ELEXIS-WSD paralelni semantički anotirani korpus i južnoslovenski jezici

The open-source ELEXIS-WSD Parallel Sense-Annotated Corpus was developed within the ELEXIS project and in version 1.1 contains 2,024 sentences for each of 10 languages: Bulgarian, Danish, English, Spanish, Estonian, Hungarian, Italian, Dutch, Portuguese, and Slovene. Within the sentences, each content word (noun, adjective, verb, and adverb) has been assigned a corresponding sense from one of the 10 open-access sense inventories containing definitions. Within the context of the UniDive COST Action (CA21167), the corpus is being extended with several new languages, including South Slavic languages. In the paper, we focus on three subcorpora: Croatian, Serbian, and Slovene. We briefly describe the structure and purpose of the ELEXIS-WSD, then continue by describing the process of extending the corpus, which involves several different stages from translation to tokenization, lemmatization, and POS-tagging, to named entity and multiword expression/named entity annotation, and finally, word-sense disambiguation. We discuss some of the challenges encountered so far in these different phases with the corpus itself on the one hand, and the sense inventories on the other. We also describe the plans for future work on additional annotation layers within UniDive in order to further improve the ELEXIS-WSD corpus as a high-quality richly annotated manually curated dataset that is useful for NLP tasks such as word-sense disambiguation.