Ontology in computer science is a formal representation of concepts within a domain. With the emergence of the semantic web, ontologies will take the role of the anchor to which all content can and should relate.
Behind the unspoken conceptualizations in our heads lie formal ontologies. Which, being necessarily man-made, pose a question.
Who has ownership over ontologies?
If their purpose is indeed to serve as the lighthouse on the sea of information, ontologies must be unambiguous, and therefore be defined and maintained by a single entity. Will this single entity be a company, a consortium, a committee, an organization or perhaps a government agency? What guarantees that formal ontologies will follow the changes that may occur in the instance domain?
There is one guarantee: collective ontology management. And I’m not thinking of Wikipedia-style collaboration, but real collective effort where everyone throws in his/her two cents.
Take a look at this very simple comparison between equivalent fractions of a content map and an ontology. (A content map is based on collective definition of probabilistic relations between content elements.)
The resemblance is hard to miss. It’s no surprise, both deal with concepts and instances bound into a network through different sorts of relations. But as we take a closer look, it becomes obvious that content mapping is fundamentally different.
- There’s no distinction between elements such as classes, instances, attributes, et cetera. They’re all content. What constitutes a class from an ontology point of view depends solely on the relation. One element may be instance and class simultaneously.
- There are fewer, more general types of connections. You can extend an ontology with new relations that specify the way certain elements are connected to each other. Content mapping defines only a few, from which new, implicit ones can be derived via machine learning.
- Domains don’t have definite borders. It is very likely that elements have connections leading out of a domain, superseding what we call ontology alignment. As an element may be instance and class at the same time, it can also belong to more than one domain. In fact, these are the connections through which cross-ontology relationships emerge.
- Dynamics is inherently embedded into the system. As content changes, connections follow. Classes are constantly created, updated or deleted by changing generalization connections.
Content mapping creates an organic system where ontologies float on the surface.
Defining ontologies in this environment is no longer necessary, they crystallize with the natural progress. We only have to harvest the upper generalization layers to get an understanding of conceptual connections in any data set. Domains needn’t be defined beforehand either. Instead, we draw their outlines where we deem them fitting.
Clues to rely on
However flexible content mapping technology may seem in defining and following ontologies, its purpose is to connect previously unconnected content, and therefore it needs clues to follow up. Prior user input, search indexes, or existing ontologies may provide these clues. Once those clues are there, content mapping simplifies ontology management in several aspects.
- Fewer relations: Only a handful of general relations are explicit, domain specific relations are all derived from those.
- No need for focused attention: Ontology management requires no supervision as implicit connections change with content.
- No knowledge of semantics: Connections (both explicit and implicit) can be set or changed without any knowledge on the subject of semantics or ontologies.