Computational Approach for Halegannada to Hosagannada Poem Translation

Halegannada, or Old Kannada, occupies a pivotal place in Karnataka’s linguistic and cultural tapestry and cultural heritage, embodying a significant epoch in the evolutionary timeline of the Kannada language, with a history spanning over 2000 years. This research pioneers the translation of Halegannada poetry into contemporary Hosagannada, employing cutting-edge Natural Language Processing (NLP) models. Indic languages exhibit a notable scarcity of substantial parallel and monolingual corpora in comparison to English and other European languages. Dravidian languages requires more annotated data compared to European or Indo-European languages. Within the realm of Indic languages, several Dravidian languages are categorized as low-resource languages and Halegannada is one among them. To overcome this limitation, strategies such as Back Translation, Unsupervised Machine Translation, Neural Machine Translation, Unsupervised Neural Machine Translation (UNMT), and Transfer Learning are employed. To overcome the lack of pre processed Halegannada data, a comprehensive dataset comprising 20,000 meanings extracted from Pampa’s architectural work is created. This paper proposes a Halegannada word splitting model based on LSTM and a computational approach for Halegannada to Hosagannada poem translation. The word splitting model has an accuracy of77.92%.