Many diverse sources can contribute to the successful interpretation of sensory input. One fundamental problem for perception is integrating these sources into the interpretation process. We present a hybrid methodology for perception that addresses this integration problem. The approach integrates special-purpose analogical representations used to store partially interpreted sensor data with a general-purpose sentential language employed to represent more coguitively-based information about a domain. The paper describes a formal apparatus for unifying the sentential and analogical representations as well as inference mechanisms for translating between the two subsystems. We exhibit the utility of the framework for the task of integrating contingent information into maps.