Towards Best Practice in the Development and Evaluation of Speech Recognition Components of a Spoken Language Dialogue System Lori Lamel, Wolfgang Minker, Patrick Paroubek Spoken Language Processing Group LIMSI-CNRS, BP 133 91403 Orsay cedex, FRANCE October, 29th 1999 Keywords: Number of pages: 14 Abstract Spoken Language Dialog Systems (SLDSs) aim to use natural spoken input for performing an information processing task such as call routing or train ticket reservation (Lamel et al., 1995). The main functionality of an SLDS are speech recognition, natural language understanding, di- alog management, response generation and the speech synthesis. This article summarizes key aspects of the current practice in the design, implementation and evaluation of speech recognition components for spoken language dialogue systems. It is based on the frameworke used in the European project DISC. 1 Introduction Recent years have seen the development of an increasing number of Spoken Language Dialogue Systems (SLDSs), including both commercial and research systems (Peckham, 1993; Guss et al., 1998; ETRW 1999). Most SLDSs are designed to enable a dialogue between a human (user) and a computer (the SLDS) with no outside intervention of any kind, but when the dialog fails some systems provide operator fallback. This means that the system functionality requires not only an accurate transcription or recognition of the words uttered by the user but also the understanding of the utterances in the context of the application. In the end, such system must make a response as appropriate as possible, be it dialing the correct telephone number, making the correct train reservation or translating a sentence. In many cases several exchanges between the user and the computer are required justifying the term spoken language dialogue system. The DISC project (Dybkjaer et al., 1998; Bernsen et al., 1999) aims at building an in-depth de- scription of the state-of-the-art in SLDSs development and evaluation with the purpose of developing a first best practice methodology in this field accompanied by a series of development and evaluation This paper is based on research carried out within the ESPRIT 4th Framework LTR Concerted action projects 24823 and 29597 DISC - Spoken Language Dialogue Systems and Components Best Practice in Development and Evaluation. submitted to Natural Language Engineering 1