2014 TrackingBeliefsandIntentionsint

From GM-RKB
Jump to navigation Jump to search

Subject Headings: Linguistic Dialog, Belief Change, Intention Change, Agent Intention.

Notes

Cited By

Quotes

Abstract

We propose a model of belief and intention change over the course of a dialogue, in the case where the decisions taken during the dialogue affect the possibly conflicting goals of the agents involved. We use Situation Calculus to model the evolution of the world and an observation model to analyze the evolution of intentions and beliefs. In our formalization, utterances, that only change the beliefs and intentions, are observations. We illustrate our formalization with the game of Werewolf.

1. Introduction

Agents are engaged in dialogues according to goals they can satisfy by talking to other agents (Cohen and Perrault 1979), (Perrault and Allen 1980). A dialogue is a sequence of utterances, produced by agents based on their beliefs, in order to reach intended states of facts by causing other agents to change their beliefs and thus also their intentions. Our goal is to model this interaction between beliefs, intentions, and utterances. The ability to predict decisions resulting from the dialogue is used as a performance measure.

An example of such a domain is the Werewolf game. Here, players are assigned the roles of either villagers or werewolves. The game proceeds in alternating day and night stages, overlooked by an impartial judge. By night, unseen by villagers, the werewolves choose and kill a victim. The victim’s identity is announced by the judge at the beginning of the next day. Then, the rest of the villagers discuss and vote to execute one person who they agree is a werewolf. The problem is to predict the outcome of each player’s vote.

The game can be modeled with utterances as actions in Situation Calculus (McCarthy 1983), based on Austin’s theory of performatives (Austin 1975). A problem is accounting for the indefinitely many perlocutionary acts, or possible effects an utterance may have on the hearer. Belief Revision (Alchourron, Gardenfors, and Makinson 1985) enables modeling and reasoning about changes of beliefs, including as a result of knowledge producing actions in Situation Calculus (Shapiro et al. 2000). Applying Belief Revision to our domain is hard because the agents need to do reasoning about beliefs over beliefs.

References

;

 AuthorvolumeDate ValuetitletypejournaltitleUrldoinoteyear
2014 TrackingBeliefsandIntentionsintRoxana Girju
Codruta Girlea
Eyal Amir
Tracking Beliefs and Intentions in the Werewolf Game2014