TLDR: This paper introduces a new family of ‘gradual semantics’ for Assumption-Based Argumentation (ABA), a popular AI framework for reasoning. Unlike traditional methods that simply accept or reject arguments, gradual semantics assign a numerical ‘strength’ to assumptions, reflecting their degree of acceptability. The authors use Bipolar Set-Based Argumentation Frameworks (BSAFs) to model these interactions and show through experiments that their proposed approach is more reliable and converges faster than existing methods adapted from abstract argumentation. This fills a significant gap, allowing for more nuanced evaluation in real-world applications where assumptions carry varying weights.
In the fascinating field of Artificial Intelligence, particularly within knowledge representation and reasoning, computational argumentation plays a crucial role. One of the most prominent frameworks in this area is Assumption-Based Argumentation (ABA). ABA provides a versatile way to model and evaluate argumentative reasoning, finding applications in diverse domains such as healthcare recommendations and explainable causal discovery.
Traditionally, ABA frameworks have been evaluated using methods that either determine which sets of assumptions can be jointly accepted (extension-based semantics) or categorize each assumption as accepted, rejected, or undefined (labelling-based semantics). While effective, these approaches often provide a binary or categorical outcome, which might not be sufficient for scenarios requiring a more nuanced understanding of argument strength.
This is where ‘gradual semantics’ come into play. Gradual semantics offer a more fine-grained evaluation by assigning a numerical ‘dialectical strength’ or ‘acceptability degree’ to arguments. Imagine a scenario where assumptions, like a weather forecast or friend availability for a picnic, carry different levels of reliability or weight. Standard ABA semantics struggle to process this quantitative information. Until now, there has been a significant gap in applying gradual semantics directly to ABA, especially when assumptions themselves carry intrinsic weights.
A recent research paper, titled “On Gradual Semantics for Assumption-Based Argumentation,” by Anna Rapberger, Fabrizio Russo, Antonio Rago, and Francesca Toni, addresses this very challenge. The authors introduce a novel family of gradual semantics specifically designed for ABA frameworks. Their approach focuses on equipping assumptions, which are the core components in ABA, with these dialectical strengths.
To achieve this, the researchers utilize ‘Bipolar Set-Based Argumentation Frameworks’ (BSAFs) as an abstraction. BSAFs provide a concise way to represent the collective attacks and supports between assumptions within an ABA framework. By abstracting the problem in this manner, they can generalize state-of-the-art modular gradual semantics, which are typically used for other forms of argumentation frameworks, to the ABA context.
The proposed gradual ABA semantics are designed to satisfy important properties like balance and monotonicity, ensuring their logical consistency and predictable behavior. The paper also explores an alternative, argument-based approach as a baseline for comparison, leveraging existing gradual semantics from Quantitative Bipolar Argumentation Frameworks (QBAFs).
The experimental evaluation, conducted using synthetic ABA frameworks, yielded compelling results. The new gradual ABA semantics, based on BSAFs, demonstrated superior performance in terms of convergence. Specifically, the BSAF-based approach converged in approximately 90% of scenarios, significantly outperforming the QBAF baseline, which converged in only about 70% of scenarios. Furthermore, the BSAF approach converged faster, typically within 30 iterations compared to 45 iterations for the baseline.
This research marks a foundational step towards a more sophisticated evaluation of weighted assumptions in ABA. By providing a method to assign numerical strengths, it opens up new possibilities for applications where a simple ‘accepted’ or ‘rejected’ is not enough, and a degree of acceptability is required. For more in-depth technical details, you can refer to the full research paper available at arXiv.org.
Also Read:
- Unpacking AI Decisions: A New Logic for Contrastive Explanations
- Balancing Logic and Scale: New Grounding Methods for Neural-Symbolic AI
Future work in this area includes exploring the correspondence of gradual ABA semantics to other evaluation paradigms, extending the semantics to ABA with preferences, and investigating settings where non-assumptions or rules also possess intrinsic strengths. This work paves the way for more nuanced and applicable computational argumentation systems.
