TL;DR
This paper introduces "reformulation," a novel belief change operation for rational agents. It allows agents to make implicit beliefs explicit without altering the logical consequences of their belief base, addressing limitations of existing AGM models when dealing with belief bases that distinguish between explicit and implicit beliefs.
What Problem Does This Paper Address?
The standard AGM (Alchourron, Gärdenfors, and Makinson) model of belief change represents an agent's beliefs as a "belief set," which is a logically closed set of sentences. However, this model struggles to capture nuances where the way beliefs are represented matters. Specifically, it doesn't distinguish between beliefs that are explicitly stated and those that are merely logical consequences.
The paper argues that using "belief bases" (sets of sentences not necessarily closed under logical consequence) offers greater expressivity. For example, two belief bases can have the same logical consequences but represent different underlying assumptions. This difference becomes crucial during belief change operations like revision or contraction. Existing AGM-style operations, when applied to belief bases, often treat these distinct bases identically if they share the same logical consequences, losing valuable information. The paper aims to bridge this gap by proposing an operation that can transform a belief base into an equivalent one, making implicit beliefs explicit, thereby enhancing the expressivity and dynamic behavior of belief base models.
Methods
The core contribution is the introduction of a "reformulation" operation. This operation takes a belief base $B$ and a sentence $p$ as input. If $p$ is already a logical consequence of $B$, the reformulation operation aims to make $p$ an explicit belief within $B$ without changing the overall set of logical consequences of $B$.
The proposed method for reformulation, $R(B, p)$, is constructed by first performing a base contraction of $B$ by $p$ (denoted $B-p$). This removes $p$ and potentially other sentences that implicitly lead to $p$. Then, $p$ is explicitly added back, along with sentences of the form $p \rightarrow q$ where $q$ is a sentence that was in $B$ but not in $B-p$, and $q$ is logically equivalent to $p$. This ensures that $p$ becomes an explicit belief while preserving the logical consequences of the original base. The paper also provides axiomatic characterizations for this operation, relating it to existing partial meet and kernel base contraction functions.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Belief Bases vs. Belief Sets: The paper highlights that while belief sets are logically closed, belief bases offer a richer representation by distinguishing between explicit and implicit beliefs. This distinction is crucial for capturing nuanced belief change.
- The Need for Reformulation: Existing belief change operations can lead to the same outcome for belief bases that are logically equivalent but structurally different. Reformulation addresses this by allowing for internal restructuring of a belief base to make implicit beliefs explicit, without altering the overall knowledge.
- Making Implicit Explicit: The core idea is to "upgrade" the epistemic status of a belief from implicit to explicit. This is achieved by removing implicit pathways to $p$ and then reintroducing $p$ as an explicit belief.
- Preserving Logical Equivalence: A key constraint is that the reformulation operation must not change the set of logical consequences of the belief base. This ensures that the agent's overall knowledge remains the same, only its internal representation is modified.
- Connection to Contraction: The reformulation operation is built upon existing base contraction mechanisms, leveraging their properties to achieve the desired outcome.
Key Results
The paper defines a reformulation operation $R(B, p)$ and provides two main results:
- Theorem 12: Characterizes reformulation when based on partial meet base contraction. It shows that $R(B, p)$ satisfies specific postulates (including logical equivalence, explicit success, isolation success, relevance, and recovery) if and only if the underlying contraction function is a partial meet base contraction.
- Theorem 13: Characterizes reformulation when based on kernel base contraction. It shows that $R(B, p)$ satisfies a slightly different set of postulates (including logical equivalence, explicit success, isolation success, core-retainment, and recovery) if and only if the underlying contraction function is a kernel base contraction.
The paper also proposes several alternative versions of the reformulation operation that further refine the notion of non-redundancy.
Limitations and Future Work
- Axiomatic Characterization of Alternatives: The paper leaves the full axiomatic characterization of the proposed alternative reformulation operations (RM2, RM3, RM4) for future work.
- Computational Complexity: While the paper focuses on the logical properties, the computational complexity of these operations, especially for large belief bases, is not deeply explored.
- Comparison with Other Operations: The paper briefly mentions related operations like "replacement" and "reconsideration" but suggests further study is needed to fully understand their relationship to reformulation.
- Practical Applications: While the authors suggest applications in legal texts, further exploration of real-world scenarios where reformulation would be particularly beneficial is a potential avenue for future research.
Summary updated