RE: Changing the language of gnus menu entries

[Drew Adams <drew.adams@oracle.com>]

> >> Well, the translation sucks. :-) Word order is so messed up that it's
> >> often difficult to make heads or tails of a sentence...
> >
> > Often, or sometimes?  Difficult, perhaps, but not always impossible,
> > unlike the case for someone who does not understand the source language.
> 
> Well, *I* would not have been able to understand enough to learn Emacs
> basics from it.

Even if your only alternative were in a language you could not read at all?
Thai?

I'm guessing you're thinking that your alternative is reading English, which
you happen to understand well.  Put yourself in the place of someone who
does not know English well.  That's the use case this is for.

> >> Such a translation would probably scare off more people than having no
> >> translation at all.
> >
> > I doubt it.  That's quite an exaggeration.
> 
> Perhaps. But I was thinking of people who understand enough English to
> get through e.g., the tutorial with some effort, who'd prefer a
> translation in their own language and are then put off by the Google
> Translate translation. It creates a bad first impression, which might
> rub off on Emacs as a whole.

That's not the use case I'm talking about.  Such a person would not try
an on-the-fly Google translation.  Circulez, il n'y a rien a voir.

And I care much less about Emacs docs making a "bad impression" than I
do about Emacs providing some help to users who currently have *none*.
This extra help would be optional, better than nothing, and advertised
clearly as such.  Anyone would be free to refer to only the English doc,
just as now.

[And even some people who don't need this feature, such as yourself,
might experiment with it anyway and send in corrections (just as you did,
*immediately*, signaling that the "key" translation for German was off).]

> > Even someone such as yourself, whose English is excellent, should not be
> > "scared off" by the translated text shown above.  You might laugh a bit,
> > and you might scratch your head a bit here and there.
> 
> The English translation you provided was a lot better, I'll admit.

Consider too the possibility that you might find it better because you
are better at German than English?  I cannot judge which is better, knowing
little-to-no German.

> However, part of the problem is that the quality of the translation is
> so unpredictable. Sometimes you get something that is quite usable,
> other times it's, well, slightly less usable.

Right.  Which is why this better-than-nothing option would be just that:
(a) better than nothing and (b) optional, on-demand.
 
> For a human, translating from English to French is not more or less
> difficult than from English to German. For a human, word order is hardly
> an issue, provided s/he is fluent in the relevant language. For a
> computer, things are different, however.

> > Google translation is not the same thing as having a professional
> > translator concentrate on Emacs docs, of course, but it's not the same
> > thing as automatic translation either.
> 
> I'm inclined to call any translation that is done by a computer
> "automatic". The fact that Google has access to large parallel corpora
> and probably has some very nifty machine learning algorithms to exploit
> those corpora doesn't really change that fact. I'm not saying it doesn't
> improve the translation, I'm sure it does, but it's still automatic IMO.

Doesn't matter much what we call it.  What matters is how helpful it is.

The fact that people use Google translate (and they do) is proof that it
can be helpful.  And that's the point: it often - in fact typically, I
think - is better than no translation at all.

> > I wonder how you might feel if the original Info doc were ONLY in Chinese
> > or Japanese or Korean or Thai, and you had a choice between ONLY that or
> > also a translation to German (or English or...)?  Think about it.
> 
> Rather than think, let's do a little experiment. For fun, I tried
> translating a bit of Japanese. Taking the third paragraph from
> <http://cx4a.org/pub/emacs-is-dead.ja.html> (after the header "Emacsの思
> 想") and feeding it to Google Translate produces the following:
> 
> ,----
> | And I should say first, but the "thought of Emacs" to speak here is
> | what I personally made ​​up on its own. I have never as far as I know,
> | I saw no one from telling "thought of Emacs", including Stallman. It will
> | place for such Stallman continues to exercise toward the "spirit of
> | freedom" is a great more his goal in particular, and said not to Ganchu
> | small problem of such "thought of Emacs". I want to say anyway, is that
> | ideological value is significantly lower because no one is sponsored by.
> | Please take care only that point.
> `----
> 
> Which is complete gibberish to me. Not a coherent thought in sight.

;-)

But all you've shown is that you can find Japanese text that Google cannot
translate well.  Instead, compare apples with apples & oranges with oranges.

Let's take the exact same Emacs tutorial text (your suggestion) as we used
for the last experiment, and this time translate from Japanese to English:

  You can enter the command of Emacs, in general control key (key top
  META meta key (the key top and and ) that says CTL such as such as CTRL CONTROL
  You use it to ) that says EDIT such as such as ALT . So , META such as CONTROL
  You decide instead of writing such as , to use a notation such as the following .

   Press and hold the C-< > character control key , and then press the < >
  character key . For example, Cf is to press the key of f while holding down the
  control key .

   Press and hold the M-< > character meta key , and press the < > character key .
  Na meta key if If not , you press and release the Escape key , then the < >
  character key I press . I write and <ESC> things escape key later .

  ! Important ! : If you want to exit Emacs, and then type Cx Cc.
  ** Cc has been rebound, but you can use Mx mode-specific-command-prefix instead
  [More] **
  Lines beginning with ">>" is , tells what to do at that time . For example,

  >> Now continue to the next screen by typing to see (see the next page ) Cv.
  ( Come on , it is v while press . Control key Let's do it )
  Later , read on to the next screen in the same manner each time you finish
  reading one screen .

  In the next screen and the previous screen , there is overlap of some lines to
  what you see . This Re is because that what is displayed is continuous is to
  be seen immediately .

  First , you need to know how to go to move through the file . And by the C-v
  That you go to the next screen Te have found already . To return to the
  previous screen , M-v ( meta-key It is a v) press and release the <ESC> or
  v, hold down the .

  >> Try C-v and M-v several times .

That's nowhere near as good as the German-to-English translation, but it is
also nowhere near the gibberish of the passage you offered.  There are some
"coherent thoughts in sight", for example: "For example, Cf is to press the
key of f while holding down the control key."

And it would be good to hear from a Japanese speaker familiar with English:
perhaps s?he could guess some things from this English, based on word order
etc., that you and I cannot guess.  Dunno.

But these experiments also miss the point, in a big way.  We are using them
as a substitute test, only because we understand English.  But the use case
is not translating to English; it is in translating from English.  We (I, at
least) cannot judge how helpful Google translate is when going from English
to Japanese.

Let's assume, as is apparently the case, that Google translation takes
advantage of oodles of existing human translations that are out there.

My guess is that there is a *lot* more such translation out there from
English to Japanese than there is from Japanese to English.  Especially
for technical info.  And especially *quality*, professional translation.

(And I've seen plenty of lousy English that is presumably human-translated
from Japanese.  Such stuff surely does not help Google DTRT, does it?)

Based on that, I'd conclude that it is probable that Google translate
does a (much!) better job translating English to Japanese than vice versa.
(I'd be curious whether I'm right about this guess.)

And the more there is a skew wrt the hypotheses, i.e., the more there is
a difference in available quality translations from English to Japanese
vs from Japanese to English, the more support for my argument that Google
translation *from English* can be helpful.

> > I think the result is generally USABLE,
> 
> And I think that the quality of the result is too unreliable. It'll
> be useful sometimes, but completely unusable at other times.

If it is unusable sometimes, then it simply won't be used those times.
No problem.

"Unreliable" means little (nothing) here: we are not basing nuclear power
plant security on such optional, on-the-fly translation being "reliable".

If this translation is USABLE SOMETIMES, then it CAN BE HELPFUL.  Users
won't expect wonders from such translation, but some will sometimes find
it helpful.

The point is (still) that this would be more helpful than having no
translation at all.