Ontolog Forum

Purpose This track examines the role that upper ontologies play for specifying contexts.

Here is a deck of slides that set out the main issues and findings.

The slides summarize the activities of this track, as follows:

  • business motivations: Taking implicit contexts of data models and making them explicit in an ontology
  • Session presentations and insights
  • Upper Ontologies gallery
  • Issues arising: Treatments of context
  • Context of ontology versus ontology of context - considerations for the overall context in which an ontology is seen.

Upper Ontologies and Context

We start with the premise that one role for a kind of ontology is to take contexts that are implicit and make them explicit. That is, we consider the contexts implicit in some stand-alone data source, typically defined by the application context or use case for which that data resource was created. For example, a company such as a bank may have multiple databases for different customer or product relationships, supplier relationships, personnel data and so on. In these examples, the implicit contexts may be the customer relationships, product types or instances and so on.

Interoperability among data resources, or the reusability of data, can only be made possible if these implicit contextual matters are dealt with. In the above examples, the contexts are roles, relationships or products. Other kinds of contextual matter may run to the whole range of ‘The Ws’: Who What When Why hoW. The kind or use of ontology considered here is the ontology as concept model; a common language across the enterprise. In order to enable interoperability among applications or re-use of data across the enterprise, what were implicit contexts for each set of data must become explicit ontological classes within the ontology.

We consider the ontological modeling of kinds of context, or concepts that may be considered to be kinds of context. That is, the representation of context. For this track, we considered the possible role of upper ontologies or top level ontologies (TLOs), and whether or not these may be used to provide a set of common organizing theories (partitions) of the overall concept ontology, for kinds of context. Would different TLOs provide kinds of partition that were explicitly identifiable as kinds of context, or would some (or all) TLOs provide partitions for different kinds of context, for example the Ws itemized above.

We consider three possible approaches to this question of the representation of context, which may or may not prove to be convergent or equivalent. These are:

1. Context as Class: is there some specific upper ontology category that is identifiable as an overarching category of ‘Context’?

2. Kinds of Context: can we identify specific high level categories of contextual matter for the Who What When Why and hoW?

3. Everything as Context: for any element in some ontology, is it not simply the case that the contextually defined ‘meaning’ or semantics of that class is the sum of all the other classes to which it is related?

In terms of (1) ‘Context as Class’ we considered one set of upper ontology partitions that represent Peircean Firstness, Secondness and Thirdness, as exemplified in one of the upper sets of mutually exclusive partitions in the KR Lattice upper ontology. In that view, any ‘Secondness’ category of thing is that which brings two or more things together in some context, that context being the ‘Thirdness’ category. In the earlier examples, the categories of customer, supplier etc. exist in the contexts of customer relationship, supplier relationship and so on.

In terms of the possible kinds of context, this track heard a number of presentations describing how different upper ontologies dealt with different categories of thing that may be considered to be contexts. In addition to roles, relationships, products and the like, we explored a range of contexts or perspectives in which the semantics of some model element, or the overall application and use of some ontology or some set of terms, could be contextualized. These included:

Where: regions and places as the context for something. Niches, environments or more generally ‘holes’ as a general organizing principle for the Where context of a thing;

When: considering the commonly used upper ontology partitioning of ‘Continuant’ versus ‘Occurrent’ (things which exist in all their parts or in their identity across time versus things that are a feature of time), one can consider some continuant as being the context in which some occurrent is understood, and conversely the occurrent as the context in which some continuant is viewed. That is, continuant and occurrent may each be considered as viewpoints for something of the other kind. An example would be that a person’s life and times would be the context in which to understand that person.

What: systems may be considered as the context for something, with environments being considered as kinds of system, so that any organism needs to be understood within its environment

Who: roles are considered as kinds of context, within a general upper ontology partitioning or Player – Role – Context. Role itself, as an upper ontology construct, may be divided into Relational Role, Processual Role and Social Role. In the social case this reflects Searle’s Ontology of Social Constructs, where the formulation of ‘X counts as Y in C’ is a specialization of this general pattern

A further type of contextualization touched upon (but not beginning with W) was the notion of granularity, either in time or space. The way some part of the world is conceptualized by some agent (and therefore the way some concept would be stood up in some formal ontology), that is the way the world is ‘carved up at the edges’ would itself depend on the appropriate scale or granularity at which that carving up takes place – atomic, molecular, cellular, animal, regional, galactic and so on. This kind of contextualization would potentially be a further feature of the Where, the When or the What.

