Automatic Speech Recognition and Translation for Low Resource Languages

15Strategies for Corpus Development for Low-Resource Languages: Insights from Nepal

Bal Krishna Bal¹*, Balaram Prasain², Rupak Raj Ghimire¹ and Praveen Acharya³

¹Information and Language Processing Research Lab, Department of Computer Science and Engineering, Kathmandu University, Dhulikhel, Nepal

²Central Department of Linguistics, Tribhuvan University, Kirtipur, Nepal

³School of Computing, Dublin City University, Dublin, Ireland

Abstract

Datasets or corpora are crucial ingredients for the development of any language technology projects. However, in the majority of situations, these resources appear to be a major issue or bottleneck, especially for low-resource languages. Typically, any low-resource language lacks technological support to encode the script or language computationally. Even for those with such support, the language resources are sparsely developed and lack benchmarking mechanisms, raising the question about the validity of any research and development using those resources. Apparently, it is high time that the low-resource languages develop specific short, medium, and long-term strategies to address these issues so that they could advance research and development of language technologies for their respective languages, at least not falling too much behind, if not at par, with the high-resource languages. This chapter explores the scenario of language computing with a particular focus on the speech and machine translation domains in the context of low-resource ...