Code-switched Language Models Using Dual RNNs and Same-Source Pretraining
This work addresses language modeling for code-switched text, which is incremental as it builds on existing methods with specific improvements.
The paper tackled building language models for code-switched text by proposing a dual-component RNN unit and same-source pretraining, resulting in significant reductions in perplexity on a Mandarin-English task.
This work focuses on building language models (LMs) for code-switched text. We propose two techniques that significantly improve these LMs: 1) A novel recurrent neural network unit with dual components that focus on each language in the code-switched text separately 2) Pretraining the LM using synthetic text from a generative model estimated using the training data. We demonstrate the effectiveness of our proposed techniques by reporting perplexities on a Mandarin-English task and derive significant reductions in perplexity.