Andrea,
Thanks for the thoughtful reply.
I will extract the main issues for discussion:
[AW] > Please refer to Kingsley's post (http://ontolog.cim3.net/forum/ontology-summit/2014-02/msg00117.html). It gives a very good explanation.
For the case of finding all instances of any given equivalent class in two or more ontologies , that would work just fine; this application uses only the ‘instance’ (and perhaps ‘subclass’)relation. My browser was unable to view the examples referenced in Kingsley’s note, so I can’t see how broadly applicable this tactic is.
[AW] > An upper ontology could be used to help create the mappings, or serve as the mediation point for the mappings. But, in my experience, the current set of upper ontologies are too generic to be useful in any specific application domain. And, at the end of the day, you want detailed equivalencies and axiomatic definitions in your integrating ontologies, not relationships to generic concepts.
Of course, no upper ontology will have all the *domain* concepts one uses in an application, but the point is to use a ‘foundation ontology’ (I prefer that term) that has **all** of the fundamental (‘primitive’) concept representations that are required to logically specify the meanings of all of the domain concepts in the communicating domain ontologies. For any given set of ontology-based applications there will necessarily be **some** finite set of primitive concepts that are sufficient to describe all of the concepts. As new applications need to communicate, it may be necessary to add new primitives, but these should not “break” the earlier applications, since the legacy systems don’t care about (i.e. use) those new primitives. As a practical matter, my initial goal is to try to identify all of the primitives that anyone may need, so that a broad range of applications can communicate accurately, and the principle can be thus demonstrated. It is likely that new primitives will be required over time, but my experience thus far suggest that they will not be so numerous as to make the tactic unworkable. In any case, it is thus far the *best* hope I have been able to identify for *accurate* automated computer communication of complex concepts among independently developed systems.
[AW] > Also, the intent of an under-specified construct (Location details in my "Person has a Location" example) is not to fully define it - but to leave it open to reuse and clarification by other modules of reusable content (whose "micro-theories" align with the needs of the re-user).
The foundation ontology should have the capability to specify concepts as broadly or narrowly as one wishes.
[AW]> Is the intent to redefine domain concepts from the ground up, in terms of the COSMO primitive concepts? That is a huge undertaking.
Yes, that appears to be necessary for any accurate communication. This presents a “triage” scenario, once a foundation ontology has been agreed to by multiple groups:
(1) Those newly building their database or ontology use the foundation ontology – this will actually save time, since a lot of well-defined basic needed concepts will already be available. (after a few years the majority of applications will be new)
(2) For legacy applications that need to communicate, where the communication task is important enough to justify the cost, each communicating database or ontology will need to be specified in terms of the foundation ontology
(3) For legacy systems that can’t bear that cost, they will communicate less effectively with alternatives such as creating equivalence relations and carefully watching test cases to be sure that doesn’t generate nonsense – or at least, too much nonsense.
[AW] > . BTW, I am a fan of using the WordNet synsets and relationships to help to elucidate semantics. But, even WordNet needs extensions to its lexicon to be close to useful in a specific domain
Yes, specialized domains will require vocabulary additions. There is a lot of good semantic structure in WordNet, but it is sketchy in the relations, and many synsets are not homogenous unique concepts, and other different synsets may represent indistinguishable concepts (i.e. are essentially synonymous, except in some cases for part-of-speech). I am creating mappings to WordNet to allow mapping and perhaps reuse of much work already done using WordNet in the Natural Language community, but do not believe WN itself is useful as is for reasoning. (It was not originally created for that purpose). It just happens to be the best free comprehensive semantic network with which much text has already been tagged.
[PC} >> The great virtue of using an interlingua ontology and translations to and from, for interoperability is that this allows local users of databases or ontology-based applications to use their own terminology and procedures freely, and only translate what needs to be communicated. So that in any user group, no matter how large, only one person needs to be bilingual in both the local terminology and the common foundation ontology.
>>
[AW] > That is also the goal of ISO 15926 and the iRING tools. Unfortunately, we then get into debates over whose upper ontology is more useful.
Yes, and that depends on the use. I like Cyc, but much of the essential structure of the top level is still proprietary, and I need one available for free public use. In any case, none of the existing upper ontologies is designed to have a primitives-based conceptual defining vocabulary, which is the tactic that I think is optimal for interoperability. In COSMO I use as much of OpenCyc and SUMO as I can , and add what else seems needed for the purpose. Other ontologies have little not already in OpenCyc and SUMO, but I use a few concepts from some other ontologies as well. At some point, I will want to create mappings to the upper levels of the OBO group – I know where the BFO concepts belong in COSMO, but have not created a comprehensive formal mapping yet.
[PC] >> This may seem somewhat large to be “the most easy to learn and use”, but one needs to realize that the function of being able to describe any concept one wants to use in an application requires a vocabulary that has the full range of the basic vocabulary of a human language.
[AW] > Don't we essentially come back to natural language then?
At the earliest (e.g. current) stage, agreement on the meanings of the entities in the ontology will need careful documentation (I use English) to minimize the chances of ambiguity and misunderstanding between independent users. Ultimately (long term goal) the meanings will be grounded in perception and action, which may need some robotic functionality. Some grounding in reality already is available by access to the web, which can be used to verify or negate inferences that are uncertain, if some language understanding is part of the system. But I am particularly intrigued by the possibility of “grounding” the meanings with virtual reality, so that the computer can present a virtual reality scenario to explain its interpretation of texts. A VR interpretation verified by a person may substitute for actual robotic sensors and effectors to give the computer very good approximations to human understanding.
