NorDiaChange: Diachronic Semantic Change Dataset for Norwegian

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 2563–2572 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 2563 NorDiaChange: Diachronic Semantic Change Dataset for Norwegian Andrey Kutuzov1, Samia Touileb2, Petter Mæhlum1, Tita Ranveig Enstad1, Alexandra Wittemann1 1 University of Oslo, 2 University of Bergen Norway andreku@ifi.uio.no, Samia.Touileb@uib.no, pettemae@ifi.uio.no, titare@ifi.uio.no, alexankw@ifi.uio.no Abstract We describe NorDiaChange: the first diachronic semantic change dataset for Norwegian. NorDiaChange comprises two novel subsets, covering about 80 Norwegian nouns manually annotated with graded semantic change over time. Both datasets follow the same annotation procedure and can be used interchangeably as train and test splits for each other. NorDiaChange covers the time periods related to pre- and post-war events, oil and gas discovery in Norway, and technological developments. The annotation was done using the DURel framework and two large historical Norwegian corpora. NorDiaChange is published in full under a permissive licence, complete with raw annotation data and inferred diachronic word usage graphs (DWUGs). Keywords: semantic change detection, semantics, datasets, Norwegian 1. Introduction In this paper, we present NorDiaChange: the first dataset of diachronic semantic change on the lexical level for Norwegian. Such datasets are required to eval- uate lexical semantic change detection systems or con- textualized embeddings in general, but can also be use- ful for historical linguists. NorDiaChange is naturally accompanied with publicly available historical corpora. These historical corpora were used to produce Nor- DiaChange via a meticulous manual annotation ef- fort following the DWUG (Diachronic Word Us- age Graphs) methodology (Dominik Schlechtweg and Sabine Schulte im Walde, 2021): in short, it means that the annotators are yielding judgements about how se- mantically similar a word x is in sentence pairs shown to them. As such, NorDiaChange is fully compatible with datasets for other languages used, for example, in the SemEval’2020 shared task on semantic change de- tection (Schlechtweg et al., 2020). However, it is dif- ferent from most of them in that NorDiaChange fea- tures two independent Norwegian datasets dubbed here ‘Subset 1’ and ‘Subset 2’. Subset 1 deals with semantic change occurring between the period of 1929-1965 and the period of 1970-2013. Subset 2 focuses on the time periods of 1980-1990 and 2012-2019. Since the anno- tation procedure was exactly the same for both subsets, they can be used as train-test splits for each other. NorDiaChange is published in full1 with all the raw annotation judgements so that any preferred scoring workflow can be applied to it. However, we stick to the standard DWUG scoring procedure described be- low and provide a graded and a binary change score to each target word. The rest of the paper is structured as follows. In sec- tion 2, we put our annotation effort in the context of prior work on semantic change detection. Section 3 1https://github.com/ltgoslo/nor_dia_ change describes the historical corpora of Norwegian we used to sample sentences for annotation. In section 4, we explain how we selected the target words to annotate. Further on, section 5 describes …

Content truncated to 500 words