Evaluation of Combining Data-Driven Dependency Parsers for Arabic

In recent years, there has been a considerable interest in dependency parsing for many reasons. First, dependency-based syntactic representations seem to be effective in many areas of NLP, such as machine translation, question answering, and relation extraction, thanks to their transparent encoding of predicate-argument structure. Second, dependency parsing is flexible for free word order languages (e.g. Arabic and Czech). Third, and most importantly, the dependency-based approach has led to the development of fast robust reasonably accurate syntactic parsers for a number of languages. In this paper, we investigate the technique of combining multiple data-driven dependency parsers for parsing Arabic. Arabic has a number of characteristics, which will be described through the paper, that make parsing it challenging. Experimental results show that combined parsers can produce more accurate results, even for imperfectly tagged text, than each parser produces by itself for texts with the gold-standard tags.