An Agent-based Architecture for a Multimodal Interface. in Working Notes of the Aaai-94 Spring Symposium on Intelligent

mouse, and speech for input, and audio and screen-based feedback for output. Multimedia presentation. The system makes use of a number of dierent media for human-computer interaction; in particular text, graphics, animation, and prerecorded video (the latter acting as a \cheap", but nevertheless very eective, substitute for in-oce video cameras and/or active badges 3) Communications-oriented capabilities. The system enables transparent communication across dierent computer platforms (Macintosh, Unix) and facilitates the interconnection of system users via telephone, email, and voice messaging. A number of features of the CALVIN architecture have proven useful for developing multimodal applications that integrate a number of distributed media resources. In particular, rapid responses to users' commands are facilitated through integration of appropriate reactive behaviors in the system's Interface and User agents (Ferguson and Davlouros 1995); in addition , blending of complementary input modalities is facilitated through the execution of multiple concurrent agents (which in turn are able to execute multiple concurrent, task-specic behaviors). Current work already underway includes porting the graphical user interface portion of the PeopleFinder to both PC and Unix platforms (in the interest of extending the tool's audience and ensuring a more thorough testing and empirical evaluation phase of the project); integrating a number of other software applications such as teleconferencing, voice dictation, and video camera-based face recognition; extending agents' capabilities for autonomously resolving run-time con BLOCKINicts resulting from shared access to the dierent presentation and communications resources used by the system (see Werkman's KBN negotiation-based con BLOCKINict resolution work for related issues (Werkman 1994)); and formalizing the various rules used by the PeopleFinder to combine multiple media with multiple modalities for both human-computer interaction and computer supported human-human communication, much along the lines of the work of Arens et al. (Arens et al. 1993) on allocating multiple media. The tool is implemented using a variety of dier-ent scripting languages (AppleScript, Quickeys, and C-shell) and runs on a Macintosh Quadra 840 AV. The tool also makes use of the Macintosh's Apple Phone tool and Geoport Telecom Adapter for performing its various computer-telephony integration tasks.