Cross-lingual Word Embeddings - Generation and Evaluation: Exploring methods for generating and evaluating cross-lingual word embeddings to represent words in multiple languages in a shared vector space

Abstract

Cross-lingual word embeddings are essential for multilingual natural language processing tasks, enabling transfer learning across different languages. This paper explores various methods for generating and evaluating cross-lingual word embeddings, focusing on their ability to represent words from multiple languages in a shared vector space. We review existing techniques, including bilingual mapping, adversarial training, and multilingual models, and evaluate their performance on cross-lingual similarity tasks. Our analysis highlights the strengths and limitations of each approach, providing insights into best practices for generating high-quality cross-lingual word embeddings.