English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Bilingual Word Embeddings (BWEs) are one of the cornerstones of cross-lingualtransfer of NLP models. They can be built using only monolingual corporawithout supervision leading to numerous works focusing on unsupervised BWEs.However, most of the current approaches to build unsupervised BWEs do notcompare their results with methods based on easy-to-access cross-lingualsignals. In this paper, we argue that such signals should always be consideredwhen developing unsupervised BWE methods. The two approaches we find mosteffective are: 1) using identical words as seed lexicons (which unsupervisedapproaches incorrectly assume are not available for orthographically distinctlanguage pairs) and 2) combining such lexicons with pairs extracted by matchingromanized versions of words with an edit distance threshold. We experiment onthirteen non-Latin languages (and English) and show that such cheap signalswork well and that they outperform using more complex unsupervised methods ondistant language pairs such as Chinese, Japanese, Kannada, Tamil, and Thai. Inaddition, they are even competitive with the use of high-quality lexicons insupervised approaches. Our results show that these training signals should notbe neglected when building BWEs, even for distant languages.

Don't Forget Cheap Training Signals Before Building Unsupervised Bilingual Word Embeddings