Musical improvisation is one of the most widely practised forms of composition and performance. It is evident in most musical cultures and can be traced back to some of the earliest artistic and religious musical practices. It is a style that incorporates such concepts as change, adjustment, experimentation and ornamentation. Musical improvisation can demand cohesion and participation, a communicative and cooperative relationship between performers. We propose a method of capturing and reshaping an instrumental gesture on the Reactable, with the aim of establishing a computer improvisation system that will respond to the performer with variations on their contributed, physical gesture. A method of sampling this gesture, specifically the physical rotation of a tangible object, is detailed. We look to techniques used in Computer Aided Algorithmic Composition (CAAC) for a practical approach to data generation from a given sampled gesture in real-time. A stochastic procedure is designated as this algorithmic approach, one that generates new gesture data sequences from the performers original instrumental gesture. Bi-directional communication limitations within the Reactable are encountered and overcome in the construction of this system. A method of continuous play-mode activation based on a probabilistic routine is put forward. The testing of this system is described, with user responses and experiences discussed. Finally, plans for future development are outlined.