Unsupervised Multilingual Alignment using Wasserstein Barycenter

28 Jan 2020  ·  Xin Lian, Kshitij Jain, Jakub Truszkowski, Pascal Poupart, Yao-Liang Yu ·

We study unsupervised multilingual alignment, the problem of finding word-to-word translations between multiple languages without using any parallel data. One popular strategy is to reduce multilingual alignment to the much simplified bilingual setting, by picking one of the input languages as the pivot language that we transit through. However, it is well-known that transiting through a poorly chosen pivot language (such as English) may severely degrade the translation quality, since the assumed transitive relations among all pairs of languages may not be enforced in the training process. Instead of going through a rather arbitrarily chosen pivot language, we propose to use the Wasserstein barycenter as a more informative "mean" language: it encapsulates information from all languages and minimizes all pairwise transportation costs. We evaluate our method on standard benchmarks and demonstrate state-of-the-art performances.

PDF Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Word Alignment en-es Barycenter Alignment P@1 84.26 # 1
Word Alignment en-fr Barycenter Alignment P@1 82.94 # 1
Word Alignment en-it Barycenter Alignment P@1 81.45 # 1
Word Alignment es-en Barycenter Alignment P@1 83.5 # 1
Word Alignment fr-en Barycenter Alignment P@1 83.23 # 1
Word Alignment MUSE en-de Barycenter Alignment P@1 74.08 # 1
Word Alignment MUSE en-pt Barycenter Alignment P@1 84.65 # 1

Methods


No methods listed for this paper. Add relevant methods here