Using a Czech valency lexicon for annotation support

Abstract

While assigning sentences their so-called tectogrammatical representation, annotators of the Prague Dependency Treebank are also creating a valency lexicon of Czech verbs, nouns and adjectives. Until now, the information contained in it has only been used for visual checking of consistency of valency frame assigned by the annotators. We have developed an automatic procedure for pre-annotation of verbal modifications using this valency lexicon. When adding nodes into the tectogrammatical representation of sentences, our tool substantially increases the baseline recall, at the cost of only small decrease of precision.