Automating Signature Evolution in LogicalTheories

摘要

We have argued for the importance of automated evolution of logical theories to adapt to new circumstances, to recover from failure and to make them better suited to the current problem. We argue that this requires more than just belief revision - although this is part of the story. We also need signature revision, i.e., changes to the underlying syntax of the theory. We have begun the work of automating signature evolution in the ors and galileo projects. We have developed repair plans over multiple theories, which address some of the tough problems of partial definedness, combinatorial explosion, coping with inconsistency and ensuring minimality, that beset this endeavour. In the future, we plan to: develop additional repair plans, research additional case studies from the history of physics, refine our currently rather ad hoc logical theories, thoroughly evaluate our repair plans on a significant corpus of case studies, and explore notions of minimal repair and other aspects of a theory of signature evolution. Ideas for future repair plans include: the converses of wms and Inconstancy; the use of analogy to create new theories; and the correction of faulty causal dependencies.