The goal here is to transpose the concept of IRMS to a compositional context, without losing its fundamental reflexive nature. This software will notably introduce linguistic components in the interaction to allow users to create structured musical objects. Composition IRMS will also include the possibility of expressing, collecting and exploiting various types of user feedbacks on their compositions, e.g. through tagging systems. The main innovation of this software will be to allow the user to use a linguistic component to tag musical material, and reuse this material for composition. This module will use machine learning techniques such as Support Vector Machines in complement to the other machine-learning techniques used in the project such a Markov models. Note that the term "linguistic" here refers to the use of "symbols" in general. These symbols may be textual, but these may not be adequate for young children. Other forms of symbols will be considered, notably "drawings" made by children, of symbols selected from a pre-programmed library of images. The use of "gestures" will also be investigated.