A Probabilistic Approach for Argument Interpretation

Abstract

We describe a probabilistic approach for the interpretation of user arguments, and investigate the incorporation of different models of a user's beliefs and inferences into this mechanism. Our approach is based on the tenet that the interpretation intended by the user is that with the highest posterior probability. This approach is implemented in a computer-based detective game, where the user explores a virtual scenario, and constructs an argument for a suspect's guilt or innocence. Our system receives as input an argument entered through a web interface, and produces an interpretation in terms of its underlying knowledge representation -a Bayesian network. This interpretation may differ from the user's argument in its structure and in its beliefs in the argument propositions. We conducted a synthetic evaluation of the basic interpretation mechanism, and a user-based evaluation which assesses the impact of the different user models. The results of both evaluations were encouraging, with the system generally producing argument interpretations our users found acceptable.

Inferences, suppositions and explanatory extensions in argument interpretation

“…Our domain of implementation is a murder mystery, for which we developed a series of BNs (Zukerman et al 2003;Zukerman and George 2005). The examples in this paper are drawn from two similar 32-node binary murder-mystery BNs.…”

“…In our initial work, we defined an interpretation as a subnet of this BN, and applied a probabilistic approach to select a subnet that fits the user's argument well. The main contributions of that research were: (1) an anytime algorithm for proposing candidate subnets of the domain BN ; and (2) a formalism for calculating the probability of these subnets-this probability encodes how probable are the structure and beliefs of these subnets in the context of the domain BN (Zukerman et al 2003;Zukerman and George 2005). Our formalism is applicable to a variety of BNs (which is orthogonal to how well the BNs model their domain).…”

An Anytime Algorithm for Interpreting Arguments

“…Calculate the probability of a configuration -For node configurations, this probability is the product of the probabilities of the entries in a configuration. For assumption configurations, this probability is a function of the "cost" of making a set of assumptions (the higher the product of the probabilities of the entries in an assumption configuration, the lower the cost) and the "savings" due to a closer match between the beliefs stated in an argument and those in the BN as a result of making the assumptions (the calculation of this component is described in [12]). For instance, for the example in Fig.…”

“…Our evaluation focuses on the anytime algorithm, i.e., the time BIAS takes to produce a good interpretation, rather than on BIAS' ability to produce plausible interpretations (which was evaluated in [12]). …”

“…The propagated beliefs in the nodes in these arguments were then altered to achieve different levels of belief distortion. Our hand-generated arguments were based on arguments entered by people (obtained in previous system trials [12]). These arguments were manually modified in order to obtain enough sample arguments to test the above three factors.…”