Generating Multilingual Text for Video Documentaries

A significant challenge for NLG is integrating generation technology with real world applications, especially multimedia and multimodal applications which have much more stringent synchronization constraints. We have implemented a system which creates narrative text that accompanies artificial documentaries (either as subtitles or to be synthesized for voiceovers), involving coordination problems for ordering and structuring the narratives in synchrony with an image animation component. In this paper we present a flexible, multilingual generation architecture for producing documentary narratives that allows us to tackle the problems related to the selection and organization of content to be coupled with visual feedback, the selection of referring expressions that correctly identify and describe visual details, and the linguistic realization in different target language