Scenario Argument Structure vs Individual Claim Defeasibility: What Is More Important for Validity Assessment?

摘要

We conduct comparative analysis of two sources of argumentation-related information to assess validity of scenarios of interaction between agents. The first source is an overall structure of a scenario, which included communicative actions in addition to attack relations and is learned from previous experience of multi-agent interactions. In our earlier studies we proposed a concept-based learning technique for this source. Scenarios are represented by directed graphs with labeled vertices (for communicative actions) and arcs (for temporal and attack relations). The second source is a traditional machinery to handle argumentative structure of a dialogue, assessing the validity of individual claims. We build a system where data for both sources are visually specified, to assess a validity of customer complaints. Evaluation of contribution of each source shows that both sources of argumentation-related information are essential for assessment of multi-agent scenarios. We conclude that concept learning of scenario structure should be augmented by defeasibility analysis of individual claims to successfully reason about scenario truthfulness.