Improving Assessment of Students through Semantic Space Construction

Assessment is one of the hardest tasks an Intelligent Tutoring System has to perform. It involves different and sometimes uncorrelated sub-tasks: building a student model to define her needs, defining tools and procedures to perform tests, understanding students' replies to system prompts, defining suitable procedures to evaluate the correctness of students' replies, and strategies to improve students' abilities after the assessment session.In this work we present an improvement of our system, TutorJ, with particular attention to the assessment phase. Many tutoring systems offer only a limited set of assessment options like multiple-choice questions,fill-in-the-blanks tests or other types of predefined replies obtained through graphical widgets (radio-buttons, text-areas). This limited set of solutions makes interaction poor and unable to satisfy the users' needs. Our interest is to enrich interaction with dialog in natural language. In this respect, the assessment problem is strictly connected to natural language understanding. The preliminary step is indeed to understand questions and replies of the student.We have reviewed the system design in the framework of a cognitive architecture with the aim to reach a double result: the reduction of the effort for the construction of the knowledge base and the improvement of the system capabilities in the assessment process. To this aim a new common semantic space has been defined and implemented. The entire architecture is oriented to intuitive and natural interaction.