SUO: reply to John Bateman
JB: John Bateman
SF: Scott Farrar
IN: Ian Niles
JB:
> follow up to some follow ups...
>
> first I was reassured by the:
>
> >IN wrote:
> >
> >>>I'm not sure I follow all of this. The assertion '(subclass Sentence
> >>>Clause)' is a goof, and I'll delete it from the SUMO.
> >>
JB:
> then things went all wonky again... :-)
>
> >
SF:
> >I would argue that a Sentence is a kind of Clause just like
> >'that Scott lives in Tucson' is a kind of Clause (a relative clause). This
> >noted, Clause is probably not so relevant to the SUMO, only for the specialized
> >linguistics domain. I would suggest deleting Clause altogether.
> >
JB:
> it may all come down to the dreary definition of terms, and this is going to
> vary according to discipline (and then again, theory). But here we go.....
>
SF:
Pinning down what exactly a sentence is remains problematic because
the characterization is intimately tied to
whatever theory you're working with. The structuralists (Bloomfield)
argued that a sentence was independent, that is, could stand alone as
the largest grammatical unit. This came before we knew much about larger
units like discourses and texts. Generativists (Chomsky) view the sentence
compmletely in terms of syntax and the formal devices used to construct
them. The Prague School stressed that the sentence was based more on
information content. Discourse people would say it relates more to turn
taking.... This could go on ad infinitum unfortunately. But see below:
JB:
> A relative clause is a subtype of clause, that is fairly clear; so are
> independent clauses
> ("Scott lives in Tuscon", "live in Tuscon" [imperative]), dependent
> clauses (a supertype of relative clause, a subtype of clause in
> general), finite
> clauses ("he is going to Tuscon", "that Scott lives in Tuscon"), nonfinite
> clauses ("to go to Tuscon", "having been living in Tuscon for several
> years",
> "Scott's boarding the bus to Tuscon": whole bunch
> of stuff about this last one and the cline from nonfinite clause to
> nominalization
> that we can very safely skip for purposes of the SUMO), etc. This is
> probably
> a fairly straightforward notion of subtype or subclass.
>
> Now lets take sentences, examples of which include:
>
> "Scott lives in Tuscon and sends emails to the list."
> "Scott lives in Tuscon, whereas John did not until he found out about
> the great climate,
> which is very healthy."
>
> (sorry about the examples; it's getting late here).
>
> Maybe you want to take these as examples of clauses too.... although I
> already find this difficult to get my head around while still ending
> up with a useful notion of clause. But, is the following a sentence?
>
> "Never!"
>
> as in "Never!", he cried.
>
SF:
This can analyzed at an instance of ellipsis, where some structure has
been deleted: (I'll) Never! (make my bed.) But again, that's theory
dependent.
JB:
> I suspect that many corpora annotations are going to say "yes!" (another
> sentence). Are these clauses? Obviously not. You may or may not
> disagree with the annotation as a sentence, but if the SUMO intends
> to start legislating how corpora should be re-annotated, then it
> is probably on a long trip to nowhere in that particular area.
>
SF:
You're right. SUMO shouldn't have anything to say about how a corpus is
annotated. But:
We have a current paper on getting around standardization
by using a linguistic ontology. I could send it on if you're interested.
I'll be presenting it at the Knowledge Technologies conference in Seattle
next week.
JB:
> On these readings then, sentence can obviously not be a subclass of clause.
>
> A sentence can, however, be seen as another name for a clause or a
> grammatically valid combination of clauses. Call the latter a clause
> complex. (Is a clause complex a subtype of clause?
> I guess; but I'll leave that to all you logic experts.)
>
> A written sentence can then be defined as (a) a name for an
> orthographically signalled
> independent clause or clause complex or (b) a name for a complete speech act
> also signalled orthographically. (And probably a bunch of other things too.
> But as a subclass of clause is a new one on me.) What a clause is can
> be defined formally (I mean in terms of linguistic form). What a sentence is
> appears almost orthographic. If it weren't so plain contrary, I would almost
> suggest that the SUMO dump sentence and use clause instead!
> Clause is nice and simple; sentence is not.
