RE: SUO: RE: CYC event vs. SUMO Process -- really different?
Dear Pat,
I think this discussion is precisely the sort of thing we need to
make progress on if we are to move forward with the SUO, so I am
reflecting it back to the list.
See comments below.
Matthew West
Principal Consultant
Shell Information Technology International Limited
Shell Centre, London SE1 7NA, United Kingdom
Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
Email: matthew.west@shell.com
Internet: http://www.shell.com
> -----Original Message-----
> From: Patrick Cassidy [mailto:pcassidy@bellatlantic.net]
> Sent: 17 June 2003 17:27
> To: West, Matthew R SITI-ITPSIE
> Subject: Re: SUO: RE: CYC event vs. SUMO Process -- really different?
>
>
> Matthew,
> This is a little extended answer to your question. I am
> sending it to you directly so as not to complicate the already
> multithreaded SUO discussion right now, but feel free to
> copy any part or all of this to a public note to the SUO list:
>
> > MW: I notice that you are already assuming that an
> activity/process/
> > event is extended in space-time. Why do you think this?
> >
> [PC] See next section and paper.
> >
> > MW: Personally I find a 3D ontology incoherent, and I really can't
> > make sense of more than small isolated parts, so I couldn't
> > possibly comment. It is not even clear to me what a 3D event
> > might be. (lets stick with actual historical ones for the moment).
> >
> > Does it (necessarily) extend in time as well as space?
> [PC] Yes.
>
> > If so, what does it consist of? (not temporal parts of anything
> > since continuants don't have temporal parts).
> > If not, what is it at all?
> [PC] A group of changes in attributes, each occurring
> in some interval of time. See details.
MW: I thought attributes were universals and not extended in space-time.
How do you account for this?
> >
>
> I will answer the other specific question below, after I
> have outlined my view of 3D Process/Event.
>
> My view of what a 3D process and a 3D event
> are is summarized briefly in a paper I presented
> in February at a meeting (CICLing-2003) in Mexico City:
>
> Dimensional Analysis to Clarify Relations Among the
> Top-Level Concepts of an Upper Ontology:
> Process, Event, Substance, Object. pp 333-346 in
> Computational Linguistics and Intelligent Text Processing:
> 4th International Conference, CICLing 2003,
> Mexico City, February 2003.
>
> available at:
> http://micra.com/process/DimensionsOfProcessAndEvent.doc
>
MW: I will try to read the paper over the next few days. Here I
respond to your summary below.
>
> Outline here. **BUT FIRST**: ignore the labels, they are
> for my own convenience in relating these concepts
> to natural language structures. If you don't like
> the label for any logical structure,
> call it a "banana" ;-)
> The point is that each of these logical structures
> is related to concepts used in natural language.
> Notice that I use "grouping" or "group" rather than
> "set", because I want to represent processes and events
> as occurring in Space-Time, and some mathematicians
> consider "sets" as being eternal and outside of
> space-time. Fine. So states, processes, and
> events aren't "sets" in the mathematical sense, but
> they are aggregates that behave in some ways like sets.
MW: The difference between aggregates and sets is that
being part of an aggregate is transitive, but membership of
a set is not. So it is easy for you to work out what you
mean (almost certainly whole/part).
> It should be made more formal. All of the below
> refer to 3D objects.
>
> =======================================================
> (1) An "Object" is something that has an ObjectAttribute.
MW: Can an ObjectAttribute have an attribute? (severly limiting
if it can't).
> An "ObjectAttribute" is something that can be attached
> to an Object by the hasAttribute relation. These
> concepts are too primitive to be specified by anything
> other than their relation to each other, and by reference to
> instances of each. They are both disjoint with Process, Event,
> State, System, PersistentObject (i.e. a 4-D "Object"),
> and Space (an abstract concept which includes Time).
> Processes and Events can have attributes, but they
> are different from ObjectAttributes.
> An Object is the only thing that can occupy a location,
> i.e. take up space. Processes and Events have a
> locational attribute by virtue of the Objects which
> participate in them.
MW: How can events and process have a location in space time
without taking up space time?
> A PhysicalObject is an Object that has mass and a
> location in our physical universe.
