Re: [uos-convene] UOS Agenda and Logistical Details

[Aldo Gangemi <gangemi@xxxxxxx>]

Dear all, maybe the last chance to have a decent connection until next Thursday!

I provide here some thoughts at the generic level, since the technical level is not going to be the main focus of the summit anyway. I'll do my best to participate in the two sessions as agreed.

At 23:59 +0100 10-03-2006, Nicola Guarino wrote:

> >     There are eight invited panelists
> > and any decisions by the panelists needs to be taken as a group. 
> > At this point, it does not appear that a majority of the eight
> > panelists will agree that the "three formal upper ontologies"
> > should have some priority in Tuesday's meeting.  If five of the
> > panelists agree to Adam's Tuesday agenda, I would support those
> > changes.

Raising too many issues on which ontologies should be discussed and on the technical (un)feasibility of relating them is not the best idea for me. As a matter fact, the major (justified) complaint against the UO solution to interoperability is that none will (hopely) impose the best partition of the world in logical terms, and that UOers like discussions more than real world applications ;).

On the other hand, showing that there exist good practices, with practical applications, of UO creation and use, seems a reasonable argument.

> Well, concerning the other five UO
> besides OpenCyc, SUMO and DOLCE, let me clarify that:
>
> 1. DOLCE is rather different from BFO,
> and developed totally independently of it. Indeed, in our final
> WonderWeb Deliverable http://wonderweb.semanticweb.org/deliverables/documents/D18.pdf,
> we present BFO as *ONE* of the modules of the WonderWeb library of
> foundational ontologies, and we attempt a comparison with DOLCE, wich
> turned out to be difficult at that time since the BFO formalization
> work was still in progress. Since then, I know that BFO has evolved,
> but I am not aware of a stable formalized version. Despite this, there
> are certainly strong similarities between the two ontologies, and the
> two research groups have been pretty much in touch.
>
> 2. D&S is a specialization of DOLCE
> especially developed to account for "socially constructed"
> entities. It adds to DOLCE some high-level entities such as
> "Descriptions", and a number of middle-level entities useful
> for a wide range of applications. In addition, while working on
> D&S Aldo has introduced the notion of "Ontology Design
> Patterns", which a compact renderings of a set of interrelated
> notions useful for a certain class of applications.

These are two different issues.

--D&S

D&S is a constructivist ontology, meaning that it does not put restrictions on the type of things and relations that someone wants to postulate, either as a domain specification, or as an upper ontology. Postulated types and relations are said to be the "ground vocabulary".

On the contrary, D&S provides means to "redescribe" things and relations from another viewpoint.

This approach has cognitive foundations like the RR (Representational Redescription) theory by Annette Karmiloff-Smith, Nozick's "invariances", schemas from cognitive semantics, etc.

From the logical viewpoint, D&S takes seriously the task of providing a vocabulary for logical reification, e.g. a description can be seen as the reification of an (even polymorphic) intensional relation, a situation can be seen as the reification of an extensional relationship, a concept can be seen as the reification of an intensional class, a collection can be seen as the reification of an extensional class, and so on (reification of intensional vs extensional entities must be interpreted as in Galton's type- vs. token-reification).

From the traditional upper ontology talk, D&S quantifies on social entities: methods, plans, norms, situations, cases, organizations, collectives, tasks, roles, parameters, etc.

The current OWL encoding of D&S assumes DOLCE as a ground top-level vocabulary (DOLCE seems well suited to accomodate for the large reification vocabulary of D&S, see Masolo et al. KR2004 paper on social descriptions), and in this form D&S has been applied to several ontology projects.

--Content Ontology Design Patterns (Codeps)

Codeps are recently proposed creatures (there is a paper of mine at ISWC 2005, but some preliminary material has circulated for two years), which provide a framework to annotate  "focused" fragments of a reference ontology.

A focused fragment is a part of an ontology that only contains the types and relations that underly "expert reasoning". Notably, Codeps can be very general, like the *Participation* pattern in DOLCE; at the core level, like the *Norm<->Case* pattern in legal ontologies like CLO (an extension of DOLCE+DnS); or very specific, like a certain protein interaction pattern in a bio ontology.

Codeps can be specialized, composed, etc., but they remain dependent on the original reference ontologies for full blown reasoning.

Codeps can be partly related to other ideas and theories, like Peter Clark's "knowledge patterns", data modelling patterns, and the OWL design patterns proposed by the SWBPD W3C working group. They will be exploited in the newly started NeOn EU project: http://www.neon-project.org, which also addresses the issues that I mention herewith.

Concerning some of the communication aspects discussed in these days, I strongly favour a communique that makes tolerant claims concerning UO.

In my opinion, provided their suitability to a certain task, and their structural/formal well-formedness (at least to a certain degree), more than one UO can be more or less equally exploitable.

Differences will be made by varied aspects: user feedback, availability of use cases driven by them, accessibility in tractable languages, etc.

I see several ways of relating ontologies.

At a macro-level, two ontologies can be both valid for a certain task if they can act as references for the relevant patterns of expertise (similar Codeps in my terms), if they provide the structure to draw certain inferences or to answer certain queries, etc. etc. At the macro-level, ontologies can be related across their focused fragments.

At a micro-level, two ontologies can only be related if their fine axiomatic granularity is taken into account. The micro-level can be addressed within different time frames from those of the macro-level.

In other words, relating ontologies can be also undertaken as the pragmatic issue of having selection criteria for them, which paramtrize their patterns or "hubs" of types and relations, e.g. by matching them to user needs.

Sorry for the many issues raised into one email, but that's my two cents by now, on which I can tell something next week.

Good luck

Aldo

-- 

Aldo Gangemi
Research Scientist
Laboratory for Applied Ontology
Institute for Cognitive Sciences and Technology
National Research Council (ISTC-CNR)
Via Nomentana 56, 00161, Roma, Italy
Tel: +390644161535
Fax: +390644161513
aldo.gangemi@xxxxxxxxxxx
http://www.istc.cnr.it/createhtml.php?nbr=71