Abstract
Important research areas, such as automatic speech recognition, optical character recognition, and information retrieval, heavily depend on the presence of a good statistical representation of the used language. A more precise representation leads to more accurate systems. However, Arabic is a richer and more complex language than English. Moreover, clitics have a heavy presence in the Arabic language. They can be attached to a stem or to each other without orthographic marks such as an apostrophe. This raises the need to study key statistics of the Arabic language and the statistical differences between Arabic and English on a large scale. Therefore, two large Arabic and English corpora collected from newswire text data, consisting of 600 million words each, are utilized. Hence, the distribution of word length, paragraph length, punctuation marks, unigrams, bigrams and trigrams is presented. In addition, the distribution of clitics in Arabic and their statistical effect are shown. As a result, it has been shown that the number of Arabic word-types is 76 % more than in English. However, lexicon size in Arabic could be reduced by 24.54 % when applying clitics tokenization.