CONCEPT
Understanding discourse structure in online text-based forums is often difficult. Posts generally refer back to previous comments (Whittaker et al 1998) and users must also keep track of multiple "threads," which may not have an obvious relationship through time or by subject header. Participants must keep track of any number of other participants by less intuitive means than a body or voice, such as tags or icons. Cues used by face-to-face interlocutors, such as gaze and interactional gestures are not available nor are the pauses and backchannel used in face-to-face and phone interactions. As a result of the limited information channel, it is often difficult to keep track of discourse structure in chat rooms and discussion groups. And, in addition, previous research on conventional discourse does not explain issues such turntaking, backchannel, and copresence in online environments (Cherny 1995.)

Because of lack of discourse research applicable to the online environment, limited theoretical grounding is available online interaction interface designers. Even with out this research, designers have created many novel and helpful interface ideas. For instance, designers may allow users to create icons or particular coloring schemes to represent themselves or to emphasize important points. In addition, users have invented novel methods of interaction to improve turn-taking and coherence. Participants may explicitly quote previous posts to create context. Still, many attempts to interact in online forums fail to produce any reaction from other users and sometimes require multiple iterations to establish mutual understanding (Whittaker et al 1998). One way to create more successful interfaces and extend the discourse research is to work on both problems simultaneously, building upon existing research in both fields. That this approach is still relatively novel may be a function of the lack of communication between the disciplines rather for theoretical reasons.

Current discussion group interfaces provide few interface cues that define discourse structure and little support for users to display structure themselves. Though some segmentation based on threads in discussions may evolve in the subject header, they are not formalized and may actually suggest an inaccurate discourse structurene. For instance, conversation may evolve away from the original concept listed in the header. That users often rename subjects with a new title and "was:" the former title suggests that this is the case.

As a result of the limited information channel, it is often difficult in chat rooms and discussion groups to keep track of the structure of the posts. Chat rooms may feel robust and alive but it is hard to relate different posts to each other without any other discourse cues. A system that kept track of a model of the participants shared plan of the intentions and goals for the discussion may help the lacking structure.

PROTOTYPE
I built a prototype of an interface for an asynchronous online forum based on Grosz and Sidner's discourse model and upon this researchers investigations of several discussion group interfaces. NOTE: this prototype only functions properly in Internet Explorer. It doesn't work in Netscape.) This interface provides a visualization of Grosz and Sidner's three components of discourse: intention, discourse structure and the current state of focus. The design of this interface allows users to define their own intentions and structure but could also support a computational system that assessed topic and discourse structure dynamically, as described above. It also could be used for synchronous or asynchronous forums.

The interface is composed of three primary windows. The first window is typical of conventional chat interfaces, a place to type in your post. Inspired by Grosz and Sidner's attentional component, the next window functions as a focus stack; it lists the most recent posts on each topic. Each topic corresponds to a color. When users wish to post a new comment, they can either select the "new topic" button or select an existing item in the focus stack. If they select a new topic, their post appears in the focus window highlighted in a new color. If they are following up on a previous topic, their post is listed at the bottom of the focus window and the old post is removed from the window. It moves to the posts window, where a history of the discourse sits. The discourse structure is visualized in two ways. First, the structure of topics can be identified through their corresponding colors. Second, each user is indicated by an icon, so it is easier to keep track of who said what. In addition users can view their posts in one of two ways: chronologically or by topic. The chronological method is the conventional way to view dialogue whereas the topic method allows users to follow particular threads that interest them while still being able to see the discourse as a whole through the focus window. This view of the discourse also provides a useful filtering method when users face the challenging task of finding a particular post or thread in an archive.

Without doing some user testing, it is unclear at this point whether it is better to list entire posts or just key ideas in the posts window. One future application of Grosz and Sidner's dominance theory may be to use intentions and their dominance, along with key words to determine which words should appear in the posts window.

[back to classes]