Re: SUO: RE: CYC event vs. SUMO Process -- really different?
This note will respond to several comments made regarding
this thread. First, I would like to make clear what I think
can be accomplished by this thread.
1. Now that John Sowa's proposal has been approved,
and effort will proceed to develop some form of
merged ontology, I am trying to discover some procedure
that will enable us to make substantive progress in
constructing such an ontology. The issue of creating
mappings to SUMO, CYC, and other ontologies is related,
but not central to building a "merged" ontology, which
will be able to stand on its own and might eventually
come to be viewed by some as "the" standard with the
largest community of users. I don't think we need to
wait for a formal organization of the effort before
exploring processes by which we might proceed.
2. I suggest that we try to find some degree of
consensus on what would constitute an "Event" in a
merged ontology, since events (or the 4-D correlate
"activity") are central to all ontologies dealing
with real-world problems. If agreement can be found
on that class, the discussion can be expanded to
related classes.
3. If a different process for building such a merged
ontology appears preferable to any of the participants,
I hope they will make a suggestion for the alternative.
===================================
The main point at issue right now is whether an Event
(called Process in SUMO) must have a location in space
in the ontology we would be constructing ("SCMO").
The location can be a default or "unknown".
It seems to be agreed that an Event must have a
location in Time.
(1) As best I can determine, neither CYC nor SUMO
currently has axioms specifying that an Event **must**
have a location in space, though CYC has an axiom
specifying that a subtype of Event, "Event-Localized"
must have a (relatively fixed) location in space.
I have proposed that all events that are located in
time (events affecting properties of physical
objects or mental states) in fact do have a location
in space, and the representation of Event should have
a relation specifying that every Event has some necessary
spatial location. Applications that have no need to
consider the spatial location of an Event can ignore that
relation, which could be given an innocuous default value
of "inThePhysicalUniverse" or "Unknown".
(2) One participant has objected to the requirement that
Events must be defined as *necessarily* having a location in
space. If we have only one general kind of Event, then
whichever option we choose, there will
be potential users who either:
(a) have to remove the "necessary" axiom to the relation, or
(b) have to add the "necessary" axiom to the relation.
If this is the correct interpretation of the discussion
thus far, it may be necessary to put this question to
a vote.
(3) An alternative could be to have, somewhat similar to CYC,
a higher-level class of "TemporalEvent" which has no spatial
location required, and as a subclass of that "SpatioTemporalEvent"
which has both spatial and temporal locations as necessary
relations. Since I consider all temporally located Events to
be necessarily spatially located (and that the CYC distinctions
are not well motivated), I think this is unnecessarily
complicated, but I could live with it, if the majority of our
participants consider it the better solution. But I think
it would create problems mapping to other ontologies.
In contemplation of the need for voting at some point
in this process, I would suggest that we form a Working Group
of the IEEE-SUO group to specifically deal with the details
of creating a merged ontology. Before proposing such a
working group in a formal motion, I would like to know if
others think it is the best way to proceed. (if almost all
voting members wish to participate, a committee may be
unnecessary).
Meanwhile, I would like to reply to some of the comments
made about this issue, and we may be able to reach a tentative
consensus on the immediate question and proceed to related
questions.
============================
R.1.
Matthew West has responded:
>
> In a 4D ontology what SUMO calls Process, CYC calls event,
> in EPISTLE we call activity, it does necessarily have a spatial
> location. This arises because an activity consists of the temporal
> parts of the participants in the activity.
>
> Of course the problem here is that a 3D ontology does not admit
> temporal parts, only spatial parts, so there is a different sort of
> inconsistency that is not so easy to overcome.
>
> Of course that an activity necessarily has a location does not mean
> that you need to know or be interested in it.
>
[PC] Yes, the spatial location of Events (Activities) in 4-D is
to me a good reason to make spatial location a necessary relation,
as I think it will make the translation between the two
paradigms easier. Ignoring the (default) spatial location should
not create any problems for those who don't need it.
I hope it is clear that having spatial locations as necessary
relation attached to the EVENT class does not require that
anyone actually fill in the value, or do anything with it.
It is just there if one wants it.
---------------------
R.2.
Chris Partridge has written [in part]
[snip]
> This raises for me the question about the motivation for
> the necessity of the different types of relations. Wouldn't
> you agree that any attempt to synthesise these candidate
> 'ontologies' merely by examination of the axioms without
> an understanding of their motivation (or whether they are
> aware of or address the relevant issues) is a thankless task?
>
> I know from many discussions with Nicola that these kinds of
> concern are part and parcel of his work in trying to build an
> ontology. Not being particularly familiar with SUMO or CYC,
> I wonder whether it is possible to divine their motivations
> from the available documentation for dealing with
> the hypothetical Mary's birthday example. Maybe a SUMO or
> CYC person could comment.
[PC] Since SUMO and CYC people monitor this list, I also hope
that they will provide suggestions regarding specific issues.
If we make a suggestion here (e.g. that a spatial location
for #$Event/&%Process is a necessary relation) that they
think is not an optimal structure for an ontology, I hope
they will let us know and explain their views.
I also hope that someone familiar with the DOLCE ontology
will contribute to these discussions. As best I can
tell, it does appear that in DOLCE an event (one type of
"Perdurant") has a necessary spatial location, since
Perdurants have spatial parts (which depends on the
spatial locations of their participants). But that may
not be the correct interpretation.
R.3.
Chris Menzel responded
>>
>> 1. Require events to occur at locations (in regions) because
>> there are events that are not spatial at all, e.g.,
>> "what if Bob thought about Mary's birthday party?"
>> is a hypothetical, with no spatial relationships at all.
>
> It's also not an event in the relevant sense. In your question
> "what if Bob...", you are wondering about what would have been
> the case had a certain event *type* been realized. Types are
> abstract, so indeed nonspatial and nontemporal. But I take it
> OpenCyc and SUMO are talking about event *tokens*, all of which
> (plausibly, anyway) occur in a region of space-time. Thinking
> events, in particular, occur between the ears.
[PC] from which I conclude that he agrees that instances
(tokens) of Events must have some spatial location. That is
indeed the issue that we are discussing.
R.4.
Richard Cooper responded:
> If no 4-space specification is required in either events or
> processes, yet a 4-space specification can optionally be attached
> to both events and processes, then I am satisfied with the
> representations. But I don't want to have to deal with defaults
> unless I somehow represent that 4-space information is important
> in the situation(s) I want to represent.
[PC] I am not quite sure whether having some default location
("InThePhysicalUniverse") in the representation of Event and then
ignoring it completely means that one has to "deal with" it in
the sense Richard intended. Depending on how one implements a
program using the ontology, I can foresee essentially zero
additional computational overhead when having a required default
location.
R.5.
Adam Pease wrote:
>
> Rich,
> I agree with you on these points, and SUMO does not force you into the
> situation that you are concerned about.
>
[PC] I am not sure what Adam means here. Is it that SUMO does not
require that one specify the exact location of every instance of
Event (neither does any other ontology that I know of), or that
SUMO does not require that an Event even have a location? I think
that the latter is true, but I would like to know if Adam has any
objection to a merged ontology requiring a spatial location
for Events, as I have suggested.
Pat
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
internet: cassidy@micra.com
=============================================