Newsgroups: comp.ai.nat-lang
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!pipex!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.781878865@tyr.diku.dk>
Sender: flanhart@tyr.diku.dk
Date: Tue, 11 Oct 1994 12:34:25 GMT
References: <36vhr1$gk2@mango.aloha.com> <flanhart.781443385@tyr.diku.dk> <37c9fr$qir@oahu.cs.ucla.edu>
Organization: Department of Computer Science, U of Copenhagen
Keywords: semantic primitives
Lines: 100




I wrote (flanhart@diku.dk (Rene Andersen)):
>>
>>The biggest problem seem to be: What should the atoms of the
>>interlingua be? 
>>

jperry@oahu.cs.ucla.edu (John Perry) writes:

>So who ever said that interlingual atoms had to map one to one onto words
>of any language?  One of the essential properties of an interlingua is
>that it is independant of the morphemes of a language, yet is able to
>map into those morphemes cleanly.  Thus, the problem is not choosing one
>interlingua atom per language-distinct-item, as you seem to be indicating,
>but selecting both atoms and a combination scheme which allow you to 
>represent each of these distinctions.

I never said, that interlingual atoms had tomap one-on-one onto words
of any language. In the original posting i summed it up as follows:

  The crux of the matter is, that the more languages you include in your
  interlingua, the finer the distinction of categories should be. No
  natural or god-given categorization seems 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
  practically impossible to construct an interlingua!!!

So my point is that the categories you would have to have in your
interlingua, would not correspond to categories in any real world
language nor to the way we think in categories (as you mention
yourself later).  The categories in the interlingua would be
completely artificial!! I think we agree on this point. 

Then you illustrate how to handle the leg-example:

>To address your example, let's say our interlingua represents leg as
>an atom representing one of several things which support something.  The
>link between the leg-atom and the supported-thing-atom would indicate
>the correct selection of a morpheme in the target language.

>For example,  (Phys-Obj-37 (class human)
>			   (name John))

>	      (Phys-Obj-38 (class leg)
>			   (part-of Phys-Obj-37))

>Here, to represent Phys-Obj-38 in Spanish, the language gen program would
>have a rule that said something like

>  if class(OBJECT?) is leg then

>      SUPPORTED-OBJECT? = part-of(OBJECT?)

>      if class(SUPPORTED-OBJECT?) is human then

>	SURFACE-FORM? = pierna

>      else if class(SUPPORTED-OBJECT?) is chair then

>	SURFACE-FORM? = pie

>      else if class(SUPPORTED-OBJECT?) is animal or table then

>	SURFACE-FORM? = pata


And here I agree with you, that this would in fact solve the problem
(theoretically).  But -- and this is a huge BUT -- take a step
backwards, and have a look at the artillery required to deal with such
a simple problem as the one posed. What if I had used given words like
"love" or "democracy". Words that have a much more complex
structure. In these cases I find it completely impossible not just for
one person to define these words intewrlingual, but also for a group
of people to AGREE on these definitions.  I could even be mean and ask
you to define "right" or "wrong" interlingual.

For these reason I say it is practically impossible to construct an
interlingua -- even though some may consider it theoretically
possible.  As a further argument for this position, consider the fact
that no-one has yet succeded in doing so.


>The goal of interlingua is not to find a set of inter-words which will 
>map one-to-one in each language.  Furthermore, the goal should not be to 
>find a single interlingua which correctly and succinctly represents the way 
>people think in all languages, but to throw together something that is fine 
>grained and flexible enough in its representation to allow systematic 
>translation between structures in the interlingua and sentences/phrases/words 
>of the languages it accounts for.

Again, I agree with you in principle, and it would be nice if we could
construct a suitable interlingua. But, based on the arguments I have
given, I still say it is impossible of practical reasons.


best

Ren'e Andersen