>
SF:
I think what's needed is a generic unit, call it Sentence (or Clause if
you want), that can later be expanded in a domain-specific ontology.
1) It should be smaller than a Text or Discourse
(subclass Sentence LinguisticExpression)
(disjoint Sentence Text)
2) It should express a Proposition.
(containsInformation ?SENTENCE ?PROP)
3) It should contain a Subject (optional) and VerbPhrase
(part ?SUBJECT ?SENTENCE);;how can we state something is optional?
;;just not include it?
(part ?VERBPHRASE ?SENTENCE)
Proposing the subclass Subject is problematic, like JB points out below.
We can change this to NounPhrase (already present in the SUMO) and, again,
allow sub-ontologies to define their own Subject category.
So we have for Sentence:
(=>
(instance ?SENTENCE Sentence)
(and
(exists ?NOUNPHRASE ?VERBPHRASE ?PROP)
(and
(instance ?NOUNPHRASE NounPhrase)
(instance ?VERBPHRASE VerbPhrase)
(part ?NOUNPHRASE ?SENTENCE)
(part ?VERBPHRASE ?SENTENCE)
(instance ?PROP Proposition)
(containsInformation ?SENTENCE ?PROP))))
> But in practice I don't think
> sentence can really be avoided from the SUMO perspective... it is
> such a nice familiar term....
> There are probably only problems for those of us who cannot parse the
> following
> clause without getting a category error:
>
> >A Clause is similar to a Sentence,
> >
You're right. This was a very sloppy statement on my part. I hope I
explained myself in the first paragraph of this message.
> for me, sentences can be made up out of clauses; the clause is more the
> material out of which I can make a sentence if I choose. I can sometimes use
> other materials to make sentences, and sometimes I can use several
> clauses to make up my sentence. But that does not entail
>
> Wood is similar to a chair.
>
> nor:
>
> chair is a subclass of wood.
>
> If one has an alternative linguistic ontology and associated theoretical
> concepts which do not give rise to a category error in parsing the
> above clause, then you are probably fine and do not see a problem; I
> think that problems do arise later, when you actually try to use these
> concepts for doing linguistics, but that is a theoretical debate.
> And these discussions of sentence/clause are drawing theoretical
> distinctions.
>
> The following draws some other theoretical distinctions:
>
> >=>
> > (instance ?SENTENCE Sentence)
> >
> > (and
> > (exists ?SUBJECT ?VERBPHRASE ?PROP)
> >
> > (and
> > (instance ?SUBJECT Subject)
> > (instance ?VERBPHRASE VerbPhrase)
> > (part ?SUBJECT ?SENTENCE)
> > (part ?VERBPHRASE ?SENTENCE)
> > (instance ?PROP Proposition)
> > (containsInformation ?CLAUSE ?PROP)))
> >
> (although this appears to be wanting to commit the SUMO to
> believing that sentences are all described by S-> Subject . VP (or are
> these rules much weaker than I thought?). Now, you don't
> see rules like S->Subject . VP in grammar books, unless they
> are examples in the exercises sections with a question like
> "say what is wrong with the following rule". So I don't
> see why they should be in the SUMO either.
Here, here. Let's keep syntactic rules out of SUMO.
>Yes, Predicate
> would have been better, and yes it is nice to link up the
> semantic/logical with the linguistic, and no it is not easy to do this
> -- it still keeps a whole bunch of linguists in work and
> they've been going at it for quite a while; so why should
> the SUMO be putting its big feet there when there are
> enough hard upper-level issues to go round?)
>
> Even if there is some theory of linguistics which is happy with
> the kind of statement embodied in the above rule
> (what is the class "Subject" by the way?), why should
> it (and not some other: or do all get thrown in?) be engraved in SUMO.....?
>
>
> >IN:
> >
> >... but I would argue for including them in the
> >linguistics-specific extension of the SUMO rather than in the SUMO itself.
> >As a general rule, I think domain-specific theoretical posits don't belong
> >in an upper-level ontology.
> >
> This should be the bottom line I think; saves a lot of time and keystrokes.
>
> John B.
>
>
>
>
>
>
>
>
>