PuK-Workshop : Planen / Scheduling und Konfigurieren / Entwerfen

OpenCCG sentence generation is a prominent approach to surface realization – choosing the formulation of a sentence – via search. Such searches often visit many infeasible composites (partial sentences), not part of any complete sentence because their grammar category cannot be extended to a sentence in a way covering exactly the desired sentence meaning. Formulating the completion of a composite into a sentence as finding a solution path in a large state-transition system, we exhibit a connection to AI Planning, and we design a compilation from OpenCCG into planning allowing to detect infeasible OpenCCG composites via AI Planning dead-end detection methods. Our experiments show that this can filter out large fractions of infeasible states in, and thus benefit the performance of, complex surface realization processes.