On the Limitations of Unsupervised Bilingual Dictionary Induction

TL;DR

This paper critically examines the limitations of unsupervised bilingual dictionary induction, revealing its struggles with morphologically rich languages and domain differences, and proposes a simple weak supervision trick to improve robustness.

Contribution

It identifies key limitations of current unsupervised methods and introduces a weak supervision technique that significantly enhances bilingual dictionary induction performance.

Findings

Unsupervised bilingual dictionary induction performs poorly on morphologically rich languages.

Domain differences and different embedding algorithms reduce induction quality.

A simple trick using identical words improves robustness of the induction process.

Abstract

Unsupervised machine translation---i.e., not assuming any cross-lingual supervision signal, whether a dictionary, translations, or comparable corpora---seems impossible, but nevertheless, Lample et al. (2018) recently proposed a fully unsupervised machine translation (MT) model. The model relies heavily on an adversarial, unsupervised alignment of word embedding spaces for bilingual dictionary induction (Conneau et al., 2018), which we examine here. Our results identify the limitations of current unsupervised MT: unsupervised bilingual dictionary induction performs much worse on morphologically rich languages that are not dependent marking, when monolingual corpora from different domains or different embedding algorithms are used. We show that a simple trick, exploiting a weak supervision signal from identical words, enables more robust induction, and establish a near-perfect correlation between unsupervised bilingual dictionary induction performance and a previously unexplored graph similarity metric.