The impact of different types of impurity on theory-softening

Abstract

Theory-softening is a novel way of utilization of imperfect domain theories. Soft interpretation of given flawed theories provides a way of classifying examples despite the presence of impurities in the theory. However, theory-softening is not equally successful for all possible types of flaws present. This paper investigates the performance of theory-softening and finds that while it is useful for impurities introducing over-specificity, it can become unhelpful and cause degraded performance for impurities introducing over-generality.