From scenario to segment - the controlled elicitation, transcription, segmentation and labelling of spontaneous speech

Abstract

(Introduction) Recent Years have witnessed a growing interest in the controlled elicitation of spontaneous speech (Levelt, 1989; Swerts and Collier, 1992; Anderson et al., 1991). This method of data collection allows cetrain interactive and linguistic phenomena of spontaneous speech to be investigated without losing control over what participants in talk produce, as is so ofthe the case in recordings of free, unscripted talk. The degree of control can be caried. It can be relatively high, keeping a tight rein on the material produced. This is the case with the material gathered both by Levelt (1989) and Swerts and Collier (1992) where speakers are required to relate the position of coloured shapes in a network. Setups such as the map task (Anderson et al., 1991) are more complex involving two participants navigating across fictious maps: it is the job if one of the participants whose map contains a path, to describe this path to the other speaker. And it is a scenario of similar complexity to the map task, in the realm of appointment-making, that we employed to gather spontaneous speech in this study. We describe our efforts at IPDS Kiel to elicit spontaneous speech between two dialogue partners in a cotrolled fashion, together with a description of these steps involved in providing various symbolic representations of the verbal and nonverbal content of the elicited speech material. Some of the work in this study, in particular the conventions for segmenting and labelling speech signals, has its origins in extensive work carried out on the PhonDat90 and PhonDat92 databases. (Kohler, 1992, 1994a). The elicitation platform together with the development of conventions for the transliteration of the elicited dialogues as well as the necessary elaboration of conventions to deal with the segmental and prosidic labelling of spontaneous speech has been carried out in the framework of the Verbmobil project (Karger and Wahlster, 1994). We hope that this study will not only be of passive interest, but may serve in whole, or in part, as a manual for those who are considering embarking on a similar enterprise themselves.