| 1 | Basic facts about the task (directly available from performing MOO actions) |
| 2 | Inferences about the task (derived from the directly presented facts) |
| 3 | Aspects of the problem solving strategy |
| 4 | MOO positions of the collaborators |
| 5 | Knowledge representation conventions used in Whiteboard diagrams |
| 6 | Mode of interaction itself (who communicates when and how). |
time place who action args said text typed command 1:10:05 Priv S *page* H oscar could have done it because he left the kitchen at about 10:30 ' oscar could have done it because he left the kitchen at about 10:30 1:10:17 1 H *page* S Yep, that could be, but did you check weapon and opportunity ? page sherlock Yep, that could be, but did you check weapon and opportunity ? 1:10:46 Priv S *page* H and he could have taken the weapo n because i5t was left in the bar ' and he could have taken the weapo n because i5t was left in the bar 1:12:20 Priv S read jacq fom dn1 1:12:23 1 H *page* S The weapon was in the bar ? page sherlock The weapon was in the bar ? 1:12:26 Priv S *read* dn1,JS read jacq from dn1 1:12:56 1 H *read* dn2,all read all from dn2 1:13:55 Priv S read gun from dn1 1:14:24 Priv S *read* dn1,KVS read kol from dn1 1:15:02 1 H *moveto* 1,LC join sherlock 1:15:03 Priv H *moveto* LC,Priv 1:15:20 Priv H *say* S Well Sherlock, it's all a bit confusing say Well Sherlock, it's all a bit confusing 1:15:32 Priv S *page* H sorry I misunderstood what the kolonel said. the gun was in his room ' sorry I misunderstood what the kolonel said. the gun was in his room 1:15:43 Priv H *say* S Yes, the gun was his say Yes, the gun was his 1:16:00 Priv H *say* S Can you show me your notebook? say Can you show me your notebook? 1:17:10 Priv S *show* dn1,H show dn1 to herc 1:17:28 Priv H *say* S Can we quickly exchange our notebooks ? say Can we quickly exchange our notebooks ? 1:17:39 Priv S *show* dn1,H show dn1 to herc 1:17:42 Priv H *give* dn2,S give dn2 to sherlock 1:18:07 Priv H *say* S I was unable to read your note book. Please give it to me say I was unable to read your note book. Please give it to me 1:18:10 Priv S *read* dn2,all read all from dn2 1:19:45 Priv H *say* S Dear Sherlock, please give me your note book for a moment. say Dear Sherlock, please give me your note book for a moment. 1:21:04 Priv H *page* S I don't see the reason why Claire couldn't be the murderer page sherlock I don't see the reason why Claire couldn't be the murderer 1:21:17 Priv H *look* Priv look 1:21:29 Priv H *say* S Are you here ? Can you hear me ? say Are you here ? Can you hear me ? 1:23:27 Priv S *read* dn1,JS read jacq from dn1 1:23:36 Priv H *page* S Probably you are reading my note book. Please hand me yours so that I can also read page sherlock Probably you are reading my note book. Please hand me yours so that I can also read 1:23:55 Priv S *page* H i thought i did ' i thought i did 1:23:59 Priv H *look* H look me 1:24:12 Priv S *show* dn1,H show dn1 to herc 1:24:29 Priv H *say* S Please type: give dn1 to hercule say Please type: give dn1 to hercule 1:24:44 Priv S *give* dn1,H give dn1 to herc 1:24:50 Priv H *say* S Thank you say Thank you 1:24:59 Priv H *read* dn1,all read all from dn1 1:27:51 Priv S *page* H claire didnt leave the room to kill
ML. marie saleve left at 8:30, when the colonel wasnt in his room, so
she could have stolen the gun. Then at 10.00, on her way to the bar
she could have killed ML. The motive is the insurance:she knew about
it. ' claire didnt leave the room to kill ML. marie saleve left at
8:30, when the colonel wasnt in his room, so she could have stolen the
gun. Then at 10.00, on her way to the bar she could have killed
ML. The motive is the insurance:she knew about it. 1:28:20 Priv S *moveto* Priv,LC walk to lobby 1:28:21 Lobby S *moveto* LC,Lobby 1:28:31 Lobby S ask marie about lisa 1:28:47 Lobby S ask marie about insurance 1:29:40 Priv H *page* S But what did you mean when you said the gun was at the bar? Who brought it there from the Colonels room? page sherlock But what did you mean when you said the gun was at the bar? Who brought it there from the Colonels room? 1:29:45 Lobby S *page* H Marie just admitted that she kenew something was wrong with the insurance ' Marie just admitted that she kenew something was wrong with the insurance 1:30:21 Priv H *page* S What's wrong with the insurance. The painting is a fake. Do you mean that ? page sherlock What's wrong with the insurance. The painting is a fake. Do you mean that ? 1:30:42 Lobby S *page* H i was wrong when i said it was at the
bar. it was in colonel's room ' i was wrong when i said it was at the
bar. it was in colonel's room 1:33:02 Priv H *page* S Ok page sherlock Ok
