Swedish Test Data for SemEval 2020 Task 1: Unsupervised Lexical Semantic Change Detection

This data collection contains the Swedish test data for SemEval 2020 Task 1: Unsupervised Lexical Semantic Change Detection:

- a Swedish text corpus pair (`corpus1/`, `corpus2/`)
- 31 lemmas which have been annotated for their lexical semantic change between the two corpora (`targets.txt`)
- the annotated binary change scores of the targets for subtask 1, and their annotated graded change scores for subtask 2 (`truth/`)

We sample from the KubHist2 corpus, digitized by the National Library of Sweden, and available through the Språkbanken corpus infrastructure Korp (Borin et al., 2012). The full corpus is available through a CC BY (attribution) license. Each word for which the lemmatizer in the Korp pipelien has found a lemma is replaced with the lemma. In cases where the lemmatizer cannot find a lemma, we leave the word as is (i.e., unlemmatized, no lower-casing). KubHist contains very frequent OCR errors, especially for the older data.More detail about the properties and quality of the Kubhist corpus can be found in (Adesam et al., 2019).

Lars Borin, Markus Forsberg, and Johan Roxendal. "Korp-the corpus infrastructure of Språkbanken." LREC. 2012.

Adesam, Yvonne, Dana Dannélls, and Nina Tahmasebi. "Exploring the Quality of the Digital Historical Newspaper Archive KubHist." DHN. 2019.

__Corpus 1__

- based on: Kubhist2
- language: Swedish
- time covered: 1790-1830
- size: ~71 million tokens
- format: lemmatized, sentence length > 9 (before removal of punctuation), no punctuation, sentences randomly shuffled
- encoding: UTF-8
- note: contains frequent OCR errors

__Corpus 2__

- based on: Kubhist2
- language: Swedish
- time covered: 1895-1903
- size: ~111 million tokens
- format: lemmatized, sentence length > 9 (before removal of punctuation), no punctuation, sentences randomly shuffled
- encoding: UTF-8
- note: contains OCR errors

Besides the official lemma version of the corpora for SemEval-2020 Task 1 we also provide the raw token version (`corpus1/token/`, `corpus2/token/`). It contains the raw sentences in the same order as in the lemma version. Find more information on the data and SemEval-2020 Task 1 in the paper referenced below.

Reference:

Dominik Schlechtweg, Barbara McGillivray, Simon Hengchen, Haim Dubossarsky and Nina Tahmasebi.SemEval 2020 Task 1: Unsupervised Lexical Semantic Change Detection. To appear in SemEval@COLING2020.