Ali,
Re:
[Ali Hashemi]
Ø Putting
aside our philosophical differences, perhaps this forum is a place where a
concrete strategy can be formulated to show how a 10?/30? mill investment can
be spent on generating a minimal yet necessary family of interlingua ontologies
to serve as a backbone for all the domain ontologies - hence actually realize
one of the main purported benefits of ontologizing computing :P. In my mind, it
doesn't seem wise to wait for the ground to be fully populated before
generating such a tool. I fear ontology is going the way of AI in the 1980's,
with a lot of promise, but little to show for it. Too many of my (smart)
friends are turned off by ontologies because so much of what is out there is
.... disappointing / infantile?
Ø Pat
C, you've proposed this $ figure -- how do you see it being spent? Assume that
this money is there - what would be your priorities?
The project proposal I have been discussing in bits and
pieces on this forum is described in a bit more detail below. Until we can
identify a possible source of funding, I imagine that this will merely be the
subject of endless debate. Our experience of the past fifteen years with upper
ontologies is that nothing serious happens without a significant commitment of
money.
Pat
Patrick Cassidy
MICRA, Inc.
908-561-3416
cell: 908-565-4053
cassidy@xxxxxxxxx
===================================================================
FoundationOntologyProject
Proposal for a Collaborative Effort to Create a Common
Foundation Ontology and a Community of Users
The goal is to create an open-source common Foundation
Ontology (FO) that is used by some substantial community of users who will test
it in practical applications and share their lessons and results. Such a
common FO should have eventually enough users to motivate third-party
developers to create utilities to make it easier to use, and applications that
demonstrate its utility. To accomplish this, it is proposed that a
collaborative project be organized with about 100 funded participants.
Motivation:
Based on the experience of the past 15 years of discussion
and effort to develop a common foundation ontology (often referred to as a
common “upper ontology”), the problems encountered appear to be:
(1) no one view of the foundation ontology elements is satisfactory
to a majority of ontology developers
(2) the FOs thus far developed have each reflected only one
viewpoint on ontology structure, out of many possible views
(3) no impressive **publicly observable** practical
applications have been developed to provide motivation to learn any of the
existing foundation ontologies
(4) learning how to use existing foundation ontologies is
time-consuming; in the absence of strong motivation (such as impressive
applications), few are willing to undertake the effort
(5) developing impressive applications for an ontology
(those that go beyond data in – data out demonstrations) is very
time-consuming – even more so than developing the FO itself. This is one
reason why Cycorp has not fielded an impressive online demo of the use of their
ontology.
(6) it is likely that impressive demonstration applications
such as Natural Language understanding will require the coordinated
collaborative efforts of many developers, but that has been inhibited by the
absence of widespread agreement on a foundation ontology as the basis for such
an effort, and by the current NL focus on corpus-based statistical methods.
The solution appears to be to create a common FO by the
efforts of a large collaborative team that will also form a user community that
can evolve the FO and learn from each other’s efforts. Based on past
experience, this is likely to require about 100 participants representing at
least 50 separate teams of developers or users of ontologies. The effort
required to create such an ontology is considerable, and adequate funding
(average ½ of full-time) for each participant will be required. The project is
likely to extend over three years, and funding for such a project is likely to
require at least 10 million dollars per year (average 100K/yr per participant).
The success of such a project will depend on prior agreement
concerning the goals, timetable, and process by which the project will be
conducted. The anticipate phasing is:
(1) preliminary on-line discussion, after some commitment of
funding is obtained, so that participants can be solicited and identified, and
can get acquainted with each others’ views and goals for the project. On
this basis, they will make their decision whether or not to participate: No
funding will be required for this discussion, but the topic is not likely to be
taken seriously unless some probable source of funding has been identified and
has made a tentative commitment.
(2) an in-person meeting of a preliminary group of funded
participants, for one or two weeks, to prepare a formal set of goals,
timetable, and procedural rules. At this point some potential participants may
decide not to participate in the full project. Travel funding will be needed
for this meeting.
If there are not at least 50 participants who agree to the
conditions of funding (see below), the project will not be funded. This is to
guarantee that there is in fact a substantial community of users that result
from the project. Generating a community of users is one of the main goals of
the project.
