A NATURAL LANGUAGE VIRTUAL TUTORING SYSTEM

ABSTRACT In this paper we present a natural language virtual tutoring system that has been developed to assist students during the learning process. For this purpose, we have used several technologies such as conversational agents and ontologies. Our system presents two advantages over traditional Content Management Systems (CMS). On the one hand, passive teaching material turns into active learning objects which students can interact with. On the other hand, our system assists students by means of a conversational agent that simulates a tutor who explains the lessons of the subject and helps students with those concepts that they find difficult to understand. KEYWORDS Virtual tutor, conversational agent, e-learning, ontology. 1. INTRODUCTION Nowadays, information and communication technologies have more and more importance in different sectors of the society. One of the most active fields is education [Cunha], where these technologies are being applied to provide students with personalized multimedia contents. There are many e-learning platforms based on CMS where students can download and study lists of topics that tutors adapt for these systems (Moodle [moodle.org], Atutor [www.atutor.ca], LON-CAPA [lon-capa.org], and Dokeos [www.dokeos.com]). In general, all these systems can be used to put the contents of the subjects at the disposal of students. However, the educational contents of these systems, in spite of their richness, are passive by nature since they do not allow involving students in the learning process beyond the display of contents. Several automatic tutoring systems have been detailed in the literature. They can be classified according to two different learning models: instruction-based models, and explanation-based models. Instruction-based models are focused on the contents to be learnt. In this case, students must solve a task or problem specified by the system, which has a model of the task to evaluate the possible mistakes that can be made in each step. To this category belong ITS-CREM [Kaklauskas], also COMET [Suebnukarn] and Slide-Tutor [Crowley] (for the diagnosis in medicine), IMITS [Butz] (for circuit analysis) and MAEVIF [Ramirez] (for virtual environments for entertainment). They focus on the sequence of steps to carry out, and the system always takes the initiative, not allowing the student to express his doubts about a concept anytime. These limits can be solved using the explanation-based models, which follow a conversational structure in which the tutor asks the student about any concept, so that a conversation is established to lead the attention to particular topics that require further explanation. In this case, the initiative can be taken by both the tutor and the student. These systems that include explicit models of the knowledge state of students entail an improvement regarding traditional CMS because the learning process is attended and actively performed, since it is possible to focus on those concepts that each particular student finds difficult. Therefore, this technology allows the system to carry out different models or tutoring styles, adapting the level of difficulty depending on the particular student, as a real tutor would do. Although conversational systems are widely used in other areas like telephone systems or automatic question answering systems, their use is not very extended in tutoring systems. We have only found two very topical systems employing this technology. CIRCSIM-Tutor [Woo] helps students of medicine to solve physiology problems whereas Auto-Tutor [Graesser] is focused on students of physics.