Grounding in computer-supported collaborative problem solving
|
|  Yüklə 1.41 Mb.
|
Related Research at TECFAThese experiments revealed the great potential of MOO environments as tools for research on collaboration: They enable the researchers (1) to design and implement a task, (2) to tune at a great level of detail the problem solving tools and the communication tools, and (3) to trace automatically all actions with any tool (using log files). Several research projects, undergraduate and master theses, have grown out of the Bootnap project, they are briefly described below. In the total, more than170 subjects have been involved in these various experiments. A new project has recently been funded by the Swiss National Science Foundation: it aims to design artificial agents able to coordinate their behavior in the MOO with human agents with implicit cues as those observed in Bootnap37. New graduate theses are starting now, aiming to compare how subjects solve the same task when the whiteboard is individual rather than shared. Other experiments were conducted using the MOO as a tool for collaboration between primary schools from different countries38. In other words, the TecfaMOO environment created by Daniel Schneider at TECFA is more than a simple tool for communication, it created a new research area in which the fundamental issues of psychology and education can be addressed from a new perspective.
The study of Montandon (1996) aims to validate the above mentioned model for one type of information: MOO positions. In Bootnap, we observed very few interactions such as "where are you". This can be explained by our 3 parameters:
In her experimental design, L. Montandon played with these variables:
The experiment was run with 20 pairs40. We consider here only the interactions regarding the current position, since information regarding future positions constitute information with respect to the strategy. The number of utterances for grounding current positions is lower with MOOrich than with MOOpoor (T19=-2.29, p< .05). These results confirm our model since if a (x-b)/y : if (1) Need.grounded [Position] is constant and high (2) Prob.grounded [Position, MOOpoor] < Prob.grounded [Position, MOOrich] (3) Cost.grounding [Moo] is constant then the model predicts that: (4) Prob.grounding.act [Position, MOOpoor] > Prob.grounding.act [Position, MOOrich]
During the preparatory stage of this research, we conducted experiments with undergraduate students in psychology. The goal of these experiments was to analyze the gestures performed by two subjects in front of a graphical display. The subjects were using Memolab, an intelligent learning environment for the acquisition of methodological skills in experimental psychology (Dillenbourg, Mendelsohn & Schneider, 1994). The subjects had to build a virtual experiment on human memory. When the experiment was created, the system simulated its results. A virtual experiment involves different groups of subjects, each group doing different activities (encoding, delay and recall). The virtual experiment is represented on the screen: the activities of each group are aligned vertically, each column corresponding to the activities of a same group. We observed 8 pairs of subjects creating and simulating 3 virtual experiments with Memolab. The subjects were sitting side by side in front of the machine. Usually, the subject who had the mouse in hands designated screen locations with the mouse, while the other used his or her hand. We observed many gestures: 878 for the whole, i.e. more than 100 per pair. The number of gestures decreases with time (Ohayon, 1996) probably because the subjects had established common grounds. However, the number of gestures increases when the task becomes more difficult41 , probably because the established grounds are not sufficient any more to cope with the new situation. The main finding is that 87% of the observed gestures have a simple deictic function (Roiron, 1996): for instance, one subject says "put it here" and clicks on an empty cell on the screen display or one subject says "the short group" and the other answers "This one?" by pointing on a group of subjects. Oehler (1996) observed that when the users pointed to a column in the display, they often (57% of cases) refer to it as "this group", thereby linking the concept (an experimental group) with its representation (a column with three boxes representing the activities of a same group). The non-deictic gestures include simple emphatic gestures and metaphoric gestures. The latter often expressed the dimensions of a two-factors experimental plan, the hands moving along two perpendicular axes in front of the screen (Roiron, 1996) In question-response dialogues, the gesture was more often with the question than with the answer. As Meier (1996) suggested, when the gesture makes explicit what the question refers to, this reference has not to be re-established for the answer. This behavior varies however according to the type of questions (open such as "where?" versus close such as "this one or that one?").
|