RE: [ontac-forum] Future directions for ontologies and terminologies

["Cassidy, Patrick J." <pcassidy@xxxxxxxxx>]

Dr. Russel,

 

  Thank you for joining into the discussion.  It is very good to have experts in uncertain reasoning participating in our working group.  I believe that ultimately such reasoning will be essential to solve the more complex reasoning tasks that we will want the computers to do properly.

 

Since this is a working group, rather than a general interest group, the method I have recommended for making progress is to create and use computational artifacts - e.g., ontologies  -- and resolve the questions of optimal structure by observing the performance of alternatives in test applications.   We have by now a number of very competent examples of ontological practice from which to take pieces and reuse them for our purposes.  Among the upper ontologies, OpenCyc, SUMO, and DOLCE are well-known, and the BFO and ISO15926 have additional useful structure from which to learn.  General questions of the type you have advanced may be best resolved by examining specific examples to see how alternatives could be used (either by "thought experiments" or by actual pilot programs).  These ontologies can themselves be used as is for such experiments.  For resolving which upper ontology to adopt, there are a couple of proposals under considerations within our COSMO-WG.  See:

            http://colab.cim3.net/cgi-bin/wiki.pl?CosmoWG

 

 

If you have any ideas about how your preferred techniques can be implemented in concrete test cases that this volunteer group may be able to investigate, please make alternative suggestions.  There is plenty of room for more ideas, but with our very limited time, it will be necessary to be as immediate and specific as possible.

 

 Regarding your note, a few specific points:

(1)     The dichotomy you present for an "upper ontology" is not clear to me from your note:

 

[LR]  One of the decisions one needs to make about an upper Ontology is whether it is needed at all.  If the answer is “yes” then the choice is to create a single upper Ontology that categorizes all human knowledge into a small number of categories. This is by definition static. The other alternative is to have a large number of categories which can be combined on the fly; this is by definition dynamic. This decision entails recognizing the form it would take.  

Could you illustrate what you mean by combining categories "on the fly"?   I think that we all agree that any common upper ontology will have to grow, evolve, and be supplemented over time;  Does this nevertheless qualify as "static"?  Has anyone actually recommended a "small number of categories" - is this a serious proposal?  How small?   It sounds like Schank's "Conceptual Dependency Theory" from the early 70's.

(b) An upper ontology is **needed** to efficiently relate terms or concepts in multiple knowledge classification systems to each other.  In other applications it may still be very useful, but not essential.

(2) [LR] First off, let me recommend to those who know of WordNet but have never read the book to do so

   Good recommendation.  I have followed the progress of WordNet since it was first released and watched it evolve from the barest of semantic networks, without definitions, into a sophisticated lexical resource with increasing numbers of  semantic relations and definitions whose words have been disambiguated.  I greatly admire that work, and in fact the major upper ontologies have had mappings created from their concepts into WordNet.  But WordNet isn't an ontology in the sense of having semantic relations whose meanings are specified by axioms describing the inferences associated with each semantic link - at least not yet.  It is a very useful lexical resource.  It doesn't substitute for an upper ontology.

The WordNet has been called an "ontology".  Within the ONTACWG we are free to be more specific about what we mean by the terms we use.  We do not now have an official "glossary", but since we are dealing with multiple knowledge classification systems, it is useful to distinguish "semantic networks" which have terms related by other terms, from "ontologies" in which the meanings of the relations are specified by some logical formalism that permits inferencing beyond the mere existence of a labeled link.  WordNet is still at this time a semantic network -- perhaps the most sophisticated one in existence -- and may yet evolve into an ontology, by this definition.

(3) [LR] 1. Should a class name in an Ontology, especially an upper level class, be restricted to a word or phrase?

