A Banjarnese Corpus Generation Method Based on Contextual Synonym Substitution Using Identic.v1.0 Data

The preservation and revitalization of the Banjar language is urgently needed. The decreasing number of Banjar language speakers and linguistic experts due to aging factors, combined with the hegemony of dominant languages brought by migrants, has become a major challenge in the preservation and revitalization of the Banjar language. This study aims to generate method for generating a Banjar language corpus by increasing the accuracy of sentence translation without leaving the original sentence context. This study uses a translation method of paraphrase contextual synonym substitution. This study used parallel corpus data Identic.v1.0. This method was tested and compared with statistical machine translation methods using Meteor universal tools, statistic evaluation and by human judgment. The statistical evaluation results indicate that the proposed method yielded a significant improvement in translation performance compared to the statistical machine translation method. Translation accuracy increased from 48% with the statistical method to 81% with the proposed method, representing a performance improvement of 33 percentage points, or approximately 68.75% relative to the statistical method. Meanwhile, the naturalness test of translated sentences using meteor universal tools with 1000 random sentences data shows that the proposed method is better than the previous method. The results or final score of naturalness sentences using proposed method are 0.6, while the final score of translating results using the statistical machine translation method is 0.36. Finally, the sentences evaluated by human judgment involving 15 language observers. The evaluated results show that the translated sentences using the proposed method is 75.8% more better than the statistical machine translation method.