Holographic Neural PCFG cuts rule-scoring parameters by 99.94% and achieves state-of-the-art unsupervised parsing

Researchers propose Holographic Neural PCFG (Hol-PCFG), a method that recasts PCFG rule scoring as algebraic relation modeling among grammar-symbol embeddings using Holographic Embeddings.

• Hol-PCFG scores left-child, right-child, and lexical-emission relations over torus-constrained embeddings via circular correlation.
• The approach gives every rule probability a closed form that carries the intrinsic structure of grammar rules by construction.
• It achieves state-of-the-art parsing performance in six languages while cutting rule-scoring parameters by 99.94% relative to the baseline model.
• The model trains more stably and can parse Japanese directly from characters without any morphological segmentation, retaining nearly the same morpheme-level performance.

Hol-PCFG provides an interpretable mathematical form for rule probabilities while significantly reducing parameter count and improving training stability.