Abstract
This work investigates the use of probabilistic logic programming as a tool for enhancing argumentation structure parsing, with a focus on Deep Probabilistic Answer Set Programming (dPASP). The objective is to replace the integer programming-based coherence enforcement step in Stab and Gurevych's argumentation parsing pipeline with a dPASP-based approach. By leveraging dPASP's capabilities for probabilistic modeling and integration with neural networks, the project aims to develop a more flexible and potentially more accurate method for ensuring the coherence of extracted argumentation structures.
The thesis explores how symbolic reasoning, enhanced with probabilistic semantics, can offer more explainable and flexible scene classification compared to traditional purely statistical models. In addition to studying the theoretical foundations, the project includes an implementation and evaluation of the proposed method.