Rambling Thoughts on Future Research
My masters project has demonstrated two important principles for me:
- The flexibility of Rules systems, along with the detailed manner in which rules must be defined.
- The power of Ontology languages like OWL, along with the limitations.
A problem with Rules systems is the detail with which a domain must be defined. Even the simplest operations need to be spelled out in detail. This means that every new domain that a system is applied to must have specific rules designed specifically for that application. However, rules are flexible enough to define the most obscure operations.
Conversely, an ontology simply makes a declaration, and all required operations will be performed (in the case of the above project, those operations will be performed using rules). All of the work for performing calculations for an ontology is done up front (such as a rules system for implementing OWL). A problem with ontologies is that all known languages have restrictions which do not allow for certain types of relationships. The classic example in OWL is an inability to define an "uncle" relationship, based on the relationships of "parent" and "brother". Another example is the inability to define a value range for a data type. Other ontology systems allow for this, but at the expense of different functionality.
A solution to the problem of inferring across knowledge domains may be a more complete ontology language. Such a language might have the ability to describe what a family looks like, and to then apply this knowledge to a group of people to determine who is related to whom, and in what way. Along with these complex relationship, data types will also need to be considered. Since rules can already handle many advanced relationships, perhaps there is some mechanism to automatically generate the kind of rules required for flexible systems.
It may be possible to define a language, complete enough to describe numerous differing systems in such a way that inferences can be made involving data from several domains at once. The kind of inferencing to aim for is already performed by people, as was demonstrated in the classic "Swanson fish oil" story.