(3) The main project will consist of development of the FO
containing representations of fundamental elements satisfactory to each of the
participants, allowing alternatives, and providing translation axioms or
procedures. In parallel, utilities to help make the FO easy to use, and
applications to demonstrate its use will be developed. It is anticipated that
the third year of the project will be devoted mostly to testing the utilities
and applications, with modifications of the FO made to supply any missing
functionality.
The goals, timetable, and process will be finalized in a
preliminary meeting of the participants, taking perhaps one or two weeks after
preliminary online discussions have acquainted the participants with each
others’ views and individual goals for their participation.
An initial set of proposals for the goals and process might
be the following. The development consortium will decide which if any of these
to adopt, and which to add.
(1) The distinctive and indispensable goal for the FO will be
to serve as a means to *translate* among the local representations (terminology
and format) of all of the participants, so that each individual can use his/her
own preferred local representation and still interoperate accurately with the
others. A second important function of the FO is serve as the basic set of
ontology elements with which the meanings of domain ontology elements can be
described (this function is served by existing FOs). For both of these
functions, the FO should have an adequate inventory of fundamental concept
representations to both create new specialized ontology elements and to
translate among the local representations of the participants.
(2) The ontology and any utilities develop by the project
will be open-source and open to suggestions from any interested party. A web
site indicating the current state of development of the components of the
project will be maintained so that suggestions can be received from any party
with an interest in the effort.
(3) setting and adhering to a timetable for development will
be essential. For this reason it is not anticipated that consensus will be
reached on most of the controversial issues, and when necessary disagreements
will be resolved by a quick vote of the executive committee (which may be the
whole set of participants, or a subset selected by the whole). Those who are
aware (based on prior discussion) that their own views will not prevail in such
a vote may choose not to participate, or may drop out after participating for
some time.
(4) the need for resolution of disagreements will be
mitigated by allowing the inclusion of all desired ontology elements from any
participant. Those elements that are logically consistent (after translation
by bridging axioms or translation functions) can be included in the base FO,
and those that are logically inconsistent will be included in one or more
extensions. The set of base FO and extensions may form a “lattice of
theories” or a simple hierarchy, as decided by the executive committee.
Importantly, *no* participant can veto the inclusion of an element from some
other participant. If it occurs that at least one of two or more logically
inconsistent representations must be included in the base FO, a choice of
*which one* to include will have to be made by vote, to keep the base FO
logically consistent.
(5) as a condition of funding, participants will agree to
test the FO in at least one application, and report to the consortium regarding
its performance and suitability in that application; if another ontology is
also tested in that application, a comparison of the performance of the two
ontologies will be reported, with an evaluation of the reason for the
difference, and any improvements that may be needed for the common FO.
Participants that drop out after less than two years may not be held to this
obligation. Funded participants are allowed and encouraged to collaborate on
developing applications in fulfillment of this condition. There may also be
non-funded participants, who will not be under any obligation to report their
use of the FO. Whether non-funded participants will have any voting rights for
resolving disagreements will be decided by the funded participants.
(6) together with the FO, it will be necessary to develop
utilities to make the resulting product easier to use. (a) One utility will be
a program to extract from the base FO a subontology that contains only those
elements required for a particular application. For example, since the base FO
will have alternative representations of some concepts, the extraction utility
will usually extract only one view, to maximize computational efficiency. The
full set of alternate views in the FO will be required only to *translate*
among the alternative views when information transfer is desired between
systems linked to the FO. By supporting extraction of subontologies from the
FO, the problem of computational complexity will be minimized.
(b) another utility will consist of a set of translation
utilities that can translate assertions from one alternative view (local
representation) to another;
(c) another utility will be a natural-language interface
that can both search for concepts described using a controlled subset of
English, and automatically translate concept specifications from English into
the logical representation. The controlled English vocabulary will be
maintained as a component of the NL utility, and can be expanded as needed to
make use of the FO easier by allowing more natural descriptions of the meanings
of new elements.
(7) At least two demonstration applications need to be
developed along with the FO, to illustrate its use. One of those applications
should be a natural-language understanding program, which will expand on the NL
interface utility developed, to include interpretation of text (by conversion
to the logical specifications of the FO) in at least one specialized domain,
and support question answering about any text interpreted.