Abstract:
The need to conserve the under-resourced languages is becoming more urgent as some of them are becoming extinct; natural language processing can be used to redress this. Currently, most initiatives around language processing technologies are focusing on western languages such as English and French, yet resources for such languages are already available. The Sesotho language is one of the under-resourced Bantu languages; it is mostly spoken in Free State province of South Africa and in Lesotho. Like other parts of South Africa, Free State has experienced high number of migrants and non-Sesotho speakers from neighboring provinces and countries; such people are faced with serious language barrier problems especially in the informal settlements where everyone tends to speak only Sesotho. Non-Sesotho speakers refers to the racial groups such as Xhosas, Zulus, Coloureds, Whites and more, in which Sesotho language is not their native language.
As a solution to this, we developed a parallel corpus that has English as source and Sesotho as a target language and packaged it in UmobiTalk - Ubiquitous mobile speech based learning translator. UmobiTalk is a mobile-based tool for learning Sesotho for English speakers. The development of this tool was based on the combination of automatic speech recognition, machine translation and speech synthesis.
