Postscript Version

STIMULATE: Human-Computer Communication and Collaboration

Barbara J. Grosz
Stuart M. Shieber

Division of Engineering and Applied Sciences
Harvard University

CONTACT INFORMATION

Engineering Sciences Laboratory
40 Oxford Street Cambridge, MA 02138
Phone: (617) 495-3673
Fax : (617) 496-1066
Email: shieber@eecs.harvard.edu
Email: grosz@eecs.harvard.edu

WWW PAGE

http://www.eecs.harvard.edu/~shieber

PROGRAM AREA

Adaptive Human Interfaces

KEYWORDS

Collaboration
Human-Computer Interaction
Multi-Modal Dialogue
Interactive Design
Context

PROJECT SUMMARY

Currently, the dominant metaphor in the design of human-computer interaction is the view of computer as factotum. Although interfaces based on this view have proved useful in a variety of settings, qualitative progress in the area of human-computer interaction awaits the ability to interact with computers as collaborators. Our research aims to provide the scientific and technological base for a new paradigm, one that enables the principled design of multi-modal dialogue-supporting interfaces.

When people communicate with a system, whether using some formal language (e.g., shell languages, query languages, menu-based systems) or a natural language, in so doing they execute actions. These user actions typically result in the system itself acting, and the resulting system actions are central to people achieving their goals. Thus, users participate in dialogues with systems about their goals and the task they want done. However, most current interfaces do not keep track of many features that are central to the success of dialogues in human collaborations; for instance, they do not use information about what a person is trying to achieve nor adequately model the dialogue context. As a result, they do little to support user-system collaboration on a task and most of the work of keeping track of the context of their communication is left to the user.

Our research is based on the results of two strands of previous research: (1) research towards the building of novel interfaces for particular problems in automated graphic design that exemplify and explore the view of communication with computers as a collaborative activity; and (2) the development of explicit theories of collaboration. It addresses a key missing element of prior research: an understanding of how theory can be integrated with practice so that our theoretical understanding of collaborative activity in the abstract can inform in a principled manner the design of concrete software interfaces, and thus greatly improve the dialogue capabilities, in multiple modalities, of current systems.

Our efforts center around three activities:

  • The implementation of concrete collaborative interfaces for particular problems, extending out from preliminary work on automated graphic design problems to a variety of other problems that joint human-computer efforts are needed to solve.
  • Specialization of the general theory and formalization of collaborative activity to provide a framework for the design of collaborative interfaces. The focus of this effort will be on analyzing the needs that arise in typical multi-modal communication situations, using the experience with implementation of interfaces for specific problems, and identifying a repertoire of system actions and processes that provide an appropriate foundation for interface design.
  • Integrating theory with practice to develop a set of design principles and conceptual tools that will enable system designers to design and construct software elements that communicate well in multiple modalities because they have the capability to serve as collaborative participants in solving problems.

    For one of our first concrete collaborative interfaces (developed before this project was started), we chose to address the problem of network diagram layout. Network diagrams are a common form of informational graphic, and constitute an area in which it may be difficult for a person to define the aesthetic qualities desired in the final design. Most research in network diagram layout has focused on automatic layout. We have created a collaborative, interactive constraint-based editor for network diagram layout, GLIDE, which improves upon general constraint-based editors by providing a small but powerful vocabulary of specialized constraints specifically designed for drawing network diagrams. GLIDE is the first system to support people in interactively specifying the visual organization of a diagram, and currently supports thirteen types of constraints, including: Alignment, Clustering, Zone, Even Spacing, Symmetry, T-Shape, and Hub-Shape. Using the collaborative approach, we take advantage of a user's expertise at globally designing the layout, and the computer's computational superiority; the user is responsible for an approximate layout of the nodes and for specifying any desired visual organization. The system then attempts to solve a constraint-satisfaction problem by calculating a local minimum using a mass-spring physical simulation. The process is intended both to be interactive and iterative. GLIDE provides simple interface mechanisms for the user to create, view, manipulate and remove constraints. Under this paradigm, the user has the flexibility to create interesting designs, without the burden (and tedium) of having to precisely place every object in the layout. It exemplifies the kind of system we will build under the first type of activity.

