Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM

Bo Zhang; Zhenghua Li; Min Zhang

doi:10.1007/978-3-030-32236-6_41

Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM

Bo Zhang
, Zhenghua Li^*
, Min Zhang

^*Corresponding author for this work

Soochow University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

As a method for exploiting multiple heterogeneous data, supervised treebank conversion can straightforwardly and effectively utilize linguistic knowledge contained in heterogeneous treebank. In order to efficiently and deeply encode the source-side tree, we for the first time investigate and propose to use Full-tree LSTM as a tree encoder for treebank conversion. Furthermore, the corpus weighting strategy and the concatenation with fine-tuning approach are introduced to weaken the noise contained in the converted treebank. Experimental results on two benchmark datasets with bi-tree aligned trees show that (1) the proposed Full-Tree LSTM approach is more effective than previous treebank conversion methods, (2) the corpus weighting strategy and the concatenation with fine-tuning approach are both useful for the exploitation of the noisy converted treebank, and (3) supervised treebank conversion methods can achieve higher final parsing accuracy than multi-task learning approach.

Original language	English
Title of host publication	Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings
Editors	Jie Tang, Min-Yen Kan, Dongyan Zhao, Sujian Li, Hongying Zan
Publisher	Springer
Pages	456-465
Number of pages	10
ISBN (Print)	9783030322359
DOIs	https://doi.org/10.1007/978-3-030-32236-6_41
State	Published - 2019
Externally published	Yes
Event	8th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2019 - Dunhuang, China Duration: 9 Oct 2019 → 14 Oct 2019

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11839 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	8th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2019
Country/Territory	China
City	Dunhuang
Period	9/10/19 → 14/10/19

Keywords

Concatenation with fine-tuning
Corpus weighting
Full-tree LSTM
Multi-task learning
Supervised treebank conversion
Treebank exploitation

Access to Document

10.1007/978-3-030-32236-6_41

Cite this

Zhang, B., Li, Z., & Zhang, M. (2019). Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM. In J. Tang, M.-Y. Kan, D. Zhao, S. Li, & H. Zan (Eds.), Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings (pp. 456-465). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11839 LNAI). Springer. https://doi.org/10.1007/978-3-030-32236-6_41

Zhang, Bo ; Li, Zhenghua ; Zhang, Min. / Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM. Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings. editor / Jie Tang ; Min-Yen Kan ; Dongyan Zhao ; Sujian Li ; Hongying Zan. Springer, 2019. pp. 456-465 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{29b1556fd27f4c658ae192b488f97c57,

title = "Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM",

abstract = "As a method for exploiting multiple heterogeneous data, supervised treebank conversion can straightforwardly and effectively utilize linguistic knowledge contained in heterogeneous treebank. In order to efficiently and deeply encode the source-side tree, we for the first time investigate and propose to use Full-tree LSTM as a tree encoder for treebank conversion. Furthermore, the corpus weighting strategy and the concatenation with fine-tuning approach are introduced to weaken the noise contained in the converted treebank. Experimental results on two benchmark datasets with bi-tree aligned trees show that (1) the proposed Full-Tree LSTM approach is more effective than previous treebank conversion methods, (2) the corpus weighting strategy and the concatenation with fine-tuning approach are both useful for the exploitation of the noisy converted treebank, and (3) supervised treebank conversion methods can achieve higher final parsing accuracy than multi-task learning approach.",

keywords = "Concatenation with fine-tuning, Corpus weighting, Full-tree LSTM, Multi-task learning, Supervised treebank conversion, Treebank exploitation",

author = "Bo Zhang and Zhenghua Li and Min Zhang",

note = "Publisher Copyright: {\textcopyright} 2019, Springer Nature Switzerland AG.; 8th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2019 ; Conference date: 09-10-2019 Through 14-10-2019",

year = "2019",

doi = "10.1007/978-3-030-32236-6\_41",

language = "英语",

isbn = "9783030322359",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "456--465",

editor = "Jie Tang and Min-Yen Kan and Dongyan Zhao and Sujian Li and Hongying Zan",

booktitle = "Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings",

address = "德国",

}

Zhang, B, Li, Z & Zhang, M 2019, Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM. in J Tang, M-Y Kan, D Zhao, S Li & H Zan (eds), Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11839 LNAI, Springer, pp. 456-465, 8th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2019, Dunhuang, China, 9/10/19. https://doi.org/10.1007/978-3-030-32236-6_41

Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM. / Zhang, Bo; Li, Zhenghua; Zhang, Min.
Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings. ed. / Jie Tang; Min-Yen Kan; Dongyan Zhao; Sujian Li; Hongying Zan. Springer, 2019. p. 456-465 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11839 LNAI).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM

AU - Zhang, Bo

AU - Li, Zhenghua

AU - Zhang, Min

PY - 2019

Y1 - 2019

N2 - As a method for exploiting multiple heterogeneous data, supervised treebank conversion can straightforwardly and effectively utilize linguistic knowledge contained in heterogeneous treebank. In order to efficiently and deeply encode the source-side tree, we for the first time investigate and propose to use Full-tree LSTM as a tree encoder for treebank conversion. Furthermore, the corpus weighting strategy and the concatenation with fine-tuning approach are introduced to weaken the noise contained in the converted treebank. Experimental results on two benchmark datasets with bi-tree aligned trees show that (1) the proposed Full-Tree LSTM approach is more effective than previous treebank conversion methods, (2) the corpus weighting strategy and the concatenation with fine-tuning approach are both useful for the exploitation of the noisy converted treebank, and (3) supervised treebank conversion methods can achieve higher final parsing accuracy than multi-task learning approach.

AB - As a method for exploiting multiple heterogeneous data, supervised treebank conversion can straightforwardly and effectively utilize linguistic knowledge contained in heterogeneous treebank. In order to efficiently and deeply encode the source-side tree, we for the first time investigate and propose to use Full-tree LSTM as a tree encoder for treebank conversion. Furthermore, the corpus weighting strategy and the concatenation with fine-tuning approach are introduced to weaken the noise contained in the converted treebank. Experimental results on two benchmark datasets with bi-tree aligned trees show that (1) the proposed Full-Tree LSTM approach is more effective than previous treebank conversion methods, (2) the corpus weighting strategy and the concatenation with fine-tuning approach are both useful for the exploitation of the noisy converted treebank, and (3) supervised treebank conversion methods can achieve higher final parsing accuracy than multi-task learning approach.

KW - Concatenation with fine-tuning

KW - Corpus weighting

KW - Full-tree LSTM

KW - Multi-task learning

KW - Supervised treebank conversion

KW - Treebank exploitation

UR - https://www.scopus.com/pages/publications/85075817192

U2 - 10.1007/978-3-030-32236-6_41

DO - 10.1007/978-3-030-32236-6_41

M3 - 会议稿件

AN - SCOPUS:85075817192

SN - 9783030322359

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 456

EP - 465

BT - Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings

A2 - Tang, Jie

A2 - Kan, Min-Yen

A2 - Zhao, Dongyan

A2 - Li, Sujian

A2 - Zan, Hongying

PB - Springer

T2 - 8th CCF International Conference on Natural Language Processing and Chinese Computing, NLPCC 2019

Y2 - 9 October 2019 through 14 October 2019

ER -

Zhang B, Li Z, Zhang M. Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM. In Tang J, Kan MY, Zhao D, Li S, Zan H, editors, Natural Language Processing and Chinese Computing - 8th CCF International Conference, NLPCC 2019, Proceedings. Springer. 2019. p. 456-465. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-32236-6_41

Conversion and Exploitation of Dependency Treebanks with Full-Tree LSTM

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this