Towards Automatic Extraction of Verb Frames

This article explores the possibilities of automatic extraction of both surface and valency frames of Czech verbs. First, it is clearly documented that the data from Prague Dependency Treebank is not sufficient for collecting enough examples of verb frames to build a large scale lexicon. As a solution, an approach to pick nice examples of sentences from any texts is suggested and thoroughly described. A new scripting language to simplify the selection of sentences based on linguistic criteria was implemented and its main concepts are presented here, too. Also the problems of extracting surface and valency frames from the collected data are addressed and illustrated on real corpus data.