Ensemble Distillation for Unsupervised Constituency Parsing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shayegh, Behzad, Cao, Yanshuai, Zhu, Xiaodan, Cheung, Jackie C. K., Mou, Lili
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913329514020864
author Shayegh, Behzad
Cao, Yanshuai
Zhu, Xiaodan
Cheung, Jackie C. K.
Mou, Lili
author_facet Shayegh, Behzad
Cao, Yanshuai
Zhu, Xiaodan
Cheung, Jackie C. K.
Mou, Lili
contents We investigate the unsupervised constituency parsing task, which organizes words and phrases of a sentence into a hierarchical structure without using linguistically annotated data. We observe that existing unsupervised parsers capture differing aspects of parsing structures, which can be leveraged to enhance unsupervised parsing performance. To this end, we propose a notion of "tree averaging," based on which we further propose a novel ensemble method for unsupervised parsing. To improve inference efficiency, we further distill the ensemble knowledge into a student model; such an ensemble-then-distill process is an effective approach to mitigate the over-smoothing problem existing in common multi-teacher distilling methods. Experiments show that our method surpasses all previous approaches, consistently demonstrating its effectiveness and robustness across various runs, with different ensemble components, and under domain-shift conditions.
format Preprint
id arxiv_https___arxiv_org_abs_2310_01717
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Ensemble Distillation for Unsupervised Constituency Parsing
Shayegh, Behzad
Cao, Yanshuai
Zhu, Xiaodan
Cheung, Jackie C. K.
Mou, Lili
Computation and Language
Artificial Intelligence
Machine Learning
We investigate the unsupervised constituency parsing task, which organizes words and phrases of a sentence into a hierarchical structure without using linguistically annotated data. We observe that existing unsupervised parsers capture differing aspects of parsing structures, which can be leveraged to enhance unsupervised parsing performance. To this end, we propose a notion of "tree averaging," based on which we further propose a novel ensemble method for unsupervised parsing. To improve inference efficiency, we further distill the ensemble knowledge into a student model; such an ensemble-then-distill process is an effective approach to mitigate the over-smoothing problem existing in common multi-teacher distilling methods. Experiments show that our method surpasses all previous approaches, consistently demonstrating its effectiveness and robustness across various runs, with different ensemble components, and under domain-shift conditions.
title Ensemble Distillation for Unsupervised Constituency Parsing
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2310.01717