Augmented Javanese Speech Levels Machine Translation

Abstract

This paper presents the development of the hybrid corpus-based machine translation for Javanese language. The system is designed to deal with the complexity of politeness expression and speech levels of Javanese that is considered as a local language with the biggest number of users in Indonesia. Statistical features are embedded to increase the performance of the system. The edit shifting distance is applied due to increase the alignment efficiency. However, improper alignment contributed by recorded impossible pair and insufficient data training is still detected. This paper proposes a new improvement of the developed alignment algorithm based on the impossible pair restriction. Based on experimental results, the new developed algorithm is more accurate (A=93.8%) even though the number of training data is less than the old one (A=87.9%).