Traditional approaches to information extraction assume that many elements of the task are static --- the user’s query, and the description of domain and corpus, for example. We seek to extend current approaches for information extraction to handle dynamic elements of the problem. We consider such issues as: - appropriate "modalities" for an interface to large amounts of natural language text; - appropriate opportunities when the system or user should take initiative to interact with one another or to modify the current state; - features for knowledge representation; - features for machine learning. We intend to investigate how such choices affect performance. Specifically, our initial focus will be on how best to support evolving information models (query and domain) interactively.