TL;DR
This paper introduces a novel argumentation framework called AAFD, which allows arguments to be associated with a specific domain of application. This innovation enables a more nuanced representation of exceptions and partial acceptance, moving beyond the all-or-nothing nature of traditional argumentation frameworks. AAFDs offer a more intuitive and compact way to model commonsense reasoning, bridging the gap between abstract argumentation and non-monotonic logic.
What Problem Does This Paper Address?
Traditional argumentation frameworks, like Dung's Abstract Argumentation Frameworks (AAFs), treat arguments as atomic units. This means an argument is either fully accepted or rejected. However, real-world reasoning often involves arguments that apply only to specific situations or objects, and can have exceptions. For instance, the argument "This fruit is an apple, and since apples are typically red, this fruit is red" is generally true, but fails for green apples like Granny Smiths. Existing frameworks struggle to capture this "partial acceptance" – where an argument is valid for some instances but not others – without creating an explosion of arguments, one for each specific case. This paper addresses the need for a formal mechanism to represent and reason about the scope of arguments and their exceptions.
Methods
The core contribution is the Abstract Argumentation Framework with Domain Assignments (AAFD). An AAFD extends standard AAFs by equipping each argument with a domain assignment. This assignment specifies the set of elements (from a universe $U$) to which the argument applies.
Key aspects of the AAFD approach include:
- Domain Assignments: A function $DU: A \rightarrow 2^U$ maps each argument to its prior domain – the set of elements it's intended for.
- Attack Relation Modification: In AAFDs, an attack $(a, b)$ doesn't necessarily invalidate argument $b$ entirely. Instead, it limits the domain of application of $b$. This means an attack can restrict the scope of an argument, which is its posterior domain – the subset of its prior domain for which it is accepted.
- Semantics: The paper defines semantics for AAFDs that focus on determining the scope of arguments, rather than just their overall acceptance. These semantics are designed to be analogous to classical AAF semantics (e.g., admissible, complete, grounded) but operate on the domain level.
- Mapping to AAFs: The authors provide two mappings from AAFDs to standard AAFs.
- $\Phi_{nat}$: This mapping creates a new argument for each element in an argument's domain. While semantically equivalent, it can be computationally inefficient if the universe $U$ is large.
- $\Phi_{[\text{nat}]}$: This mapping is more sophisticated and leverages the concept of indistinguishable elements (elements that always appear together in argument domains). It creates a single argument for each equivalence class of indistinguishable elements, making it more efficient when $U$ is large and the number of arguments is manageable. This mapping requires canonical scopes.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Partial Acceptance is Key: The most significant insight is that arguments can be "partially accepted" based on the specific elements they apply to. This is a crucial step towards modeling real-world reasoning where exceptions are common.
- Attacks as Domain Restrictors: Rethinking the role of attacks from "invalidating" to "restricting the domain" is a fundamental shift that allows for more nuanced reasoning.
- Domain Assignments as a Compact Representation: AAFDs offer a more compact and intuitive way to represent commonsense reasoning scenarios compared to creating numerous specific arguments in traditional AAFs.
- Equivalence of Semantics: The ability to map AAFDs to AAFs demonstrates that the proposed semantics are compatible with established argumentation theory, providing a solid theoretical foundation.
- Canonical Scopes for Efficiency: The introduction of canonical scopes and the $\Phi_{[\text{nat}]}$ mapping highlight the importance of identifying and exploiting structural regularities (like indistinguishable elements) for computational efficiency.
Key Results
- The paper formally defines AAFDs and their semantics, including notions of compliant, conflict-free, admissible, complete, grounded, preferred, and stable scopes.
- It establishes that AAFD semantics are equivalent to AAF semantics through two mapping functions ($\Phi_{nat}$ and $\Phi_{[\text{nat}]}$).
- Complexity analysis shows that AAFDs can be represented by AAFs, with the efficiency of the mapping depending on the relative sizes of the argument set and the universe. The $\Phi_{[\text{nat}]}$ mapping is particularly efficient when the universe is large.
- The paper demonstrates that AAFD semantics satisfy desirable properties like domain capping, scope maximality, and scope consistency, and introduces the concept of canonical scopes to capture the idea of indistinguishable elements.
Limitations and Future Work
- Background Theory Extension: The paper acknowledges that the current AAFD model uses explicit domain assignments, which can be cumbersome. Future work includes enhancing AAFDs with a background theory (using predicates) to describe domains more abstractly, allowing for reasoning about attributes and categories rather than just individual elements.
- Computational Efficiency: While mappings to AAFs are provided, further algorithmic improvements for computing AAFD extensions are needed, especially for handling large domains and complex argument structures.
- Dialectical Frameworks: Extending AAFDs into dialectical frameworks is proposed as a way to further leverage their expressiveness for commonsense reasoning problems.
- Implementation: The authors plan to develop and implement algorithms for computing AAFD extensions.
Summary updated