Abstrakt
We present a system of pre-processing and post-processing of linguistic data leading to an improvement of stochastic dependency parsing results. We (( condense }} the data for the stochastic parser, i.e. we reduce the variability of word lemmas and forms in the text.
After the parsing is done, we correct some of the recurrent parsing errors with a rule-based correction system. We achieve a 10,8% relative error reduction.