The most basic information is that provided directly from the MOO, either through observations of rooms or objects, reading notes, or as suspect's answers to questions. This information is generally fairly easy to understand and not subject to disagreement. Even so, miscommunication can still take place. Table 4 includes an example, which also illustrates the mechanisms for repair. At 1:10:46, S conveys inaccurate information about the position of the gun, (because of either misreading or mis-remembering an answer by a suspect). When this information is questioned (1:12:23), S checks his data (1:14:24) by consulting what the kolonel told him about the gun. S clears up his own misconception, reading the correct answer, and tries to repair the miscommunication at 1:15:32. However, H misunderstood this repair, acknowledging the gun ownership but failing to register the information about the position. This leads to the later request for clarification at 1:29:40.
These facts are always accessible to the other agent (through the same means that the first agent learned it), but it is probably not already perceived/known unless the other agent is (or has been) in the same MOO room. The grounding criterion for these facts varies greatly, both with respect to the facts themselves, and depending on what else is known or suspected at the time. Early on, many potentially important facts are put on the whiteboard, both to ground them and to keep them accessible for combining with other information. Later on, only facts which are important for confirming or disproving the current hypothesis are communicated via the MOO, while others are left uncommunicated (and ungrounded), or made accessible implicitly through the notebooks.
Inferences about the task generally have a much higher grounding criterion - they are steps along the path from the basic facts to the final solutions. They are generally communicated through MOO messages, through whiteboard postings, or through combinations of the two. Unlike the direct facts, there is room here for negotiation and disagreement about the inference itself, although understanding what the inference itself is generally straightforward. The excerpt in Table 4 shows a couple of inferences as well. The same misunderstanding described above also includes a miscommunication about an inference. Knowing that the gun was in the Kolonel's room, H deduces from S's mis-statement that someone brought the gun to the bar. Other expressed inferences serve as points of negotiation through which misconceptions and missing information can be cleared up. This room for negotiation sometimes increases the cost of (successfully) conveying inferences.
Another locus of grounding is the problem solving strategy - who
should do what, and when, in order to gather the information and make
the deductions to solve the mystery. Collaborators can work in a
number of different styles, including close collaboration - finding
and discussing and the information together, or a more distributed
strategy - splitting up the task of inspecting the rooms and
interrogating the suspects, while meeting to discuss only at
intervals, or even non-collaboratively.
Often the strategy changes several times during the
collaboration process. While explicitly grounding the strategy can
lead to more effective collaboration, the grounding criterion is
fairly low, since mis-coordinated strategies do not generally harm
the solution process. Thus we often notice pages describing what a
subject will do next, but if this is not acknowledged, it is rarely
repeated or repaired. Often individual sub-strategies can be
inferred by noticing what information and inferences have been
expressed, both on the whiteboard and through the MOO.
The excerpt in Table 4 shows several shifts in strategy. First, as this excerpt begins, the subjects are moving from a data-gathering phase to a discussion and analysis phase. As the conversation proceeds (and increases in difficulty), Hercule decides to reduce the costs of MOO communication by joining Sherlock in the Private residence (1:15:02). Then Hercule suggests that they each review the facts collected by the other, rather than just introducing them as relevant to the inferences (1:16:00, 1:17:28). This plan leads to miscommunication, however, as Sherlock is not sure how to do this. The first attempt, (1:17:10) merely shows H what the notebook looks like, but does not convey the information recorded inside. H repairs, first by asking for the correct operation, and then more explicitly describing the action (1:18:07). When S merely repeats the showing, H gets more direct with the requests, checking also to see if contact has been lost (1:21:17,1:21:29). At S's reply, H checks to see if somehow he did have the notebook (1:24:12), and finally gives S even more explicit instructions, showing exactly what should be typed to realize the command. As H finally reads the info from S's notebook, S decides (from his prior viewing of H's notebook) to go back and gather more data (making direct discussion temporarily more difficult).
