[ontolog] Scheduled Technical Discussion - Thu 2005-04-28    (C48)

Conference Call Details    (C49)

Attendees    (C8O)

Background and Resources    (C8V)

 Prepared by: NicolasRouquette / 2005.04.27; 
 dicussion content updated during session on 2005.04.28    (CA4)

Postground:    (CCI)

AdamPease pointed out to several incorrect statements made about SUMO, particularly with respect to the modularity of SUMO itself. The notes below should be read with the understanding that they may represent incorrect views about SUMO or about other ontologies. (NicolasRouquette)    (CCJ)

Background:    (C7D)

The popularity of the "semantic web" and furious pace of "ontology" development in particular have created a very exciting bubble of research and development activity. From a pragmatic perspective, bleeding edge research has the potential to simplify and improve application-specific ontology development. Bleeding-edge technology has a strong risk stigma associated to it. However, ontology-based development is perhaps a unique kind of bleeding-edge technology where the risk of using it has turned into a fuzzy, bleeding-edge concept of its own.    (C7E)

Goals:    (C7F)

Assemble a panel of practitioners and experts with a pulse the bleeding edge to review new developments and advances in the field that pertain to the adoption of recommended practices, particularly, modular ontology development and the role of annotation tags in this process.    (C7G)

Resources about modular ontologies:    (C7N)

- SUMO (AdamPease)    (CBM)

  http://www.ontologyportal.org/    (CCK)

- DOLCE    (C7O)

  http://www.loa-cnr.it/DOLCE.html    (C7P)

- PSL (Process Specification Language)    (C7Q)

  http://www.mel.nist.gov/psl/ontology.html    (C7R)

- openEHR    (C7S)

  http://www.openehr.org/repositories/archetype-dev/latest/index.html    (C7T)

- openGALEN ontology for Human anatomy    (C7U)

  http://www.opengalen.org/open/crm/crm-anatomy.html    (C7V)

- a repository of factored ontologies    (C7W)

  http://www.mindswap.org/2004/multipleOnt/FactoredOntologies/    (C7X)

Resources on tools supporting modular ontology development:    (C7Y)

- PROMPT    (C7Z)

  http://protege.cim3.net/cgi-bin/wiki.pl?ProtegePluginsLibraryByType#nid3UO    (C80)

- E-connections    (C81)

  http://www.mindswap.org/2004/multipleOnt/    (C82)

- FCA    (C83)

  http://protege.cim3.net/cgi-bin/wiki.pl?ProtegePluginsLibraryByType#nid3U6    (C84)

- Problem-Solving-Methods    (C85)

  http://protege.cim3.net/cgi-bin/wiki.pl?ProtegePluginsLibraryByType#nid3UW    (C86)

Agenda & Proceedings    (C8W)

1) Welcome & confirmation of agenda    (C8X)

2) Appointment of secretary to take minutes    (C8Z)

3) Roll-call of participants    (C92)

4) Announcements, Urgent Follow-ups and Special Items    (C94)

5) Discussion of Main Topic - "Ontologies and Tagging"    (C95)

7) Other business - none    (C9A)

8) Next meeting date and adjourn    (C9B)

Discussion:    (C9O)

(1) monolithic vs. factored ontology development    (C7H)

Developing a single, monolithic ontology for a particular application is both time consuming and is fast becoming both unwise and unecessary.    (C7I)

a) This is unwise because there are many factored ontologies available now, most of which are the product of rigorous but time consuming development.    (C7J)

b) Monolithic ontology development is unecessary if alternative modular ontology development approaches are practical    (C7K)

Are these sustainable arguments for practioners? Should the rationale be approached differently?    (C7L)

What ontologies are we talking about?    (CCW)

(PeterYim) Ask to clarify what kinds of ontologies is this discussion about.    (CCX)

(NicolasRouquette) Focus on OWL for various reasons:    (CCY)

- it is a popular standard that achieves a reasonable common ground to describe how different ontologies relate to one another    (CCZ)

- it was designed with a careful tradeoff between expressiveness and tractability    (CD0)

These reasons matter to the risks factors associated to ontology development.    (CD1)

- an ontology developed independently of any other and without any explicit relationship to any other ontology is not a reasonable proposition.    (CD2)

