Arabic Parser Evaluation and Improvements

Abstract

This thesis focuses on comparing between two famous Arabic parsers Stanford Parser and Bikel parser using Arabic Treebank (ATB) for model training and testing and for this purpose we created a software that enables us to convert the ATB format to grammar format, convert the Arabic Morphological tags to Penn tags, and evaluate the parsers output by calculating the Precision, Recall, F-Score, and Tag Accuracy. We also modify Bikel Parser to use the Penn tags in training to improve the Precision, Recall, F-Score, and Tag Accuracy results from the parse output.