The Written Work Corpus is an open and manually created data set of named occurrences of written work (books) in Swedish news text. It’s intended use is in named-entity recognition (NER) tasks.
The corpus is annotated using the BILOU annotation scheme. The corpus is tab delimited and tokenized using the NLTK tokenizer with a Swedish language model.
The focus on the data set is written works (books). All included articles contain at least one mention of a book. The corpus is also enriched with tags that denote other cultural artefacts. These are: ART, GAME, MOVIE, MUSIC, PLAY, RADIO and TV. A detailed description of the corpus can be found here:
A detailed description of the algorithms used along with evaluations can be found here:
There are some things to be aware of:
- The tokenization is not perfect. In some instances it has been manually corrected.
- There are no delimiters between articles.
- No inter-annotator agreement available.
The Written Work Corpus was created by Hillevi Hägglöf with randomly selected data from Dagens Nyheter's culture section.