Since the ontologies we are discussing are intended for use in computers, as far as the computer inferences are concerned, the names of the classes are of no consequence whatever.  The computer programs that use ontologies (other than ontology-mapping programs) do not attempt to interpret the strings used as class labels - those are mostly for the computer.  But if the ontology is to be used to specify the meaning of a terminology or vocabulary, then there must be mappings from the ontological entities -- classes, relations, functions, instances -- to the terms in the vocabulary.  This can involve a very straightforward relation from concept to word(s).  The correspondences will in general be many-to-many: several words may refer to the same concept (synonymy), and several concepts may be referred to by the same word (less often, by the same phrase) -- i.e., ambiguity.   But since ontologies are created by and used by people, in order to preserve sanity and avoid wasting time, the concepts are best labeled by strings of characters that a person can recognize, which indicate the intended meaning of the concept.  If we are labeling a concept that is usually referred to by a word that is in fact ambiguous, it is good practice to use that word, but with a qualifier that disambiguates the sense.  This practice is used in thesauri as well as in ontologies.  The method for adding a qualifier is solely a matter of taste, as long as the term is unique in any given namespace.  In OpenCyc we see examples like "Tree-thePlant" and "GeometricThing-Abstract".  In other ontologies one may see "AbstractObject" versus "PhysicalObject".  Because of ambiguity, in large ontologies one word is often not a good enough label to provide a good idea of the intended meaning of a concept.  Long compound names for variables in computer programs serve much the same purpose.

Since we anticipate that an ontology may be used for applications in more than one language, it is clear that there must be a mechanism to associate any number of terms, ideally in specified contexts, with the logically-specified concepts in the ontology.

(4) A further “yes” argument is that if a concept in the Ontology is not lexicalized then we have to create a name for that concept’s class. But to name it is not to claim it: we cannot claim that our work is done just because we have stung two or more words together. Lacking a linguistic history of our new phase we need to either (1) enumerate the subclasses (in set theory an extensional description) or (2) define a set of inclusion rules (in set theory an intensional description).

All of these considerations have been discussed in works on ontology, and are echoed in the structure of the existing ontologies we have on hand as examples of accepted practice.  If you think that the way these issues are handled in SUMO, OpenCyc, or DOLCE are not adequate, it will be very interesting to learn what other techniques would serve better.

If there are any publicly available examples of the use of uncertain reasoning that we can access and experiment with, I hope you can point them out to us.  Although our work on selecting a Common Semantic Model is only starting, having such an example should be very instructive in helping us to be sure that any ontology structure we choose will be able to smoothly accommodate uncertainty. 

Pat

Patrick Cassidy
MITRE Corporation
260 Industrial Way
Eatontown, NJ 07724
Mail Stop: MNJE
Phone: 732-578-6340
Cell: 908-565-4053
Fax: 732-578-6012
Email: pcassidy@xxxxxxxxx

 

---

From: ontac-forum-bounces@xxxxxxxxxxxxxx [mailto:ontac-forum-bounces@xxxxxxxxxxxxxx] On Behalf Of Lucian_Russell/ESI/CSC
Sent: Friday, December 30, 2005 9:55 AM
To: ONTAC-WG General Discussion
Subject: Re: [ontac-forum] Future directions for ontologies and terminologies

Static vs. Dynamic Queries

Hello from a former lurker. My name is Lucian Russell and my specialty is applying the principles of uncertain reasoning to data, all kinds. I have been in the computer science field for several decades.

One of the decisions one needs to make about an upper Ontology is whether it is needed at all.  If the answer is “yes” then the choice is to create a single upper Ontology that categorizes all human knowledge into a small number of categories. This is by definition static. The other alternative is to have a large number of categories which can be combined on the fly; this is by definition dynamic. This decision entails recognizing the form it would take.  

This initial submission will lay out one issue: the need or not for concept lexicalization for the highest level of a static Ontology. As John Sowa cited the philosopher Aristotle’s timeless insight, I will cite the timeless insight of Gloria Estphan who in the 1980s sang the hit “The Words Get in the Way”.

First off, let me recommend to those who know of WordNet but have never read the book to do so, especially Chapter 1 (nouns) and Chapter 3 (verbs); subsequently read Chapter 2 (adjectives and adverbs) and re-read the two surrounding chapters. To those who have read the book, congratulations for being so well organized and prescient. For those not following WordNet it is up to Release 2.1 and is completing a mapping of words in explanations to senses in WordNet. This latter work will be finished up in April. The current release is available for downloading; it is also on-line for any single word.

