Grounding in computer-supported collaborative problem solving

Role of the whiteboard in problem solving

1. Supporting inferences: spatial organization

We review some 'graphical ' objects observed in the whiteboard. Simple objects (list, marks,...) are addressed in the next section.

• 
  Timelines. Four pairs drew a timeline: pair 2 filled it exclusively, adding on the top of it the information (e.g. motive) which is important for the task but does not fit into the timeline. Pairs 4, 16 and 22 develop partially a timeline, but drop it before completion (figure 12). The timelines take several graphical forms. The actual form does not really matter, the point is these lines support the general function of comparing/sequencing numeral values, especially intervals. It is indeed difficult to reason about intervals without visualizing them.
  

• 
  Tables. Two pairs draw tables containing one row for each suspect and three columns containing the 3 criteria we provide them for identifying the murderer: the motive to kill, the opportunity to get the gun and the opportunity to kill. In pair 10, we provided the table at the outset, while pair 12 drew spontaneously the same table. Pair 12 filled table cells with simple marks, just writing a few annotation around the table. Pair 10 filled the table cells with small notes. In pair 12, a mark means 'yes', while in pair 10, the note can lead to conclude 'yes' or 'no'. They use an explicit 'no' note in several cells. Tables are not only very efficient to organize data and to detect missing information, but they also speed up the solution process: since the subjects are instructed that the murderer possess these 3 attributes (the motive to kill, the opportunity to get the gun and the opportunity to kill), as soon as one attribute is missing, they can discard the suspect. In Pair 12, since cells are fills with a simple mark, it is easy to identify rows with an empty cell and hence to discard the suspect. The detectives then draw a line across that row (see figure 14).
  

• 
  Maps. Several pairs draw the map of the Auberge that we provided them on the instruction sheet. This map does not really help to find the solution, because the solution of the enigma does not imply any spatial reasoning such as "Hans could not got from X to Y without crossing this room and meeting Rolf". They may have drawn these maps because they thought the task would be intrinsically spatial, either because there is a focus on spatiality in the MOO or because the warm-up task precisely consisted in drawing the map of a few rooms. However, these four maps provide detectives with formation which was not on the printed map. Pairs 19, 20 and 21 note on the map all the objects they found (while the provided map only included the suspects). Pair 6 drew on the map the moves of all suspects during the evening of the murder, probably hypothesizing a spatial solution such as one mentioned above. The maps of pair 21 and 19 do also support management: in pair 19, subject move a letter (H/S) to indicate to each where they are; in pair 21, they write a note "done" in each visited room.
  

• 
  Graphs. Three pairs draw a graph, i.e. the relate a set of notes with arrows. In Pair 5 (figure 1) and Pair 18 (figure 16), the graph synthesizes social relations among a few suspects. These graphs have been abandoned before to be completed. In Pair 14, the graph is a simple linear sequence between the elements of the solution.
  

• 
  Areas. The most common use of space consist of associating two or more items because two notes close to each other or because two objects overlap (a box around a note, a cross on a note, ...). This process is central to the mechanisms described in the next sections.
  

2. Supporting individual and group memory.

In the meanwhile, we simply selected the 4 whiteboards which seem richest in information (pairs 6, 7, 10 and 12) and discovered that they have rather high redundancy rates (respectively 16, 26, 29 and 10, while the average redundancy for all pairs is 12). One interpretation is that the subjects do not necessarily look at the whiteboard whether information X is present before to go and look for this information, because finding information on a whiteboard full of information, without a clear spatial organization, might take longer than finding the same information in the MOO. Another interpretation would be that the information on the whiteboard leads the detectives to look for more information, i.e. to ask again the same question for analyzing the answer under the liht of other information. To verify this hypothesis we should in the future compare with greater detail which information is on the whiteboard and which new information is collected.

In a task with potential disagreement, a memory tool should not archive data, it should also archive who brought the data on the whiteboard. This is for instance the case in the Belvedere systems (Suthers et al, 1995), which aims to support argumentation. It is also more important when more than 2 people collaborate on the whiteboard, since, when only 2 people are involved, if A can remember that she put on the whiteboard, she can also infer what B put. In our experiments, knowing who brought some data, was less important for facts than for inferences. In three pairs (7, 11 and 15) each detective uses a different color, e.g. black for Sherlock and yellow for Hercule. We are however not sure that this was a deliberate decision: it may simply be that the color was selected for some reason (playing a little bit with the tool before the beginning), and that the just kept writing in the same color.