- but, building an ontology from other ontologies is not simple either, particularly in terms of explaining what concepts and relations from other ontologies are used in the context of the new ontology developed.    (CD3)

(PatCassidy,AdamPease,NicolasRouquette) The expressiveness of an ontology language    (CD4)

- KIF is highly expressive and there is certainly evidence it is practical    (CD5)

- Are the tradeoffs a specific ontology language makes (e.g., OWL) really necessary?    (CD6)

- What matters is to "use" an ontology language whose expressiveness does not get in the way of describing the important concepts & relationships that matter (on first order)    (CD7)

- Reasoning matters also (on second order), however, it is clearly a tool / implementation issue. There seems to be an unspoken consensus and agreement that description does and should take precedence over reasoning in the simple sense that there is no reasoning without representation.    (CD8)

- (NicolasRouquette) Alternatives to OWL mentioned:    (CD9)

-- The Handbook of Description Logic    (CDA)

   http://portal.acm.org/citation.cfm?id=885746    (CDB)

This book includes an appendix that describes several classes of description logics, including logics that feature "role value maps". Including such constructors yields a significant increase in the description logic's expressiveness. Intuitively, role value maps allow the logic to express constraints where an instance can be reached by two different paths, each path defined as a composition of a role whose results are applied to another role and so on. This handbook is a particularly significant resource because it includes not only a whole slew of complexity results but also very good explanations where this complexity comes from in terms of which language features a description logic.    (CDC)

So, if you ever wanted to know what "SHIQ" and "ACLN" really mean, this book is definitly the place to look. The authors even include tips on "pronunciation"!    (CDD)

-- There is a large class of formal languages outside the realm of description logics.    (CDE)

Notable examples include:    (CDF)

Maude (a rewriting logic)    (CDG)

   http://maude.cs.uiuc.edu/    (CDH)

SpecWare (a formal specification system based on abstract algebra)    (CDI)

   http://www.specware.org/    (CDJ)

(AdamPease) Focus on tradeoffs:    (CB3)

(PatrickCassidy) - OWL has tradeoffs (expressiveness vs. tractability of reasoning). It's almost certain that we will see more and more work in OWL.    (CB4)

(AdamPease) - unless we have evaluated these tradeoffs w.r.t. other languages, then we do not have an objective basis to compare the worthiness of OWL's tradeoffs.    (CB5)

(NicolasRouquette) - unless we have a set of application/domain benchmarks, we cannot objectively compare OWL's merits w.r.t. other ontology languages    (CB6)

(AdamPease) - OWL embodies a choice to limit expressiveness to content that can be handled efficiently for inference. Content which might result in slow computation is not expressible.    (CB7)

(NicolasRouquette) Dual-level formal ontology development    (CDK)

There is an inherent conflict between two different but related issues:    (CDL)

- It is important to leverage highly expressive languages, especially if such languages are particularly well suited to describing important aspects of a particular domain of interest.    (CDM)

The consequence of this proposition is that we have to confront issues that arise from having multiple ontologies, particularly from describing a thing of interest in terms of different descriptions that will not be perfectly aligned.    (CDN)

- It is important to capture some information about how different domains of interest relate to one another if for no other reason than to provide a rigorous basis for comparing different descriptions of the same concept in different ontology languages/formalisms    (CDO)

Example of SUMO & Differential Calculus    (CB8)

(PatrickCassidy) use procedural extensions to define the connection between, e.g., the notion of "partial derivative" in diff. calculus and the notion of "change" in its KIF definition in SUMO.    (CB9)

- what does this procedure compute?    (CBA)

c.1) KIF = there is a "translation" between measures.    (CBB)

a procedure operationalizes this "translation"    (CBC)

c.2) Diff. calculus = operations: partial differentials, integration,...    (CBD)

- Is OWL reasonable?    (CBE)

(PatrickCassidy) OWL does not have function terms.    (CBF)

OWL DL's datatype properties are not usable as function terms.    (CBG)

(AdamPease) OWL does not have a function.    (CBH)

- composite term denotes the concept of the of the united states government.    (CBI)

(government_fn united_states)    (CBJ)

- this is creating an instance on the fly.    (CBK)

d) "Modularity" = SUMO vs. DOLCE    (CBL)

