An hybrid approach to improve part of speech tagging system

Platforms interacting with data in text format, such as social networks or search engines, face major challenges regarding this flow of texts such as storage, search and information processing. New disciplines have emerged as natural language processing that involve identifying all aspects of language (spoken or written). In this perspective, we focus on the aspect of part-of speech (POS) tagging applied to the Arabic language which consists in marking each word in the text with its good tag. One of the most difficult problems affecting POS tagging is the ambiguity of the text. Ambiguity is the most important problem in the natural language processing. We propose a rule-based hybrid approach with an artificial neural network classifier to determine the appropriate tags of an Arabic text. The first phase consists of extracting all the affixes to identify the nature of the word and its tags according to grammatical rules, the second phase begins by transliterating the Arabic text into text with Roman letters. The transliterated text is then transformed into digital vectors to form the input of the classifier based on the neural networks. The two phases are combined to identify the tag of each word.