[om-list] On the Analytical Semantics of Natural Languages

[Thomas L. Packer at home]

Om people

    I can't remember if I said this here or not (but it's worth repeating):
it would do us all very well to research what other people have done,
particularly the work on interlinguas:

http://www.ontolog.com/Ontolog/MathesisAsInterlingua.html#ExampleInterlinguas

    Have you guys looked at SUO-KIF or Ontic?

    If you have found any other interlingua-attempts, please let me know.
I'll add them to my list.

tomp


~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Víðar sum quem nihil obstat.
Omnia apud me Mathesis fiunt.
www.Ontolog.Com
~~~~~~~~~~~~~~~~~~~~~~~~~~~~


----- Original Message ----- 
From: "Luke Call" <lacall at onemodel.org>
To: "Thomas L. Packer at home" <ThomasAndMegan at Middle.Net>
Cc: "One Model List" <om-list at onemodel.org>
Sent: Friday, 10 October, 2003 05:33
Subject: Re: [om-list] On the Analytical Semantics of Natural Languages


I don't understand all of this at the moment but I'd like it if OM could
do #'s 1 & 2 for you--because it could represent any noun as an entity
with behavior and properties, and other parts of speech as the behavior
and/or properties. OM would let us map any word to those entities,
behavior, or properties, so it's not tied to a single language (unlike
what I understand Cyc etc to do).

I don't know when or if this will all work, but I'm working on it, slowly.

Human language is so darned inconsistent. A self-consistent interlinked
object model that lets you define taxonomies on the fly that represent
your understanding, seems useful. Such activity wouldn't be for the
purpose of a taxonomy for its own sake, but just a side-effect of
recording your knowledge in OM like you'd record words in a word-processor.

BTW I'm studying the language scheme again. I looked hard at Python as a
scripting & testing language, & as a way to write code w/ less typing
(at least for tests) and really liked it, but Scheme is so powerful in
its ability to represent and manipulate ideas that I may go that
direction; we'll see.

All this is such a learning experience.

Luke

Thomas L. Packer at home wrote:
> Mark
>
>     Sounds very compatible with my three-year goal, which may be a
> subset of what you are talking about (since I can't do everything).
>  ....
 >
>     I agree that we need a much more thoroughly defined and complete
> theory of the meaning of language, which I see as a mapping function
> between a statement in natural language and a corresponding statement in
> a very well defined idea-language, i.e. a semantic-explicating language,
> that is, an interlingua.
>
>     My plan has four pieces:
>
>     (1)  Develop my currently vague and ill-defined ideas of Mathesis
> into an interlingua, or as Mark calls, a megalingua.  That is, make a
> notation (kind of like a programming language, like scheme, lisp,
> prolog, etc.) that is capable of representing any idea in a logically
> usable way.  (Potentially, this Mathesis will be able to represent the
> meaning of a statement written or spoken in any language, English,
> German, Swahili, mathematical notation, first-order predicate calculus,
> C++, etc.)
>
>     (2)  Write a long list of translation-pairs by hand: each pair would
> consist of a statement in the target language (probably English) and its
> representation in Mathesis.
>
>     (3)  Develop or adapt a general-purpose inference/learning system
> that is able to learn a mapping function from the target language to
> Mathesis in the form of productions/rules or some other format.  For
> this part I might use some existing cognitive modelling architecture
> like Soar, Act-R, ICMAUS, etc., if I can find a way to make it
> fit.  Otherwise, I'll have to make something of my own.
>
>     (4)  Give the translation pairs to the learning system as a training
> set and a testing set and see if it can translate the testing set.
>
>     If the mapping function is in a suitable format, then we can learn a
> lot about semantics, assuming this system actually works.  Even if we
> can't understand the mapping function, if Mathesis is actually able to
> perform well as both an idea-language and as a logical system (something
> akin to lambda calculus), then we will still learn a lot about
> semantics, and have something very useful whether we learn from it or not.
>
>     To make it useful, we could write a query capability (like in
> prolog) or write a natural language generator to another target
> language, and have the best MT (machine translation) available.  :-)
>
>     Big dreams, as always.
>
>     ciao,
> tomp

> ...

>     ----- Original Message -----
>     *From:* Mark Butler <mailto:butlerm at middle.net>
>     *To:* One Model List <mailto:om-list at onemodel.org>
>     *Sent:* Thursday, 02 October, 2003 17:53
>     *Subject:* [om-list] On the Analytical Semantics of Natural Languages
>
>
>     I have lately discovered that although analytical linguistics is a
 > ....