I. Lokuge and S. Ishizaki. Geospace: An interactive visualization system for exploring complex information spaces. In I. R. Katz, R. L. Mack, L. Marks, M. B. Rosson, and J. Nielsen, editors, CHI95: Human Factors in Computing Systems, CHI 95 Conference Proceedings, pages 409–414, Denver, Colorado, USA, May 7-11 1995. ACM, Addison-Wesley. [pdf]
This paper present a seminal attempt to use the Spreading Activation Theory for information retrieval in multi-model user interfaces. The main reason for them to use this technique is that it follows the natural learning path of the user, which find it difficult to formulate his/her information seeking goal. Additionally, the advantage is that the ASN (activation spreading network) preserve’s the user previous state of exploration.
The system was tested with an interface called GeoSpace, that allowed the user to formulate vocal query on spatial features of the Boston’s area. The interface was conceived as a series of chained layers containing the relevant information. Each layer’s opaquity was regulated by the spreading retrieval engine.
Additionally, the author were trying to incorporate a simple learning mechanism that could associate different layers based on the retrieval’s order as suggested by the user.
3-4-2007 -> Geospace was commented by Frohlich  as a dynamic homefinder system you can talk to, since it combines natural language input with a map formalism output to five users access to a database of area and property information.
This form of mixed mode interaction shows how the characted of model workd or conversational interfaces can be altered by introducing elements of the complementary mode into the input or output interface.