Ulrich Wille - Revision of Feature Structures

Belief revision studies the rational change of information states in the light of new evidence, where the interesting case is when the new evidence contradicts former beliefs, forcing the agent to give up some of these beliefs in order to maintain consistency. On the other hand, a rational agent will try to keep as much information as possible, since information is a valuable good. Hence, belief revision is by no means trivial.
In giving a formal account of belief revision, the first important question to be answered is, how information shall be represented. In AGM (Alchourrón et al. 1985, Gärdenfors 1988), a standard approach to the theory of belief revision, information states are represented by epistemic sets, i.e., sets of sentences of a language of propositional logic, closed under sentential deduction. There are more sophisticated ways of modelling information states, e.g. as sets of possible worlds or as probability functions. In this talk, I will represent information states as sortal frames, aka feature structures (Carpenter 1992). 
A feature structure is a function from features to their values, where the value of a feature in a feature structure can itself be a feature structure.
Modelling information states as feature structures makes a more elaborated approach to belief revision possible. Not only information states to be revised are represented as feature structures, but also the new information which leads to the revision.
I will discuss some possibilities of defining belief revision of feature structures. In the simple case where the new information is consistent with the original information state, not genuine revision is performed but what is called “expansion” in the belief revision literature. Expansion of feature structures comes down to the operation known as “unification”. 
The case of genuine belief revision on the other hand, where the new information is not consistent with previous beliefs, is similar to an operation known as “default unification” (Lascarides et al. 1996). The relations and differences between default unification and feature revision will be discussed as well as several topics in the belief revision framework as they present themselves from the point of view of feature structure revision. 
In this way my talk aims to show that on the one hand, representing information by feature structures leads to an interesting approach to belief revision, while on the other hand taking into account topics in the field of belief revision might shed new light to operations on feature structures like default unification.

References

Alchourrón, Carlos E. / Peter Gärdenfors /David Makinson (1985): On the logic of theory change: Partial meet contraction and revision functions, Journal of Symbolic Logic 50: 510 – 530.

Carpenter, Bob (1992): The Logic of Typed Feature Structures with Applications to Unification Grammars, Logic Programsand Constraint Resolution. Cambridge: University Press.

Gärdenfors, Peter (1988): Knowledge in Flux. Modeling the Dynamics of Epistemic States, Cambridge: MIT Press.

Lascarides, Alex / Ted Briscoe / Nicholas Asher / Ann Copestake (1996): Order Independent and Persistent Typed Default Unification, Linguistics and Philosophy 19:1, 1 – 89.