Three systems that generate real-time natural language commentary on the RoboCup simulation league are presented, and their similarities, differences, and directions for the future discussed. Although they emphasize different aspects of the commentary problem, all three systems take simulator data as input and generate appropriate, expressive, spoken commentary in real time.

The information it provides a dynamic, real-time environment units resulting from such an analysis in which it is still relatively easy for tasks to be encode a deeper understanding of the timevarying classified, monitored, and assessed. Moreover, scene to be described. They include a commentary system has severe time restrictions spatial relations for the explicit characterization imposed by the flow of the game and is of spatial arrangements of objects as well thus a good test bed for research into real-time as representations of recognized object movements.