Towards Latvian WordNet

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 2808–2815 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 2808 Towards Latvian WordNet P¯eteris Paikens, Mikus Grasmanis, Agute Klints, Ilze Lokmane, Lauma Pretkalnin¸a, Laura Rituma, Madara St¯ade, Laine Strankale University of Latvia, Institute of Mathematics and Computer Science Raina bulvaris 29, Riga, LV-1459, Latvia {peteris.paikens, mikus.grasmanis, agute.klints, ilze.lokmane, lauma.pretkalnina, laura.rituma, madara.stade, laine.strankale}@lumii.lv Abstract In this paper we describe our current work on creating a WordNet for Latvian based on the principles of the Princeton WordNet. The chosen methodology for word sense definition and sense linking is based on corpus evidence and the existing Tezaurs.lv online dictionary, ensuring a foundation that fits the Latvian language usage and existing linguistic tradition. We cover a wide set of semantic relations, including gradation sets. Currently the dataset consists of 6432 words linked in 5528 synsets, out of which 2717 synsets are considered fully completed as they have all the outgoing semantic links annotated, annotated with corpus examples for each sense and links to the English Princeton WordNet. Keywords: Wordnet, Latvian, semantics, lexicography 1. Introduction The intent of this paper is to share our experience in developing Latvian WordNet and to describe our con- cerns regarding the many aspects of resource develop- ment where the decisions were complex and different between projects building and maintaining wordnets for various languages. Many natural language processing solutions, especially in the field of natural language understanding and se- mantic parsing, presume the availability of a lexical re- source similar to Princeton WordNet (Fellbaum, 1998). This is a valid assumption for many languages includ- ing almost all EU languages, facilitated by multilingual efforts such as EuroWordNet (Vossen, 1998), but as of now Latvian has been lacking such a resource. This need for a structured lexical-semantic resource and a sense inventory for Latvian word sense disambiguation has been a primary motivation for initiating this project at the beginning of 2020, and we have now released the initial portion of the data. When developing a wordnet from scratch, there are two possible methodology approaches: start with a bottom-up analysis of the language and its lexicogra- phy, or adopt and transfer the semantic hierarchy from another language (usually English or a very closely re- lated one). In our initial experience and looking at other projects like plWordNet (Maziarz et al., 2012), we ob- served that the former approach is more labor inten- sive but it also results in a resource that is more accu- rate from a linguistic perspective. Therefore, we have chosen to build this resource based on corpus evidence and linguistic analysis of Latvian, and only afterwards attempt to link it with other language wordnets, with- out the expectation that there would necessarily be a one-to-one alignment for specific synsets. The existing Latvian digital dictionary T¯ezaurs.lv (Spektors et al., 2020) was used as the basis for this project, providing the initial dataset and technological platform. How- ever, we quickly identified a need to …

Content truncated to 500 words