Chunking

This project is developing a mathematical and algorithmic account of chunking, a ubiquitous mental phenomenon that causes continuous phenomena to be perceived and stored in memory as discrete chunks. A simple rule seems to account for chunking in one (and probably two) dimensions: Chunks are subsequences (or regions) that have low internal entropy and high entropy at their boundaries. Various algorithmic implementations of this rule have produced good chunking behavior in several domains, including finding boundaries in text, finding boundaries in robot behavior, and identifying two-dimensional characters. Chunking appears to work hierarchically, so that the chunks discovered at one level are themselves organized into higher level chunks. For an overview, see the segmentation paper below.