An Active Walk through Semantic Space (Poster)
Teachers College, Columbia University
Last modified: March 30, 2006
When moving through a physical space, human beings marks the landscape with physical footprints. When moving through a semantic space, human beings mark the "landscape" with cognitive footprints: scientific papers, computer code, as well as poems and musical compositions. Subsequent walkers can choose to forge new paths or follow in the footsteps of their predecessors. Whether moving through a physical or semantic space, the walker incurs lower costs by choosing the latter option: foot trails ease one's physical effort, and cognitive trails ease one's meaning-making effort. This choice amplifies the trail. Oft-travelled paths become pedestrian thoroughfares, scientific paradigms, or musical genres. Taken as a whole, the system of trails--whether physical or semantic--serves as a collective memory from which and upon which successive walkers read and write information about the space.
Physical trail systems have been well described by the Active Walker Model; the model can account for patterns in river formation, the movement of heat-seeking missiles, as well as the pheromone trails of ant swarms and the foot trails of human crowds. This study explores the extent to which the Active Walker Model can account for pattern of cognitive trails in a semantic space. Specifically, I consider the semantic space created during the course of ten MATLAB programming contests, where participants freely viewed, reused, and extended each other's computer code in order to optimize ten algorithmic problems. I combine Latent Semantic Analysis and Multidimensional Scaling to induce and map the contours of the semantic space. This map, together with data about the paths taken by contest participants, serves to parameterize the Active Walker Model. I compare the pattern of trails formed and followed by simulated walkers to the pattern formed and followed by contest participants. A good fit would reveal deep ties between group-level cognitive processes and other natural and artificial processes. Through further investigation, one could use the Active Walker Model to better understand how human groups generate and optimize meaning.