Previous: A task independent description of role limiting methods Up: A task independent description of role limiting methods Next: Cover & differentiate
While other chapters in this deliverable focus on modelling operators that represent inference-function expansion and method expansion, this chapter uses modelling operators to represent the translation of model ontologies. Model ontologies have an internal, formal structure and an extension: the group of things the model refers to. In order to the apply a method on a different type of task the extension of the domain ontology has to be redeveloped: the types of information that have to be acquired need to be defined for the new context. This is a process of translation and analogical reasoning between the application of the ontology and task in the original and the new context. We have represented the result of this mapping has been in tables. These tables translate the ontology from the original generic model to a model specific for a task type. In CommonKads library terminology, these tables are therefore modelling operators.
Role Limiting Methods ( RLM) [McDermott, 1988] are one of the best known generic models. RLM's reduce the roles of knowledge in the domain problem solving. Role limitation facilitates knowledge acquisition, knowledge representation, efficient inferencing and separately represented control. RLM require a certain domain model organisation and provide control mechanisms to apply the domain knowledge to make certain types of inferences in order to reach an end state. RLM's should not be confused with weak methods or problem solving methods.
A weak method has weaker domain requirements, and has therefore a wider range of applicability. One problem with weak methods in general is that the domain ontology is not `role limiting', a source of irregular representations. A second problem is that the provided search control is often not powerful enough. To solve this, domain knowledge has to be extended with problem solving control [Laird et al., 1986][Duursma, 1992]. This is a second source of irregular domain representations.
In CommonKads a problem solving method is described as ``a description of how to achieve a certain (sub-) task, characterised by a goal and input/output specifications, through the generations of sub-tasks'' [Wielinga et al., 1992]. A RLM extends this with a description of the domain ontology and domain schemata.
In this chapter RLM's are described by a their domain ontology, task decompositions, inference or function structure and control. 0pt 0pt
A single RLM is a description combining elements from each group.
In [Marcus, 1988] five RLM's are discussed; cover and differentiate (C&D), `qualitative reasoning and cover and differentiate', propose and revise (P&R), acquire and present and extrapolate form a similar case (ESC). Of these only C&D, P&R and the extrapolations approach are complete descriptions and independent of a specific task.
RLM's are applicable on a task type that can satisfy the domain assumptions. Not a lot of tasks share all these characteristics. This limits the applicability of shells like MOLE and SALT that have been specifically developed for one single sc rlm. However, the RLM's can be used as a starting point and later be refined for a larger set of tasks. An RLM that is directly applicable on diagnosis can not easily be used for modelling design problems, because these two tasks types share few characteristics. The next few sections show hoe tasks in between these two extremes can often be modelled using different sc rlm's.
The CommonKads library describes RLM's by the use of:
We present each of the generic model definitions of RLM's first in the terminology of the orginal task and context. We use modelling operators to make vocabulary matchings for a set of applicable task types. This makes the original RLM generic with respects to several of task types. This vocabulary is then used for the relabelling of knowledge roles and for the rephrasing of the domain model requirements.