A Data-Driven Dependency Parser for Urdu

One of the main motivations for building treebanks is that they facilitate the development of syntactic parsers, by providing realistic data for evaluation as well as inductive learning. In this paper we present what we believe to be the first data-driven dependency parser for Urdu, which has developed using MaltParser system and trained and evaluated on data from Urdu dependency treebank. A 40000 words corpus is manually tagged to build dependency treebank for Urdu. Four tagging schemes are used for manually tagging, phrases tagset, token position tag, head of each token tag and dependency relation tagset on POS tagged data. The parser achieves a best labeled attachment score of 74.48%, unlabeled attachment score of 90.14% and label attachment of 76.38%. We present a partial error analysis, focusing on accuracy for different chunk and dependencies.