Automatic Assessment of Short Free Text Answers

Assessment plays a central role in any educational process, because it is a common way to evaluate the students’ knowledge regarding the concepts related to learning objectives. Computer assisted assessment is a research branch established to study how computers can be used to automatically evaluate students’ answers. Computer assisted assessment systems developed so far, are based on a multitude of different techniques, such as Latent Semantic Analysis, Natural Language Processing and Artificial Intelligence, among others. These approaches require a reasonable corpus to start with, and depending on the domain, the corpus may require regular updates. In this paper we address the assessment of short free text answers by developing a system that captures the way the teacher evaluates the answer. For that, the system first classifies the teacher question by type. Then concerning the type of question, the system permits the teacher define scores associated with subparts of the answer. Finally, the system performs the assessment based on these sub scores. For certain types of questions, paraphrases of answers are also considered in an attempt to obtain a more precise assessment. The system was trained and tested on exams manually graded by a History teacher. Based on the results obtained, we show that there is a good correlation between the evaluation of the instructor and the evaluation performed by our system.