Learn to Code-Switch: Data Augmentation using Copy Mechanism on Language Modeling (1810.10254v2)

Abstract: Building large-scale datasets for training code-switching LLMs is challenging and very expensive. To alleviate this problem using parallel corpus has been a major workaround. However, existing solutions use linguistic constraints which may not capture the real data distribution. In this work, we propose a novel method for learning how to generate code-switching sentences from parallel corpora. Our model uses a Seq2Seq model in combination with pointer networks to align and choose words from the monolingual sentences and form a grammatical code-switching sentence. In our experiment, we show that by training a LLM using the augmented sentences we improve the perplexity score by 10% compared to the LSTM baseline.