djsaab.info

[This is not meant to be a well-written, coherent essay. Rather, it is simply my reflections of some of the concepts discussed in: Uschold, M. (2003) Where Are the Semantics in the Semantic Web? AI Magazine 24(3): 25-36.]

Talking about an agent’s ontology, Uschold says:

If it were written only for people to understand, this specification could be a glossary.

It is a simple statement that belies the complexity of human cognition and its ability to function as a connectionist network. Humans are able to read a word and its definition and recall a complex of associations and experiences to create meaning, to understand its semantic content and the ontological commitments that are shared within a community. An agent doesn’t have this ability and must rely on explicit specifications (ala Gruber) and formal languages to communicate its meaning to other agents. Successful communication is the goal, but machines are confronted with heterogeneous sources, “different ontology representation languages, different modeling styles, and an inconsistent use of terminology.”

Uschold describes the semantic continuum:

Semantics can be implicit, existing only in the minds of the humans who communicate and build web applications. They can also be explicit and informal, or they can be formal. The further I move along the continuum, the less ambiguity there is, and the more likely it is to have interoperable, robust, and correctly functioning web applications.

The difficulty with implicit semantics is their inherent ambiguity. People don’t always agree about the meaning of a term. Yet this is exactly where people live, so to speak. Humans engage in dialogue and discourse in order to clarify meaning. Through this interaction, we make the implicit semantics explicit. Engaging with machines isn’t as interactive, typically. We don’t expect our machines to be intelligent in terms of understanding semantic content and to be able to express that understanding as humans do. We also don’t expect that machines will be able to assess contextual cues that we use as humans to make meaning of our interactions with others, nor do we expect machines to remember our previous interactions with them (unless we explicitly encode that into the preferences of each particular application or service).

As Uschold continues to describe the continuum, moving towards the more formally expressed semantics for machine processing, it occurs to me that what we are doing is conforming our natural ways of interacting to the requirements of a machine. Why are we not trying to make the machine conform to our natural way of doing things? Shouldn’t the goal be to enable the machine to dynamically discover the meaning of content and how to use it? This, as it turns out, is what Uschold tackles next, machine-proccessible semantics:

For a computer to automatically determine the intended meaning of a given term in an ontology is an impossible task, in principle; it would require seeing into the mind of the author. Therefore, a computer cannot determine whether the intended meaning of two terms is the same.

Getting an agent to understand semantic content can be accomplished by hardwiring the meaning of terms and procedures into them, or through accessing external, publicly agreed to declarations such as ontologies. The heterogeneity of information and meaning makes both of these strategies limited. Information from different web sites may need to be integrated, so we require some way to map the different meanings to each other. Ontologies in conjunction with semantic mapping and translation techniques play a key role in semantic integration (Uschold citing Bradshaw et al. 2003).

Uschold speaks of human consensus regarding the use of terms. He refers to this consensus as an implicit shared semantic repository. I use the term cultural schemas to refer to this shared cognitive “repository.”

So, where are the semantics in the semantic web? Uschold offers these six explanations:

1. They are often just in the human-as-unstated assumptions derived from implicit consensus (e.g., price on a travle or bookseller web site).
2. They are informal specification documents, e.g., the semantics of UML or RDF SCHEMA.
3. They are hardwired in implemented code (e.g., in UML and RDF tools and in web shopping agents).
4. They are in formal specifications to help humans understand or write code (e.g., a modal logic specification of meaning of inform in an agent communication language).
5. They are formally encoded for machine processing, (e.g., fuel-pump has (superclasses SHO: pump)).
6. They are in the axiomatic and model-theoretic semantics of representation languages (e.g., the formal semantics of RDF).

Tags:cognition, concepts, cultural schemas, formal ontology, information, language, ontologies, semantics semantic web
Comments are off for this post

Explaining what ontology is is difficult enough. Everyone seems to have their own idea about what it is. I have my own ideas about what ontology is and how they should be constructed. Even more difficult is trying to explain what an ontologist does. The Preface to the FOIS’06 Proceedings crafts an extraordinary description of ontology and the work of ontologists as theoreticians and engineers.

