In plain words: A sentence model that builds a phrase tree one choice at a time, scoring each choice so it can parse sentences and predict the next word. It beat the best earlier tree-building model at parsing and ordinary word-by-word networks at predicting English and Chinese.

Abstract

We introduce recurrent neural network grammars, probabilistic models of sentences with explicit phrase structure. We explain efficient inference procedures that allow application to both parsing and language modeling. Experiments show that they provide better parsing in English than any single previously published supervised generative model and better language modeling than state-of-the-art sequential RNNs in English and Chinese.

Chris Dyer, Adhiguna Kuncoro, Miguel Ballesteros, Noah A. Smith
arXiv:1602.07776 · cs.CL, cs.NE · submitted Feb 25, 2016 · updated Oct 12, 2016
abstract · pdf · html · Proceedings of NAACL 2016 (contains corrigendum)