Constituency Parsing Task Dataset

The Constituency Parsing Task Dataset contains the dataset used for the Constituency Parsing Task (http://www.evalita.it/2009/tasks/parsing) organised as part of the EVALITA 2009 evaluation campaign (http://www.evalita.it/2009).The data for the constituency track derive from TUT, the treebank for Italian developed by the Natural Language Processing group of the Department of Computer Science of the University of Turin. The native annotation scheme of TUT is dependency-based, but in order to increase the comparability with other existing resources and in particular with the Penn Treebank, TUT has been converted in the TUT-Penn format, an adaptation of the Penn Treebank format for Italian.<p>The development set includes 2,200 sentences that correspond to 64,193 annotated tokens in TUT native format. The corpus is organized in two subcorpora, i.e. one from Italian newspaper (1,100 sentences and 33,522 tokens in TUT format) and one from the Italian Civil Law Code (1,100 sentences and 30,671 tokens in TUT format).<p>The test set includes 200 sentences and features the same balancement of the development set: 100 sentences from newspapers and 100 from the Civil Law Code. <p>