Toward a Taxonomy of Communication Errors

Researchers in a variety of disciples such as conversation analysis, second language acquisition, speech pathology, and computational linguistics have classified various types of communication errors that occur between humans, and between humans and computers. In this paper, we attempt to bring together insights from diverse disciplines into a common theoretical framework from which to compare and contrast various types of errors. We describe a classification scheme based on how people collaboratively resolve uncertainties by establishing the mutual belief that their utterances have been understood well enough for current purposes—a process referred to as grounding. The classification scheme not only highlights how different types of failures are related to each other through the common bond of uncertainty, but also provides guidelines for the design of error handling in automated systems.