Grounding in computer-supported collaborative problem solving
|
|  Yüklə 1.41 Mb.
|
Structure of grounding patternsThe statistics presented so far summarize grounding patterns across protocols by counting pairs of utterances. However, these pairs of utterances form more complex patterns and episodes. To describe larger patterns, we first discuss turn taking, then present different forms of irregular forms of turn taking: simultaneous turns and parallel threads.
The mechanisms of turn taking in the MOO are very different than in voice conversation. On one hand, there is no constraint to wait for one's partner answer before to say more. On the other hand, one can implicitly or explicitly refer to utterances earlier than the last one, since they are still visible on the screen. Hence a MOO conversation between two people is not a simple alternation of turns. We counted the index of complexity on 'say' and 'page' commands. This index evaluates the regularity of turn taking. Its value would be zero if, knowing the speaker at turn n we have a probability of 1 for predicting who will speak at n+1. Its value would be 1 if knowing the speaker at turn n does not give us any information regarding who will speak at n+1. Its average value in our protocols is 0.9 (SD = .06), i.e. it indicates a complete non-systematicity of turn taking! Interestingly, the average index of complexity is exactly the same if we consider the group with a high acknowledgment rate and the group with a low acknowledgment rate (both 0.9 as well). This seems to indicate that irregular turn taking does not really affect acknowledgment. It may imply that acknowledgment has not to be systematic, an utterance being acknowledged only when there is a need for acknowledging it. This point contribute for our model of grounding (see section 7) in which the need for acknowledging a piece of information constitutes a key parameter for predicting grounding efforts. In typed interactions, the absence of intonation for indicating the end of a turn is probably compensated by the fact that sending the message indicates the end of a turn. It occurs that an accidental 'return' sends a message before it is finished and this message is hence hence acknowledged prematurely, as in example 40.
Example 40: Accidental 'return' leads to premature acknowledgment (from Pair 19, translated) Sometimes, a speaker wants to express himself with several turns, i.e. to keep the floor. He could send a long message, but there are several reasons for not doing it: (1) explanations are clearer with a sequence of small messages (lines) than as a big message; (2) the text entry pane in the MOO window generally displays a few lines, 3 in these experiments, and scrolling makes difficult the edition of longer messages; and (3) it is sometimes desirable to keep the listener attention during the time necessary for entering the whole message (for instance if one types a long message with the 'say' command, the listener may leave the room in between and hence not receive the message). In examples 41, the speaker implicitly indicates that he wants keep the floor by using a ":".
Example 41: Implicit way to keep the floor (from Pair 3) Conversely, sending the message may not be enough for triggering change of speaker, as in example 41 above ("ok'" at 96.3) or in example 42 below. In this pair, Hercule had problems26 for using the MOO and namely for communicating. Sherlock explicitly indicates the end of his turn as in radio communication (106.3 and 106.4), thereby inviting Hercule to react to his previous messages.
Example 42: Explicit hand over messages 'stop' and 'over' (from Pair 3, line 105.5 translated)
In typed interaction, two users can 'talk' simultaneously without disturbing each other. There are numerous examples of simultaneous speech in our protocoles. In the example 43, the two subjects simultaneously read their notebook which lead them to reject simultaneously (88.2) Sherlock's initial proposal. Later (88.9), they simultaneously complete Skerlock's list of remaining suspects (88.7).
Example 43: Simultaneous talk indicating parallel cognitive processes (from Pair 18) In example 43, simultaneity of talk probably indicates that the two detectives perform in parallel similar cognitive processes, respectively rejecting and completing an hypothesis, on the same object, respectively utterances at 87.1 and 88.7. We do of course also observe simultaneous talk corresponding to independent cognitive processes like in example 44. However, although we did not systematically count all cases of simultaneous talk, in the majority of them, the two subjects say more or less the same thing or at least talk about the same thing.
