Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Unifying Framework for Learning Argumentation Semantics

Zlatina Mileva, Antonis Bikakis, Fabio Aurelio D'Asaro, Mark Law, Alessandra Russo

TL;DR

The paper presents LAS_arg, an inductive-logic-programming framework that learns acceptability semantics for multiple argumentation formalisms (AAF, BAF, VAF, ABA) in an interpretable way by learning ASP encodings from labeled examples. It establishes equivalence with ASPARTIX encodings for key semantics and demonstrates superior time performance and data efficiency compared to both ASPARTIX and a deep-learning baseline on ICCMA-23. The approach also supports translating ABA into AAFs to apply the same learning method, yielding compact and transparent rule sets. Overall, the work advances interpretable, data-efficient learning of formal argumentation semantics with practical implications for human-machine dialogues and reasoning systems.

Abstract

Argumentation is a very active research field of Artificial Intelligence concerned with the representation and evaluation of arguments used in dialogues between humans and/or artificial agents. Acceptability semantics of formal argumentation systems define the criteria for the acceptance or rejection of arguments. Several software systems, known as argumentation solvers, have been developed to compute the accepted/rejected arguments using such criteria. These include systems that learn to identify the accepted arguments using non-interpretable methods. In this paper we present a novel framework, which uses an Inductive Logic Programming approach to learn the acceptability semantics for several abstract and structured argumentation frameworks in an interpretable way. Through an empirical evaluation we show that our framework outperforms existing argumentation solvers, thus opening up new future research directions in the area of formal argumentation and human-machine dialogues.

A Unifying Framework for Learning Argumentation Semantics

TL;DR

The paper presents LAS_arg, an inductive-logic-programming framework that learns acceptability semantics for multiple argumentation formalisms (AAF, BAF, VAF, ABA) in an interpretable way by learning ASP encodings from labeled examples. It establishes equivalence with ASPARTIX encodings for key semantics and demonstrates superior time performance and data efficiency compared to both ASPARTIX and a deep-learning baseline on ICCMA-23. The approach also supports translating ABA into AAFs to apply the same learning method, yielding compact and transparent rule sets. Overall, the work advances interpretable, data-efficient learning of formal argumentation semantics with practical implications for human-machine dialogues and reasoning systems.

Abstract

Argumentation is a very active research field of Artificial Intelligence concerned with the representation and evaluation of arguments used in dialogues between humans and/or artificial agents. Acceptability semantics of formal argumentation systems define the criteria for the acceptance or rejection of arguments. Several software systems, known as argumentation solvers, have been developed to compute the accepted/rejected arguments using such criteria. These include systems that learn to identify the accepted arguments using non-interpretable methods. In this paper we present a novel framework, which uses an Inductive Logic Programming approach to learn the acceptability semantics for several abstract and structured argumentation frameworks in an interpretable way. Through an empirical evaluation we show that our framework outperforms existing argumentation solvers, thus opening up new future research directions in the area of formal argumentation and human-machine dialogues.
Paper Structure (15 sections, 4 theorems, 1 figure, 1 table)

This paper contains 15 sections, 4 theorems, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. Argumentation
  4. Learning Answer Set Programs
  5. Unifying Framework For Learning Argumentation Semantics
  6. Learning the Semantics of AAFs
  7. Equivalence with ASPARTIX
  8. Applying $LAS_{arg}$ to Flat ABA Frameworks
  9. Step 1: Constructing Arguments.
  10. Step 2: Finding the Attack Relations
  11. Evaluation
  12. Time Performance for Computing AAF Extensions
  13. Comparison to Deep Learning methods
  14. Compactness and Interpretability
  15. Conclusion and Future Work

Key Result

Theorem 1

Let $F$ be an ASP representation of an AAF. Let $T_{\sigma}=\langle B_{AAF},S_M,\langle E^+,E^-\rangle \rangle$ be a learning task for the $\sigma$-semantics of AAF, where $\sigma$ stands for admissible, complete or stable. Let $H_{\sigma}$ be a solution to $T_{\sigma}$, and $P_{\sigma}=B_{AAF} \cup

Figures (1)

  • Figure 1: Average PAR-2 scores for ASPARTIX and ILASP for different semantics on the ICCMA-23 dataset (the lower the better).

Theorems & Definitions (6)

  • definition 1
  • Theorem 1
  • proof
  • Lemma 3.1
  • Lemma 3.2
  • Lemma 3.3