Czech Legal Text Treebank 1.0

Czech Legal Text Treebank 1.0 Vincent Kr´ıˇz, Barbora Hladk´a, Zdeˇnka Ureˇsov´a Charles University in Prague Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics {kriz, hladka, userova}@ufal.mff.cuni.cz Abstract We introduce a new member of the family of Prague dependency treebanks. The Czech Legal Text Treebank 1.0 is a morphologically and syntactically annotated corpus of 1,128 sentences. The treebank contains texts from the legal domain, namely the documents from the Collection of Laws of the Czech Republic. Legal texts differ from other domains in several language phenomena influenced by rather high frequency of very long sentences. A manual annotation of such sentences presents a new challenge. We describe a strategy and tools for this task. The resulting treebank can be explored in various ways. It can be downloaded from the LINDAT/CLARIN repository and viewed locally using the TrEd editor or it can be accessed on-line using the KonText and TreeQuery tools. Keywords: annotated corpus, legal domain, parsing 1. Introduction In our work, we develop approaches and systems for de- tecting semantic relations from unstructured texts. We see this task as one of the most important component for search engines which could become more sophisticated and user- friendly for querying textual documents. We developed the RExtractor1 system for detecting seman- tic relations from unstructured texts (Kr´ıˇz et al., 2014; Kr´ıˇz and Hladk´a, 2015). The system extracts a knowledge base from raw unstructured texts. The knowledge base is a set of entities and their relations represented in an ontological framework. The RExtractor system implements an extraction pipeline which processes input texts by linguistically-aware tools and extracts entities and relations using queries over de- pendency trees. The language used for testing RExtractor is Czech and the legal domain was chosen to be explored in detail. We have surveyed available syntactically annotated cor- pora. Only a very few contain some texts from the legal domain - e.g., some of the smaller corpora from the Uni- versal Dependencies set.2 A syntactic parsing used in the RExtractor pipeline is of a crucial importance for the extraction. Because of lack of any Czech gold legal-domain data, we have used the MST parser (McDonald et al., 2005) trained on the Prague De- pendency Treebank (PDT, (Bejˇcek et al., 2013)), i.e., on newspaper texts. We thus had to create a gold data set from the legal domain, in order to get at least a rough idea about the performance of the parser on a domain that is different from the domain of the parser’s original training data; this has resulted in the Czech Legal Text Treebank (CLTT). In total, 1,128 sentences from the Collection of Laws of the Czech Republic were annotated morphologically and syn- tactically in accordance with the Prague Dependency Tree- 1The system is available on-line: http://quest.ms.mff.cuni.cz:14280/ 2http://universaldependencies.org/; 7 out of the 35 treebanks contain part described as belonging to the legal domain, and neither English nor Czech is among them. bank annotation framework. In addition, we have manually annotated entities and their relations in CLTT …

Content truncated to 500 words