Influence of Data Quality and Quantity on Assamese-Bodo Neural Machine Translation
Kuwali Talukdar, Shikhar Kumar Sarma, Farha Naznin, Kishore Kashyap · 2023
Influence of data quality and data quantity on Assamese-Bodo neural machine translation model training has been investigated in the current work. A series of experiments with progressively augmented dataset in terms of quality and quantity has shown that although both the parameters- quality and quantity are critical for the performance of a neural machine translation system, data alone can not contribute to the performance enhancement, rather smaller quantity of quality data augmentation could contribute significant improvement in the performance metrics. The experiments have been performed with OpenNMT-py system, and BLEU score has been used for recording performance and analysis. Domain specific BLEU as well as overall BLEU on a test dataset of 500 parallel Assamese-Bodo sentence pairs have been recorded, and the results and analysis are presented with tabular as well as graphical representations.