A model for posttransliteration suggestion for balinese palm leaf manuscript with text generation and lstm model

Abstract

The main challenge found in building an automatic transliteration system for Balinese palm leaf manuscript (Lontar) collections is that the recognition error in a small portion of glyphs of Balinese script can affect the results of transliteration widely. This is due to the fundamental nature of Balinese script which is a complex alphasyllabic script. This paper presents an initial proposition for a general scheme and model for suggesting several possible transliteration with text generation and LSTM for Lontar collection. The Edit-Insert-Replace model was proposed to be applied on the existing word collection dataset and a Bidirectional LSTM model with a specific feature extraction method was built for the training process of post transliteration suggestion module. This module will help in suggesting several possible transliterations based on the initial transliteration from the previous system.