AbstractThe expansion of the information highway has generated requirements for more effective access to global and corporate information repositories. These repositories are increasingly multimedia, including text, audio (e.g., spoken language, music), graphics, imagery, and video. The advent of large, multimedia digital libraries has turned attention toward the problem of processing and managing multiple and heterogeneous media in a principled manner, including their creation, storage, indexing, browsing, search, visualization, and summarization. Intelligent multimedia information access is a multidisciplinary area that lies at the intersection of artificial intelligence, information retrieval, human computer interaction, and multimedia computing. Intelligent multimedia information access includes those systems which go beyond traditional hypermedia or hypertext environments and analyze media, generate media, or support intelligent interaction with or via multiple media using knowledge of the user, discourse, domain, world, or the media itself. Providing machines with the ability to interpret, generate, and support interaction with multimedia artifacts (e.g., documents, broadcasts, hypermedia) will be a valuable facility for a number of key applications such as videoteleconference archiving, custom on-line news, and briefing assistants. These media facilities, in turn, may support a variety of tasks ranging from training to information analysis to decision support. In this talk I will describe our group’s efforts to provide content based access to broadcast news sources, including our use of corpus-based processing techniques to the problems of video indexing, segmentation, and summarization. In addition to better access to content, we also need to concern ourselves with enabling more effective, efficient and natural human computer or computer mediated human-human interaction. This will require automated understanding and generation of multimedia and demand explicit representation of and reasoning about the user, discourse, task and context (Maybury 1993). To this end, I will describe our work in progress that aims to fully instrument the interface and build ( automatically and semi-automatically) annotated corpora of human-machine interaction. We believe this will yield deeper and more comprehensive models of interaction which should ultimately enable more principled interface design.