Re: SUO Scope and Purpose -- representation level -- natural language
Paul,
I'd argue for a very fine grain of distinctions for the ontology which
would also be separate from linguistic classifications. I'd say that
(subclass-of Formation Event) and (subclass-of Executive Object) would be
so imprecise that those terms would not have a definition sufficient for
any range of meaningful inferences. I would also suggest that (subclass-of
Executive (or Manager Governing-body)) [leaving aside for the time being
whether disjunctions should be allowed in our language] is a poor choice in
an ontology as it conflates linguistic issues of the dual meaning of the
English work "Executive" with the ontological notions of what an Executive
is as a CompanyManager or as a GoverningBody. Issues of statistical
cooccurance in English utterances are not relevant for the ontology
although they may be for a natural language system which employs the SUO in
order to do better than statistical analysis alone.
Adam
At 03:57 PM 7/25/2000 +0200, Paul Buitelaar wrote:
>apease wrote:
>
> > Paul,
> > I may have misunderstood your example, but it seems to me an excellent
> > example in fact of how an upper ontology could assist in disambiguating
> natural
> > language. If we formulate your example sentence loosely as
> >
> > (instance-of Formulate-Token1 CreationAction)
> > (performedBy Formulate-Token1 SinnFein)
> > (performedBy Formulate-Token1 UlsterUnionists)
> > (objectActedOn Formulate-Token1 Executive-Lexeme2)
> >
> > and we need to decide which is correct
> >
> > (instance-of Executive-CompanyManager Executive-Lexeme2)
> > or
> > (instance-of Executive-GoverningBody Executive-Lexeme2)
> >
> > and we further assume that these terms have been defined with respect
> to an SUO,
> > if the SUO has very general rules that state people can't create other
> adult
> > humans, then the NLU system could pick the correct definition for the
> > instance.
> > While the meanings of Executive-GoverningBody, SinnFein etc would
> have to be
> > created in a domain specific ontology, the value of the SUO in this example
> > seems clear to me.
>
>Dear Adam, thanks for the comments. In principle I could agree with your
>argument,
>but things still depend on the level of 'grainedness' of the SUO that we are
>talking about. The definitions you give would indeed be useful, but do they in
>fact represent the level of representation used for the SUO?
>
>For instance, if the SUO is on the level of objects, attributes and events
>(the
>kind of upper level concepts to be expected perhaps), the definitions for
>formation and executive could be:
>
> formation : event
> executive : object
>
>In this case, there would be no ambiguity. An 'executive' is simply an object
>involved in some event. Not much information on what kind of object, what
>kind of
>event and what kind of relation there is between them.
>
>If the SUO would be more fine-grained, involving different kinds of
>events, and
>objects, the definitions could be:
>
> formation : action
> executive : individual OR collective
>
>In this case, there is a choice between two objects that are involved in a
>certain
>action: an individual or a collective object. On this level both
>interpretations
>would be equally acceptable (both individual and collective objects can be
>involved in actions).
>
>If the SUO would be even more fine-grained (getting to your level of
>representation), involving more specific actions, and more specific
>individual and
>collective objects, the definitions could be:
>
> formation : create-action
> executive : human OR group
>
>In this case, there is again a choice between two objects that are
>involved in a
>'create-action'. On this level, we can start distinguishing between these two
>interpretations of the word executive. For instance with rules or by
>taking into
>account the statistical probabilities of co-occurrence between the SUO classes
>'create-action' and 'human' and between 'create-action' and 'group'. These
>probabilities, however, will depend on the domain(s) for which they are
>computed.
>(The political domain will favor the second co-occurrence, whereas the first
>co-occurrence will be favored in domains dealing with birth-rates and
>similar.)
>
>Finally, distinctions could be made on the lexical (and domain specific)
>level,
>such as:
>
> formation : arrange
> executive : manager OR governing-body
>
>Such distinctions should obviously not be included in the SUO, but rather
>in lower
>level, domain specific ontologies.
>
>Let me know if this example somewhat clarifies the scope and purposes I
>see for
>the SUO. I would be interested to learn which level of representation you (and
>others) would argue for in the SUO.
