Information Theater versus Information Reenery Where Intelligence Resides Information Theater

Intelligent behavior of a system is often equated with its ability to perform logical or heuristic inferences. Intelligence is thought to reside in the ability to reene the knowledge base, and thus in the articulation of \new" facts beyond the set of assertions explicitly represented. The fragility and complexity of the inference mechanisms required for this invite attempts at imbuing other components of the system with \intelligence". We have chosen to focus on the intelligence required to locate knowledge instead of that required to infer facts. The problem we therefore start with is: what should one present in response to a query, and what sort of information can be reliably extracted from running text? Consider a text-based intelligent system that does not reene text into something resembling facts, but rather presents and possibly annotates interesting text passages for perusal by the user. This distinction between fact-inference and text presentation means replacing the requirement for deep understanding by the more feasible goal of text selection. It also implies a certain degree of forgiveness: the system is not required to be perfectly accurate since a person ultimately assists in the interpretation. Pursuing systems that are capable of dealing intelligently with texts demands consideration of (1) the role of the user; (2) scale of data requirements; and (3) the interplay between general and specially-tailored processing techniques. We will discuss each of these considerations, and set forth a vision of the \informa-tion theater," drawing on our experience in implementing systems for intelligent information access. The end product of information theater is texts or fragments of text arranged in a way meaningful to a user. High-bandwidth interaction greatly innuences the nature of the underlying search and analysis algorithms. This has been brought out in our experience with the development of a text-access paradigm that weaves together interface, presentation and search in a mutually reinforcing fashion 3, 1]. Second, because text originates in paper as well as in electronic form, the documents may have physical attributes which are as important as their content. Consequently, as much as possible of the original appearance of the document is to be retained, including layout structure and relative positioning of text-pieces. Third, in order for information theater to provide sig-niicant beneets beyond what is ooered by, say a collection of books, its reach must include a vast and varied range of material. Indeed, small scale corpora are uninteresting from our …