> (2) a "system" is any grouping of (one or more) objects plus
> the attributes, relations, and processes related to
> those objects. An ontologist can select the objects
> of a system and ignore those of no interest, even
> if they are in close proximity.
MW: This does nto seem to be more than a simple aggregate of
objects, which by implication are "atomic". Is that right?
> (3) an "instantaneous state" of a system is the grouping
> of the attributes (properties, relations, fluents,
> whatever you call them) of the objects and
> processes of the system.
How do you say for example "The room was at 20C all day." ?
> (4) a "process" is group of one or more functions describing
> the values of the instantaneous state (group of attributes)
> of a system as a function of time.
MW: This looks like a function that describes a property space,
rather than the processes that cause the changes in properties.
> Every Process
> extended over some interval of time results in an Event.
MW: In EPISTLE we have events too. For us an event is the temporal
plane of a state (see my paper for a full explanation).
http://www.matthew-west.org.uk/Documents/Spatio-temporal-Paradigm.pdf
A state (temporal part) is can have "properties",
i.e. relationships. They must be true for the whole temporal extent.
> A continuous process can give rise to an infinite
> number of Events, arbitrarily designated as the
> changes that occur in any arbitrary interval of Time.
> This is similar to Sowa's view of Process as a
> function, but he doesn't have the same view
> consistently in his paper, and doesn't relate it
> precisely to Event as I suggest in my paper.
> (5) an "event" is the grouping of all of the changes
> in attributes of a system over **some fixed interval of
> time**.
MW: Well I could have an aggregate event that was like that, but
ours describe the leading edge of a new state, and so have zero
time duration.
> The number of such changes depends on the
> granularity of the process, which may be specified
> as a non-zero number or may be considered to be zero
> for all attributes and also for time. In the latter
> case, the process is continuous and the number of
> changes in the grouping is infinite for any finite
> interval of time. Every Event is the result of
> operation of some Process. An Event may also be
> specified as a designated group of AtomicEvents.
> An AtomicEvent is one for which there is only
> a StartingState, an EndingState, and a TimeInterval.
> No intermediate states are defined (however, they
> can be specified later, and then the Event is no
> longer viewed as an AtomicEvent. Each of these
> logical structures is a representation of a reality,
> and the representation can change to suit the
> ontologist.)
MW: Sounds like atomic event is a bit of a waste of time.
> An "InstantaneousEvent" is deprecated as being
> not physically realistic. But when needed it can
> be represented by taking the starting and ending
> "points" to be the bracketing sequential TimeIntervals
> during which no changes in the attributes occur.
MW: As stated above our events have zero time thickness, but
can have non-zero duration (a boundary in space and time).
An example is a wave moving on the ocean. The wave front is
a continuous event.
> The change that is characteristic of the Event then
> occurs between Time Intervals rather than between
> Time Points. This should only be necessary to
> represent Mental Events such as change of ownership
> or change of human roles, that can be designated
> in advance as taking effect at a particular
> time point. No Event can take place at one
> time point, as that would lead to the logical
> contradiction that an object had two contradictory
> properties at the same time.
> A "ComplexEvent" may result from any number of
> processes running simultaneously, in linear or
> branched sequences. It is a series or group of
> series of AtomicEvents -- unless the Process is
> continuous. The sequential relation of the
> intervals is implicit in their time intervals,
> and need not be specified by an explicit
> relation between sequential component Events.
> (6) A "NullProcess" is a Process in which there is no change
> of any attribute as a function of time. It is the
> limiting case of a Process in which less and less
> change occurs. A NullProcess proceeding in any time
> interval gives rise to the NullEvent -- a period of
> time during which nothing (relevant) has changed.
MW: Your null process looks like our idea of state. We require
that activities cause change to distinguish them from states.
> The kind of "State" that is often referred to
> in computation would be closely analogous or identical
> to a NullEvent -- a period of Time during which
> the attributes remain constant. It can also be
> called a "PersistentState" to distinguish it from
> an "InstantaneousState", the attribute values at one
> time point.
MW: This is very analogous to our concept of possible_individual.
