A Flexible Dynamic System for Automatic Grading of Programming Exercises

The research on programs capable to automatically grade source code has been a subject of great interest to many researchers. Automatic Grading Systems (AGS) were born to support programming courses and gained popularity due to their ability to assess, evaluate, grade and manage the students' programming exercises, saving teachers from this manual task. This paper discusses semantic analysis techniques, and how they can be applied to improve the validation and assessment process of an AGS. We believe that the more flexible is the results assessment, the more precise is the source code grading, and better feedback is provided (improving the students learning process). In this paper, we introduce a generic model to obtain a more flexible and fair grading process, closer to a manual one. More specifically, an extension of the traditional Dynamic Analysis concept, by performing a comparison of the output produced by a program under assessment with the expected output at a semantic level. To implement our model, we propose a Flexible Dynamic Analyzer, able to perform a semantic-similarity analysis based on our Output Semantic-Similarity Language (OSSL) that, besides specifying the output structure, allows to define how to mark partially correct answers. Our proposal is compliant with the Learning Objects standard.