TL;DR
This paper introduces "Bi-free formulas" as a novel way to characterize formulas that are truly independent of inconsistency in knowledge bases, considering both syntactic and semantic (atom-centric) perspectives within Priest's minimally inconsistent logic (LPm). This new notion helps refine inconsistency measures by ensuring they aren't affected by these truly independent formulas.
What Problem Does This Paper Address?
Measuring inconsistency in knowledge bases is crucial for managing and handling conflicting information. Researchers have developed various "inconsistency measures" to quantify how "bad" a knowledge base is. A desirable property for these measures is "Free Formula Independence," which intuitively means that formulas not involved in the core of the inconsistency shouldn't change the measure's outcome.
However, the original notion of "free formulas" (those not in minimal inconsistent subsets) has limitations. It doesn't fully capture independence when inconsistency is viewed through the lens of paraconsistent models (which allow for some contradictions) or when focusing on the atoms involved in the inconsistency. Existing attempts to address this, like "safe formulas" or "safely consistent formulas," have also fallen short, failing to encompass all intuitively independent formulas (like tautologies) or being too restrictive. This paper aims to bridge this gap by proposing a more robust characterization of formulas truly free from inconsistency.
Methods
The authors build upon Priest's minimally inconsistent logic (LPm), a paraconsistent logic that uses a three-valued system ({True, False, Both True and False}). They introduce two key novel concepts:
- B-atom-free formula: This is a semantic notion. A formula is B-atom-free if its presence or absence doesn't change the set of "B-atoms" (atoms assigned the "Both True and False" value) in the minimal inconsistent models of a knowledge base. This captures independence from an atom-centric perspective.
- Bi-free formula: This is the core contribution. A formula is Bi-free if it's independent of inconsistency from both the syntactic perspective (not being part of any minimal inconsistent subset) and the semantic perspective (being B-atom-free). This dual requirement ensures a stronger notion of independence.
They then propose a corresponding property, Bi-free Formula Independence, which states that if a formula is Bi-free, adding or removing it from a knowledge base should not change the inconsistency measure.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- The limitations of "free formulas": The paper highlights that simply not being in a minimal inconsistent subset isn't enough to guarantee a formula is truly independent of inconsistency, especially when considering paraconsistent semantics.
- Atom-centricity matters: Focusing on the atoms involved in inconsistency (B-atoms in LPm) provides a crucial semantic dimension that syntactic approaches miss.
- Bi-free formulas are a stronger guarantee: By combining syntactic and semantic independence, Bi-free formulas offer a more reliable way to identify formulas that truly don't impact inconsistency measures.
- Bi-free formulas generalize existing notions: Tautologies, safe formulas, and safely consistent formulas are shown to be special cases of Bi-free formulas, demonstrating the broader applicability of the new concept.
Key Results
The paper establishes theoretical results showing how Bi-free formulas relate to existing concepts:
- Bi-free formulas are a subset of both B-atom-free and syntactically free formulas.
- The proposed "Bi-free Formula Independence" property implies Tautology Independence and Safe Formula Independence, making it a more encompassing property for inconsistency measures.
- The authors demonstrate that their Bi-free notion can be used to construct or evaluate inconsistency measures that satisfy desirable properties, even for measures that might not satisfy simpler independence criteria.
Limitations and Future Work
- Extended Abstract: This is an extended abstract, meaning the full proofs and detailed discussions are likely in a separate, longer publication (referenced as [Mu, 2019]).
- Focus on LPm: The work is specifically situated within Priest's minimally inconsistent LP (LPm). Extending these concepts to other paraconsistent logics or inconsistency frameworks could be a future direction.
- Practical Application: While the theoretical framework is laid out, the paper doesn't delve into specific applications or empirical evaluations of inconsistency measures using the Bi-free property. Demonstrating the practical benefits of Bi-free Formula Independence on real-world knowledge bases would be valuable.
Summary updated