The Semantic Web has popularized the concept of ontology. The usual definition one sees for the term is Gruber’s “specification of a conceptualization” (Gruber 1993). The definition could use some unpacking. Essentially, it refers to the idea that the terms used by AI agents should be defined by reference to a shared schema which classifies those terms. The shared schema, or specification, simply defines how terms are related to each other in a graph. More than a simple data model, where the relationships between entities or objects are not explicitly defined, an ontology also names the possible relationships between terms. So, instead of just having “address” as a field of the entity “person,” an ontology will explicity assert that a person has an address. So verbs (which are often buried in “predicates”) are an important part of ontologies. In fact, one way to think of an ontology is to call it a data model with verbs. (Sort of.)

Ontologies are useful for programming agents, which are just programs that consume and produce information on their own. (I call them infophages.) So, when you program an agent to respond to a term, you have the agent’s program refer to an ontology to disambiguate the term among synonyms, and to make logical inferences about the term. For example, if the agent encounters the term “Socrates,” it will be found to be specified as a member of the class “Human” and since “Human” will have the property “is Mortal,” the agent can transfer the property to “Socrates” too. (Truth, or validity, is just traceability within a reference graph.)

Or, if the agent encounters the word “Madonna,” and that term is formally specified as part of an ontology (say, by using Tim Berners-Lee’s URI method), the agent will be able to trace its class to either “Pop Singers” or “Religious Figures” or “Christian Iconography” or whatever.

Sometimes-often-ontologies are described as if they provide “meaning” for agents. Even the term Semantic Web implies this. When introduced to novices, the Semantic Web is often introduced as bringing “meaning” to the web. I find this explanation misleading at best. Meaning is much to too important of a word to describe what is going on here, which is simply a layer of classification being added to what would other wise be a list of terms. Ontologies are just systems of classification for terms, a second-order set of terms that (1) increases the probability that an agent can process a term in a way that users will expect, and (2) adds a layer of connectivity, above the raw verbiage on the web, that decreses the average distance, or degress of separation, between any two terms.

So, ontologies have two properties that make them useful and, as I want to argue, usefully viewed from a comparative perspective. First, they introduce verbs into the mix of data modelling. Second, they are just systems of classification-like Australian totemic systems, Mayan calendars, and Western philosophical ontologies such as Aristotle’s and Leibniz’s.

Comparativism will help us create better ontologies, and better systems for using ontologies, in two ways.

First, comparativism can train our attention on the great rabbit warren of words and meanings that lies at the heart of the ostensibly neat world of triples-I refer to the predicates. Although the nouns that comprise the vocabularies for subjects and objects in RDF can be neatly specified as analytic taxonomies of terms, predicates are subject to no such rules, and can contain within themselves whole sentences. Unlike the nouns, predicates mask a great deal of what Kant would call “synthetic judgements.” I propose a sociolinguistic approach to the use of verbs embedded in the predicate systems of linked data vocabularies that will provide a better basis for crafting predicates.

Second, comparativsm can help us move beyond set theory-useful as it is-to consider other totalizing schemes for organizing ideas, such as those described by anthropologists. I mentioned totemic systems, calendars, and Western ontologies. But there are many others to consider. What is more, anthropologists have done a great deal to describe, classify, interpret, and even explain these. Such an approach would be grounded in the following works, all of which define or grow out of the structuralist tradition within anthropology:

1. Emile Durkheim and Marcel Mauss, 1903, Primitive Classification.
2. Louis Dumont, 1966, Homo Hierarchicus.
3. Michel Foucault, 1966, The Order of Things.
4. Claude Lévi-Strauss, 1968, The Savage Mind.
5. Victor Tutner, 1969, The Ritual Process: Structure and Anti-Structure.
   
6. Mary Douglas, 1970, Natural Symbols.
7. Pierre Bourdieu, 1977 [1972], Outline of a Theory of Practice.
8. Alfred Gell, 1975, Metamorphosis of the Cassowaries.
9. Rodney Needham, 1975, Polythetic Classification.
10. Edmund Leach, 1976, Culture and Communication.
11. Marshall Sahlins, 1978, Culture and Practical Reason.
12. Jadran Mimica, 1988, Intimations of Infinity.

And so forth.

This entry was posted on Tuesday, April 14th, 2009 at 10:52 am and is filed under Comparative Ontology, theory. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.