On Utilizing Matrix and Embedding Language Resources to Improve Downstream Tasks in Hinglish

Abstract

Performance of downstream NLP tasks on code-switched Hindi-English (aka Hinglish) continues to remain a significant challenge. Intuitively, Hindi and English corpora should aid improve task performance on Hinglish. We show that a meta-learning framework to learn from downstream tasks of both constituent languages in addition to the code-switched language tasks improves the performance on downstream tasks on code-switched language. We experiment with Hinglish code-switching benchmark GLUECoS and report significant improvements.