This paper presents a method which discovers the structure of given open source programs from their developer mailing lists. Our goal is to help successive developers understand the structures and the components of open source programs even if documents about them are not provided sufficiently. Our method consists of two phases: (1) producing a mapping between the source files and the emails, and (2) constructing a lattice from the produced mapping and then reducing it with a novel algorithm, called PRUNIA (
ing Algorithm Based on
ttributes), in order to obtain a more compact structure. We performed experiments with some open source projects which are originally from or popular in Japan such as Namazu and Ruby. The experimental results reveal that the extracted structures reflect very well important parts of the hidden structures of the programs.