Another sort of meta-information to be archived is whether a particular piece of information has been agreed or not. We provided the detectives with agreement stamps, i.e. personal marks they could paste on whiteboard elements to mark agreement, disagreement or doubts. However, they did almost not use these stamps. Communication often relies on the assumption that what is not explicitly disagreed is agreed. Hence acknowledgement delay plays an important role, since in case of silence, it gives an idea to the speaker whether his utterance has been not noticed or not disagreed.

With respect to memory issue, it is interesting to notice that subject rarely erased notes from the whiteboard²⁹. This is partially due to the fact that subject were facing a monotonic task, i.e. a fact true at time t, was still true at time t+1. When an hypothesis about one suspect was discarded, instead of erasing it, they usually preferred to mark a cross, which maintains the information that this hypothesis was abandoned.

3. Supporting regulation.

Regarding the management of the data collection phase, the whiteboards theoretically help to see which data have been already collected. These data are displayed on the whiteboard by rooms or by suspects (there is no much difference between the two since most suspects are lone in their room). Implicit coordination results from the fact that if information X is displayed on the whiteboard, it is useless to collect this information. As mentioned in the previous section, it is however not clear that this implicit coordination is efficient. Let us compare pair 6 or 7, in which the subjects put on the whiteboard one big note for each room, summarizing all information that room, and pair 21, in which the subjects draw a map and paste a "done" label each time a room has been visited. The task management is more explicit in the latter and more organized since it reproduces the Auberge map. However, in this case the spatial organization was not so important, and the set of notes posted by pairs 6 and 7 have the advantage of relating task management with task knowledge and relating rooms to suspects. In pair 7, this task management process becomes more explicit when one partners starts to add small notes with the room number of the larger notes in which information was collected per suspect (without indicating the room). Indicating rooms is a management process since information is available on the printed map provided to subject with the instruction sheet. It does not convey new information but help to organize data.

With respect to the management of the data analysis/synthesis stage, the whiteboard provides a shared memory of who is not suspect anymore. This is done by but a line crossing notes, names or name rows in a table for pairs 2, 4,7, 12 and 16, by adding "no gun" labels on notes for pair 6 and by circling discarded suspects in red (!) for pair 14. . Sometimes the reason for discarding the suspect is also written, like in pair 14.

In one case, whiteboard regulation is performed simply by indicating the current mutual position of detectives: in pair 19, they draw a map and move respectively small 'H' and 'S' marks when they change room.

For half of the pairs, the whiteboard implicitly reifies the problem solving strategy, both during data collection and data analysis. This may be the most important role of the whiteboard in CSCW. This was however not the case for all pairs ( see in the appendix the whiteboards of 5, 13, 17, 18, 19, 20, 21 and 22). We compare the interaction between subjects in these two groups: those who make the strategy obvious ( by structuring the data in tables, by marking progressively which suspects are discarded, ...) and those which simply put a few names and inks on the whiteboard, without any organisation during data collection or data analysis. We expected the 'unstructured' whiteboard group to discuss more about strategy, since it was not reified on the whiteboard. This is however not the case: the average number of interactions concerning task management is 36.3 for the pairs with a structured whiteboard and 33 for the others. Actually, the difference between these two groups concerns the inferences: the 'unstructured' group put significantly fewer inferences on their whiteboard (in average 20.1 for 'structured' and 7.6 for unstructured, F= 6.86, df=1; p=.05) while they included the same number of facts (mean = 14 for structured group and 13.4 for unstructured). This difference is probably due to two factors. First, when the whiteboard is structured, any fact can, simply by its position, be turned into an inference: for instance if the subject writes "was out last night" in the row of Lucie Salève and in the column "opportunity to kill", this simple fact actually means "Lucie has not opportunity to kill since she was out last night". Second, a structured whiteboard generally relies on a systematic strategy in which all suspects are considered one by one, while non-structured whiteboard are not systematic, the subject pasting a few notes here and there.

Moreover, the acknowledgment rate of inferences (both in talk/talk and around the whiteboard) is significantly higher for the 'unstructured' group (mean = 36%) than for the 'structured' group (mean=55%) (F=10.7; df =1, p=.01). The difference of acknowledgment rate might be just the consequence of the difference in number of inferences: since the 'structured' group write down more inferences, he has less time/attention to acknowledge all of them. This is actually not the explanation here since the two groups also differ by the acknowledgment rate for inference in talk/talk interactions (mean= 45% for structured group, mean=56% for unstructured group, F=5.28; df=1, p=.05). Hence, the explanation could simply be that a structured whiteboard and the acknowledgement rate for inferences are two indicators of the quality of collaboration between two subjects.