The Language of Television News and Debates: Some Observations Oriented Towards Respeaking

Abstract

Respeaking is a live subtitling method not yet practised in Turkey. A respeaker dictates the verbal content of a live TV program in reduced form to an automatic speech recognition engine which renders a written text to be post-edited on a subtitling component. This study presents a number of observations on those aspects of spoken Turkish which may be relevant to prospective respeaking practices for TV news and debates as well as on possible strategies to be employed while transforming the original content into subtitles. A qualitative analysis of a transcribed audiovisual corpus (total duration 76 min. 6 sec.) comprising clips from news/economics programs and debates has shown that the following factors might influence text reduction during respeaking: sentence length, word order and ungrammaticality/disfluency on syntactic level; mood, modality and speech act expressions on semantic/pragmatic level and elements with low information content (e.g. fillers, redundantly used discourse markers) on discursive level. An evaluation of the data with respect to two text reduction strategies, namely omission and reformulation, has suggested that omission may be prioritized in order to prevent latency.

Keywords

• live subtitling
• respeaking
• automatic speech recognition
• omission
• reformulation