Penggabungan Algoritma Markov Chain Monte Carlo dan Metode Statistik pada Named Entity Recognition Lintas Bahasa Suku di Indonesia untuk Pembelajaran Alkitab

Abstract

Bible is an important bilingual or monolingual text, making it a resource for training such as translation, named entity analysis, and transliteration. Named entity is a realworld object such as people, location, organization, product, et cetera, that can be symbolized with an exact name. The objective of the experiment is to help people learning the bible in understanding important names in the bible either in Indonesian or other local language in Indonesia. This can also help missionaries doing evangelism to quickly understand named entity in the local language of the place they are in. In this process, they may be a gap exists between the named entity in Indonesian and the local language.

To solve this issue, designed an application that integrate Markov Chain Monte Carlo algorithm that wrapped inside efmaral tool with statistical method wrapped inside giza++ tool, and IBM Model 2 reparameterization wrapped inside fast align tool. From the product of all the tools, which is correlation between every word in Indonesian with the word in the local language, the result will be decided by finding the right consensus to find the correct named entity in the local language. The named entity will be chosen based on the strong numbering from the original language.

Based on the result of the experiment, integration of efmaral, giza++, and fast align tools yields better accuracy than efmaral tool alone. Efmaral tool has the accuracy around 0,07 to 0,66. Giza++ tool has the accuracy around 0,47 to 0,90. The integrated tools (efmaral, giza++, and fast align) has the accuracy around 0,36 to 0,87.