TLDR: A new research paper by Carlo Proietti and Antonio Yuste-Ginel introduces a novel concept of expressivity to compare abstract and structured argumentation frameworks dealing with qualitative uncertainty. The study finds that structured argumentation (ASPIC+) with uncertain rules or premises is more expressive than basic abstract frameworks, but less expressive than abstract frameworks with explicit dependencies. Furthermore, within ASPIC+, uncertainty in rules is shown to be more expressive than uncertainty in premises.
In the realm of artificial intelligence and logical reasoning, understanding how to handle uncertainty in arguments is a critical challenge. A recent research paper, titled Comparative Expressivity for Structured Argumentation Frameworks with Uncertain Rules and Premises, by Carlo Proietti and Antonio Yuste-Ginel, delves into this complex area, offering a novel perspective on comparing different formal approaches to argumentation with qualitative uncertainty.
The Challenge of Uncertainty in Argumentation
Imagine a debate where not all facts or rules are definitively known. This is the essence of qualitative uncertainty in formal argumentation. Traditionally, two main paradigms exist: abstract argumentation and structured argumentation. Abstract argumentation frameworks (AFs) represent arguments as nodes in a graph and attacks between them as arrows. Uncertainty in this context might mean we’re unsure if a particular argument exists or if an attack relation holds. Structured argumentation, exemplified by frameworks like ASPIC+, builds arguments from more fundamental components such as premises and inference rules. Here, uncertainty can stem from whether a premise is true or if an inference rule is applicable.
The core question the researchers address is how these different ways of representing uncertainty compare in terms of their ‘expressivity’ – essentially, their power to model various scenarios of uncertainty. This is crucial because information from structured arguments can often be lost when translated into abstract models, leading to a disconnect between theoretical models and practical applications.
A New Measure of Expressivity
To bridge this gap, Proietti and Yuste-Ginel introduce a new notion of expressivity that allows for a systematic comparison between abstract and structured formalisms. Their method involves two key steps. First, they establish a standard for uncertainty in abstract argumentation using ‘argument-incomplete argumentation frameworks with dependencies’ (dep-arg-IAFs). In these frameworks, uncertainty is about which arguments are undecided, and expressivity is measured by the number of possible ‘completions’ – the ways this uncertainty can be resolved.
The second step involves ‘lifting’ structured formalisms to this abstract level. This means giving structured argumentation frameworks dealing with uncertainty an abstract counterpart and then comparing them against different types of dep-arg-IAFs based on their relative expressivity.
Uncertainty in ASPIC+ Frameworks
The paper focuses on ASPIC+, a popular structured argumentation framework, and explores two primary sources of uncertainty within it: uncertain inference rules and uncertain premises. A ‘rule-incomplete structured argumentation framework’ (rul-ISAF) is one where some inference rules (strict or defeasible) are uncertain. A ‘premise-incomplete structured argumentation framework’ (prem-ISAF) is one where some premises (axioms or ordinary premises) are uncertain.
The researchers then define how ‘completions’ work for these structured frameworks, essentially representing the different ways the uncertain rules or premises can be resolved to form a complete argumentation framework.
Key Findings on Expressivity
The study yields several significant results, summarized in a clear diagram within the paper. These findings shed light on the relative power of different uncertainty models:
-
Structured vs. Abstract (without dependencies): Systems for structured argumentation with uncertainty (both rul-ISAFs and prem-ISAFs) are shown to be strictly more expressive than general argument-incomplete argumentation frameworks (arg-IAFs) without dependencies. This means that structured approaches can capture more nuanced forms of uncertainty than basic abstract models.
-
Structured vs. Abstract (with dependencies): However, when abstract frameworks are extended with dependencies (imp-arg-IAFs), they become strictly more expressive than structured argumentation frameworks with uncertainty. This highlights the power of explicitly modeling relationships and conditions between uncertain arguments at the abstract level.
-
Uncertain Rules vs. Uncertain Premises: Within structured argumentation, the paper demonstrates that rule-incomplete systems (rul-ISAFs) are strictly more expressive than premise-incomplete ones (prem-ISAFs). This suggests that uncertainty about the rules of inference provides a richer modeling capacity than uncertainty solely about the initial facts or premises.
Also Read:
- New AI Framework Unlocks Deeper Understanding of Argument Structures in Text
- Quantifying Disagreement in Random Permutation Sets
Implications for the Field
This research unifies previous work on argumentation with uncertainty in structured formalisms and provides a robust framework for comparing their expressive power. The introduction of a precise notion of expressivity that can handle both abstract and structured formalisms is a significant methodological contribution. The results offer valuable insights for researchers and practitioners working on computational argumentation, guiding the choice of appropriate formalisms for modeling real-world scenarios involving qualitative uncertainty. Future work aims to extend these findings to other structured argumentation formalisms beyond ASPIC+ and to explore their broader context in journal publications.