For each of the above there are usually upper ontology concepts that would form the broad categories of which different kinds of these contextual notions may be categorized – the holes, roles, occurrents and so on. Absent so far in these explorations has been the ‘Why’ and it was not clear if this represented a special challenge to this treatment, something that did not belong in an ontology, or something that could only be applied to the ontology as a whole (why are we using this ontology, this set of representations, to do this thing?)

It is possible that approaches (1) and (2) could be combined, or could be considered as the same kind of idea; that the various Ws, being partitions in many upper ontologies, might either be or (as in the case of roles, be defined within) some context. The contexts of client relationships versus specific product customer relationships point to a possible hierarchy of roles and a corresponding hierarchy of contexts; might it be that this pattern forms a common organizing principle for all or most of the identified context types (perhaps with granularity or scale as a further, separate distinguishing feature)? This idea is recommended as deserving further exploration.

The third approach suggested was that everything in a given ontology is the context for everything else. Given the wide-ranging nature of the Ws, could it be that most or all of the concepts to which any given concept relates would in any case fit under one or another of these? That would make (3) convergent with approaches (1) and (2). Again we recommend this needs further exploration, with realistic examples.

A further, seemingly quite separate approach to context emerged in discussions, and may tie in to the missing W: the Why. This is the intent which some viewer or user brings to some ontology. This relates to the question of ‘Perspective’. Is Perspective a kind of Context or vice versa? What is the perspective from which a given overall ontology is used, understood or reasoned over? What is it for?

Given our original premise that an enterprise or reference ontology exists to make previously implicit data (or other language element) concepts explicit, this task can never realistically be completed. To include every possible ‘contextual’ matter as something within the ontology would be to complete an ontology of everything – the mythical and entirely useless 1:1 scale map of the world. Therefore, having hauled a bunch of contextual things into the ontology and made them part of it, that ontology itself still exists within some context and is interpreted in that context. All we are doing is making the ontology broader or narrower.

Further discussions on this track and elsewhere suggest that this is not really a separate use of the notion of ‘Context’ at all. An exploration of the notion of ‘Micro-theories’ as exemplified by the Cyc project was helpful in providing the language with which to consider this question. Micro-theories are not, as the name might suggest, some siloed representation of subject matter that is incompatible with other representations of the same or adjacent subject matter. Rather it takes the form of a modularization of the overall ontology, such that in order to reason over some specific topic (say human liver cells or credit default swaps), one needs only the broader concepts related to that subject matter and to concepts referred to in its properties, and one does not need to make reference to ontologies of adjacent subject matter (skin cells, interest rate swaps etc.).

An analogy that has been proposed here is that if the overall reference ontology is like the walls of a darkened room, then shining a light on parts of that ontology equates to considering the context of that part of the ontology. The light may cover a broader or a narrower area but will not cover the whole ontology at one time. The features of the world that have been stood up as explicit ontological classes may not all need to be referred to at the same time. And the room itself of course ends somewhere.

On this basis, we suggest that what were seen as two separate notions of context – classes of thing that are kinds of Context arranged under some upper ontology, versus the context in which the ontology itself is used – are in fact the same basic notion of context. Different uses of ontology would take place in different contexts, and many (but not all) things that might be considered as context, might be included within some ontology in order to contextualize other elements in that ontology.

Conclusion: there is the ontology of contexts and there is also the context of ontologies. The ontology of contexts is simply an ontology; the broadest categories of the kinds of thing that may be considered as the context for other things, are what makes up the upper ontology. These are the notions of Who, What When Where and hoW, which, taken along with more constructive aspects of the ontology like the scales or granularities, hierarchies of roles, relations, descriptions of systems, processes, functions and so on make up the context for any given concept. What you do with the ontology – the Why – remains a matter of broader perspective.