An anatomy of graceful interaction in spoken and written man-machine communication

The r e have recent ly been a number of attempts to provide natural and flexible inter faces to computer systems through the medium of natural language. While such interfaces typ ica l ly pe r f o rm wel l in response to straightforward requests and questions within their domain of d i scourse , they often fail to interact gracefully with their users in less pred ictab le c i rcumstances. Most current systems cannot, for instance: respond reasonably to input not con fo rming to a rigid grammar; ask for and understand clarification if their user's input is unclear; o f fe r clarif ication of their own output if the user asks for it; or interact to reso lve any ambiguit ies that may arise when the user attempts to describe things to the system. We be l ieve that graceful interaction in these and the many other contingencies that can ar i se in human conversat ion is essential if interfaces are ever to appear cooperat ive and he lp fu l , and hence be suitable for the casual or naive user, and more habitable for the expe r i enced user. In this paper; we attempt to circumscribe graceful interaction as a f ie ld for s tudy , and ident i fy the problems involved in achieving it. T o this end we decompose graceful interaction into a number of relatively independent ski l ls: skil ls involved in parsing elliptical, fragmented, and otherwise ungrammatical input; in ensu r ing robust communication; in explaining abilities and limitations, actions and the motives beh ind them; in keeping track of the focus of attention of a dialogue; in identifying things f rom descr ipt ions, even if ambiguous or unsatisfiable; and in describing things in terms app rop r i a te for the context. We claim these skills are necessary for any type of gracefu l in teract ion and sufficient for graceful interaction in a certain large class of appl icat ion domains. None of these components is individually much beyond the current state of the art, and w e outl ine the architecture of a system that integrates them all. Thus, we p ropose g race fu l interact ion as an idea of great practical utility whose time has come and wh ich is r i pe for implementation. We are currently implementing a gracefully interacting system along the l ines presented; the system will initially deal with typed input, but is eventual ly intended to accept natural speech.