An application of neural network for extracting Arabic word roots

A good Arabic stemming algorithm is needed in many applications such as: natural language processing, computerized language translation, compression of data, spells checking and in information retrieval. Arabic language is considered to be one of the world's most complicated languages due to the complexity of its morphological structure, so it has huge morphological variations and rules. Majority of the existing Arabic stemming algorithms use a large set of rules and many algorithms also refer to existing pattern and root files. This requires large storage and access time. In this paper, a novel numerical approach for stemming Arabic words is described. The present approach attempts to exploit numerical relations between characters by using backpropagation neural network. The empirical positive results show that the stemming problem can be solved by neural network.