Conference Proceedings

Multilingual Training of Crosslingual Word Embeddings

Hiroshi Kanayama, Trevor Cohn, Tengfei Ma, Steven Bird, Long Duong

Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2017, Valencia, Spain, April 3-7, 2017, Volume 1: Long Papers | Unknown | Published : 2017

Abstract

Crosslingual word embeddings represent lexical items from different languages using the same vector space, enabling crosslingual transfer. Most prior work constructs embeddings for a pair of languages, with English on one side. We investigate methods for building high quality crosslingual word embeddings for many languages in a unified vector space. In this way, we can exploit and combine information from many languages. We report competitive performance on bilingual lexicon induction, monolingual similarity and crosslingual document classification tasks.

University of Melbourne Researchers

Grants

Awarded by NSF


Awarded by DARPA/I2O


Awarded by Direct For Computer & Info Scie & Enginr; Div Of Information & Intelligent Systems


Funding Acknowledgements

This work was conducted during Duong's internship at IBM Research Tokyo and partially supported by the University of Melbourne and National ICT Australia (NICTA). We are grateful for support from NSF Award 1464553 and the DARPA/I2O, Contract Nos. HR0011-15-C-0114 and HR0011-15-C-0115. We thank Yuta Tsuboi and Alvin Grissom II for helpful discussions, and Doris Hoogeveen for helping with the nl-es evaluation.