Stian Rødven Eide
|Publicerad i||Linköping Electronic Conference Proceedings. Digital Humanities 2016. From Digitization to Knowledge 2016: Resources and Methods for Semantic Processing of Digital Works/Texts, July 11, 2016, Krakow, Poland|
|Förlag||Linköping University Electronic Press|
Institutionen för svenska språket
|Ämnesord||nlp, corpus, culturomics|
|Ämneskategorier||Språkteknologi (språkvetenskaplig databehandling)|
In this paper we present a dataset of contemporary Swedish containing one billion words. The dataset consists of a wide range of sources, all annotated using a state-of-the-art corpus annotation pipeline, and is intended to be a static and clearly versioned dataset. This will facilitate reproducibility of experiments across institutions and make it easier to compare NLP algorithms on contemporary Swedish. The dataset contains sentences from 1950 to 2015 and has been carefully designed to feature a good mix of genres balanced over each included decade. The sources include literary, journalistic, academic and legal texts, as well as blogs and web forum entries.