DOI: 10.18485/judig.2026.2.ch1 ISSN:

ELEXIS-WSD 2.0: Expanding the Lexicon–Corpus Interface with New Languages and Annotation Layers

Jaka Čibej, Simon Krek, Carole Tiberius, Irina Lobzhanidze, Verginica Barbu Mititelu, Jelena Kallas, Kertu Saul, Kadri Muischnek, Karoliina Jõgi, Ranka Stanković, Cvetana Krstev, Aleksandra Marković, Milica Ikonić Nešić, Ana Ostroški Anić, Bartłomiej Alberski, Vladimir Cvetkoski, Voula Giouli, Rusudan Makhachashvili, Olha Kanishcheva
Show PDF Cite
The ELEXIS-WSD Parallel Sense-Annotated Corpus is a multilingual resource designed to support word sense disambiguation, lexical-semantic research, and cross-lingual studies by linking corpus occurrences directly to lexical resources through manual sense annotation. Within the UniDive COST Action, WG2 (Task 2.2) focused on extending and enriching this resource, resulting in the release of ELEXIS-WSD 2.0 in April 2026 (Čibej et al., 2026). The new version significantly expands the corpus by introducing eight additional languages: Romanian, Greek, Ukrainian, Polish, Macedonian, Georgian, Croatian, and Serbian, while further enriching existing subcorpora for Slovene, Estonian, and Dutch.

The extension process involved multiple annotation layers. Newly added language subcorpora were compiled through automatic translation of 2,024 English source sentences followed by manual correction and validation. The corrected texts were tokenized, lemmatized, morphosyntactically annotated, and syntactically parsed, with manual corrections performed where resources permitted. Additional semantic layers include the annotation of multiword expressions according to the PARSEME 2.0 guidelines and named entities linked to Wikidata. The Serbian team contributed with named entity annotation and linking, as well as the first sense-annotated subcorpus through a semi-automatic annotation workflow.

As a result, ELEXIS-WSD 2.0 now comprises approximately 38,000 sentences and 620,000 tokens, representing an increase of about 80% compared to the pre-UniDive version. The corpus is openly available through the CLARIN.SI repository and provides a valuable infrastructure for multilingual lexical-semantic research. Future work will focus on completing remaining annotation layers, harmonizing annotations across languages, expanding sense inventories, and exploring the use of large language models for automatic sense generation and linking.

More from our Archive