🚀 go-pugleaf

On Thu, 27 May 1999 12:52:29 +0200, Boyd Buchin
<boyd@informatik.unibw-muenchen.de> skribis:

>Yan King Yin's wrote:
>>                    my idea is to remove the "semantic"
>> part of this problem so that the remaining task for the
>> computer is reduced to syntax / transliteration aspects,
>> things that computers do well.  
>
>Is this possible?

Hi Boyd. I'm working on a project for that, so I think it's possible.
It's called UTL and will have a proper web-page soon.

>First of all you will obviously need disambiguation (i.e. for words and
>syntactic structures that can be interpreted in two ways you have to
>state which interpretation is intended) for this to work. Let's suppose
>this is possible for the fragment of speech and the languages you have
>in mind (e.g. mathematical texts, technical manuals or law texts and the
>languages English, French and German).

I think it'll be possible for much more, though translation quality
won't be such as that of a human translator. But serious errors will
be rather unusual. Post-editing will be necessary if target text
requires it, but anyway it will be minimum.

>But even then there are simple examples where a purely syntactic
>approach has to fail, even between relatively similar languages like
>English and German. In these two languages the most salient examples for
>this are negation and the generic use of articles. A universal language

I think you're talking here about a "generation" problem. I would like
you to give an example if not.



Saludos,
Marcos