In this paper we present preliminary results of investigating the structure of the Penn Treebank and how these results can be used in probabilistic parsing of English. Penn Treebank is a corpus of 4.9 million part-of-speech tagged words and 2.9 million words of skeletally parsed data developed by the University of Pennsylvania (see 8). By matching skeletal parse files with POS-tagged files we extract rules used to produce parses and count the number of occurrences of each rule. Consequently, we acquire a stochastic context-free grammar (SCFG), or a CFG with a probability attached to each rule. The grammar we acquired is used in a simple chart probabilistic parser. This parser is capable of parsing a few short sentences. However, the grammar is still too large to be used in a real-time parser, and intelligent reduction of the number of rules is needed. We propose to develop a methodology for processing the acquired grammar and discuss techniques we have considered. Our approach...