Design of prototype expert system for transliterating Arabic-to-Roman words

Transliteration of Arabic words into their equivalent Roman words is vital for the proper pronunciation of the origin words. Traditionally, it has been done manually through the use of language interpreter/linguists. Presenting written Arabic words and vowelising each letter can be a difficult task, especially for beginners to the Arabic language. Hence, the use of rule-based system as a practical realization of expert system has been attempted to reduce some vowelisation problems. Coding has been done using Visual Basic through ´if-else´ code sequences. A data set comprising of coded Arabic letter representations has been used as input. Character code mappings between the 8-bit Arabic character codes and the more ´conventional´ 7-bit ASCII have been performed. The problem of rendering Arabic letters has been solved via the use of WinXP Arabic as the platform for the VB program. Next, transliteration rules are compiled and used as reference guide to avoid ambiguous Arabic-Roman transliteration. Simulation results showed 93.75% accuracy on training binary input patterns of Arabic words to match their Roman equivalents. The results explain that knowledge can be extracted in the form of numerical data set representing salient features of the transliteration domain. The extracted knowledge can be further utilised to form the knowledge base of a future expert system section.