W3C home > Mailing lists > Public > w3c-translators@w3.org > April to June 2007

Machine Translation Problems; Disambiguating Pronouns, Etc.

From: CE Whitehead <cewcathar@hotmail.com>
Date: Sat, 09 Jun 2007 12:00:33 -0400
Message-ID: <BAY114-F238E4B0C4B6EE114C1A418B3240@phx.gbl>
To: w3c-translators@w3.org
Cc: christophe.strobbe@esat.kuleuven.be, duerst@it.aoyama.ac.jp, ecrire@catherine-roy.net



Examples of Machine Translations where pronouns (it/she) and the like (for 
example, a word that can be either a pronoun or article but where the choice 
should be clear from context; and also a verb that has two meanings but 
again where the meaning should be clear from context) are not disambiguated

(also one example where the word order got really convoluted in the 
translation)

1.

(Below is the sentence that got goofed up yesterday, without the subjunctive 
-- which I understand is fading from oral French; but not from written; 
perhaps that's the problem with this first machine translation example)

Il va pleuvoir le dimanche, n'est-ce pas? Y a-t-il
quelque chose que je peux faire?

Machine Translation  (I got slight variation for the two translators I used 
in this one case I think; I'll check again, but that's out of the scope of 
this email):

It Sunday, isn't this will rain? Is there something which I can do?

* * *

2.


(The same text as above, except with 'puisse' [the subjunctive form] insted 
of 'peux' [the indicative]; but note that in neither case does the machine 
translator know what to do with "le" in "le dimanche"--'le' can be a pronoun 
or a definite article in French, but here 'le' is clearly, based on the 
sentence structure, a definite article!)

Il va pleuvoir le dimanche, n'est-ce pas? Y a-t-il  quelque chose que je 
puisse faire?


Machine Translation:

It Sunday, isn't this will rain? Is there something which I can do?
* * *

3.

(The machine translations do not disambiguate 'it' from 'her' in the example 
below [where 'l' in the phrase "l'a vue" can refer to either a thing or a 
woman/girl]; there's not enough information to choose 'it' or 'her' in this 
case, of course; but actually the translators I looked do seem  -- at this 
point in time -- to ignore all surrounding text when choosing pronouns to 
translate the pronouns in a phrase embedded in that text!):

Est-il possible que personne ne l'ait vue?

Machine Translation:

Is it possible that nobody saw it?

* * *
4.

(Below, the machine translator gives two options for the verb "comprenez" -- 
thus letting the user choose which is correct; a good idea perhaps?):

Etant donné que vous ne comprenez rien de ce qui est écrit ci-dessus, est-il 
aussi possible que vous ne compreniez pas le texte entier?

Machine Translation:

Since you do not include/understand anything of what is written above, is it 
as possible as you do not include/understand the whole text?

* * *

5.
(Is the word order messed up in the translation below, or is it the 
translation that is messed up?)

Pour les traductions ci-dessus, faut-il noter les URL's des documents 
originaires?

Machine Translation:

"For is the translations above, necessary to note the URL' S of the 
originating documents?"

* * *

6.

(A Convoluted Sentence in the original French  -- probably my French is 
wrong, too -- but anyway note that the correct translation of 'faire' in 
this case is 'do' but the machine translates 'faire' as 'make'; I thought 
the machine translations could do better at disambiguating the various 
meanings of faire!)

Le plus que je peux faire en ce cas-ci sera d'y ajouter quelques petites 
remarques sur les cas où il est possible qu'il n'est pas possible!

Machine Translatotion:

"More than I can make in this case will be to add some small remarks to it 
on the cases where it is possible that it is not possible!"


* * *

--C. E. Whitehead

cewcathar@hotmail.com

_________________________________________________________________
Get a preview of Live Earth, the hottest event this summer - only on MSN 
http://liveearth.msn.com?source=msntaglineliveearthhm
Received on Saturday, 9 June 2007 16:00:47 UTC

This archive was generated by hypermail 2.4.0 : Friday, 17 January 2020 20:27:39 UTC