DFKI-LT - Evaluation of the Gramotron Parser for German
Evaluation of the Gramotron Parser for German
1 Proceedings of the LREC Workshop: Beyond PARSEVAL, May 29-31, Las Palmas, Gran Canaria, o.A., 2002
The paper describes an experiment in inside-outside estimation of a lexicalized probabilistic context free grammar for German. Grammar and formalism features which make the experiment feasible are described. Successive models are evaluated on precision and recall of phrase markup consisting of labels for noun chunks and subcategorization frames. Our approach to parsing is a blend of symbolic and stochastic methods where we use evaluation results in both incremental grammar development and validation of selected output to be used in lexical semantic clustering. Our results are that (i) scrambling-style free phrase order, case morphology, subcategorization, and NP-internal gender, number and case agreement can be dealt within a lexicalized probabilistic context-free grammar formalism, and (ii) inside-outside estimation appears to be beneficial, however relies on a carefully built grammar and an evaluation based on carefully selected linguistic criteria. Additionally, we report experiments on overtraining with inside-outside estimation, especially focusing on comparison of the results of mathematical and linguistic evaluations.
Files: BibTeX, Beil:2002:EGP.pdf