3 ms·

Me too. I didn't notice the "translate" in the URL, and just thought it was written by a non-native English speaker. I did a double take at the end when I real

Me too. I didn't notice the "translate" in the URL, and just thought it was written by a non-native English speaker. I did a double take at the end when I realized it was translated. Does Google do a particularly good job on French to English? (Bigger training corpus?) I more often used it on Chinese or Japanese, and it didn't seem as good.

jvzr 13y ago
The source material is written in a rather simple and direct way. Short sentences, very specific words. It may make things easier for Google to translate. Very well written anyway.
Robin_Message 13y ago
I believe the Canadian parliament provides an excellent French-English training corpus (http://en.wikipedia.org/wiki/Hansard#Translation http://en.wikipedia.org/wiki/Hansard#Translation) For example, see http://www.parl.gc.ca/HousePublications/Publication.aspx?Language=F&Mode=1&Parl=40&Ses=3&DocId=5072532 http://www.parl.gc.ca/HousePublications/Publication.aspx?Lan... and http://www.parl.gc.ca/HousePublications/Publication.aspx?Language=E&Mode=1&Parl=40&Ses=3&DocId=5072532 http://www.parl.gc.ca/HousePublications/Publication.aspx?Lan...
Vivtek 13y ago
French to English generally works rather well because they have similar sentence structure (except for inverted adjectival order) and neither has a lot of long-distance dependency stuff. Even Systran kicks ass on French to English. Where FR>EN fails is vocabulary. This text, as you see, is really basic terminology, nothing technical. The French categorize things very differently in technical texts sometimes - to the point of despair for the translator. Google tends to do OK on that to a point, then fail in alarming ways when you least expect it.
- TheCraiggers 13y ago
  "then fail in alarming ways when you least expect it." That's basically the M.O. of all completely automated systems though. Especially when concerning non-trivial things like language translation. Stuff like that requires intelligence in many cases to choose the right word based on the context and information the speaker is trying to communicate.
- raphaelj 13y ago
  French grammar is very similar to the English's one. A text translated "word for word" to English from French, using the later grammar, will almost look readable. Strange but not really wrong. Most errors will be from an over-usage of perfect tenses ("I have visited" vs "I visited") and wrong superlatives ("most fast" vs "fastest"), as these are the biggest differences between the two languages.
JunkDNA 13y ago
Total idle speculation, but a while back I read about how google uses statistical algorithms to do their translations instead of using the more traditional machine translation methods. They do this by running the stats on sets of identical documents that are available in multiple languages. Given that French is the other established major international language, I suspect there is an exceptionally good corpus of text to use from international organizations like the UN. I would also suspect Canada has a huge number of documents in both English and French as well.
- cschmidt 13y ago
  That's exactly what I was thinking. All UN stuff is in both English and French, and as you say Canada would be another good source.
  - vidarh 13y ago
    Pretty much all EU publications as well.
    - kryptiskt 13y ago
      EU documents are written in twenty-seven flavors of bureaucratese though, don't know how useful that is for a corpus.
      - apendleton 13y ago
        Definitely: the EU, in particular, seems to be pretty inventive in their use of English, coming up with new words, novel grammatical constructions, and novel meanings for old words, and since so many of the documents are written by non-native English speakers, this inventions end up being assumed correct by other bureaucrats and re-used in subsequent writing, such that the EU has developed its own weird English pseudo-dialect. See: http://languagelog.ldc.upenn.edu/nll/?p=4620 http://languagelog.ldc.upenn.edu/nll/?p=4620
      - gotofritz 13y ago
vidarh 13y ago
Possibly, but also consider that English and French are very close. English vocabulary is largely lifted from French (there's the well known statement ascribed to Alexandre Dumas - that English is just badly pronounced French), and while there are some differences in sentence structure, they are fairly basic when you are mostly writing in present tense like here. His post is written in very clear, straightforward French with few or no fancy flourishes that might complicate the translation. My French is decidedly not the best, and I have no problem reading the original at a decent tempo - there's the odd word missing from my vocabulary that I had to figure out, but that's it.
- DanBC 13y ago
  Wikipedia has an article with some description and a huge list. It's interesting. (https://en.wikipedia.org/wiki/List_of_English_words_of_French_origin https://en.wikipedia.org/wiki/List_of_English_words_of_Frenc...)

jvzr 13y ago

The source material is written in a rather simple and direct way. Short sentences, very specific words. It may make things easier for Google to translate. Very well written anyway.

