Co-training an Unsupervised Constituency Parser with Weak Supervision

Abstract

We introduce a method for unsupervised parsing that relies on bootstrapping classifiers to identify if a node dominates a specific span in a sentence. There are two types of classifiers, an inside classifier that acts on a span, and an outside classifier that acts on everything outside of a given span. Through self-training and co-training with the two classifiers, we show that the interplay between them helps improve the accuracy of both, and as a result, effectively parse. A seed bootstrapping technique prepares the data to train these classifiers. Our analyses further validate that such an approach in conjunction with weak supervision using prior branching knowledge of a known language (left/right-branching) and minimal heuristics injects strong inductive bias into the parser, achieving 63.1 F 1 on the English (PTB) test set. In addition, we show the effectiveness of our architecture by evaluating on treebanks for Chinese (CTB) and Japanese (KTB) and achieve new state-of-the-art results. 1

Dance preservation archives

“…Past work has proposed incorporating other data structures specialized for hierarchical patterns into neural networks, including context-free grammars (Kim et al, 2019a,b;Kim, 2021), trees (Tai et al, 2015;Zhu et al, 2015;Kim et al, 2017;Choi et al, 2018;Havrylov et al, 2019;Corro and Titov, 2019;Xu et al, 2021), chart parsers (Le and Zuidema, 2015;Maillard et al, 2017;Drozdov et al, 2019;Maveli and Cohen, 2022), and transition-based parsers (Dyer et al, 2015;Bowman et al, 2016;Dyer et al, 2016;Shen et al, 2019a).…”