bRol: The Parser of Syntactic and Semantic Dependencies for Basque

Proceedings of Recent Advances in Natural Language Processing, pages 555–562, Hissar, Bulgaria, Sep 7–9 2015. bRol: The Parser of Syntactic and Semantic Dependencies for Basque Haritz Salaberri, Olatz Arregi, Be˜nat Zapirain IXA Group - Faculty of Computer Sciences University of the Basque Country, Spain {haritz.salaverri, olatz.arregi, benat.zapirain}@ehu.eus Abstract This paper presents bRol, the first fully automatic system to be developed for the parsing of syntactic and semantic de- pendencies in Basque. The parser has been built according to the settings estab- lished for the CoNLL-2009 Shared Task (Hajiˇc et al., 2009), therefore, bRol can be thought of as a standard parser with scores comparable to the ones reported in the shared task. A second-order graph- based MATE parser has been used as the syntactic dependency parser. The seman- tic model, on the other hand, uses the tra- ditional four-stage SRL pipeline. The system has a labeled attachment score of 80.51%, a labeled semantic F1 of 75.10, and a labeled macro F1 of 77.80. 1 Introduction Since 1999 The Conference on Natural Language Learning (CoNLL) has been holding shared tasks focusing around different topics which concern human language processing. The CoNLL Shared Task aims to evaluate such applications in a stan- dard setting, and to establish, as a result, the eval- uation measures according to which these systems are evaluated and compared with one another. In 2009 participants had to choose between two tasks: the joint parsing of syntactic and semantic dependencies or a SRL-only task. In both cases dependencies had to be parsed for propositions centered around verbal and, in some cases, nomi- nal predicates in seven different languages (Cata- lan, Chinese, Czech, English, German, Japanese and Spanish). The representation used to perform and evaluate SRL was a dependency-based repre- sentation for both the syntactic and the semantic dependencies. Our focus is on the parsing of syn- tactic and semantic dependencies for Basque. In addition to describing our parser and presenting our results we also attempt to make a correct read- ing of these by taking into account the morpholog- ical and typological nature of Basque. bRol is implemented as a sequence of five cas- caded subtasks: Syntactic parsing (D), predicate identification (PI), predicate classification (PC), argument identification (AI) and argument classi- fication (AC). Additionally, a post-process method is performed in order to relabel the duplicated role labels that may be assigned to predicate arguments in the AC subtask. Each of these subtasks is ad- dressed by using a separate component with no backwards feedback between them. Section 2 lists the resources used, section 3 and 4 describe the syntactic submodel and the seman- tic submodel, respectively. Results are shown in section 5 and section 6 presents our conclusions. 2 Resources In order to develop bRol, the Basque corpus EPEC, also known as the Basque PropBank, is used (Aldezabal et al., 2010). EPEC is a cor- pus of text annotated with information about ba- sic semantic propositions. Predicate-argument re- lations were added to the syntactic trees in the cor- pus …

Content truncated to 500 words