    As an initial effort to use a general theory of collaboration in developing an interface (and in coordination with the Ubiquitous Information Project at Harvard), we are developing a collaborative interface testbed -- named the Distributed Information Access for Learning (DIAL) testbed -- to support distance learning and multi-media information access to students who are geographically distributed. Using DIAL, students will be able to access a variety of sources of course information over the WWW such as: electronic textbooks, lecture notes, assignments, exams, solutions, library reference materials, technical reports, videos of lectures, and video links to teaching assistants and professors. The current testbed system provides students with access to a web server which maintains a hierarchical context reflecting embeddings of subgoals from previous interactions with the student. The context is then used to interpret subsequent information queries: information access is thereby modelled in terms of a sequence of successively refined queries, instead of requiring the student to enter precisely what he is searching for as in, for example, conventional web searching tools. The interface is also collaborative in the sense that the student need not know all of the details regarding locations of pertinent information nor means of accessing them. A number of elements from the SharedPlan formalism (Grosz & Kraus, 1996, 1997) are explicitly represented within the system.

    PROJECT REFERENCES

    Barbara J. Grosz and Sarit Kraus, "Collaborative plans for complex group action." Artificial Intelligence. 86 (1996) 269-357.

    Barbara J. Grosz and Sarit Kraus, "The Evolution of SharedPlans." Foundations and Theories of Rational Agencies. A. Rao and M. Wooldridge, eds. (1997) To appear.

    Kathy Ryall, Joe Marks and Stuart Shieber, "An Interactive System for Drawing Graphs", Proceedings of Graph Drawing '96. Stephen North, ed., [Springer Verlag, Berkeley, CA Series: Lecture Notes on Computer Science] 1190 (1996) October.

    Kathy Ryall, Joe Marks and Stuart Shieber, "An Interactive Constraint-Based System for Drawing Graphs." Proceedings of the 10th Annual Symposium on User Interface Software and Technology [UIST] (1997) To appear.

    AREA BACKGROUND

    Together, computers and humans can often find solutions to problems that neither could have discovered alone. To construct systems that can work together with their users requires an understanding of, and formal models for, collaboration (Grosz, 1996 contains background references). Work on collaboration and dialogue systems (e.g., Lochbaum, 1995; Rich and Sidner, 1997) has shown the usefulness of such models. In addition, systems designers must address the question of how best to balance user intervention and computer control (Wegner, 1997; Grudin, 1990, 1993). Traditional interface design has relied upon a master-slave framework, in which the interface is a means for people to control computers. Under this approach, the division of labor is such that the user issues commands to the computer; the computer plays a passive role, responding to the user's actions. Systems based on this framework range from automatic (in which the user sets parameters, and the machine computes "the answer") to manual (in which the computer merely records the actions of its user). Little attention has been given to the middle ground, or semi-automatic systems, in which the interface would be a means for the computer and its user to work together on solving some problem. Such a collaborative approach to interface design would delineate the roles of computer and user and would exploit the strengths of each collaborator. People draw upon a broad range of experience and often rely upon intuition when making decisions. Computers, on the other hand, have a narrow area of expertise, and in most cases greater computational power. A person's knowledge (and preferences) can be used to guide and help the computer come up with an acceptable solution.

    AREA REFERENCES

    Barbara J. Grosz, "Collaborative Systems: 1994 AAAI Presidential Address." AI Magazine, 2 (1996) 17 Summer.

    Jonathan Grudin, "Interface: An Evolving Concept." Communications of the Association for Computing Machinery, 36 (1993) 4 April [Special Issue on Graphical User-Interfaces] 110-119.

    Jonathan Grudin, "The Computer Reaches Out: The Historical Continuity of Interface Design." Human Factors in Computer Systems (CHI), April (1990) 261-268.

    Karen E. Lochbaum, "The Use of Knowledge Preconditions in Language Processing." Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI-95) Chris S. Mellish, ed., [Morgan Kaufmann Publishers, Inc. San Mateo, CA] 2 (1995) 1260-1266.

    C. Rich and C. Sidner, "COLLAGEN: When agents collaborate with people." In Proceedings: 1st International Conference on Autonomous Agents, Marina del Rey, CA (1997) February, 284-192.

    Peter Wegner, "Why Interaction is More Powerful than Algorithms." Communications of the Association for Computing Machinery, 40 (1997) 5 May, 80-91.

    RELATED PROGRAM AREAS

    Speech and Natural Language Understanding, Other Communication Modalities, Intelligent Interactive Systems for Persons with Disabilities.