next up previous
Next: 5 Observations on Grounding Up: 4 Experimental Setting Previous: 4.3 Collected Data

4.4 An Example of Grounding and Miscommunication

  We have noticed many examples of different sorts of miscommunication which could be classified according to the source of the problem in Table 1. Table 2 shows an example (from our warmup task) of one a persistent source of miscommunication that shows up in several of our dialogues: the relation of MOO location and communication.

One of the most basic sources of miscommunication in our setting is the fact that not all messages are noticed by the partners (a problem at level 2 from Table 1). Unlike a face-to-face setting, where it is fairly easy to see whether someone is attending to a message, one cannot determine this in the computer-mediated setting without feedback - either explicitly through some sort of acknowledgment or repair, or implicitly, by using the knowledge in some way. Although it is not necessary that the message be noticed immediately (level 1 is satisfied, since messages are accessible), since both the MOO and whiteboard messages persist for some time, one can not always be sure that the collaborator has registered the message. In the MOO, level 1 can also be an issue, since some commands (including say) will only deliver a message to someone in the same MOO room. There are a number of cases in which someone uses say rather than page when the partner is in another room and the message never reaches the partner. There are two reasons why this can happen, either the actor simply uses the wrong command, or the actor does not realize that the partner is not in the same room. Sometimes this is noticed and repaired by the actor, by simply repeating the message as a page, or, pre-emptively, by following or summoning the partner to reestablish the ``physical'' proximity. Other times, however, the actor does not seem to notice that the message was not sent.

In some cases, such as simple reports of what the actor will do next, the situation is not repaired, making it hard to tell whether the actor just decided that it wasn't necessary to have this information be part of the common ground, or whether there was an actual misunderstanding about the transmission. In other cases, the actor actually requests some reply from the partner, querying why there was no response. Sometimes this is then the source of a further misunderstanding - the partner, who never received these messages, assumes that the actor was talking about something else and ends up apologizing.

We can see much of this kind of behavior in the excerpt in table 2. The dash in the args column indicates say commands that were not heard by anyone else. The who command tells the actor which room the players are in and is thus a pro-active way of assuring that say will work. We also see a move by H at 12:47, so he can talk to S, as well as a repair from say to page at 17:27. H's answer, while perhaps accurate with respect to his immediate activities (drawing on the whiteboard) does not really address S's point - the last successful MOO communication by H was at 3:23, and since then, Sherlock has sent 4 previous messages while H made 4 unsuccessful attempts.


next up previous
Next: 5 Observations on Grounding Up: 4 Experimental Setting Previous: 4.3 Collected Data

David Traum
Thu May 23 21:12:30 MET DST 1996