Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uhog.mit.edu!sgiblab!swrinde!pipex!uknet!EU.net!sunic!trane.uninett.no!eunet.no!nuug!dkuug!news.uni-c.dk!diku!tyr.diku.dk!flanhart
From: flanhart@diku.dk (Rene Andersen)
Subject: Re: The Potential Pitfalls of Interlinguas
Message-ID: <flanhart.782746709@tyr.diku.dk>
Sender: flanhart@tyr.diku.dk
Date: Fri, 21 Oct 1994 13:38:29 GMT
References: <36vhr1$gk2@mango.aloha.com> <flanhart.781443385@tyr.diku.dk>  <flanhart.781878865@tyr.diku.dk> <1994Oct16.185436.9996@leeds.ac.uk>
Organization: Department of Computer Science, U of Copenhagen
Lines: 34

nik@scs.leeds.ac.uk (Nik Silver) writes:

>Information may not exist without language, but we may still conceive
>of an interlingua. That interlingua can be a natural language itself.
>Klaus Schubert has worked on a machine translation system using
>Esperanto as the interlingua. Having a natural language as an interlingua
>(and leaving aside the question of whether or not Esperanto is natural)
>means that (for example) words of too rough a granularity can be specified
>with relative clauses. Any theoretical problems will be exactly those
>of a human translator, which is not a bad result.

This is quite interesting, though it doesn't really get us anywhere.
I have always seen the purpose of interlingua to be: 

To resolve the ambiguity within natural language.

And if the interlingua is ambiguous itself (by being a natural
language) then we still have the difficulties of mapping, if we want
to (say) generate a translation in a new natural language. Any
arguments?

Besides, the MT-project with Esperanto (DLT) eventually was stopped,
so the success of that system can be debated. The main reason was that
Esperanto wasn't expressive enough, and further information to the
linguistic expression had to be added -- thus moving the interlingua
(Esperanto) further away from its natural language status.

As I have mentioned before, we have yet to see a suitable
interlingua.  No MT-systems based on this idea have (so far) been
successful.

best 

Ren'e Andersen (flanhart@diku.dk)
