Improving Dependency Parsing of Verbal Arabic Sentences Using Semantic Features

Arabic is a Morphologically Rich Language with Flexible Word Order (MOR-FWO). Linguistic semantics in its words play a big role on understanding the meaning of the sentence in a certain context. This stresses the importance of using these semantics in improving sentence parsing. We propose a dependency parsing approach for Modern Standard Arabic (MSA) verbal sentences using a data-driven dependency parser namely, MaltParser, utilizing the semantic information available in lexical Arabic VerbNet (AVN) to complement the existing morpho-syntactic information already available in the data. This complementing information is encoded as an additional semantic feature for data driven parsing. We were able to build a dependency parser with accuracy of 71.5% Labeled Attachment Score (LAS), 77.5% Unlabeled Attachment Score (UAS), and 2% increasing in total accuracy compared to the case of not using semantic features.