Task-specific Language Modeling for Selecting Peer-written Explanations

Students who are learning to program, often write "buggy" code, especially when they are solving problems on paper. Such bugs can be used as a pedagogical device to engage students in reading and debugging tasks. One can take this a step further and require students to explain in writing how the bugs affect the code. Such written explanations can indicate students' current level of computational thinking, and concurrently be used in intelligent systems that leverage "learnersourcing", the process of generating course material for other learners. In this paper, we discuss how to combine learning analytics techniques and artificial intelligence (AI) algorithms to help an intelligent system distinguish between strong and weak textual explanations.