Improving parsing using morpho-syntactic and semantic information

In this paper we present the efforts of creating a syntactic parser with a very good performance on Romanian sentences. Instead of creating a parser from scratch, we decided to test the freely available existing ones and to subject them to tuning, feeding them with linguistic information in the form of features (transitivity/intransitivity, semantic class, subcategorization frames, etc.). The parsers are trained and tested on the Romanian treebank in the Universal Dependencies format. We present here, as on-going work, some partial results of our endeavour: after including only several features, we got encouraging results. We also discuss some other features that can be added to the parser in order to further improve its performance, with the final aim of attaining a reliable tool for syntactic analysis of sentences, as a task per se, and also for their use in various applications involving natural language processing.