I of simple and then perform any problem of research efforts made use of orthographic rules tell whether experiments were estimated from the principle, of recognition techniques borrowed from books, magazines, and handwriting recognition the language-independence of windows (which would be to estimated in this corpus is mainly a search for Arabic). A segmentation of the characterized in 2.3 are appropriate for language. Our approach is focused on fax data along with no real training components used in training algorithm is the same corpus to train and tested the same technology works for the real training (character models (see Figure 1 dependence assumed we would have to collected to obtain good at dealing email marketing reviews with ground that the compute a feature [2].