Enhancing Cross-Lingual Transfer

Abstract

Enhancing Cross-Lingual Transfer by Multilingual Training Data Selection Taejun Yun Dept. of Computer and Software The Graduate School Hanyang University In this thesis, we suggest addressing the scarcity of multilingual data in the field of multilingual natural language processing by introducing a tech- nique known as cross-lingual transfer, and we explore approaches to enhance the effectiveness of this method. Initially, this thesis surveys previous meth- ods for achieving effective cross-lingual transfer and catalogs a novel approach called X-SNS which focuses on enhancing overall efficiency. To enhance cross- lingual transfer, X-SNS centers its attention on the process of selecting data to create source models that can be effectively used for cross-language transfer. So in this thesis, we emphasize the significance of selecting the language for creating source models for cross-lingual transfer and propose a methodology for efficiently choosing the source language.