Learning explicit rules in a neural network

The authors propose an architecture called RuleNet, which, based on knowledge of the task domain, allows for the extraction of symbolic condition-action rules from the connection strengths in a neural net. By exploiting constraints inherent in the domain of symbolic string-to-string mappings, RuleNet can learn to induce explicit symbolic condition-action rules from examples. These rules represent a powerful representational language with which to describe the inner workings of a network. In addition, they facilitate faster learning and generalize perfectly to further examples that follow the rules. This formal string manipulation task can be viewed as an abstraction of several interesting cognitive models in the connectionist literature, such as case role assignment or translating English text into phonetic symbols.<<ETX>>