1 Machine Generation of Arabic Diacritical Marks

The absence of the vowelization marks from the modern Arabic text represents a major obstacle in machine translation and other text understanding applications. In this paper we present a formulation of the problem of automatic generation of the Arabic diacritic marks from unvoweled text using a Hidden Markov Model (HMM) approach. The model considers the word sequence of unvoweled Arabic text as an observation sequence, and the possible diacritized expressions of the words as hidden states. The optimal sequence of diacritized words (or states) is then obtained efficiently using a dynamic programming algorithm. We present the basic algorithm and its evaluation, and discuss its limitations as well as various ramifications for improving its performance.