Evaluating a formal KBS specification language

Abstract

In recent years, the knowledge engineering community has begun to explore formal specification languages as a tool in the development of knowledgebased systems. These formal knowledge modelling languages have a number of advantages over informal languages, such as their precise meaning and the possibility to derive properties through formal proofs. However, these formal languages also suffer from problems which limit their practical usefulness: they are often not expressive enough to deal with real world applications, formal models are complex and hard to read, and constructing a formal model is a difficult, error prone and expensive process. The goal of the study presented in this paper is to investigate the usability of one such formal KBS modelling language, called (ML) 2. (ML) 2 is strongly based on the structure of the knowledge-models used in the KADS KBS development method. We first designed a set of evaluation criteria. We then applied (ML) 2 in two case-studies and scored the language on our evaluation criteria. (ML) 2 scored well on most of our criteria. This leads us to conjecture that the close correspondence between the informal KADS models and the formal (ML) 2 models avoids some of the problems that traditionally plague formal specification languages.

Knowledge modeling directed by situation-specific models

“…In ML(2), the declarative parts of a conceptual model are expressed using predicate calculus, facilitating a static validation of those parts. However, the developers of ML(2) admit that offering similar support for the task model is not yet possible because this model is expressed procedurally (Van Harmelen, Aben, Ruiz & Van de Plassche, 1996). Another example is MODEL-K (Voss & Karbach, 1993), an imperative language that provides a formalized model with the operational semantics needed to produce a full working system.…”