zhaoyanpeng / syntax-induction-in-deep-learning-era

Unsupervised Learning of Syntax

Dataset

WSJ, CTB, SPMRL, and UD are widely used. You can check out XCFG for treebank preprocessing.

Dependency Grammar Induction

Constituency Grammar Induction

Compound Probabilistic Context-Free Grammars for Grammar Induction, paper, code.

Unsupervised Recurrent Neural Network Grammars, paper, code.

Unsupervised Latent Tree Induction with Deep Inside-Outside Recursive Autoencoders, paper, code.

Unsupervised Learning of PCFGs with Normalizing Flow. paper. code.

Emerging Topics in Unsupervised Grammar Induction

Visually Grounded Grammar Induction

• Visually Grounded Neural Syntax Acquisition, [paper], [code].
• Visually Grounded Compound PCFGs, [paper], [code].
• VLGrammar: Grounded Grammar Induction of Vision and Language, [paper], [code].

Grammar Induction with Language Models

Specialized LMs

• Neural Language Modeling by Jointly Learning Syntax and Lexicon, [paper], [code].
• Ordered Neurons: Integrating Tree Structures into Recurrent Neural Networks, [paper], [code].

General-Purpose LMs

• Are Pre-trained Language Models Aware of Phrases? Simple but Strong Baselines for Grammar Induction, [paper], [code].
• Perturbed Masking: Parameter-free Probing for Analyzing and Interpreting BERT, [paper], [code].

Neural Lexicalized Grammar Induction

• The Return of Lexical Dependencies: Neural Lexicalized PCFGs, [paper], [code].
• Neural Bi-Lexicalized PCFG Induction, [paper], [code].

Transfer Learning for Grammar Induction

Cross-Lingual Transfer

• Multilingual Grammar Induction with Continuous Language Identification, [paper], [code].

Cross-Domain Transfer

• On the Transferability of Visually Grounded PCFGs, [paper], [code].

Language Acquisition

Bootstrapping Language Acquisition, paper, code.

Syntax for Downstream Tasks

Scalable Syntax-Aware Language Models Using Knowledge Distillation. paper. code.

Language Modeling with Shared Grammar. paper. code.

Learning to Compose Task-Specific Tree Structures, paper, code.

Learning Latent Trees with Stochastic Perturbations and Differentiable Dynamic Programming, paper, code.