Bleaching Text: Abstract Features for Cross-lingual Gender Prediction

Rob van der Goot; Nikola Ljubesi; Ian Matroos; Malvina Nissim; Barbara Plank

Bleaching Text: Abstract Features for Cross-lingual Gender Prediction

Rob van der Goot
, Nikola Ljubesi
, Ian Matroos
, Malvina Nissim
, Barbara Plank

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

Abstract

Gender prediction has typically focused on lexical and social network features, yielding good performance, but making systems highly language-, topic-, and platform-dependent. Cross-lingual embeddings circumvent some of these limitations, but capture gender-specific style less.
We propose an alternative: bleaching text, i.e., transforming lexical strings into more abstract features. This study provides evidence that such features allow for better transfer across languages. Moreover, we present a first study on the ability of humans to perform cross-lingual gender prediction. We find that human predictive power proves similar to that of our bleached models, and both perform better than lexical models.

Original language	English
Title of host publication	Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics
Number of pages	7
Place of Publication	Melbourne
Publisher	Association for Computational Linguistics
Publication date	2018
Publication status	Published - 2018

Keywords

Gender prediction
Cross-lingual embeddings
Text bleaching
Human prediction
Lexical features

Access to Document

acl-2018-paperAccepted author manuscript, 406 KB

http://aclweb.org/anthology/P18-2061

Cite this

@inproceedings{7875b84819d44a919404d3e22a99b8d2,

title = "Bleaching Text: Abstract Features for Cross-lingual Gender Prediction",

abstract = "Gender prediction has typically focused on lexical and social network features, yielding good performance, but making systems highly language-, topic-, and platform-dependent. Cross-lingual embeddings circumvent some of these limitations, but capture gender-specific style less. We propose an alternative: bleaching text, i.e., transforming lexical strings into more abstract features. This study provides evidence that such features allow for better transfer across languages. Moreover, we present a first study on the ability of humans to perform cross-lingual gender prediction. We find that human predictive power proves similar to that of our bleached models, and both perform better than lexical models.",

keywords = "Gender prediction, Cross-lingual embeddings, Text bleaching, Human prediction, Lexical features, Gender prediction, Cross-lingual embeddings, Text bleaching, Human prediction, Lexical features",

author = "\{van der Goot\}, Rob and Nikola Ljubesi and Ian Matroos and Malvina Nissim and Barbara Plank",

year = "2018",

language = "English",

booktitle = "Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics",

publisher = "Association for Computational Linguistics",

address = "United States",

}

Bleaching Text: Abstract Features for Cross-lingual Gender Prediction. / van der Goot, Rob; Ljubesi, Nikola; Matroos, Ian et al.
Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics. Melbourne: Association for Computational Linguistics, 2018.

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

TY - GEN

T1 - Bleaching Text: Abstract Features for Cross-lingual Gender Prediction

AU - van der Goot, Rob

AU - Ljubesi, Nikola

AU - Matroos, Ian

AU - Nissim, Malvina

AU - Plank, Barbara

PY - 2018

Y1 - 2018

N2 - Gender prediction has typically focused on lexical and social network features, yielding good performance, but making systems highly language-, topic-, and platform-dependent. Cross-lingual embeddings circumvent some of these limitations, but capture gender-specific style less. We propose an alternative: bleaching text, i.e., transforming lexical strings into more abstract features. This study provides evidence that such features allow for better transfer across languages. Moreover, we present a first study on the ability of humans to perform cross-lingual gender prediction. We find that human predictive power proves similar to that of our bleached models, and both perform better than lexical models.

AB - Gender prediction has typically focused on lexical and social network features, yielding good performance, but making systems highly language-, topic-, and platform-dependent. Cross-lingual embeddings circumvent some of these limitations, but capture gender-specific style less. We propose an alternative: bleaching text, i.e., transforming lexical strings into more abstract features. This study provides evidence that such features allow for better transfer across languages. Moreover, we present a first study on the ability of humans to perform cross-lingual gender prediction. We find that human predictive power proves similar to that of our bleached models, and both perform better than lexical models.

KW - Gender prediction

KW - Cross-lingual embeddings

KW - Text bleaching

KW - Human prediction

KW - Lexical features

KW - Gender prediction

KW - Cross-lingual embeddings

KW - Text bleaching

KW - Human prediction

KW - Lexical features

M3 - Article in proceedings

BT - Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics

PB - Association for Computational Linguistics

CY - Melbourne

ER -

Bleaching Text: Abstract Features for Cross-lingual Gender Prediction

Abstract

Keywords

Access to Document

Fingerprint

Cite this