RE: SUO: RE: Re: Comment #13 - 'NL'
Jon, Adam, Matthew, et al.,
My modest proposal about representing an SUO concept as
a triple has led to a lot of commentary. Before responding
directly, I'd like to say a bit more about my motives for
making that suggestion.
The first reason for replacing the English word "concept"
with a formal structure is that no one can agree on what
a concept "really is". Rather than spend endless e-mails
in trying to reach a consensus, I proposed that the two-word
phrase "SUO concept" be defined as a triple (p,t,s). Anyone
who wants to offer other proposed definitions of the English
word is free to do so, but the term "SUO concept" would be a
formal object whose "meaning" is defined by the KIF statements
associated with p.
The second reason is related to my ongoing work on controlled
natural languages, which I proposed earlier as a means of
representing the formal KIF axioms and definitions in a more
palatable form for people who are not native speakers of KIF.
The idea of expressing an SUO concept as a triple (p,t,s) is
the thin opening wedge that would enable a controlled NL
generator to translate KIF statements into an English-like
(or French-like, German-like, Japanese-like, ...) form.
Matthew West wrote:
>Your proposal below binds the concept to some linguistic based thing. This
>is certainly not how I see concept.
I'm not trying to impose any philosophical, psychological,
or linguistic theory of concepts on anyone. I am merely
suggesting a formal structure (p,t,s) as a replacement for
the vague terminology about concepts that has been used in
various e-mails. The term I propose, "SUO concept", is a
name for a triple (p,t,s), not a name for anything that goes
on in the human mind.
> So, a concept is a
>non-redundant reference to an object, where an object may be e.g. an
>individual, a class, or a relation.
That is precisely the purpose of the triple (p,t,s). In my
note, I suggested that p be a KIF predicate, but I have no
objection to generalizing it to refer to any KIF object.
>On the other hand the concept itself is not a particular term.
No, an SUO concept is formally defined by the KIF object p.
The term t is not the concept, but a word or phrase that can
be used in the English comments to refer to the denotation
of p. As an example, p might be a monadic KIF predicate
named LifeInsuranceCoEmp. The corresponding triple would be
(LifeInsuraceCoEmp, "life insurance company employee", "noun")
In any use of the ontology, the English phrase would be an
unambiguous reference to the corresponding KIF predicate,
whose axioms and definitions would constitute the total
"meaning" of that phrase within the scope of any SUO document.
It would also be possible to have a German term associated
with the same predicate:
(LifeInsuranceCoEmp, "Lebensversicherungsgesellschaftsangestellter",
"noun, masculine")
In this example, a four-word English phrase and a one-word (so
to speak) German phrase have the same "meaning" p. Since German
has distinct grammatical forms for different genders, the third
member of the triple, which is just "noun" in English, has an
extra qualifier in German: "noun, masculine".
Jon Awbrey asked three questions:
> 1. In your list of syntactic categories, was the omission
> of "pronouns" deliberate? In my way of thinking pronouns
> are even more basic than nouns, so I started to worry
> a little bit here.
I agree that pronouns are very important. But they are
indexicals, whose purpose is not to express a particular
"concept" or predicate, but to point to some other term
which does. I would assume that any version of controlled NL
would support pronouns (or their equivalent as variables), but
I don't believe that the SUO ontology would define new KIF
objects that would be expressed by pronouns.
> 2. What sorts of predicates correspond to prepositions,
> and I do not mean just the easy ones like "membership"
> for "in", which is really more like "belongs to" for
> "is in"? Are you going to have notions of partially
> saturated predicates, that is, of applying predicates
> to incomplete argument lists as a way of putting
> constraints on the tuples that can fill the other
> positions? Relative terms, rhemes?
I would allow phrases to be introduced, which might not look
like traditional prepositions or verbs. The phrase "for
the purpose of" might be considered a preposition. One
might also define the following triple:
(memberOf, "belongs to", "transitive verb")
This would say that the KIF predicate named memberOf could be
expressed by the English phrase "belongs to", which behaves
syntactically like a transitive verb. I would also consider
generalizing the third position s of the triple to something
like a "case frame" or in CG terms a "canonical graph", which
would allow quite general kinds of constructions, including
type constraints and defaults on the related terms.
> 3. Do you have in mind triples of the form <p, t, s1> and
> <p, t, s2> where s1 =/= s2, in other words, a polymorphic
> type regime?
Yes. There are many possible elaborations of the t and s
entries in the tuple. For example, the dyadic predicate
Possess might have two different triples:
(Possess, "?x1 has ?x2", "transitive verb")
(Possess, "?x2 of ?x1", "preposition")
As another example, a triadic predicate Between might have
a triple something like the following:
(Between, "?x1 is between ?x2 and ?x3", "intransitive verb")
This would, of course, raise all sorts of questions about
binding variables, etc. For the moment, I just wanted to
propose the idea of replacing the vague word "concept" with
a technical term "SUO concept", which could be elaborated
to support the means for translating KIF into controlled NL.
John Sowa