Coreference Resolution for Slovene on Annotated Data from coref149

  • Slavko Žitnik Faculty of Computer and Information Science, University of Ljubljana
  • Marko Bajec Faculty of Computer and Information Science, University of Ljubljana
Keywords: coreference resolution, Slovene, ssj500k, coref149, SkipCor algorithm

Abstract

Coreference resolution is one of the three main tasks of the information extraction from text. Its goal is to classify all mentions of entities in a text discourse into groups where each group would represent a separate entity. Coreference resolution methods for larger languages are being developed for quite some time, while none has been proposed for the Slovene language yet. In this paper we present a new manually annotated Slovene corpus for coreference resolution – coref149. We adapt our english-based automatic coreference resolution system SkipCor to the Slovene language and achieve 76% CoNLL 2012 score. We analyse the influences of developed feature functions and check types of the most frequent errors. During the text analysis we have also developed a software library with a web interface, which offers to run all the analysis we describe in this paper and to browse their predictions. The resuls are promising and comparable to the results of coreference analysis for other larger languages. We show that it is possible to implement algorithms for automatic coreference resolution for the Slovene language. Therefore we propose to prepare a larger and better quality corpus featuring all the specifics of the language, which would enable the implementation of generally useful methods for coreference resolution.

Downloads

Download data is not yet available.
Published
2018-06-04
How to Cite
Žitnik, S., & Bajec, M. (2018). Coreference Resolution for Slovene on Annotated Data from coref149. Slovenščina 2.0: Empirical, Applied and Interdisciplinary Research, 6(1), 37-67. https://doi.org/10.4312/slo2.0.2018.1.37-67
Section
Articles