        Ismael Ávila 1,2 , Ricardo Gudwin 1 , 1 University of Campinas, UNICAMP, Av. Albert Einstein $ 400, 13083$970 Campinas, Brazil 2 CPqD R&D Center, Rod SP, 340, 13086$902 Campinas, Brazil {ismael.avila@gmail.com, rgudwin@gmail.com}  We briefly describe an ITS CALL tool in which the domain (L2) is modeled in terms of its lexemes and their usual interrelations. Each L2 lexeme behaves as an autonomous agent that attempts to be learned in a multi$agent environment. So as to be learned, it competes for a limited resource, the course interface, where it can present its meaning by means of images and other cues. Once presented, the lexeme helps other lexemes with which it can form valid syntagms to create example sentences. Each agent’s individual goal is subject to global goals which, depending on the adopted teaching strategy, may prioritize some agents, and then favor them in an action selection mechanism. We discuss the lexeme$agent structure and advance some implications of such approach.  CALL, ITS, multiagent, domain model, NLP.   We introduce here an ongoing research in the area of ITS language teaching. Its main contribution is the approach in which every L2 lexeme (pedagogical unit) acts as an autonomous agent whose goal is to be learned. This creates a bottom$up arrangement that is very flexible to adapt to changes in the pedagogical goals and that reflects the incremental nature of the L1 acquisition [1]. Furthermore, the relations among the agents simulate the typical dependencies that the lexemes present in the L2 grammar. We briefly describe the implementation and its implications.  !  " # #  The ITS CALL tool described herein is aimed to teach about 2700 L2 lexemes: 1500 nouns, 800 verbs, 300 adjectives, and 80 connectors (prepositions and conjunctions). Each lexeme is programmed to act as an autonomous agent that seeks to be learned, in an action selection process as defined in [2]. In other words, each lexeme$agent has the individual goal of teaching its own meaning to the learner, both in isolation and in combination with other lexemes. A presentation on the screen is a necessary condition