Cory,
I realized I didn’t address your point on modularity in my response – but had intended to.
I agree that a focus on use (and reuse) of ontologies, and their sustainment would improve modularity in such ontologies. I believe that there are at least two significant dimensions of variability that would inform desirable modularity. One dimension is variability across the scope of intended application contexts, especially those involving binary variability in which some entity or property is present or absent in different portions of the target context/purpose scope (as opposed to, say, differences in degree of detail or value/namespace ranges, although these also may suggest some types of modularity). The second significant dimension is that of variability over time, i.e., the anticipatable or likely evolutionary trajectory within the target domain scope. The latter is, of course, more difficult/risky, but I can give examples from real-world system engineering and architecture decisions where design teams ignored some pretty obvious trends, to their detriment. Since one purpose of modularity is to allow an artifact to adapt to future changes without requiring a complete redesign, determining which portions of an ontology are like to change at different rates will suggest ways to modularize so as to minimize coupling among those portions – or, alternatively, to make that coupling more dynamic, which would, I believe, require more explicit representation of that coupling in the ontology itself.
Hans
From: ontology-summit-bounces@xxxxxxxxxxxxxxxx [mailto:ontology-summit-bounces@xxxxxxxxxxxxxxxx] On Behalf Of Hans Polzer
Sent: Tuesday, January 28, 2014 7:36 PM
To: 'Ontology Summit 2014 discussion'
Subject: Re: [ontology-summit] [Reusable Content] Characterizing or measuring reuse
Ed, Cory:
These are all good points, but it isn’t just trust in supplier’s content and quality – it is also the issue of appropriateness of the supplier’s content and quality for the would-be user’s application scope and purpose. Arguably, “quality” and “content” may include attributes which characterize the scope of the application and possible purposes it was intended for, but in my experience they fall short in this area. The examples of successful reuse that Cory points out are the result of a business model on the part of the supplier that considers the needs of a broad range of potential users – broad enough to justify (over time) the investment in development required to consider that range of requirements, and to support the sustained relationship and evolution of the product over a timeframe that meets the expectations of potential users related to their application/use of the product. This usually requires some venture capital and/or a somewhat obsessed individual with sufficient “vision” to make the ROI case that the up-front investment in time and effort is justified, and then sufficient sustained effort to reach potential users and convince them that a long term relationship makes mutual sense (and then deliver on that relationship).
As I mentioned in my other post on this topic, developing some artifact for a range of possible institutional and environmental contexts and some range of application purposes and scope requires one to be clear about what the salient attributes and attribute values associated with those ranges are. It may well be OK to make something with limited institutional/domain application scope, or to make something with very broad scope, but one should be explicit about what that scope is. Otherwise you run the risk of “false advertising” (as in “my ontology is very flexible and extensible”), and breeching any trust relationship you might have established with prospective users. A key issue I see on this forum (as well as in other contexts), is that people are often unaware of the implicit scope assumptions they make. This is not a criticism – it is human nature. But it is exacerbated by a tendency to want to make artifacts that are scope and context independent – and thereby obviating the need for messy/imprecise/vague specifications of domain, institutional, and environmental context scope boundaries relevant to the artifact in question. And that puts most of the burden on the potential users of the artifact to figure out whether the artifact in question will actually work in their context for their purposes. Some of that will always be unavoidable, and it is always prudent to “trust but verify”. But creators of artifacts intended for broad use (or “reuse”) would be well advised to be more explicit and expansive about the context scope ranges over which their artifact is intended to and verified to apply/work as advertised.
Hans Polzer
Ed,
I would add to your list:
· The would-be user trusts the supplier’s content and quality
· The would-be user trusts that the resource will be maintained and always available
There is a lot of “reuse failed” talk, but I would suggest that what failed is “casual reuse”. There is tremendous reuse of software libraries – that are of high quality and well supported by an organization or community. There is fantastic reuse of applications. There is good reuse of languages. There is some reuse of established Ontologies but rather poor reuse of some other kinds of models, e.g. UML. For anything to be reused substantially takes effort, commitment and marketing on the part of the creator. It also requires a business model for sustainment.
Most models/ontologies start either from a blank page or reverse engineered from technology specifications. There is very poor concept reuse and this tendency to re-invent causes semantic islands and unnecessary redundancy. Perhaps this is more due to the trust and content issues than representations. Certainly that models/ontologies are not modular creates a problem. Perhaps that is a problem because the commitment to design for reuse and sustained support has not been the focus of modelers.
-Cory Casanave
Andrea,
There is a big difference between ‘opportunities for reuse’ and ‘reuse’. Reuse is a behavior. The so-called “opportunities” are just resources that could be reused, but of themselves they are just available resources. A wealth of information doesn’t have a clear relationship to the use of that information, especially since some of it contradicts others. But I think your last question is the key one:
> Or, is content present but it is just very difficult to use/re-use?
Reuse is dependent on all of the following:
- the content is present (on the Internet/Web)
- the would-be re-user knows that the content is present, i.e., can find it
- the would-be re-user is motivated to find and examine the content
- the content is in a form suitable for the planned re-use, or can be “readily” converted to a useful form
- the would-be re-user knows how to convert the form, if necessary
- the content is consistent with the micro-theory adopted by the re-user
- the re-user is able to determine that the content is consistent with his/her theory
That all of these factors must be present makes it the nature of the beast that content is difficult to reuse.
What we tend to see is
- “reuse by direction”: Use this so that your model will be consistent with ours/hers.
- “reuse by social pressure”: I use the BFO because I know a lot of knowledge engineers who use/like it.
And of course, both of these lead to the Catch-22 problem: the content has to have been reused in order to be reused.
Murray Burke (DAML) once commented that upper ontologies would be reused because most knowledge engineers would be too lazy to roll their own. Experience suggests that the latter is true, but the former is not necessarily a consequence.
In sum, I think reuse is both a technical problem and a social problem. (And if I knew how to overcome either, I would be rich, or something. J)
-Ed
Other important questions in the "reusable content" arena are how to ascertain and improve the amount of reuse.
It "seems" that reuse is low, but there are many sites offering reusable content and therefore many opportunities for reuse. For example, in the Ontology Design Pattern (ODP) space, there are:
- W3C'S Ontology Engineering and Patterns Task Force (OEP) [1]
- Ontology Design Patterns org wiki [2]
In addition, there are foundational ontologies available, as discussed in the Upper Ontology Summit (2006) [4], as well as domain ontologies like FIBO.
So, does the wealth of information contradict the perception?
Or, is content present but it is just very difficult to use/re-use?
Perhaps we need to refine our engineering approaches and abilities to better find and evaluate reusable content? This is discussed in a paper by María Poveda-Villalón, Mari Carmen Suárez-Figueroa and Asunción Gómez-Pérez [5] that I found quite interesting.
I personally would love to see a review and recommendation system put in place for ontologies, patterns, linked data models, etc. Is this something that we could achieve?