JADE AND AIML: A MULTI-AGENTS TUTORING SYSTEM M. Gentile 1 , A. Augello 2 , M. Allegra 1 , G. Fulantelli 1 , G. Pilato 2 1 Italian National Research Council Institute for Education Technology Via Ugo La Malfa, 153 – 90146 Palermo - ITALY 2 Italian National Research Council Istituto di Calcolo e Reti ad alte prestazioni Viale delle Scienze 90128 Palermo – ITALY {manuel.gentile,allegra,fulantelli}@itd.cnr.it augello@csai.unipa.it pilato@pa.icar.cnr.it Abstract In this paper we present a Multi Agent Tutoring System made up of a community of chat agents which have specific competences and are able to carry out natural language conversation with students. Animated conversational agents can be used in the tutoring systems to produce incremental gains in learning. The agents of the proposed system have been developed by integrating two emerging technologies: Java Agent Development Environment (JADE) and ALICE technology. The system includes a set of Chat Agents (CA), each with a specific knowledge base, which is implemented through a set of the Artificial Intelligence Markup Language (AIML) files. The agents’ knowledge is stored in a tree managed by an agent called GraphMaster (GMA). When a user queries an expert agent it forwards the request to the GMA which produces the associated response as its output. We have also built a Student Gui Agent (SGA) and the Tutor Gui Agent (TGA); the former is the interface through which the students can request support in a specific topic, while the latter allows the tutor to monitor the students’ activity and stimulate conversation between students and the appropriate expert CA. Through the TGA, the human tutor can analyze the students’ conversation and run the Targeting cycles in order to produce an increasingly refined knowledge base of the various CAs. The TGA also allows the tutor to create new CAs based on a new set of AIML files. Keywords Tutoring System, Natural Language Process, Multi Agent System, JADE, FIPA, AIML, ALICE 1 INTRODUCTION The conversational agents represent an original approach to Human-Computer interaction problems; in particular, the chat-bots systems make it simple to create a dialogue system based on natural language. For this reason they can be used as interfaces for a wide range of applications. Natural language processing technologies have matured enough for us to begin applying them in educational software [1]. In recent years tutorial dialogue systems have become more and more prevalent; these systems help students actively construct knowledge through text based conversations [15][17][2][7]. An example is a system that takes a knowledge-based approach to natural language understanding and uses a statistical text classifier as a backup [1]. This system uses a logic-based representation of semantic content and a representation of pedagogical content knowledge in the form of a hierarchy of partial and complete explanations. Another example is given by AutoTutor [9], developed by the Tutoring Research Group (TRG) at the University of Memphis to simulate the dialogue patterns of typical human tutors. Many other systems have yielded successful evaluations with students [10][5][9]. Another class of tutoring system is designed through a multi agent implementation. These systems interconnect autonomous agents that realize complex system functionality through communication and collaboration. An example is the Multi-Agent Tutoring System (MATS), that supports web-based tutoring agents which can collaborate over a teaching task. The agents have been defined as artificial teachers, distributed over the Internet, heterogeneous in terms of knowledge structure and content, supporting various interaction paradigms [16]. Our approach has been to create a multi agent tutoring system by integrating a FIPA-compliant platform and a well-established chat-bot technology: the Java Agent Development Environment (JADE) [11] and the ALICE free software technology [4]. The aim is to build a tutor system with expert chat agents able to carry out natural language conversation with students and support them in the learning process by answering question on topics of specific interest. The motivation for this choice is that an MAS system can solve problems that are too large for a centralized single agent, due to resource limitations or the sheer risk of having one centralized system [12].