Error in natural language dialogue between man and machine

Abstract

A great deal of research has been done in the field of error processing, but the difficulty of the problem and the diverse ways in which it has been approached have made synthesis and methodological reflection necessary at this point. The present article begins by proposing a typology of errors in natural language dialogue between man and machine. The discussion of possible correction strategies that follows shows that the correction of user competence errors should take precedence over the correction of performance errors. The discussion also emphasizes that error prevention, although generally neglected, is a major way to avoid dialogue dead-ends, and should be taken into account in natural-language interface design. It is proposed that if dialogue systems cannot as yet be adapted to users, then they must be transparent and consistent enough so that users can adapt to them. As an illustration, the paper gives an overview of various methods developed from these methodological principles for correcting and preventing errors at the lexical, syntactic and semantic levels.