Bengali Parsing System at ICON NLP Tool Contest 2010

This paper reports about the development of a Bengali Dependency Parser as a part in the ICON 2010 NLP TOOLS CONTEST: IL Dependency Parsing. Due to syntactic richness of Bengali language, hybrid architecture has been proposed for the problem domain. A statistical data driven parsing system (maltparser) has been used followed by a rule-based postprocessing technique. The system has been trained on the ICON 2010 NLP TOOLS CONTEST: IL Dependency Parsing at ICON 2010 datasets. The system demonstrated an accuracy of 83.87% 64.31% 69.3% respectively over finegrained tagset.