DL-based Support to Domain Engineering

1 Overview and Motivations We believe that an automated support tool for Domain Engineering (DE) should behave as an Intelligent Knowledge Management Environment (IKME) and not as an expert system. It should only help the analyst in the discovery and in the explicitation of contradictions, redundancies and hidden properties detected in the Domain Knowledge Base (DKB) under development and/or analysis. In other words, it should not suggest modii-cations to (or worse, directly modify) the description of concepts in the considered DKB. Present IKMEs for Domain Engineering suuer, in our opinion, from their mixing up two representation levels: 1. a user-oriented conceptual level, where domain-speciic elements are represented (e.g., data, processes and ows in Data-Flow Diagrams 9, 11]); 2. an underlying logic-oriented epistemological level, where inferences are drawn from general-purpose elements (e.g., DL concepts and roles). The explicit separation of operational tasks between humans and computer systems should imply a corresponding separation of levels (1) and (2) in support tools. We claim that in most of present IKMEs one of the two above levels is not explicitly present, i.e., it is \em-bedded" in the other level. A few IKMEs (e.g., 1]), directly implement specialized reasoning procedures based on the distinct conceptual primitives of the modelled domain, such as entities, relationships and so on. Conversely , the majority of IKMEs (e.g., 6, 2] directly \pro-gram" general-purpose reasoners 1. In other words, they describe domain knowledge in terms of epistemological primitives (such as concepts and roles, sets and relations, nodes and arcs and so on). We believe that both approaches suuer from some drawbacks. In the direct implementation of specialized reasoning procedures, only a particular concept model is supported. This means that variants of the (already existing) reasoning algorithms 1 Here and in the following, general-purpose reasoners are also called inferential engines or KR systems. of general-purpose reasoners have to be re-implemented. In the direct usage of general-purpose systems, the conceptual level is implicit, making it diicult to assess the representation adequacy. Moreover, since modelling is not performed in terms of domain-speciic concepts, an intelligent explanation of drawn inferences, which is part of the support, must be separately built. We propose an engineering approach to the development of IKMEs for conceptual modelling and analysis which explicitly deals with both representation levels at the same time. This gives rise to a multilevel approach which is more suitable from an engineering perspective. More precisely, a …