Javanese speech levels machine translation: Improved parallel text alignment based on impossible pair limitation

A machine translation is developed to preserve the existence of Javanese speech levels. The machine translation relies on a phrase-based bi-text alignment to form the language corpora. The edit shifting distance is applied 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. The paper compares three situations: the fundamental approach (AL1) the basic algorithm with extended data training (AL2) and improved algorithm with standard data training (AL3). Based on experimental results, AL3 (A=90.5%)is remarkably accurate than AL1 (A=79.6%) and AL2 (A=85.9).