In this thesis we are concerned with the role of formal argumentation in artificial intelligence, in particular in the field of knowledge engineering. The intuition behind argumentation is that one can reason with imperfect information by constructing and weighing up arguements intended to give support in favour or against alternative conclusions. In dynamic argumentation, such arguements may be revised and strengthened in order yo increase to decrease the acceptability of controversial positions. This thesis studies the theory, architecture, development and applications of formal arguementation systems from the procedural perspective of actually generating argumentation processes. First, the types of problems that can be tackled via the argumentation paradigm in knowledge engineering are characterised. Second, an abstract formal framework are built from an underlying set of axioms, represented here as executatble logic programs. Finally an architecture for dynamic arguementation systems is defined, and domain-specific applications are presented within different domaind, thus grounding problems with very distinctive characteristics into a similar source in argumentation. The methods and definitions desribed in this thesis have been assessed on various bases, including the reconstruction of informal arguements and of arguments captured by existing formalisms, the relation between our framework and these formalisms, and examples of dynamic argumentation applications in the safety-engineering and multi-agent domains.

Formal argumentation aims to provide a structured framework for cognitively compatible automated reasoning in the context of artificial intelligence; however, due to its roots in mathematical logic, formal argumentation research is typically focused on formal, ‘object-level’ aspects. It has yet to be studied how environmental, ‘meta-level’ structures, can affect human intuitions regarding the formalized model. To bridge this gap, we examined whether bold text used in visualization of argumentation frameworks, the core structures of formal argumentation,affects human assessment of the acceptability of the arguments and the confidence in the assessment. 48 participants divided into four condition groups evaluated the acceptability of four sets of arguments with simple and simplified floating reinstatement. We put different arguments(or none) into bold print to nudge a decision (and to form a control group, respectively). The results show limited evidence in favor of bold text having an increasing effect on the acceptability of the topic argument but no significant changes in the confidence in the answer. A replication study with a larger sample size is warranted to increase confidence in the results.

Argumentation and reasoning have been an area of research in such disciplines as philosophy, logic and artificial intelligence for quite some time now. In the area of AI, knowledge needed for reasoning can be represented using various kinds of representation systems. The natural problem posed by this fact is that of possible incompatibility between heterogeneous systems as far as communication between them is concerned. This imposes a limitation on the possibility of extending smaller knowledge bases to larger ones. In order to facilitate a common platform for exchange across the systems unified formalisms for the different approaches to knowledge representation are required. This was the motivation for Dung [11] to propose in his 1995 paper an approach that later came to be known as an abstract argumentation framework. Roughly speaking, Dung's arguments are abstract entities which are related to each other by the means of conflicts between them. An intuitive graphical representation of Dung style framework is a graph whose nodes stand for arguments and whose edges stand for conflicts. A framework postulated this way is on one hand too general to be used on its own, but on the other hand it is general enough as to allow for varied extensions increasing its expressiveness, which indeed have been proposed. They include value-based argumentation frameworks by Bench-Capon et al. [6], preference-based argumentation frameworks by Amgoud and Cayrol [1] and bipolar argumentation frameworks by Brewka and Woltran [7], to name a few. The present thesis is concerned with yet another variation of Dung's framework: the concept of splitting. It was developed by Baumann [4] with one of the underlying purposes being that the computation time in frameworks which have been split into two parts and then computed separately may show some improvement in comparison to their variant without splitting. It was one of the main tasks of my work to develop an efficient algorithm for the splitting operation based on the theoretical framework given in [4]. On the other hand I hoped to confirm the expectation that splitting can indeed make a computation perform better.