Live subtitling with speech recognition. Causes and consequences of text reduction

Abstract

Speech technology has made it possible to use speech recognition for the simultaneous subtitling of live television broadcasts using the technique of respeaking. Analyses show that live subtitles, like pre-recorded subtitles, are nearly always a reduced form of the spoken comments. However, the live-subtitling process in itself may have an effect on the reduction strategies used by live subtitlers. The aim of this study is to gain a better understanding of the causes and consequences of quantitative text reduction in live subtitling. Three excerpts of an infotainment talk show were subtitled by twelve respeakers of the Flemish public television channel, VRT. They were instructed to complete the task using three different reduction conditions. Various subtitle features, such as reduction percentages and delay, as well as measures of the respeakers’ working memory were collected. In a hierarchical multilevel analysis we defined which external factors affect the degree of reduction. The results show that reduction is not a random process. In contrast, its occurrence and form are largely determined by a number of external factors, viz. delay, amount of source text and the proportion of ‘full’ deletions. A large volume of evidence suggests that respeakers opt to omit certain comments rather than reducing them. It also appears that the decision to delete a comment seems not to be primarily based on the amount of input, while the decision to reduce partially is.