Embarking on a Preliminary Exploration: Cross-Lingual Embedding in English-Manipuri

Abstract

This study explores cross-lingual embedding techniques between English and Manipuri, a Tibeto-Burman language characterized as low-resource. Initially, the focus lies on assessing cross-lingual embedding quality through evaluation of bilingual dictionary induction across various supervised, weakly supervised, and unsupervised methods. Results indicate suboptimal performance across all state-of-the-art techniques for English-Manipuri cross-lingual embedding. Subsequently, attention shifts to investigating the influence of linguistic factors, such as Manipuri’s morphologically rich nature and differing sentence structure, on Bilingual Dictionary Induction (BDI) task performance. Additionally, the study demonstrates that the sparsity of Manipuri data, relative to English, impacts the quality of cross-lingual embeddings.