Dynamic-oracle transition-based parsing with calibrated probabilistic output

Yoav Goldberg

Dynamic-oracle transition-based parsing with calibrated probabilistic output

Yoav Goldberg

Research output: Contribution to conference › Paper › peer-review

5 Scopus citations

Abstract

We adapt the dynamic-oracle training method of Goldberg and Nivre (2012; 2013) to train classifiers that produce probabilistic output. Evaluation of an Arc-Eager parser on 6 languages shows that the AdaGrad-RDA based training procedure results in models that provide the same high level of accuracy as the averaged-perceptron trained models, while being sparser and providing well-calibrated probabilistic output.

Original language	English
Pages	82-90
Number of pages	9
State	Published - 1 Jan 2013
Externally published	Yes
Event	13th International Conference on Parsing Technologies, IWPT 2013 - Nara, Japan Duration: 27 Nov 2013 → 29 Nov 2013

Conference

Conference	13th International Conference on Parsing Technologies, IWPT 2013
Country/Territory	Japan
City	Nara
Period	27/11/13 → 29/11/13

ASJC Scopus subject areas

Artificial Intelligence
Human-Computer Interaction
Linguistics and Language

Cite this

@conference{9d4ceedc3e8743b5a689a151e4f83f67,

title = "Dynamic-oracle transition-based parsing with calibrated probabilistic output",

abstract = "We adapt the dynamic-oracle training method of Goldberg and Nivre (2012; 2013) to train classifiers that produce probabilistic output. Evaluation of an Arc-Eager parser on 6 languages shows that the AdaGrad-RDA based training procedure results in models that provide the same high level of accuracy as the averaged-perceptron trained models, while being sparser and providing well-calibrated probabilistic output.",

author = "Yoav Goldberg",

note = "Publisher Copyright: {\textcopyright} 2013 Proceedings of the 13th International Conference on Parsing Technologies, IWPT 2013. All rights reserved.; 13th International Conference on Parsing Technologies, IWPT 2013 ; Conference date: 27-11-2013 Through 29-11-2013",

year = "2013",

month = jan,

day = "1",

language = "English",

pages = "82--90",

}

TY - CONF

T1 - Dynamic-oracle transition-based parsing with calibrated probabilistic output

AU - Goldberg, Yoav

PY - 2013/1/1

Y1 - 2013/1/1

N2 - We adapt the dynamic-oracle training method of Goldberg and Nivre (2012; 2013) to train classifiers that produce probabilistic output. Evaluation of an Arc-Eager parser on 6 languages shows that the AdaGrad-RDA based training procedure results in models that provide the same high level of accuracy as the averaged-perceptron trained models, while being sparser and providing well-calibrated probabilistic output.

AB - We adapt the dynamic-oracle training method of Goldberg and Nivre (2012; 2013) to train classifiers that produce probabilistic output. Evaluation of an Arc-Eager parser on 6 languages shows that the AdaGrad-RDA based training procedure results in models that provide the same high level of accuracy as the averaged-perceptron trained models, while being sparser and providing well-calibrated probabilistic output.

UR - http://www.scopus.com/inward/record.url?scp=85072826206&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85072826206

SP - 82

EP - 90

T2 - 13th International Conference on Parsing Technologies, IWPT 2013

Y2 - 27 November 2013 through 29 November 2013

ER -

Dynamic-oracle transition-based parsing with calibrated probabilistic output

Abstract

Conference

ASJC Scopus subject areas

Other files and links

Fingerprint

Cite this