4.1 Grounding mechanisms are use to built shared sub-spaces
4.1.4 Grounding spatial location
Do I know where my partner is and does she know where I am? These are basic questions enabling collaboration and interaction. In these experiments, the physical location is fixed (in front of the computer), but the "virtual" position of the player in the MOO can be important. This information is generally shared and monitored in a non-intrusive way for several reasons.
- Level 1 of sharedness is permanent: at any time, my partner can type 'who' to check where I am. Level 2 of sharedness is quasi-permanent: many MOO commands provide side-information on the position of the agent who executed the command. For instance, when A pages B from the lobby, B receives a message "You sense that A is looking for you in lobby" before getting the actual message sent by A. In addition, when the detectives report data on the whiteboard, they can often infer where the other has been from the nature of the information. The first step in grounding, i.e. providing the information, is often performed implicitly while the agents are performing other action.
- This information is simple. If A can infer that B has perceived this information (Level 2), he can deduce that B understood it. There is no real need for monitoring understanding (level 3). Actually, this claim is probably only true for pairs who are familiar with the MOO. For instance, A must know that, when he pages to B from the lobby, B receives the message "You sense that A is looking for you in the lobby".
- Position is non-persistent. It would be a waste of energy to constantly monitor where a partner is, since this information can change frequently.
- This information is only useful in two cases: (1) when one wants to use a MOO command which only works if the partner is in the same room (e.g. say, give, follow, show, read); (2) to infer which actions and effects have been observed.
- In case of erroneous knowledge about respective positions, the consequences are not dramatic. There are several cases where one detective uses 'say' instead of 'page' (hence his message does not reach his partner) which are not noticed or not repaired. If the consequences are important, it is easy to repair, the player can just page "I am in room 4"
For these reasons, explicit grounding acts for position information are rare, and mostly repairs. In example 14, Sherlock deduced that a sent message was not read, because it contained a request for acknowledgment which had not been satisfied. Actually, in this protocol Hercule previously used 'say' 6 times when Sherlock was in another room, without performing any repair. Conversely, in example 15, the "who" command is used in a pro-active way in order to coordinate action.
Repairing spatial knowledge is preferably performed through MOO actions ('who', 'look', 'where') rather than through verbal interactions. Verbal interactions are used to ground future positions, which is quite logical since the MOO cannot provide this information. In this case, more than the position is grounded: in example 16, it is the problem solving strategy (What are you looking for), and in example 17, it is the interaction mode.
Regarding MOO actions, there is no functional necessity that, when performing some action, A knows which action B is doing at the same time. Once again, it is more important to know which actions (especially questions) the partner has previously performed and which ones he will perform. In this case partners really ground collected facts (see 4.1.1) and the problem solving strategy (see 4.1.3).
Grounding in Multi-modal Task-Oriented Collaboration - 3 SEP 1996
Generated with Harlequin WebMaker