>
>Thanks,
>
>
> Paul
>
>
> > At 02:59 PM 7/21/2000 +0200, Paul Buitelaar wrote:
> > >Content-Type: text/plain; charset=iso-8859-1
> > >X-MIME-Autoconverted: from 8bit to quoted-printable by ruebert.ieee.org id
> > >IAA22705
> > >
> > >
> > >Dear SUO list members,
> > >
> > >Given the recent discussion on this list concerning ontologies and
> > >natural language processing/understanding, I would like to share my vote
> > >and comments on the SUO proposal. Please let me know about any comments.
> > >
> > >I'm sorry this vote comes without much introduction to my backgrounds,
> > >but I refer to the material available from my webpage mentioned below.
> > >
> > >Thanks,
> > >
> > >
> > > Paul Buitelaar
> > > DFKI Language Technology
> > > Saarbrücken, Germany
> > >
> > > http://www.dfki.de/~paulb/
> > >
> > >
> > >Received: from dfki.de (lost [134.96.187.60])
> > > by clichy.dfki.uni-sb.de (8.8.8/8.8.8) with ESMTP id TAA09824;
> > > Thu, 13 Jul 2000 19:59:42 +0200 (MET DST)
> > >Sender: paulb@dfki.de
> > >Message-ID: <396E038E.F79381A9@dfki.de>
> > >Date: Thu, 13 Jul 2000 19:59:42 +0200
> > >From: Paul Buitelaar <paulb@dfki.de>
> > >Organization: DFKI
> > >X-Mailer: Mozilla 4.72 [en] (X11; U; SunOS 5.6 sun4u)
> > >X-Accept-Language: en
> > >MIME-Version: 1.0
> > >To: "Schoening, James R CECOM DCSC4I"
> > > <James.Schoening@mail1.monmouth.army.mil>
> > >Subject: Re: Call for vote on SUO Scope and Purpose
> > >References:
> <272086090786D211853D0000F808137102911C6A@mail5.monmouth.army.mil>
> > >Content-Type: text/plain; charset=iso-8859-1
> > >X-Mozilla-Status2: 00000000
> > >X-MIME-Autoconverted: from 8bit to quoted-printable by ruebert.ieee.org id
> > >IAA22705
> > >
> > > > To vote, send an email to me (not this list), by July 26, 2000,
> > > with
> > > > one of the following votes:
> > > >
> > > > YES (with or without comments)
> > >
> > >... although I doubt if the SUO development will have much impact on the
> > >last *
> > >of your list (about natural language understanding
> > -- see below). Typically,
> > >influence on word meaning (a major component in NLU) is on a lexical or
> > >(domain) specific semantic level. This means that abstract semantic
> classes on
> > >a level high up in a given ontology (assuming it is hierarchically
> organized)
> > >will be of not much use in NLU.
> > >
> > >Perhaps as an example, in identifying the meaning of "executive" in the
> > >following sentence, it will probably not be useful to identify merely
> events,
> > >properties, attributes, classes, actions and similar. Instead, one
> needs at
> > >least domain specific knowledge about possible concepts (and their
> linguistic
> > >realizations) in the field of "politics".
> > >
> > >"On his arrival in Belfast, there was no sign of an end to the deep
> divisions
> > >between Sinn Fein and the Ulster Unionists on the formation of a new
> executive
> > >for Northern Ireland and on the decommissioning of all paramilitary
> weapons."
> > >
> > >Nevertheless, I understand the objective of the SUO project is to define
> > >higher
> > >levels first which will give anchors for lower level, domain specifc
> > >ontologies. This is a reasonable objective, although it may be more
> effective
> > >to start from the lower levels and combine, merge them at an
> appropriate time
> > >into a more abstract higher level ontology.
> > >
> > > > * Natural language understanding tasks in which a knowledge
> > > based
> > > > reasoning system uses the ontology to disambiguate among likely
> > > > interpretations of natural language statements.
> > >
> > > Paul Buitelaar
> > > DFKI Language Technology
> > > Saarbrücken, Germany
> > >
> > > http://www.dfki.de/~paulb/
> > >
> > >
> >
> > -----------------
> > Adam Pease
> > Teknowledge
> > (650) 424-0500 x571
-----------------
Adam Pease
Teknowledge
(650) 424-0500 x571