How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers

Zeljko Agic; Natalie Schluter

doi:10.18653/v1/P17-2107

How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers

Research output: Conference Article in Proceeding or Book/Report chapter › Article in proceedings › Research › peer-review

Abstract

In dependency parsing, jackknifing taggers is indiscriminately used as a simple adaptation strategy. Here, we empirically evaluate when and how (not) to use jackknifing in parsing. On 26 languages, we reveal a preference that conflicts with, and surpasses the ubiquitous ten-folding. We show no clear benefits of tagging the training data in cross-lingual parsing.

Original language	English
Title of host publication	Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics
Number of pages	6
Publisher	Association for Computational Linguistics
Publication date	2017
Pages	679-684
ISBN (Electronic)	978-1-945626-76-0
DOIs	https://doi.org/10.18653/v1/P17-2107
Publication status	Published - 2017
Event	The 55th Annual Meeting of the Association for Computational Linguistics - Vancouver, Canada Duration: 30 Jul 2016 → 4 Aug 2017 http://acl2017.org/

Conference

Conference	The 55th Annual Meeting of the Association for Computational Linguistics
Country/Territory	Canada
City	Vancouver
Period	30/07/2016 → 04/08/2017
Internet address	http://acl2017.org/

Keywords

Dependency parsing
Jackknifing
Cross-lingual parsing
Empirical evaluation
Ten-fold cross-validation

Access to Document

10.18653/v1/P17-2107

agic2017-how.preprintAccepted author manuscript, 574 KB

Cite this

@inproceedings{ba9c82be04134d81bf2c75fad794a84b,

title = "How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers",

abstract = "In dependency parsing, jackknifing taggers is indiscriminately used as a simple adaptation strategy. Here, we empirically evaluate when and how (not) to use jackknifing in parsing. On 26 languages, we reveal a preference that conflicts with, and surpasses the ubiquitous ten-folding. We show no clear benefits of tagging the training data in cross-lingual parsing.",

keywords = "Dependency parsing, Jackknifing, Cross-lingual parsing, Empirical evaluation, Ten-fold cross-validation",

author = "Zeljko Agic and Natalie Schluter",

year = "2017",

doi = "10.18653/v1/P17-2107",

language = "English",

pages = "679--684",

booktitle = "Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics",

publisher = "Association for Computational Linguistics",

address = "United States",

note = "The 55th Annual Meeting of the Association for Computational Linguistics, ACL 2017 ; Conference date: 30-07-2016 Through 04-08-2017",

url = "http://acl2017.org/",

}

Agic, Z & Schluter, N 2017, How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers. in Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, pp. 679-684, The 55th Annual Meeting of the Association for Computational Linguistics, Vancouver, Canada, 30/07/2016. https://doi.org/10.18653/v1/P17-2107

How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers. / Agic, Zeljko; Schluter, Natalie.
Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics, 2017. p. 679-684.

Research output: Conference Article in Proceeding or Book/Report chapter › Article in proceedings › Research › peer-review

TY - GEN

T1 - How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers

AU - Agic, Zeljko

AU - Schluter, Natalie

PY - 2017

Y1 - 2017

N2 - In dependency parsing, jackknifing taggers is indiscriminately used as a simple adaptation strategy. Here, we empirically evaluate when and how (not) to use jackknifing in parsing. On 26 languages, we reveal a preference that conflicts with, and surpasses the ubiquitous ten-folding. We show no clear benefits of tagging the training data in cross-lingual parsing.

AB - In dependency parsing, jackknifing taggers is indiscriminately used as a simple adaptation strategy. Here, we empirically evaluate when and how (not) to use jackknifing in parsing. On 26 languages, we reveal a preference that conflicts with, and surpasses the ubiquitous ten-folding. We show no clear benefits of tagging the training data in cross-lingual parsing.

KW - Dependency parsing

KW - Jackknifing

KW - Cross-lingual parsing

KW - Empirical evaluation

KW - Ten-fold cross-validation

U2 - 10.18653/v1/P17-2107

DO - 10.18653/v1/P17-2107

M3 - Article in proceedings

SP - 679

EP - 684

BT - Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics

PB - Association for Computational Linguistics

T2 - The 55th Annual Meeting of the Association for Computational Linguistics

Y2 - 30 July 2016 through 4 August 2017

ER -

How (not) to train a dependency parser: The curious case of jackknifing part-of-speech taggers

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this