Large-scale CCG Induction from the Groningen Meaning Bank

In present CCG-based semantic parsing systems, the extraction of a semantic grammar from sentence-meaning examples poses a computational challenge. An important factor is the decomposition of the sentence meaning into smaller parts, each corresponding to the meaning of a word or phrase. This has so far limited supervised semantic parsing to small, specialised corpora. We propose a set of heuristics that render the splitting of meaning representations feasible on a largescale corpus, and present a method for grammar induction capable of extracting a semantic CCG from the Groningen Meaning Bank.