The purpose of this chapter is to present in accessible form a set of recommendations for the specification, design and assessment of interactive dialogue systems in which spoken language plays a significant part. The chapter is designed with simplicity and clarity of presentation in mind; its primary focus is on practical recommendations which the reader can easily apply in dialogue system development and comparison.
Work in this area is still at a very early stage of development. The recommendations contained herein must therefore be regarded as no more than provisional, born of the experiences and informed intuitions of those who have contributed to the chapter so far. References to the supporting literature are provided where appropriate but the literature on interactive spoken language systems is still rather sparse.
The technical focus of this chapter is interactive dialogue systems, defined most generally as computer systems with which humans interact on a turn -by-turn basis. Our attention will thus be confined to those interactive systems in which ``natural'' language plays an important part in the communication process. Indeed, the recommendations are principally designed to apply to those systems in which natural language supports the majority, or the totality of the communication. More than this, we will focus particularly on those systems in which spoken natural language is the primary means of communication. Because there are many points of similarity between interactive dialogue systems based on spoken and written language, the recommendations seek to crystalise lessons learned from the written language domain. Also, some recommendations arising directly from experience in the spoken language domain can be expected to be portable to the text domain. However, readers whose primary interest is in text-based interactive systems should bear in mind that any material contained in this chapter which is of relevance to their concerns will not be marked explicitly as such.
Spoken language dialogue is, in many ways, the most challenging of all the language technologies because even the most basic dialogue system needs to subsume significant parts of most of the major fields of language engineering. The primary object of this chapter is to address issues which particularly arise when the various component technologies (such as speech recognition , parsing , database management, linguistic generation, speech synthesis, etc.) are integrated in a single functional system. Readers would do well to consider whether they require more detailed information relating to any of these component technologies. If so, other chapters in this volume, and other volumes in the EAGLES series, will provide good entry points to these fields.
In an interdisciplinary technical field such as interactive dialogue processing, the importance of mastering the terminology should not be underestimated. Technical terms are defined where they first appear in the text.