- (AdamPease) SUMO is has 11 different modules explicitly defined and the dependencies among them are also explicitly identified    (CBP)

- (NicolasRouquette) DOLCE has some modularity with, e.g., DOLCE-Lite, TemporalRelations, SpatialRelations, Descriptions&Situations, Actions, etc    (CCL)

- (AdamPease,NicolasRouquette) We lack a precise definition of what "Modular" is as a predicate on an ontology. Until such a definition is available, it meaningless to postlate how SUMO and DOLCE differ in terms of modularity.    (CCO)

e) How do we compare modular ontologies?    (CBQ)

Assuming that SUMO and DOLCE are examplars of "modular" ontologies according to some well-defined criteria; i.e:    (CCP)

if "modular-ontology-p" is a boolean predicate and:    (CCQ)

  (modular-ontology-p SUMO) = t    (CCS)
  (modular-ontology-p DOLCE) = t    (CCR)

then, a reasonable second-order question would be to look for a useful criteria for comparing and contrasting their modular organization?    (CCM)

(AdamPease) one basis of comparison is "reified" vs. "defined" concepts    (CBR)

- reified concept    (CBS)

-- the concept has a name; e.g. "state"    (CBT)

- defined concept    (CBU)

-- the concept has a name    (CBV)

-- a formal definition    (CBW)

-- e.g. state = the set of values of the properties of that a thing has a specific time.    (CBX)

(NicolasRouquette) ontological commitment defined somewhere    (CBY)

(AdamPease) the meaning of the concept is precisely that of its defining axioms, no more, no less    (CBZ)

(PatrickCassidy) ok for "concepts" but for "instances", especially real-world instances, axioms are not sufficient.    (CC0)

Example: "the Eiffel tower" -- it is an instance (of the class of "towers")    (CC1)

f) example    (CC2)

"state" in SUMO vs. "state" in DOLCE    (CC3)

- are they the same "thing"?    (CC4)

(PatrickCassidy, NicolasRouquette)    (CC5)

Consensus:    (CC6)

Equivalence requires still logical equivalence along with a "precision" equivalence comparison function to make judgements where the two ontologies agree "up to a degree of precision"    (CCB)

- The subset where the two ontologies overlap and are consistent is difficult to describe    (CCE)

- (PatrickCassidy) That subset is, for practical purposes, too difficult to describe and won't be done. This leads to the conclusion that, for practical purposes, it would be better to avoid these issues of partial overal altogether and focus on developping a separate ontology (whethe it is modular or not) from scratch.    (CCF)

- (NicolasRouquette) This reinforces the point of this discussion -- developing an ontology from scratch is difficult, fraught with many dangers, etc... Therefore, making "modular" ontology development something practical and easier than "scratch" ontology development is a non-trivial problem.    (CCG)

This is the end of the scheduled discussion time. The discussion did not get much into issues of tagging as described below.    (CCH)

(2) tagging concepts with their ontological origin and interepretation    (C7M)

The pedigree of an application-specific ontology developed according to recommended best practices, guidelines and principles might have the following breakdown:    (C9P)

a) a big chunk of the ontology comes from direct reuse of third-party and legacy ontologies already factored,modularized and properly organized for ease of reuse    (C9R)

b) a small chunk of the ontology comes from specific development    (C9S)

c) another small chunk of the ontology consists of linkages that describe how the new ontology concepts and relations relate to various reused parts and how the different new and reused parts constitute a new hybrid ontology    (C9U)

What guidelines specifically address development concerns about:    (C9V)

- selecting and choosing which modular ontologies to start from (a)    (CA1)

- scoping what new modular development is required (b,c)    (CA2)

- assessing how well the ontology thus produced makes sense (c)    (CA3)

- characterizing what kinds of inferences an ontology provides as a reusable component (a) and produces as a new capability (c)    (C9X)

- metrics about the value obtained of the work required/involved to develop the new ontology vs. the usefulness of the inferences and reasoning capabilities the new ontology produces (a,b,c)    (C9Z)

Session Recording of this Discussion Session    (C9E)

 (Thanks to PeterYim & KurtConrad for getting the session recorded.  -ppy)    (C9G)