Second Messenger¹ is a collection of applications that reveal simple observations in group dynamics as they occur. The intention of each of these applications is to provide a way for a group to reflect upon its interaction in such as way as to encourage the group to consider a more diverse set of viewpoints in its discussion.

My interest is in building tools that support team collaboration and coordination, with a consideration of how mutual awareness of behavior increases the effectiveness and fluidity of the team's interactions.

Towards this, a core focus of this project is evaluating the impact of these applications on group behavior. Some of the motivating questions: How do new types of information displays become incorporated into a group's collaboration? How do these applications change the collective behavior of a group? Do these applications improve (or deteriorate) a group's decision-making ability?

Relevant Publications

Joan Morris DiMicco, Kate J Hollenbach, Anna Pandolfo, Walter Bender. (2007) "The Impact of Increased Awareness while Face-to-Face." Special Issue on Awareness Systems Design, Human-Computer Interaction. Volume 22 (2007), Number 1.

Joan Morris DiMicco. (2005) "Changing Small Group Interaction through Visual Reflections of Social Behavior." PhD Thesis, Massachusetts Institute of Technology, May 2005.

Joan Morris DiMicco, Walter Bender. (2007) "Group Reactions to Visual Feedback Tools." Full Paper, Proceedings of The Second International Conference on Persuasive Technology, April 26-27, 2007, Stanford, CA.

Joan Morris DiMicco, Kate J Hollenbach, Walter Bender. (2006) "Using Visualizations to Review a Group’s Interaction Dynamics." Extended Abstract, Conference on Human Factors in Computing Systems (CHI 2006), Montreal, Quebec, Canada, April 2006.

Joan Morris DiMicco, Kate J Hollenbach. (2006) "Visualization of Audio: A social tool for face-to-face groups." Social Visualization Workshop, Conference on Human Factors in Computing Systems (CHI 2006), Montreal, Quebec, Canada, April 2006.

Joan Morris DiMicco. (2005) "Evaluating Collaboration Technology Using User-Centered Design and Input-Process-Output Methodologies." Workshop on User-Centered Design and Evaluation of Services for Human-Human Communication and Collaboration, International Conference on Multimodal Interfaces (ICMI 2005), Trento, Italy, October 2005.

Joan Morris DiMicco, Anna Pandolfo, and Walter Bender. "Influencing Group Participation with a Shared Display." in Proceedings of the Conference on Computer Supported Cooperative Work (CSCW 2004). Chicago, IL, November, 2004.

Michael I. Norton, Joan Morris DiMicco, Ron Caneel, Dan Ariely. "AntiGroupWare and Second Messenger -- Simple Systems for Improving (And Avoiding) Meetings." BT Technology Journal. Vol 22, No 4, October 2004.

Joan Morris DiMicco. "Designing Interfaces that Influence Group Processes." Doctoral Consortium Proceedings of the Conference on Human Factors in Computer Systems (CHI 2004). Vienna, Austria, April 2004.

Joan Morris DiMicco, Walter Bender. "Second Messenger: Increasing the Visibility of Minority Viewpoints with a Face-to-face Collaboration Tool." ACM Conference on Intelligent User Interfaces (IUI'04). Madeira, Portugal, January 2004.

The Technology

The following sections present the Second Messenger applications:

1. Capturing and Amplifying Verbal Content

A common flaw of group decision making is ignoring or discounting non-majority-held opinions. To amplify minority voices and ideas, this application captures the words spoken in a face-to-face conversation, using voice recognition, and displays a filtered version of them based on what was said and who said it.

As the group converses, the phrase snippets from the conversation fall down the center of the screen and users can move the phrases around the screen to organize their ideas. Untouched snippets collect at the bottom of the display. Click on the image to view an AVI movie of the application.

The application uses IBM's ViaVoice to perform speech recognition and uses QTag to perform a part-of-speech filtering on the detected words. Each participant gets categorized by how much they are speaking, and depending on this categorization, more stringent filtering rules are applied to their words.

After building this application, the Second Messenger project went in two directions. The first was to explore a very different way of representing the verbal content of a meeting. This was done by my UROP Kate Hollenbach and we wrote a description of it here: Second Messenger: A display for face-to-face conversations. The other direction was to simplify and take away the verbal content, focusing on representations of behavior patterns....

