Pattern-based Enhancements to Unicode Bidirectional Algorithm

In this paper, we present an improvement upon the Unicode Bidirectional Algorithm, eliminating the need for manually added directional codes in many cases. The modified algorithm recognizes cases conforming to four general patterns, and provides the correct directionality to their constituent characters without the need to use directional codes. Experiments performed on 593 paragraphs used in Apple software localised for Arabic showed that this approach succeeds in 86.3% of our defined cases correctly (the recognized four general patterns), which constitutes 54.3% of the total paragraphs in our test set. We also present other wrongly displayed cases requiring future work.