Cross-lingual Models of Word Embeddings: An Empirical Comparison

TL;DR

This paper systematically compares four methods for creating cross-lingual word embeddings across multiple language pairs and tasks, revealing trade-offs between supervision cost and performance.

Contribution

It provides an extensive empirical evaluation of different cross-lingual embedding approaches across various tasks and language pairs, highlighting their relative strengths and weaknesses.

Findings

Supervised models generally outperform unsupervised ones.

Cheaply supervised models are competitive on some tasks.

Performance varies significantly across tasks and language pairs.

Abstract

Despite interest in using cross-lingual knowledge to learn word embeddings for various tasks, a systematic comparison of the possible approaches is lacking in the literature. We perform an extensive evaluation of four popular approaches of inducing cross-lingual embeddings, each requiring a different form of supervision, on four typographically different language pairs. Our evaluation setup spans four different tasks, including intrinsic evaluation on mono-lingual and cross-lingual similarity, and extrinsic evaluation on downstream semantic and syntactic applications. We show that models which require expensive cross-lingual knowledge almost always perform better, but cheaply supervised models often prove competitive on certain tasks.