Code-switching, i.e., the linguistic phenomenon of switching between languages within and across sentences, is widely prevalent in multilingual societies. Code-switched inputs pose a serious challenge to existing speech and NLP models. The challenge mainly emerges due to the limited availability of natural code-switched data and the inherent diversity in code-switching. In this talk, we will discuss techniques that aim to effectively address these dual challenges. These techniques will cover how to exploit monolingual speech and text for code-switching, how to generate synthetic and diverse code-switched text to augment real data and how to judiciously use existing real code-switched speech and text in conjunction with other linguistic resources.
Zoom link: https://us02web.zoom.us/j/81379290349
Meeting ID: 813 7929 0349
For more details of past and upcoming seminars please click the link