If there's one assumption that comes naturally to cognitive scientists, it's that mental activities can be divided into parts. When these psychologists try to explain some complex activity—say, reading or problem solving—they typically do so by specifying the components that make it up. An analysis of reading might consist of a description of individual steps like word recognition, parsing, and interpretation, together with an account of how these steps are temporally and causally combined. An analysis of problem solving might proceed by breaking the process into subparts such as encoding of the given information and the goal, assessing differences between givens and goal, and applying operators to reduce the differences. This type of analysis might be repeated with respect to the parts. For example, we might decompose word recognition into letter encoding, memory access, and decision. At some point in this successive partitioning we reach the level of atomic mental activities—"elementary information processes," as Newell and Simon (1972) call them—which cannot themselves be decomposed within a cognitive framework. Exactly which activities one takes as atomic depends on one's theory, but they might include simple operations of adding a new piece of information to a location in memory, comparing one piece of information to another, and issuing motor commands.