Two-Level Probabilistic Grammars for Natural Language Parsing

This paper presents a novel parsing algorithm based on PCW (Probabilistic Context-Free) grammars. It explores theoretical aspects of the algorithm, including capacity and computational complexity, specifically focusing on the distinction between the most probable derivation tree and the most probable w-tree. The implementation handles large grammars efficiently by optimizing the retrieval of rules, and the results indicate that although the parser might not always return the most probable tree, it effectively does so for the grammars utilized in the experiments.