Unsupervised Recurrent Neural Network Grammars

TL;DR

This paper explores unsupervised learning of recurrent neural network grammars using variational inference, demonstrating competitive performance in language modeling and grammar induction without requiring annotated parse trees.

Contribution

It introduces an unsupervised training method for RNNGs using an inference network as a neural CRF parser, enabling grammar induction without labeled data.

Findings

Unsupervised RNNGs match supervised models in language modeling tasks.

They perform competitively in constituency grammar induction.

The approach works well for English and Chinese datasets.

Abstract

Recurrent neural network grammars (RNNG) are generative models of language which jointly model syntax and surface structure by incrementally generating a syntax tree and sentence in a top-down, left-to-right order. Supervised RNNGs achieve strong language modeling and parsing performance, but require an annotated corpus of parse trees. In this work, we experiment with unsupervised learning of RNNGs. Since directly marginalizing over the space of latent trees is intractable, we instead apply amortized variational inference. To maximize the evidence lower bound, we develop an inference network parameterized as a neural CRF constituency parser. On language modeling, unsupervised RNNGs perform as well their supervised counterparts on benchmarks in English and Chinese. On constituency grammar induction, they are competitive with recent neural language models that induce tree structures from words through attention mechanisms.