Figure 7.

(Top) The participants at generation i are individually trained on dynamic set i − 1. They then communicate about two novel sets of triangles: a randomly generated set (dynamic set i) and a set that remains constant for all participants (the static set). The labels applied to the dynamic set become the training input to generation i + 1. (Bottom) During communication, the director is shown a triangle and is prompted to type a label to describe it. The label is then displayed on the matcher's screen along with an array of six triangles to choose from. The matcher's task is to click on the triangle that his or her partner is trying to communicate. As feedback, both participants see the target triangle and the selected triangle.