As described in Section 4.4, and above, the relative MOO position is sometimes important for the success of certain commands (e.g.,say, give), or for facilitating easier conversations. The grounding criterion for position is relatively low, since much of task performance can proceed without this knowledge. Also, the current groundedness is generally fairly high - several commands (such as page) provide information about where the actor is. The only tricky point is that this information is fairly transient, as players can move to a different room at any time. The cost of grounding this information is also fairly low, often it is provided implicitly. Also, a single short MOO command will suffice to determine positions (who) or move to the same room as the other (join).
In these experiments, we have left the organization of data on the whiteboard completely at the discretion of the subjects. Even with a limited tool such as this whiteboard, there is still a wide range both in how the whiteboard is used, and how meaning conventions can be expressed. For instance, color, arrows, size, and absolute and relative position of objects can all be used to denote meaning, given a suitable convention (which may be implicit). Some pairs use the whiteboard mainly as a repository for ``post-it'' style notes, while others draw conceptual graphs of social relationships or time-tables indicating the suspects' locations at the relevant times. Negated information can be indicated either by removing it, or by covering with crossed lines.
The grounding criterion for the conventions depends on the complexity of the displayed information, as well as the grounding criterion of the information, itself. For straight reporting of facts, the criterion can be fairly low, but for organized data such as in a chart, the conventions must be understood if it is to be of use. The cost depends on how the conventions are grounded - whether through MOO discussion, posting a legend on the whiteboard, or implicitly, through use of the conventions. The prior groundedness depends on how ``natural'' the convention is: conventions such as overlap of text boxes for a functional relationship between the entities, or overlaying with crossed lines to indicate removal from consideration require little (if any) explicit grounding. On the other hand, use of specific color codes to represent information types requires more explicit grounding if it is to be used consistently by both collaborators. Earlier in the same dialogue as that from which Table 4 is excerpted, the subjects established a color code for information type, in which green indicated opportunity. Later, when (perhaps by mistake) H places motive information in brown, S changes the color to green, repairing the drawing to conform to the code.
One final bit of coordination that must take place is how the collaborators will manage their interactions - will they just type at the same time, or try to approximate the turn-taking rules of spoken language. Will whiteboard usage be egalitarian, or primarily managed by one participant? How will dialogue and action be interleaved? How much attention will they pay to grounding information from any of the above realms, vs. spending time on problem solving? Much of this coordination proceeds implicitly, until problems occur. Because of the persistence of presentations in both MOO and whiteboard communication, like the problem-solving strategy and the white-board conventions, this information is only crucially part of the common ground when it inhibits understanding. We notice that strict turn-taking is not necessary - towards the end of the excerpt in Figure 4, at 1:29:40, both subjects were typing at the same time. S's happened to come out first, but each is still able to reply to the query of the other with no problem, due to the persistence of information on the screen. This phenomena of interleaved topics is prevalent throughout MOO-communication (and similar styles of textual chat programs).
Summarizing, the grounding criterion for each of these information types will be subordinate to the overall task - finding the killer. There will be changes in urgency for the different aspects of the common ground from Table 3 as the collaboration proceeds. For example, facts and inferences which relate directly to producing, confirming, or disproving current hypotheses about the murder will have a high criterion, while others may be fairly low. Similarly, the positions of the collaborators in the MOO is generally only important if they want to work together. The degree of prior groundedness will be due, in large part, to the types of action performed and features of the collaborative tools. Putting something on the whiteboard achieves a high degree of groundedness even without explicit acknowledgment, due to presence in a shared (visual) situation, as indicated by Lewis [Lewis1969]. Also, some MOO commands, such as page, give an automatic acknowledgment when the messages are sent to a recipient. The cost of grounding will also be highly dependent on the tools. For instance, writing a note in the whiteboard takes much more effort (in terms of time and mouse manipulations and editing ability) than sending the same message through the MOO. Similarly, some messages are easier to recognize or understand than others.