Example 44: Simultaneous talk indicating independent cognitive processes (from Pair 10 ) For the subject themselves, synchronous talk can be interpreted as an indicator that they partner his following them closely. In example 45, Hercule realizes this simultaneous thinking (76.3).
Example 45: Hercule notices simultaneous talk (from Pair 13) In the example 45 (above), the two messages about Marie are not exactly synchronous. The actual delay between the first and the second is 8 seconds. We have to take into account that the time indicated in protocols (log files) refers to the moment where the user sends his message (i.e. hits the 'return' key). Hercule started to type his sentence before Sherlock sent his message, they have been typing simultaneously but Sherlock finished his message and sent it 8 seconds before Hercule. When the user introduces a MOO command, (s)he usually watches the text entry zone which is located at the bottom of the MOO window and may not see the messages of his/her partner which are displayed a few lines above. If (s)he does, the incoming message may make irrelevant the message being typed and hence lead the subject to cancel it. We cannot quantify this process from our log files, since no MOO command is performed, we can only infer such phenomena in cases such as the examples 46 and 47.
Example 46: Hercule cancels his message (71.7) because of simultaneous talk (from Pair 20, translated )
Example 47: Sherlock cancels his message (162.5) because of simultaneous talk (from Pair 16) We do also observe cases of synchronous interactions on the whiteboard, and especially the case which are interesting with respect to shared understanding: when the two subjects produce the same note as the same time like in example 48.
Example 48: Synchronous writing on the whiteboard (from Pair 22)
An acknowledgment in the MOO does not always refer to the last turn of the other speaker. This phenomena disturbs new MOO users. This is very often the case when more than 2 people interact, but it did also frequently occur in our experiments with only two subjects. In example 49, the pattern includes two interwoven acknowledgment patterns: H:88.5-S:89.3 and S:88.8-H90.3. In example 50, the pattern S43.6-H43.9 is indented inside the pattern H:43.5-S:44.1. It is the semi-persistence of talk in the MOO which enables subjects to cope with such situations.
Example 49: XYXY turns (from Pair 11, translated )
Example 50: XYYX turns (from Pair 12 ) Sometimes however, the speaker makes the effort of pointing to which utterance he refers to; in example 51, the reference is established by repeating a part of it of the acknowledged utterance..
Example 51: In 83.4, "not Lucie or Heidi " makes explicit which utterance is acknowledged by "Right"(from Pair 18) The MOO could be designed to support 'acknowledgment with reference' as in utterance 83.4 of the last example. We implemented a dialogue room in which the user could specify which utterances he was answering to. We did not further develop this approach but consider it as a promising way for future research.
The notion of interwoven turns may sound pejorative. We prefer to emphasize the subject skill to maintain two conversations in parallel. In some cases, subject S1 talks about topic T1 and subject S1 about topic T2, without any acknowledgment. This is probably the worst case with respect to the elaboration of a shared solution. However, it is not certain that the absence of acknowledgment means that one detective is not following the other's work in the background, but does not find useful to comment his partner work and hence simply report his own work. Our data do not enable us to study the cognitive consequence of such 'parallel and independent' talk. More interesting are the cases of pairs who conduct together two different conversations at the same time, each conversation being acknowledged as a normal conversation, like in example 50 (above). In example 51p, Hercule is talking in the MOO about the Helmut Von Schneider and simultaneously taking notes in the whiteboard about Oscar Salève.
Example 51: Parallel threads across modes: talking about one thing the MOO and another thing on the whiteboard (from Pair 12). These results open a new avenue for research which was not originally included in this project. They raise two questions: (1) how newcomers to MOO environments progressively adapt themselves to models of dialogue which are very different from the usual voice conversation, (2) how models of dialogue in which several lines of argument may be pursued simultaneously affect problem solving. We would postulate that users take the best benefit from MOO interactions if they do not stick rigidly to voice dialogue patterns but learn how to use the more complex dialogue patterns which emerge in MOO conversations. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||