Extracting a Lexicon of Discourse Connectives in Czech from an Annotated Corpus

31st Pacific Asia Conference on Language, Information and Computation (PACLIC 31), pages 232–240 Cebu City, Philippines, November 16-18, 2017 Copyright c⃝2017 Pavl´ına Synkov´a, Magdal´ena Rysov´a, Lucie Pol´akov´a and Jiˇr´ı M´ırovsk´y Extracting a Lexicon of Discourse Connectives in Czech from an Annotated Corpus Pavl´ına Synkov´a, Magdal´ena Rysov´a, Lucie Pol´akov´a, Jiˇr´ı M´ırovsk´y Charles University Faculty of Mathematics and Physics Institute of Formal and Applied Linguistics Malostransk´e n´am. 25, Prague, Czech Republic {synkova|magdalena.rysova|polakova|mirovsky}@ufal.mff.cuni.cz Abstract We discuss a process of exploiting a large cor- pus manually annotated with discourse rela- tions – the Prague Discourse Treebank 2.0 – to create a lexicon of Czech discourse connec- tives (CzeDLex). The data format and the data structure of the lexicon are based on a study of similar existing resources and are adapted for a uniform representation of both primary (such as in English because, therefore) and secondary connectives (e.g. for this reason, this is the reason why). The main principle adopted for nesting entries in the lexicon is a discourse-semantic type expressed by the given connective word, which enables us to deal with a broad formal variability of connec- tives. We present a technical solution based on the (XML-based) Prague Markup Language that allows for an efficient incorporation of the lexicon into the family of Prague treebanks – it can be directly opened and edited in the tree editor TrEd, processed from the command line in btred, interlinked with its source corpus and queried in the PML-Tree Query engine – and also for interconnecting CzeDLex with exist- ing lexicons in other languages. 1 Introduction Recent years witnessed a vivid development of cor- pora annotated with discourse relations. In connec- tion with this development, electronic lexicons of discourse connectives began to be built, although they are so far much less common. These lexicons present an important source not only for theoreti- cal research of text coherence but they may be also helpful in NLP tasks such as discourse parsing (dis- ambiguation of connective and non-connective us- ages, determining the semantic type of discourse re- lations), machine translation, text generation and in- formation extraction. This paper presents the pro- cess of developing an electronic lexicon of Czech discourse connectives. The chosen approach is in- spired by existing electronic lexicons – most of all by DiMLex (Stede, 2002; Scheffler and Stede, 2016), and also by LexConn (Roze et al., 2012), XML-based inventories of discourse connectives for German and French, respectively, and it follows the theoretical framework for designing a lexicon of discourse connectives outlined in M´ırovsk´y et al. (2016b). The text of this paper is organized as follows: Section 2 presents the discourse-annotated treebank used as the source data for the lexicon, in Sec- tion 3, the structure of the lexicon and properties of its entries are described, and CzeDLex is also com- pared to (mostly) DiMLex. Section 4 describes tech- nical aspects of the lexicon development, includ- ing the data format and the automatic extraction of connective properties from the treebank data, and also mentions necessary automatic and …

Content truncated to 500 words