> (7) The kind of "Process" that is represented in PSL
> and can be represented by a Petri Net is classified
> as a "Procedure", which is a subclass of Plan, which
> is a subclass of "MentalObject". It is a set of
> propositions specifying a series of steps that
> are designed to achieve a particular result.
MW: This is equivalent to our class_of_activity. A particular
activity is a member of this class when it exhibits the various
steps. (classes are sets).
> Unlike real *physical* processes and Events, such
> propositional structures may include contingencies
> and loops. The **execution** of such a Procedure
> in some time interval gives rise to physical processes
> and their associated Events.
>
> ====================================
> That's the general idea. An outline of an ontology that
> includes those concepts and relates them to other concepts
> of interest is in Protege text file format at:
>
> ftp://micra.com/process/PUOCYC3.pins
> ftp://micra.com/process/PUOCYC3.pont
> ftp://micra.com/process/PUOCYC3.pprj
>
MW: Too much for me to easily get my head round. Lets settle
for NL at present.
> This is only an outline, and I have not yet been
> able to spend a lot of time on it. It is useful now
> primarily to illustrate how the process and Event
> representations would fit in with a broader ontology.
> I have taken some branches directly from CYC, which
> I think has a lot of useful structure at the middle
> levels, but is not to my taste at the higher levels.
> although I think I have a good idea of how I would
> want to build an ontology for my own purposes --
> mostly language understanding and generation --
> I have been reluctant to set off on my own and do
> it both because it is too much for one person, and
> also because even if I did that and built a system
> using it that could do impressive things, I would
> expect it to not be reusable elsewhere because the
> ontology would differ too much from that of others.
> I feel strongly that in order to enable the kind
> of knowledge and result-sharing that will help
> our field grow we need significant agreement on a
> widely used upper ontology, and that is why I am
> spending time trying to find a mechanism to build one.
MW: I appreciate the motivation for your efforts.
> =========================================
>
> As to the specific issues in the IEEE discussion
> group, I am trying to find some procedure that
> will allow us to make real progress in building
> something we can all (or most) use. I obviously
> haven't yet found the formula, since, no matter how
> much I try to get opinions on a specific issue, I keep
> getting comments (some very intelligent and perceptive),
> but not addressing the issue I hope to resolve.
MW: I think that will continue to happen until you address
foundational issues, though by all means use activity/process/
event as a place to start exploring these.
>
> =================================================
>
> Before getting to the question you asked, one more
> item while we're talking here:
>
> I have just started looking at the Epistle ontology,
> and am unclear how to interpret "PossibleIndividual" and
> relate it to the classes. In most ontologies I have
> seen, the hierarchy is implicitly a hierarchy of
> classes, but in the Epistle hierarchy it seems that
> "class" is explicit in the name. But then "PossibleIndividual"
> sort of looks like a class, but when "subtypes" of those
> ?classes? are mentioned, they seem to be referring to
> instances. And I still can't quite grasp the relationship
> between "activity" and "class of activity". I couldn't
> find documentation specifically explaining the usage and how
> it varies from what (I think) is more common usage.
> Can you point me to or send me some document that explains
> the rationale behind the general structure of the
> hierarchy?
MW: Some general principles:
1. All entity types are classes by definition.
2. Possible_individual is the class of all things that exist
in space time - possibly or actually.
3. There is a classification relationship between class and
thing that means that any instance of class (or its subtypes)
can take members from any instance of thing (or its subtypes -
which of course include both possible_individual and class).
4. Our convention is that say class_of_individual has as
instances classes whose members are all members of possible_
individual. So instances of class_of_activity has as members
instances of activity.
As an exmple of this, "pump" would be an instance of class_of_
individual. A particular pump would be an instance of possible_
individual (or one of its subtypes) and have a classification
relationship to "pump".
In turn, "pump" would have a classification relationship with
"equipment type" which would be an instance of class_of_class_
of_individual.
>
> ============================================================
>
> To answer the specific question:
>
>
> >
> >>The temporal parts of the 4D activity would have corresponding
> >>translations to the temporal parts of the Event.
> >
> >
> > MW: How would you relate the temporal parts of the event to the
> > continuants that participate in the activity/process/event?
> >
> [PC] The Event describes changes in the attributes of the
> continuants during a time interval. The Temporal Parts
> are each themselves Events, describing the changes in
> the attributes of the continuants during those time
> subintervals.
