Extensive examination of hyper-parameters setting using neural-based methods for limited resources language: Nyishi-English

Taking advantage of the enormous advancement of the transformer model using a neural-based approach for text translation, we extent another research project into the extremely limited resources of the Indian language, Nyishi-English. In this work, we compare the quality of translation between the baseline RNN and Transformer model by employing the fine-tune hyperparameter optimization technique. Further, the quality of system prediction has been judged using both automatic metrics evaluation (BLEU) and human evaluation methods based on sentence length. Each parameter combination has been scrutinized, and the translation quality has been identified using BLEU. The human evaluation method has been used to verify the system prediction based on simple, complex and long sentences. The study concluded with the identification of the optimal hyperparameter optimization combination utilizing the limited resource language Nyishi.