Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!EU.net!sunic!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.781443385@tyr.diku.dk>
Sender: flanhart@tyr.diku.dk
Date: Thu, 6 Oct 1994 11:36:25 GMT
References: <36vhr1$gk2@mango.aloha.com>
Organization: Department of Computer Science, U of Copenhagen
Lines: 70

devin@aloha.com writes:

>I am planning to develop an interlingua for machine translation and knowledge
>representation, and I need advice.  The plan of attack I am contemplating is
>to create a set of symbols, essentially nouns, that can be modified for use as
>verbs or modifiers by setting attribute bits.  Does this sound like a good
>approach, or have others tried the same thing and failed; and if so, then for
>what reason(s)?

As you may know, almost all MT-groups has abandoned the use of
interlingua for several reasons, of which the most important seem to
be: People can't agree on WHAT the interlingua should be like. So
although interlingua is attractive from a theoretical point of view,
in real life it works out to be impossible (objections anyone?) to
construct. But let me elaborate on this.

The biggest problem seem to be: What should the atoms of the
interlingua be? 

Consider the word 'leg'. Should this be an atom? Well, if we only look
at English then certainly. But assume that Spanish is also
involved. Spanish divides 'legs' in three categories:

1) 'pierna' -- human leg
2) 'pata'   -- legs of animals and tables
3) 'pie'    -- leg of chairs.

So in order to include Spanish we will have to use the three as atoms.
So far so good. Now include French, which have the following
categories:

1) 'jambe' -- human leg
2) 'patte' -- legs of animals (and insects)
3) 'pied   -- legs of furniture.

Note, how the categories goes ACROSS the spanish. This means we wouild
have to design the following five atoms in our interlingua, in order
to represent any leg-meaning in any of the languages:

1) human leg
2) legs of animals
3) legs of tables
4) legs of chairs 
5) legs of furniture exclusive tables and chairs.

The crux of the matter is, taht the more languages you include in your
interlingua, the finer the distinction of categories should be. No
natural or god-given categorization seem to appear at any stage. All
you get is a finer and finer distinction. For this reason, there is no
atoms on which to build an interlingua, meaning it is
theoretically/practically impossible to construct an interlingua!!!

So in my opinion, give up the interlingua idea. For further reference,
have a look at :

        W. John Hutchins & Harold L. Somers:
        An introduction to machine translation.
        Academic Press Limited. 1992.

Here, some fundamental problems with interlingua are discussed.


best wishes

Ren'e Andersen

PS Comments are welcomed.



