Semantic technologies for the production, dissemination, retrieval and browsing of multimedia information all require the generation of machine-processable descriptions of the meanings conveyed by the multimedia information, i.e. its content. Typically, metadata is generated by the automatic analysis of raw multimedia data and is then used to structure, index, summarise and browse multimedia data collections. Two fundamental questions arise. What forms should the descriptions of multimedia content take? And how can the descriptions be generated automatically? In this seminar I will argue that recent studies of narrative and multimodality have important contributions to make to the specification and design of data structures and algorithms for multimedia content analysis. After reviewing relevant theories and analytic frameworks from the fields of narratology, semiotics and multimedia discourse analysis, I will focus on findings from recent research into the extraction of narrative structures in feature films, and the modeling of image-text relations in web pages. Potential applications of this work include film retrieval based on story similarity and hypervideo based on story structures, and enhanced web search engines that fuse meanings from the image and text components of web pages.

By way of background… Narrative is an important concept because much multimedia data exists to tell a story, be it a feature film, a news story, or somebody’s life story realized in their personal media collection. To date however, most techniques for multimedia content analysis describe the topic, or aboutness, of media items in terms of the entities and events referred to / depicted, without describing the connections between them that make them into a story. That multimodality is an important concept for multimedia content analysis should go without saying – a defining characteristic of multimedia is the fact that the multimedia whole conveys more than the sum of its parts. However, much research has concentrated on the analysis of individual media types (image, video, text, audio) in isolation - their integration, if addressed at all, is ad-hoc. Recent advances in multimedia discourse analysis have produced insights into how different media types, e.g. image and text, combine to convey meaning in print, film, and new media forms.

Download presentation slides

Watch the webcast replay >>