But, no doubt, there will be some effort necessary to create the logical specifications. From what I have seen, the effort needed will be a lot less than already being expended to achieve less useful results, as far as automated interpretation of communications is concerned.
Pat
Patrick Cassidy
MICRA Inc.
cassidy@xxxxxxxxx
1-908-561-3416
From: ontology-summit-bounces@xxxxxxxxxxxxxxxx [mailto:ontology-summit-bounces@xxxxxxxxxxxxxxxx] On Behalf Of Andrea Westerinen
Sent: Friday, February 07, 2014 1:46 PM
To: Ontology Summit 2014 discussion
Subject: Re: [ontology-summit] Ontolog Ontology of Ontologies
Patrick, Inline ...
On Thu, Feb 6, 2014 at 7:28 PM, Patrick Cassidy <pat@xxxxxxxxx> wrote:
Andrea,
I’m not sure what you had in mind for an “integrating ontology”,
but for the purpose of interoperability, among which is the function of relating ontologies to each other, I expect that that purpose would be served optimally by a foundation ontology (“upper ontology”) that has all of the logical structures necessary to create logical specifications of all of the terms in all of the ontologies that want to interoperate.
An upper ontology could be used to help create the mappings, or serve as the mediation point for the mappings. But, in my experience, the current set of upper ontologies are too generic to be useful in any specific application domain. And, at the end of the day, you want detailed equivalencies and axiomatic definitions in your integrating ontologies, not relationships to generic concepts.
Also, the intent of an under-specified construct (Location details in my "Person has a Location" example) is not to fully define it - but to leave it open to reuse and clarification by other modules of reusable content (whose "micro-theories" align with the needs of the re-user).
Such a foundation ontology would serve as an “interlingua” into which any other ontology (or data base of data structure) could be translated, and subsequently re-translated into any other ontology whose terms are logically specified using the terms and structures of the foundation ontology.
I see that is a goal of COSMO (from the OntologWiki page [1]) - to "expand it until it includes representations of all of the "defining vocabulary" words in the Longman Dictionary of Contemporary English (LDOCE)". However, even with COSMO in place, there would still need to be tooling and/or human intervention to map to it.
There are a lot of Ifs here (and a lot of work to be done):
1. If all the necessary terms and axioms are in place in COSMO (since this is more than just taking terms from a dictionary)
2. If reusable semantic content is defined using those terms, or with a mapping to them
3. If tooling exists to exploit this information for search and mapping
Then mapping and reuse would be much improved.
That is the intended function of the COSMO ontology which I am building. The COSMO is intended to include representations of at least the full set of primitive concepts that are needed to specify the intended meaning of any elements of a domain ontology that needs to integrate with others. Focus on the primitives keeps the foundation ontology to the smallest possible size that will serve its function; and that small size makes it easier to learn and use. The COSMO is presently in OWL, and needs to be translated into FOL and supplemented with additional axioms.
Is the intent to redefine domain concepts from the ground up, in terms of the COSMO primitive concepts? That is a huge undertaking.
I am still creating mappings from elements to WordNet synsets of similar meaning, and to the LDOCE defining vocabulary to make it useful for Natural Language applications. The current version of the ontology (“COSMO.owl”) and some documents describing it can be obtained from the directory at:
http://micra.com/COSMO
I will take a look. BTW, I am a fan of using the WordNet synsets and relationships to help to elucidate semantics. But, even WordNet needs extensions to its lexicon to be close to useful in a specific domain.
The great virtue of using an interlingua ontology and translations to and from, for interoperability is that this allows local users of databases or ontology-based applications to use their own terminology and procedures freely, and only translate what needs to be communicated. So that in any user group, no matter how large, only one person needs to be bilingual in both the local terminology and the common foundation ontology.
That is also the goal of ISO 15926 and the iRING tools. Unfortunately, we then get into debates over whose upper ontology is more useful.
The first versions of COSMO is intended to include at least all representations of the senses used for the 2200 words used in the Longman dictionary to define the terms in that dictionary. This is considered (by me) as a reasonable first approximation to a full set of primitives. There are more than one sense used for a number of those words, and to avoid forcing users to define common complex terms, some non-primitive terms are included. So that the current version contains almost 8000 classes (types) and 1000 relations.
This may seem somewhat large to be “the most easy to learn and use”, but one needs to realize that the function of being able to describe any concept one wants to use in an application requires a vocabulary that has the full range of the basic vocabulary of a human language.
Don't we essentially come back to natural language then?
Learning this is eminently feasible when a person is properly motivated, as one can easily see from the number of non-native English speakers who learn English so as to be able to communicate within any of many international communities (including ontolog). And remember, only one person per user group needs to be adequately bilingual.
Pat
What would an "integrating" ontology actually look like? I mean basically we're naming ontology design patterns here, correct? With the idea of aligning said patterns? In this case, I'm hearing an "integrating" ontology is composed of nothing but owl:imports, does not define its own classes, properties or individuals? Whether that's "integrating" may be in the eye of the beholder but what is certain is that it's totally derived from other ontologies, a useful factoid when trying to sort one's way through a thicket.
thx/jmc
On 2/5/2014 2:57 PM, Andrea Westerinen wrote:
John and John,
In as much as the "integrating" ontology is an "ontology about ontologies (and their concepts)" ... yes, it is a meta-ontology. However, I find that people handle the term, "integrating ontology", much better. And, it conveys the purpose (as opposed to being open-ended).
Given that some of my customers view ontologies with some skepticism (they call ontology, the "o-word"), I much prefer avoiding adding scary modifiers onto a scary word. (That is just my experience.)
--