2. Histogram Display
(with Anna Pandolfo, research staff member)

The simple display below shows how much each person has spoken during a meeting in relation to the other speakers. The display has dynamically adjusting histogram bars and, along the top of the screen, a visual representation of who has spoken over the previous 30 seconds. When the application detects that someone is speaking, a color-coded identifying circle moves across the top of the display and the bars on the histogram adjust by their relative percentage participation. Click on the image to view an AVI movie of it in action.

So this is very simple in concept, but how do you think it would make you behave in a meeting? Would you talk more or talk less? Or just ignore it? These were the questions we had when we ran a behavioral study to observe how people altered their participation during a decision-making task. In brief, the results showed that over-participators talk less and under-participators do not increase the amount they speak. Those with critical information for the task are unaffected by the display while those who have non-critical information decrease their participation.

3. Visualizations of Simple Group Dynamics
(with Katherine Hollenbach, undergraduate researcher)

The next step of this research has been to build more complex visualizations of the same information ("who spoke when"). We built an application that focuses on visualizing the speaking patterns of a conversation in many different ways, showing you who's overlapping who, who's contributing a lot of backchannel comments, and who's generally dominating.

We are giving these visualizations to both laboratory groups and real-world groups and observing how it changes their behavior and what they learn about themselves by using the application after a meeting. Stay tuned for more information on these studies!

Screenshots of the application (Histogram, Bouncing Balls, Group Circle, and Timeline):

 

 

 

 

If you have account access, you can read more about these visualizations in the August/September 2004 issue of Frames.

4. Private Display
(with Christopher Reed and James Houghton, undergraduate researchers)

One of the findings from the Histogram Display study was that introducing the display half-way through the group's interaction disturbed the group's trust dynamics. Also, certain subjects rated the accuracy much lower than others. This led us to wonder if a private display of similar information could be more useful to individuals than a public display. Here's a picture of some LED lights that act as private messaging displays to individuals in a meeeting.

 

The Research Contributions

The outcomes of this research are:

I. In a face-to-face setting, a dynamic display, positioned peripherally to a discussion, influences the behavior of the individuals in the group. When this display reveals information about participation in a meeting, the following was shown to occur:

• Those at the highest levels of participation decrease the amount they speak. Individuals who do not have critical information to contribute to the discussion decrease the amount they speak.
• Individuals who speak the least do not spontaneously increase their contributions, possibly because they are unaware they are contributing below than the average.

II. When a group watches a visualization of its previous interaction, there are multiple effects on the perceptions of the decision outcome, on individual behavior, and on the group-level decision-making process:

• Groups that anticipated seeing a replay of their interaction found the task more challenging than others. They also felt they had performed significantly better than those groups that did not anticipate this replay. After viewing the interaction visualization and performing a second task, groups felt they were significantly more efficient during the task than those who had not seen a replay. Individuals who spoke the least in the visualized interaction demonstrate a significant increase in contributions after viewing the visualization.
• Groups performing an information-sharing decision task demonstrate a change in decision-making strategy, which greatly benefits those that were ineffective at information sharing during the visualized task. Those that had a good information-sharing practice were harmed by their subsequent change in strategy.

III. When pre-existing groups incorporate a real-time display of participation into their regular meetings, they demonstrate consistent patterns week-to-week, and also a desire for change:

• Groups consistently state that they wish to hear more from the quietest individuals. The group members who consistently speak the most in meetings initially react to the display by speaking less, but over time express less desire to decrease their contributions.
• After having a display of participation present in multiple meetings, the quietest members of real-world groups were observed to speak more.

The implication of these findings is that providing any automated feedback to groups will cause them to adapt their behavior to accommodate to the normative pressure imposed by the feedback. In the case of the Second Messenger interfaces, the implied ideal of balanced participation led to an adjustment of participation behavior. Without mediating the communication and without explicitly instructing groups on how to change, individuals altered their personal behavior and this influenced their perceptions of the task and the group¬íˆÑˆ¥s decision-making strategy.

The challenge put forth by this implication is to find other measures of group behavior that can provide informative feedback to groups about their behavior in collaboration settings. The following section outlines different ways of approaching this challenge.