Formal Ontology in Information Systems, Proceedings of the Fourth International Conference (FOIS 2006), Brandon Bennett and Christiane Fellbaum (eds.), IOS Press

Preface

Since ancient times, ontology, the analysis and categorisation of what exists, has been fundamental to philosophical enquiry. But, until recently, ontology has been seen as an abstract, purely theoretical discipline, far removed from the practical applications of science. However, with the increasing use of sophisticated computerised information systems, solving problems of an ontological nature is now key to the effective use of technologies supporting a wide range of human activities. The ship of Theseus and the tail of Tibbles the cat are no longer merely amusing puzzles. We employ databases and software applications to deal with everything from ships and ship building to anatomy and amputations. When we design a computer to take stock of a ship yard or check that all goes well at the veterinary hospital, we need to ensure that our system operates in a consistent and reliable way even when manipulating information that involves subtle issues of semantics and identity. So, whereas ontologists may once have shied away from practical problems, now the practicalities of achieving cohesion in an information-based society demand that attention must be paid to ontology.

Researchers in such areas as artificial intelligence, formal and computational linguistics, biomedical informatics, conceptual modeling, knowledge engineering and information retrieval have come to realise that a solid foundation for their research calls for serious work in ontology, understood as a general theory of the types of entities and relations that make up their respective domains of inquiry. In all these areas, attention is now being focused on the content of information rather than on just the formats and languages used to represent information. The clearest example of this development is provided by the many initiatives growing up around the project of the Semantic Web. And, as the need for integrating research in these different fields arises, so does the realisation that strong principles for building well-founded ontologies might provide significant advantages over ad hoc, case-based solutions. The tools of formal ontology address precisely these needs, but a real effort is required in order to apply such philosophical tools to the domain of information systems. Reciprocally, research in the information sciences raises specific ontological questions which call for further philosophical investigations.

Tags:concepts, fois, formal ontology, informatics, information, language, ontologies, philosophy, science semantics
Comments are off for this post

Ontology is the theory of being. It asks the fundamental questions: What exists? What is real? Philosophers have been trying to work out the answer to these questions since at least the ancient Greeks first posed them. Ontology has been understood by philosophers as being singular. Ontology, if one were to be devised, would delineate all things that exist in the universe and their relationships with one another. From a philosophical perspective, there can be only one Ontology.

Things changed in the twentieth century. Scientists, in particular a guy named Quine, began to conceive of Ontology a theory of a specific domain. Since there were many domains and types of domain knowledge, it stood to reason that there were more than one ontology. The concept of multiple ontologies, each being a way of explaining the knowledge of a different domain, caught on among scientists. And ever since, scientists and philosophers have been trying to devise ontologies for their respective domains of research. They are trying to capture not only the things that comprise the domain, but also the concepts and processes each domain deems pertinent.

Information and computer scientists have devoted a lot of time and resources to developing ontologies. The impetus behind such research is the promise of interoperability and the sharing of data and information between information systems. The problem with computers is that they don’t grasp meaning. They can manipulate data and identify patterns of information, but they can’t create meaning for themselves or for humans. Ontologies are seen to be the keys to the kingdom, as it were, for the creation and sharing of meaning among the bits and bytes of data and information we have floating around in our information systems.

Researchers in information systems have been working on developing ontologies for a while. And they’ve failed miserably. There have only been a few successful ontologies developed, but they’re not generalizable. Why have ontology engineers and researchers failed? For many reasons, I think. First among them is their failure to understand what an ontology is–a taxonomy, a concept model, a concept map is not an ontology. Yet, time and again, researchers attempt to create rigid hierarchical structures that they think are ontologies. Second, they give primacy to their own biased worldview. They are convince that their scientific paradigms–to the exclusion of other paradigms, scientific or other–are the only valid or meaningful ways of understanding the world. They attempt to reduce the concepts and entities of a given domain to some form of quantitative measure–the only “real” measure of what exists in the world.

In combining the first two mistakes, they commit a third: neglecting schematic forms of cognition in favor of a symbolic processing paradigm. Their attempts to make ontologies interoperable amount to vocabulary matching strategies. Anyone who has ever done any language translation understands that there is often a lot of cultural filling-in that must happen in order to make the translation meaningful. There is no such thing as word-for-word translation. Yet, we keep searching for strategies that allow our machines to do just that.

Ontologies are not taxonomies. Ontologies are not intelligible as discreet quanta of information. And ontologies are not the result of symbolic processes. Ontologies are concepts, variable and schematic in nature. I also contend that they are emergent and akin to cultural schemas. How we can develop ontologies for information systems that reflect their emergent and adaptable nature is the focus of my research. I’ll have lots more to post about my theoretical perspectives and research. I just wanted to get the ball rolling.

Tags:concepts, cultural schemas, emergent, ontologies taxonomies
Comments are off for this post