Automatic and Interactive Search in Flexible Dispute Derivations for Assumption-Based Argumentation

Master's thesis by Piotr Gorczyca

Supervisor Sarah Alice Gaggl, Martin Diller
Logische Programmierung und Argumentation
20 April 2022 – 5 Juli 2022
Download

Formal argumentation (FA) is a branch of knowledge representation and reasoning in artificial intelligence, which offers ways of resolving potential conflicts within a knowledge base and inferring which claims can be trusted. FA can be further divided between abstract- and structured argumentation, with the first one treating arguments as atomic entities and the latter allowing to further investigate their internal structure. Assumption-based argumentation (ABA) is one of the main general structured argumentation frameworks, in which dispute derivations (DDs) are methods for determining the acceptance of claims in dialectical manner. With DDs arguments and counter-arguments are constructed interchangeably between, as can be conceived, two fictitious players – the proponent and the opponent of a set of claims under scrutiny.

Among the many formalisations of DDs for ABA, which have been proposed throughout the past decades, flexible dispute derivations (FlexDDs) are the latest. FlexDDs offer a number of solutions unprecedented in the former versions, including reusing one player’s arguments by the other player or support for complete and stable argumentation semantics to name a few. Most notably however, alongside the regular, backward, top-down reasoning from claims to premises they also allow for construction of forward, bottom-up arguments, from premises to claims. Moreover, FlexDDs comes in two versions – a more high-level and abstract argument-based version as well as a more concrete and implementation-focused rule-based version. In this thesis we focus on automatized search of successful dispute derivations for flexible dispute derivations. We devise procedures and examine the properties of strategies tailored for specific semantics, with the aim of isolating features impacting the efficiency, among other concerns. Furthermore, we investigate the influence of forward reasoning on the performance of the search procedures. We have performed a thorough empirical evaluation, the results of which are presented and interpreted, to back up our claims and hypotheses. Moreover, our implementation for FlexDDs has been significantly extended, currently capable of following the defined search strategies in an automatic search for successful DDs as well as having rich support for interactive reasoning, besides many other notable features which we report on in this work.