MW: But attributes do not even exist in space time ... do they?
>
> >
> >>The
> >>"temporal part" relations would not be identical because they
> >>would have different arguments, but they would be translatable.
> >
> >
> > MW: Temporal part of what? Activity/process/event or participant.
> >
> [PC] Events have temporal parts, which are themselves component
> Events.
MW: That is not good enough. How do events occupy space-time? Since
they do occupy space-time, I must be able to see them mustn't I?
> Processes, since they are functions in the PUO ontology,
> do not have Temporal Parts. But a SUMO "Process" looks a lot to
> me like what I would call an Event. The SUMO Process has not been
> described in enough detail for me to know if there is any difference.
>
> > MW: If you are proposing a dualist (continuant AND occurent) based
> > approach with an eternalist view of space-time then the occurent
> > elements might be considered the same as the 4-dimensional view.
> [PC] I do propose a dualist ontology. I think the views can
> be made to translate precisely, but as I have said, it depends on the
> specific axiomatization of the 3D and 4D views.
MW: Of course a poor 3D or 4D view will just demonstrate their
inadequacy when you try to do the mapping.
> I have suggested
> a 3D "Event" be the group (physical analog of set) of all changes
> in attributes during a time interval.
MW: Changes to what?
> I am not sure if an "Activity"
> or a "history" (Pat Hayes's term) would, if explained in detail,
> also amount to an analogous grouping. It could, for example, if
> in 4D it is considered as the set of *changes* in attributes of
> the instantaneous time-slices of the 4D objects participating in
> the Event. It can't just *be* a time-slice of an object, since the
> essential nature of a PUO Event is the *change* in attribute from
> time to time.
MW: It can be if you take the edge. There is a poitn where the
change has happened and the new state starts.
> I haven't seen enough discussion of changes in
> attributes over time in a 4D ontology to know if there can be
> a direct translation. I could imagine a formulation in 4D that
> would correspond, but I don't know if anyone uses such a
> formulation.
MW: What you have sounds close to what we do. We have states of
individuals when some property does not change, we have the temporal
edges of those states that we call events, and we have activities
that cause those events.
> The way "Activity" is glossed in Epistle sounds to me a lot
> like it would be a Process in PUO -- i.e. something that
> "brings about change". The changes then wold be the "Events".
MW: Correct. Your events seem to cover the period over which change
happens, rather than the point where the new state has come into
being.
> =============== gloss of Activity =====
> An <activity> is a <possible_individual> that brings
> about change
> by causing the events that mark the <beginning> or <ending> of a
> <possible_individual>.
> An activity consists of the substates of those members of
> <possible_individual> that participate in the activity. The
> participating substates will be classified by the <role_and_domain>
> that indicates the role of the substate in the <activity>.
> EXAMPLE Pumping a fluid with a mechanical pump can be
> represented
> by an instance of <activity>.
> =======================================
>
> . . . **but** in PUO a Process is the function that is active
> throughout the lifetime of an event. The way "Activity" is glossed,
> it sounds like Activities and Events alternate, and are not
> concurrent. I couldn't tell from my reading. If it is concurrent
> with Event, then it may be closely analogous to a PUO Process.
MW: Your events seem to be equivalent for us to a temporal part where
change is taking place.
>
> Notice that both CYC and SUMO conflate Process and Event. What
> they seem to be representing are only Events, and processes which
> are functions are totally absent from their ontologies -- which
> makes me wonder how they can deal with differential functions on
> physical processes.
>
> > Since all continuants have dual object that are their
> > life-histories,
> > a 4-dimensionalist will argue that this is just redundant.
> >
> In 3D we are really, **really** interested in how things
> change over time. If there were a lot of emphasis on that
> in a 4D ontology, 3D might not be necessary.
MW: 4D is all about how to deal with change efficiently. You just
need to read Peter Simon's "Parts" to see how complicated it gets
to take acccount of change in a 3D world.
> But when
> I read papers on the 4D issue, they tend to go on and on
> about whether something is "wholly present" at a point
> in time. Jeez, who the hell cares??