Prof. Christiane Fellbaum is the current Principal Investigator (PI) on the WordNet project (Prof. George Miller who started WordNet is old enough that I had him as an instructor for a General Education course while an undergraduate, so is no longer working full time).  She makes the following distinction between WordNet and other Ontologies. WordNet is a lexical ontology. That means that it is based on words in the English Language, though some noun phrases are also included (the words are separated by the underscore character). Although this is generally true, even WordNet contains some non-lexicalized nodes, which brings me to the key question:

1. Should a class name in an Ontology, especially an upper level class, be restricted to a word or phrase?

It would seem that there is something to be said on both sides of this issue.

On the “no” side, for example, WordNet shows that in English verbs there are two senses of the verb “move”. One is movement with a translation in space and one is movement without this change in special position (e.g. spinning). The two senses are not lexicalized but are rather distinct senses of a word with a given spelling (homograph).

On the “yes side” if for a static Ontology there is a single word (or a well known noun phrase) that is used for the name of a class then we have a sound basis for looking at sub-classes and the class membership of instances. In this case we say that the concept is lexicalized.  

A further “yes” argument is that if a concept in the Ontology is not lexicalized then we have to create a name for that concept’s class. But to name it is not to claim it: we cannot claim that our work is done just because we have stung two or more words together. Lacking a linguistic history of our new phase we need to either (1) enumerate the subclasses (in set theory an extensional description) or (2) define a set of inclusion rules (in set theory an intensional description).

Of course we could lexicalize the concept by taking a word from another language that does lexicalize the concept, but if the Ontology is meant for English speakers that new word still needs to be explained to them (the same is true for any language whose words are used to name concepts).

In practice the class membership criteria for instances of an Ontology are specified by a set of constraint rules on property values. Using Descriptive Logic one must be able to reason about the assertion “if X is a class and x in an instance of a thing with property values (a1, a2 …..an) then x is a member of the class X “. The answers are “Yes” and “No”. So for an Ontology a non-lexical class name is really a substitute for a mathematical description.

Static Ontologies, therefore, should probably be ones that primarily use lexicalized concepts in classes. That being the case, it would seem that the logical consequence is that for the top Ontological categories one should use WordNet’s “beginning names” unless there was a REALLY good reason not to do so.

Some researchers in Artificial Intelligence claim they have a better top level Ontology, and maybe they are justified. However, WordNet has some 15 years behind it, so the set-descriptions of the any different concepts should be carefully spelled out so that we may judge.

In the end we can choose what we want; we just should be sure of why we made the choices and know the consequences.

If I have repeated well worn arguments, I will take my deserved newbie lumps and be suitable chastened :-)

Lucian Russell, PhD
___________________________
Lucian Russell, PhD
Sr. Architect - Principal
Federal Consulting Practice
Computer Sciences Corporation
3110 Fairview Park Drive
Falls Church VA, 22042
Ph: (703)-645-5237
Fax: (703)-645-5233
Cell:(703)-340-5807


----------------------------------------------------------------------------------------
This is a PRIVATE message. If you are not the intended recipient, please delete without copying and kindly advise us by e-mail of the mistake in delivery. NOTE: Regardless of content, this e-mail shall not operate to bind CSC to any order or other contract unless pursuant to explicit written agreement or government initiative expressly permitting the use of e-mail for such purpose.
----------------------------------------------------------------------------------------

_________________________________________________________________
Message Archives: http://colab.cim3.net/forum/ontac-forum/
To Post: mailto:ontac-forum@xxxxxxxxxxxxxx
Subscribe/Unsubscribe/Config: 
http://colab.cim3.net/mailman/listinfo/ontac-forum/
Shared Files: http://colab.cim3.net/file/work/SICoP/ontac/
Community Wiki: 
http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/OntologyTaxonomyCoordinatingWG    (01)