Abstract

In this paper, a grammatical-inference algorithm is developed with finite sets of sample graphs as inputs and hyperedge-replacement grammars as outputs. In particular, the languages generated by inferred grammars contain the input samples. Essentially, the inference procedure iterates the application of an operation which decomposes hyperedge-replacement rules according to edge-disjoint coverings of the fight-hand sides of the rules. The main result is a characterization of the inferred grammars as "samplescomposing" meaning that each sample can be derived and each rule contributes to the generation of samples in a certain way.