Robin_Message 13y ago

I believe the Canadian parliament provides an excellent French-English training corpus (http://en.wikipedia.org/wiki/Hansard#Translation http://en.wikipedia.org/wiki/Hansard#Translation) For example, see http://www.parl.gc.ca/HousePublications/Publication.aspx?Language=F&Mode=1&Parl=40&Ses=3&DocId=5072532 http://www.parl.gc.ca/HousePublications/Publication.aspx?Lan... and http://www.parl.gc.ca/HousePublications/Publication.aspx?Language=E&Mode=1&Parl=40&Ses=3&DocId=5072532 http://www.parl.gc.ca/HousePublications/Publication.aspx?Lan...

Vivtek 13y ago

French to English generally works rather well because they have similar sentence structure (except for inverted adjectival order) and neither has a lot of long-distance dependency stuff. Even Systran kicks ass on French to English. Where FR>EN fails is vocabulary. This text, as you see, is really basic terminology, nothing technical. The French categorize things very differently in technical texts sometimes - to the point of despair for the translator. Google tends to do OK on that to a point, then fail in alarming ways when you least expect it.

TheCraiggers 13y ago
"then fail in alarming ways when you least expect it." That's basically the M.O. of all completely automated systems though. Especially when concerning non-trivial things like language translation. Stuff like that requires intelligence in many cases to choose the right word based on the context and information the speaker is trying to communicate.
raphaelj 13y ago
French grammar is very similar to the English's one. A text translated "word for word" to English from French, using the later grammar, will almost look readable. Strange but not really wrong. Most errors will be from an over-usage of perfect tenses ("I have visited" vs "I visited") and wrong superlatives ("most fast" vs "fastest"), as these are the biggest differences between the two languages.

JunkDNA 13y ago

Total idle speculation, but a while back I read about how google uses statistical algorithms to do their translations instead of using the more traditional machine translation methods. They do this by running the stats on sets of identical documents that are available in multiple languages. Given that French is the other established major international language, I suspect there is an exceptionally good corpus of text to use from international organizations like the UN. I would also suspect Canada has a huge number of documents in both English and French as well.

cschmidt 13y ago
That's exactly what I was thinking. All UN stuff is in both English and French, and as you say Canada would be another good source.
- vidarh 13y ago
  Pretty much all EU publications as well.
  - kryptiskt 13y ago
    EU documents are written in twenty-seven flavors of bureaucratese though, don't know how useful that is for a corpus.
    - apendleton 13y ago
      Definitely: the EU, in particular, seems to be pretty inventive in their use of English, coming up with new words, novel grammatical constructions, and novel meanings for old words, and since so many of the documents are written by non-native English speakers, this inventions end up being assumed correct by other bureaucrats and re-used in subsequent writing, such that the EU has developed its own weird English pseudo-dialect. See: http://languagelog.ldc.upenn.edu/nll/?p=4620 http://languagelog.ldc.upenn.edu/nll/?p=4620
    - gotofritz 13y ago

vidarh 13y ago

Possibly, but also consider that English and French are very close. English vocabulary is largely lifted from French (there's the well known statement ascribed to Alexandre Dumas - that English is just badly pronounced French), and while there are some differences in sentence structure, they are fairly basic when you are mostly writing in present tense like here. His post is written in very clear, straightforward French with few or no fancy flourishes that might complicate the translation. My French is decidedly not the best, and I have no problem reading the original at a decent tempo - there's the odd word missing from my vocabulary that I had to figure out, but that's it.

DanBC 13y ago
Wikipedia has an article with some description and a huge list. It's interesting. (https://en.wikipedia.org/wiki/List_of_English_words_of_French_origin https://en.wikipedia.org/wiki/List_of_English_words_of_Frenc...)