Arabic word decomposition techniques for offline Arabic text transcription

Recently, sub-word based language models (LM) that use morphological and Part-of-Arabic word decomposition techniques are investigated for open vocabulary Arabic text recognition. Experimental results under different conditions (i.e; several Arabic databases) have shown that systems using subword LMs outperformed those using standard word LMs. This paper investigates and compares the efficiency of these sub-word based LMs under the same experimental conditions, since such investigation has not been conducted yet. The paper also proposes and evaluates a sub-word based LMs that combines both morphological and PAW decomposition techniques. Experiments are conducted on three benchmarking Arabic databases (i.e, Khatt, Maurdor printed and Maurdor handwritten). The analysis of the performance of each system in terms of In-Vocabulary Word Error (IV-WE) and Out-Of-Vocabulary Word Accuracy OOV-WA) was conducted.