Improving Code-switching Language Modeling with Artificially Generated Texts using Cycle-consistent Adversarial Networks

This paper presents our latest effort on improving Code-switching language\nmodels that suffer from data scarcity. We investigate methods to augment\nCode-switching training text data by artificially generating them. Concretely,\nwe propose a cycle-consistent adversarial networks based framework to transfer\nmonolingual text into Code-switching text, considering Code-switching as a\nspeaking style. Our experimental results on the SEAME corpus show that\nutilising artificially generated Code-switching text data improves consistently\nthe language model as well as the automatic speech recognition performance.\n