The Effects of Language and Pruning on Function Structure Interpretability

Abstract

In this research, the interpretability of function structures is evaluated through a user study in which participants are given function structures and asked to identify the product that is modeled. Two abstraction factors are controlled in the experiment: the type of functions and the specificity of the terms, thus resulting in functional models are four level of abstraction. The user study shows that free language significantly improves the accuracy and speed of human interpretability over the functional basis vocabulary. Further, pruned function structures significantly improve the speed of interpretability over reverse-engineered function structures without a loss of accuracy. It is concluded that the levels of each factor are useful for different activities and stages of design. Recommendations are made for the appropriate combinations of factor levels for various design activities.