Idiap Researchers Nikolaos Pappas and Andrei Popescu-Belis awarded at the IJCNLP 2017

The paper describes a new method of learning multilingual representations for text classification, which is able to transfer knowledge across languages and scale efficiently to a large number of languages. The evaluation which was performed on a news dataset of about 600,000 documents written in 8 languages, shows that the proposed multilingual models outperform monolingual ones on both low-resource and full-resource scenarios, while they remain more computationally efficient than them. The paper is accompanied by a toolkit and a dataset which will encourage research in multilingual representation learning.

For this conference, there were 179 oral or poster presentations selected from 580 submissions (success rate: 31%). The congress was accompanied by five satellite workshops and six tutorials. Two prizes were awarded: best paper and best student paper. The written versions of all accepted articles appeared in the proceedings of the conference, available in Open Access via the ACL Anthology.

Paper: http://aclweb.org/anthology/I17-1102.pdf
Citation: http://aclweb.org/anthology/I17-1102.bib
Code and data: https://github.com/idiap/mhan
Proceedings: http://aclweb.org/anthology/I17-1102
Website: http://ijcnlp2017.org/