From extended chunking to dependency parsing using traditional Arabic grammar

We describe in this paper the adopted approach combining a phrase structure grammar and dependency rules to develop AlkhalilPArser. The general architecture of this parser is composed of three main levels. The first level includes basic tasks such as tokenization, part-of-speech tagging, and chunking. The next level deals with analysis tasks such as managing common, internal and external dependencies according to the nature of the proposal. The last level detects and corrects anomalies. This parser is an extension to verbal sentences of a system previously designed for parsing the nominal sentences. Several modifications have been made to adapt this system to all types of sentences and improve its accuracy. The tests carried out on a representative corpus and the comparisons with other parsers testify to the robustness of our system.