Generalized Predictive Shift-Reduce Parsing for Hyperedge Replacement Graph Grammars

Abstract

Parsing for graph grammars based on hyperedge replacement (HR) is in general NP-hard, even for a particular grammar. The recently developed predictive shift-reduce (PSR) parsing is efficient, but restricted to a subclass of unambiguous HR grammars. We have implemented a generalized PSR parsing algorithm that applies to all HR grammars, and pursues severals parses in parallel whenever decision conflicts occur. We compare GPSR parsers with the Cocke-Younger-Kasami parser and show that a GPSR parser, despite its exponential worst-case complexity, can be much faster.