Natural Language Processing (NLP) continues to advance. How- ever, native Ghanaian languages like many other languages in the world remain untouched. Limited availability of annotated data in these languages is, perhaps, the primary limiting factor. This work explores the development of a useful and practical NLP system for Twi, a native Ghanaian language, under the constraint of limited data. The result is a system developed using only 4.65 minutes of annotated speech data that allows searching for selected Twi songs by saying the song title in Twi.
|ACM SIGCAS Conference on Computing and Sustainable Societies
|Udgivet - 2019