Talking to machines is one of the dreams and visions that shaped our collective imagination about the future of the computer era. Although the research areas of speech recognition, speech synthesis and natural language processing are probably as old as computer science itself, only today are we observing a modest but relentlessly increasing impact of conversational machines into everyday life. In spite of the impressive advances observed in the related disciplines during the past decade, automatic spoken dialog systems are still fairly primitive when compared to the spoken language capability of humans. One of the reasons is that spoken language communication is an extremely complex process that involves several layers of knowledge, from acoustic to semantic and pragmatic. In this talk I will show why spoken dialog systems are difficult to build, what is the science behind them, why some of the past attempts failed, and what are the most successful technical solutions adopted today. Finally I will describe how the increasing success of today's system is due not only to the solution of the related technical problems, but also to a better understanding of the human issues behind the design of speech user interfaces.

Roberto Pieraccini received the Dr.Ing degree in electrical engineering from the Università degli Studi di Pisa, in Pisa, Italy, in 1980. From 1981 to 1990 he was with CSELT (Torino, Italy), where he conducted research on algorithms for speech recognition. In June 1990 he joined the Speech Research Group at AT&T Bell Laboratories (Murray Hill, NJ) and was involved in stochastic methods for language understanding. During that period he contributed to the novel idea of stochastic language understanding. In 1995, he joined AT&T Shannon Laboratories (Florham Park, NJ) where he was Principal Member of Technical Staff and worked on new concepts for spoken language dialog. In particular he contributed to the development of a new mathematical formalization of dialog systems within the reinforcement-learning paradigm. Since November 1999, he is with SpeechWorks International, where he leads the Natural Dialog group, a NY-based team devoted to the advancement of the spoken dialog system technology. The group is currently active in the fields of natural language processing, dialog management and multi-modal dialog techniques. During his career, he authored and co-authored more than 100 publications on the subjects of speech recognition, language modeling, language understanding and dialog.