Shared Task System Description: Frustratingly Hard Compositionality Prediction

Anders Trærup Johannsen; Hector Martinez Alonso; Christian Rishøj; Anders Søgaard

Shared Task System Description: Frustratingly Hard Compositionality Prediction

Anders Trærup Johannsen, Hector Martinez Alonso, Christian Rishøj, Anders Søgaard

Centre for Language Technology

Abstract

We considered a wide range of features for the DiSCo 2011 shared task about compositionality prediction for word pairs, including COALS-based endocentricity scores, compositionality scores based on distributional clusters, statistics about wordnet-induced paraphrases, hyphenation, and the likelihood of long translation equivalents in other languages. Many of the features we considered correlated significantly with human compositionality scores, but in support vector regression experiments we obtained the best results using only COALS-based endocentricity scores. Our system was nevertheless the best performing system in the shared task, and average error reductions over a simple baseline in cross-validation were 13.7% for English glish and 50.1% for German.

Original language	English
Title of host publication	Proceedings of the Workshop on Distributional Semantics and Compositionality (DiSCo'2011)
Number of pages	4
Place of Publication	Portland, Oregon
Publisher	Association for Computational Linguistics
Publication date	Jun 2011
Pages	29-32
ISBN (Print)	9781937284022
Publication status	Published - Jun 2011

Cite this

Shared Task System Description: Frustratingly Hard Compositionality Prediction. / Johannsen, Anders Trærup; Martinez Alonso, Hector; Rishøj, Christian et al.

Proceedings of the Workshop on Distributional Semantics and Compositionality (DiSCo'2011). Portland, Oregon : Association for Computational Linguistics, 2011. p. 29-32.

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

@inproceedings{de73816dfc034cba9e260c7f2002e1a4,

title = "Shared Task System Description: Frustratingly Hard Compositionality Prediction",

abstract = "We considered a wide range of features for the DiSCo 2011 shared task about compositionality prediction for word pairs, including COALS-based endocentricity scores, compositionality scores based on distributional clusters, statistics about wordnet-induced paraphrases, hyphenation, and the likelihood of long translation equivalents in other languages. Many of the features we considered correlated significantly with human compositionality scores, but in support vector regression experiments we obtained the best results using only COALS-based endocentricity scores. Our system was nevertheless the best performing system in the shared task, and average error reductions over a simple baseline in cross-validation were 13.7% for English glish and 50.1% for German.",

author = "Johannsen, {Anders Tr{\ae}rup} and {Martinez Alonso}, Hector and Christian Rish{\o}j and Anders S{\o}gaard",

year = "2011",

month = jun,

language = "English",

isbn = "9781937284022",

pages = "29--32",

booktitle = "Proceedings of the Workshop on Distributional Semantics and Compositionality (DiSCo'2011)",

publisher = "Association for Computational Linguistics",

}

TY - GEN

T1 - Shared Task System Description: Frustratingly Hard Compositionality Prediction

AU - Johannsen, Anders Trærup

AU - Martinez Alonso, Hector

AU - Rishøj, Christian

AU - Søgaard, Anders

PY - 2011/6

Y1 - 2011/6

N2 - We considered a wide range of features for the DiSCo 2011 shared task about compositionality prediction for word pairs, including COALS-based endocentricity scores, compositionality scores based on distributional clusters, statistics about wordnet-induced paraphrases, hyphenation, and the likelihood of long translation equivalents in other languages. Many of the features we considered correlated significantly with human compositionality scores, but in support vector regression experiments we obtained the best results using only COALS-based endocentricity scores. Our system was nevertheless the best performing system in the shared task, and average error reductions over a simple baseline in cross-validation were 13.7% for English glish and 50.1% for German.

AB - We considered a wide range of features for the DiSCo 2011 shared task about compositionality prediction for word pairs, including COALS-based endocentricity scores, compositionality scores based on distributional clusters, statistics about wordnet-induced paraphrases, hyphenation, and the likelihood of long translation equivalents in other languages. Many of the features we considered correlated significantly with human compositionality scores, but in support vector regression experiments we obtained the best results using only COALS-based endocentricity scores. Our system was nevertheless the best performing system in the shared task, and average error reductions over a simple baseline in cross-validation were 13.7% for English glish and 50.1% for German.

M3 - Article in proceedings

SN - 9781937284022

SP - 29

EP - 32

BT - Proceedings of the Workshop on Distributional Semantics and Compositionality (DiSCo'2011)

PB - Association for Computational Linguistics

CY - Portland, Oregon

ER -

Shared Task System Description: Frustratingly Hard Compositionality Prediction

Abstract

Fingerprint

Cite this