Gmytrasiewicz and Durfee [25] have applied their Recursive Modelling Method to the computation of value of information of negotiation acts (see Section 2.10). The RMM uses a tree of game matrices, with a belief value at each node of the tree. Value of information computations are used to find the difference in expected utility between the total tree, which the agent must use if it does not have the benefit of information, and the weighted sum of utility over the pair of child subtrees that are obtained when the informed belief is placed at the root of the RMM tree. The communicative acts are as follows. Modelling acts are used to communicate beliefs, and so correspond with the tell act, with an equivalent pragmatic definition. Intentional acts are used to communicate preferences and so correspond with the propose act. Their pragmatic definition is that the hearer prunes all but the preferred alternative from the matrix, which is a less effective but simpler approach to the dry-land algorithm described here. Instead, the dry-land algorithm seeks an explanation for the preferred alternative. Once an explanation is found, the choice node is effectively pruned, but as well as that, the hearer has gained some information about the beliefs of the proposer. Question acts are used to declare an agent's ignorance of alternatives, allowing an autonomous response by the hearer using a modelling act. Again this is a simpler approach to the dry-land interpretation of questions that is used here. Requests are not included as communicative acts, even though they have been shown here to be necessary. Additional acts that have no correlates here are imperative acts, which are used to declare knowledge of which of a number of uncertain alternatives is the true one, and acknowledgement acts, which are used because messages are sometimes not heard over noisy communication channels.