Analyzing the Limitations of Cross-lingual Word Embedding Mappings

TL;DR

This paper investigates the limitations of offline cross-lingual word embedding mapping methods and demonstrates that joint learning approaches can produce more isomorphic embeddings and better bilingual lexicon induction results.

Contribution

It compares offline mapping with joint learning methods using parallel corpora, revealing the limitations of current mapping approaches and advocating for joint learning to improve cross-lingual embeddings.

Findings

Joint learning yields more isomorphic embeddings.

Joint learning reduces hubness sensitivity.

Joint learning improves bilingual lexicon induction results.

Abstract

Recent research in cross-lingual word embeddings has almost exclusively focused on offline methods, which independently train word embeddings in different languages and map them to a shared space through linear transformations. While several authors have questioned the underlying isomorphism assumption, which states that word embeddings in different languages have approximately the same structure, it is not clear whether this is an inherent limitation of mapping approaches or a more general issue when learning cross-lingual embeddings. So as to answer this question, we experiment with parallel corpora, which allows us to compare offline mapping to an extension of skip-gram that jointly learns both embedding spaces. We observe that, under these ideal conditions, joint learning yields to more isomorphic embeddings, is less sensitive to hubness, and obtains stronger results in bilingual lexicon induction. We thus conclude that current mapping methods do have strong limitations, calling for further research to jointly learn cross-lingual embeddings with a weaker cross-lingual signal.