UC Irvine

The overall debugging process is a complicated and troublesome task, involving several stages

and dimensions of human comprehension. Developers seek understanding of several aspects

of faults, such as, where the faults are located in the code, what sequences of actions invoke

faults that cause failures, and why the program is failing due to the faults. Despite a large

body of research for providing automation for the first two tasks, very little work has been

conducted in helping to assist in the last question of "why" -- that is, for describing the

nature of the fault. I propose an automated approach to describe software faults that can

indicate the nature of faults and their failures; thus ameliorating comprehension and reducing

manual effort. To create this automated approach I propose using a combination of dynamic

analysis techniques with information retrieval and text mining to generate natural language

clues. In this document I outline the design of my technique along with a research plan that

I have used to investigate the effectiveness of using a such an approach. In particular, I detail

five evaluations that provide a thorough assessment of the relative merits and shortcomings

of the proposed technique.