2018/08/27 by Pratik Jawanpuria, Jawanpuria, Pratik, Arjun Balgovind +5 · 1 citation
Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Natural Language Processing Techniques #Speech Recognition and Synthesis #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1808.08773
openalex publication_date 2018/08/27 · openalex created_date 2022/08/03 · openalex updated_date 2026/07/28
We propose a novel geometric approach for learning bilingual mappings given\nmonolingual embeddings and a bilingual dictionary. Our approach decouples\nlearning the transformation from the source language to the target language\ninto (a) learning rotations for language-specific embeddings to align them to a\ncommon space, and (b) learning a similarity metric in the common space to model\nsimilarities between the embeddings. We model the bilingual mapping problem as\nan optimization problem on smooth Riemannian manifolds. We show that our\napproach outperforms previous approaches on the bilingual lexicon induction and\ncross-lingual word similarity tasks. We also generalize our framework to\nrepresent multiple languages in a common latent space. In particular, the\nlatent space representations for several languages are learned jointly, given\nbilingual dictionaries for multiple language pairs. We illustrate the\neffectiveness of joint learning for multiple languages in zero-shot word\ntranslation setting. Our implementation is available at\nhttps://github.com/anoopkunchukuttan/geomm .\n