Rule Augmented Unsupervised Constituency Parsing

Abstract

Recently, unsupervised parsing of syntactic trees has gained considerable attention. A prototypical approach to such unsupervised parsing employs reinforcement learning and auto-encoders. However, no mechanism ensures that the learnt model leverages the wellunderstood language grammar. We propose an approach that utilizes very generic linguistic knowledge of the language present in the form of syntactic rules, thus inducing better syntactic structures. We introduce a novel formulation that takes advantage of the syntactic grammar rules and is independent of the base system. We achieve new state-of-the-art results on two benchmarks datasets, MNLI and WSJ. 1

Unsupervised Full Constituency Parsing with Neighboring Distribution Divergence

“…Models trained on masked language model put forward another framework for unsupervised parsing procedure. These models, like DIORA and its variants (Drozdov et al, 2019;Sahay et al, 2021), have been verified by experiment results to be efficient in discerning constituents from sentences. Unfortunately, these models fail to label the constituents after constructing an unlabeled treebanks from sentences.…”

“…Unfortunately, the need of large annotated datasets limits the performance of supervised systems on languages of low resources. As the result, many unsupervised systems have been proposed for constituency parsing (Drozdov et al, 2019;Kim et al, 2020;Shen et al, 2021;Sahay et al, 2021) by exploiting unlabeled corpus. However, current unsupervised constituency parsing systems are still far from a real full procedure, especially for the reason that most of these systems only induct an unlabeled structure of the constituency tree.…”