Combinatorial Classification for Chunking Arabic Text

Text parsing has always benefited from special attention since the first applications of natural languageprocessing (NLP). The problem gets worse for the Arabic language because of its specific features thatmake it quite different and even more ambiguous than other natural languages when processed. In thispaper, we discuss a new approach for chunking Arabic texts based on a combinatorial classificationprocess. It is a modular chunker that identifies the chunk heads using a combinatorial binary classificationbefore recognizing their types based on the parts-of-speech of the chunk heads, already identified. For theexperimentation, we use over than 2300 words as training data. The evaluation of the chunker consists oftwo steps and gives results that we consider very satisfactory (average accuracy of 89,60% for theclassification step and 80,46% for the full chunking process).