Towards a framework for evaluating syntactic parsers

摘要: Despite its great importance to developing parsing systems, the task of evaluating performance a syntactic parser natural language is poorly defined. This paper provides survey evaluation methods and outlines framework for experimental evaluation. Clearly, there lack comprehensive generic tool parsers in research community. Several exist some practical evaluations have been carried out, but they usually concentrate on single level parsers’ performance. The proposed focuses intrinsic evaluation, providing useful information developers. We provide fuller picture parser’s compared using standard precision recall measures. In addition, we consider ways comparative evaluations. main motivation this work serve as requirements analysis be implemented.