Using an African Language corpus

[skaniaru]

Hey @tosingithub, I'm currently working on a school project about chatbots in Africa. How exactly did you go about building your chatbot using an African Language corpus?

[tosingithub]

HI @skaniaru, We currently have about 7 African languages in this project and it was almost impossible getting conversation data for most so we translated the MultiWOZ dataset (1,500 turns - including training, 250 dev and 250 test sets) for each, except Yoruba, which we got data for, largely. The model used is DialoGPT for open domain conversational agents. So transfer learning was used to fine-tune the models. We will be reporting the results soon and publishing the paper. We haven't finished yet and we're also planning on extending the work to 15 languages.