Using speech recognition to create virtual dialogues is now possible, although it is still difficult to build real conversation partners. In this context, the CALL-SLT system developed by the University of Geneva offers a platform to create virtual dialogues with an avatar on the Web (http://callslt.unige.ch/). In this type of game, the student participates in a virtual dialogue by translating and properly verbalizing a speech act (a conceptual proposal that is not syntactic) provided in their native language. In order to provide a more immersive learning situation, we proposed an adaptation of the CALL-SLT system which does not use the learner's native language. In this paper we describe how we used this tool in the context of a real French foreign language (FFL) training for employees of French companies. For exploratory purposes, 10 adult learners of FFL through distance learning were given access to the tool. We collected audio and usage data, as well as reports of interviews with the participants and their trainer. In the paper, we will summarize all the results. These are very encouraging but show, amongst other things, that the interaction and the relationship with a virtual conversation partner is not entirely straightforward for adult learners.