We also identified a group of "helper" words we call "approximations" (e.g., slightly, near, almost, just). They seem to express categorization of a stimulus by opposition to another category (i.e. "not cardinal") and their use predicted recall bias away from the diagonals.