MW: No that is the 3D view, and the reason it gets talked about is
that it is the key difference in world view. However, in my view you
need to move beyond what the view is and see how it works in practice.
We have done that.
> It doesn't show up in my
> axioms, or it's just implicit in the existential operator.
MW: Being implicit is a bit problem, since it means computers don't
know what you mean.
> The main point is that **language** has a much closer mapping
> to 3D than 4D and I am primarily interested in language
> understanding and generation.
MW: On the whole I find that language itself is neutral. I can talk
4D quite happily using NL. I think the origins of 3D lie in our primary
need to carry a "current state view" of the world around with us as
what most of our mental processing is directed at. I think we tend to
swop between that and some sort of 4D view for looking forwards and
back in time. I suspect we may need to develop something similar in
the long term. The 4D would be for memory and reasoning, the 3D for
action.
> I doubt that anyone will ever
> convince NLP types to use 4D, except for perhaps some special
> cases. I think that 4D has a more concise representation of
> roles than 3D, and it may be more computationally efficient
> when reasoning about the relations of members of an
> organization. It would be nice to be able, for example,
> to use 4D for role relations and convert back to 3D when
> we are concerned with rapid changes over time -- especially
> those that can be represented by mathematical functions.
MW: See my view above.
> I know you have concluded that 4D is also better for
> representing industrial processes. I haven't seen the
> details of how it would be used in such a case, and if
> I did I might find even more uses for 4D.
MW: I think it is generally closer to a scientific/engineering view
of the world.
> Another big hang-up is not being able to see *all* of
> a 4-D "object". I can walk around a 3-D object and get a
> good idea of what it all looks like, but I can't see the
> future parts of a 4-D object.
MW: No you don't walk round a 3D object, you walk round a time
slice of a 4D object :-)
> This is very distressing.
> You really need some powerful motivation to make us
> 3D-ers believe that the baseball in my hand is only
> part of an object that extends into the future. It sure
> **feels** like its all here. But . . . If there
> are interesting (and publicly available) systems that use 4D,
> it will get more attention. I'm in favor of any representation
> that makes my computations easier.
MW: The secret is to step outside time and see your baseball as
having a trajectory through time as well as space. Take a look
at the space-time maps in my paper (see above).
>
> So, as to the timing of when we tackle such issues:
> >
> >
> > MW: I have no doubt that all the sensible things that can
> be said in
> > a 3D ontology can be said in (and translated to) a 4D ontology. The
> > issues I forsee are more that there are times when 3D
> ontologies are
> > ambiguous but 4D ones are precise, and things that can be said
> > clearly in a 4D ontology, but in extreme cases a 3D ontology must
> > deny.
> >
> >> I agree the mapping of 3D to 4D is important. I suggest a
> >>thread on the relation of 3D Events to 4D activities. But I
> >>recommend that we postpone initiation of that thread until we
> >>determine if there is any kind of predominant view on
> >>**any** aspect of the 3D "Event", since the relationship will
> >>be critically dependent on the axiomatization of the 3D Event.
> >>If there is a specific axiomatization of 3D event that you
> >>think is not translatable into your view of 4D activity,
> >>it would be good to know that and we can try to avoid it.
> >
> >
> > MW: On the contrary it would be particularly good to explore it. If
> > we don't look at the difficult things we will get nowhere. But note
> > comments above. I am not expecting anything to be more difficult in
> > a 4D ontology - after all I moved to a 4D foundation
> because I found
> > it gave greater clarity and precision of meaning.
> >
>
> OK, but:
> (1) I know that Pat hayes has looked into 3D vs. 4D and I think
> that he has some mappings, but I don't know anything more than that.
> I would suggest we ask him what he has done before starting a public
> thread on the SUO list.
MW: I doubt if Pat has something formal, but like me since he has been
translating from everyday NL to 4D NL for some time, is confident it
can be done.
> (2) Since I think that some notion of a series of changes in
> attributes is common to all 3D accounts of "Event", I would like
> to know what is the closest analogous concept in 4D. I haven't
> found "a series of changes in attributes" in any of the 4D
> ontology discussions I have read (not many, I confess).
MW: Look at my paper above.
>
> 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
> =============================================
>
>