Neural Bi-Lexicalized PCFG Induction
This work addresses a computational bottleneck in grammar induction for natural language processing, but it is incremental as it builds on existing neural L-PCFG methods.
The paper tackled the problem of neural lexicalized PCFGs ignoring bilexical dependencies due to computational complexity, and the result was a new parameterization approach that improved running speed and unsupervised parsing performance on the English WSJ dataset.
Neural lexicalized PCFGs (L-PCFGs) have been shown effective in grammar induction. However, to reduce computational complexity, they make a strong independence assumption on the generation of the child word and thus bilexical dependencies are ignored. In this paper, we propose an approach to parameterize L-PCFGs without making implausible independence assumptions. Our approach directly models bilexical dependencies and meanwhile reduces both learning and representation complexities of L-PCFGs. Experimental results on the English WSJ dataset confirm the effectiveness of our approach in improving both running speed and unsupervised parsing performance.