I was once driving down the street when I noticed an attractive woman wearing a T-shirt with that song title on it. I came very close to running off the road.
 Signature Peter Moylan, Newcastle, NSW, Australia. http://www.pmoylan.org For an e-mail address, see my web page.
J. J. Lodder - 30 Mar 2008 13:19 GMT > >>> I ran into a curious one last night. Apparently someone on a TV singing > >>> contest had decided to replace the song lyric [quoted text clipped - 37 lines] > Possibly even a list of song titles. There's a certain logic to not > translating song titles. A testable hypothesis.
> Is there a song called "Made in USA"? Can't guess,
Jan
Leslie Danks - 30 Mar 2008 13:35 GMT [...]
>> Possibly even a list of song titles. There's a certain logic to not >> translating song titles. [quoted text clipped - 4 lines] > > Can't guess, No need to - Google is your friend:
<http://www.amazon.com/Made-USA-Pizzicato-Five/dp/B0000036SZ>
 Signature Les
J. J. Lodder - 30 Mar 2008 20:55 GMT > [...] > [quoted text clipped - 10 lines] > > <http://www.amazon.com/Made-USA-Pizzicato-Five/dp/B0000036SZ> Veux pas le savoir,
Jan
Evan Kirshenbaum - 30 Mar 2008 18:44 GMT >> Probably a case sensitive dictionary of fixed expressions, > > Possibly even a list of song titles. There's a certain logic to not > translating song titles. But the title of the song is "Lady Marmalade", no?
 Signature Evan Kirshenbaum +------------------------------------ HP Laboratories |If a bus station is where a bus 1501 Page Mill Road, 1U, MS 1141 |stops, and a train station is where Palo Alto, CA 94304 |a train stops, what does that say |about a workstation? kirshenbaum@hpl.hp.com (650)857-7572
http://www.kirshenbaum.net/
Paul Wolff - 31 Mar 2008 00:12 GMT >Peter Moylan <peter@DIESPAMMERSDIEpmoylan.org> writes: > [quoted text clipped - 4 lines] > >But the title of the song is "Lady Marmalade", no? Now we only need to establish that the name of "Lady Marmalade" is called "Haddock's Eyes", and we are home and dry.
 Signature Paul
LFS - 31 Mar 2008 08:08 GMT >> Peter Moylan <peter@DIESPAMMERSDIEpmoylan.org> writes: >> [quoted text clipped - 7 lines] > Now we only need to establish that the name of "Lady Marmalade" is > called "Haddock's Eyes", and we are home and dry. Nice.
 Signature Laura (emulate St. George for email)
|
|
|