A structuralist theory of belief revision (Q548705): Difference between revisions

The author defines revisions and contractions by borrowing from three different sources: (i) the structuralist theory of science, or rather, a simplified version of it, developed by the author under the name of minimal structuralism; (ii) a form of induction that embeds the rule which infers \(A\) from \(C\) and \(A,T\models C\) in the formalism of prioritised default logic; (iii) Rott's model of revision of belief bases. An example from \textit{R. Gärdenfors}'s book [Knowledge in flux. Modeling the dynamics of epistemic states. Cambridge, MA: MIT Press (1988)] is used to provide some motivation and illustrate some of the concepts. Minimal structuralism is built from \textbf{T}-theoretical and \textbf{T}-non-theoretical entities of the form \[ y=\langle D_1,\dots,D_k,N_1,\dots, N_p\rangle\text{ and }x=\langle D_1,\dots,D_k,N_1,\dots, N_p,T_1,\dots,T_q\rangle, \] respectively, where \(D_1, \dots, D_k\) are domains of objects of a given type, and \(N_1, \dots, N_p\), \(T_1, \dots, T_q\) are relations over those domains, meant to represent empirical predicates for the \(N_i\)s, and theoretical concepts for the \(T_i\)s. So-called internal and external links constrain the theoretical descriptions, in a given theory \textbf{T} for the internal links, and with respect to two theories \textbf{T} and \({\mathbf T}'\) for the external links. Five postulates are given to formalise a predicate \(\mathbf{AE}(\mathbf{T})(x,y)\) whose intended meaning is that the \textbf{T}-theoretical structure \(x\) is an admissible extension of the representation \(y\) of an empirical phenomenon. These postulates are added to the definition of some notation and a set of accepted propositions of a particular kind to make up a set \(W\). This notation includes \(y=\mathbf{r}(\mathbf{T})(x)\) to express that \(x\) and \(y\) are related as above, \(\mathbf{I}(\mathbf{T})\) to refer to the set of intended applications of \textbf{T}, and \(\mathbf{M}(\mathbf{T})\) to refer to the set of models of \(\mathbf T\). A strict partial order \(<_R\) is defined on the set \(D\) of defaults of the form \[ \frac{b_j\in\mathbf{I}(\mathbf{T}),\;b_j=\mathbf{r}(\mathbf{T}),\;x\in\mathbf{M}(\mathbf{T}):(x,b_j)\in\mathbf{AE}(\mathbf{T})}{(x,b_j)\in\mathbf{AE}(\mathbf{T})}\,. \] The revision of \(A\) by \(\phi\) is then defined as \(\inf(W\setminus\{\phi\},D,<_R)\). The reader will probably be disappointed not to find a worked out example that illustrates how all pieces fit together and which role they play in the revision process. Such an example would help to see whether the claim that ``this integration allows for a substantial simplification of the ranking information that is necessary to define revisions and contractions in a unique manner'' is justified.