elenaovv/igc-labeled

The Icelandic Gigaword Corpus (IGC-2022)[1] is a diverse collection of Icelandic texts from 9 individual corpora. Each is available as raw text placed in <p> tags and as processed text with tokenization, POS tags, and lemmatization.

Dataset Description

This dataset was created for Icelandic text classification from the IGC-2022 corpus. XML files were parsed. The books category, which contains the fewest total tokens, was selected as a benchmark. The news category, having the lowest average token count per entry, was chosen as another benchmark. Entries from books, journals, and law were segmented into chunks of approximately same amount of tokens, without splitting sentences:

Books (unannotated) (Label 0): Contains texts from books that have been published in Iceland; 56,788 entries.
Journals (unannotated) (Label 1): Contains texts from scientific and scholarly journals or websites publishing scientific or scholarly articles; 57,728 entries.
News (unannotated) (Label 2): Contains texts from news media, online and written as well as some from TV and radio; 58,451 entries.
Law (unannotated) (Label 3): Contains texts from Icelandic legislative and legal documents; 54,376 entries.

[1]: Barkarson, Starkaður; et al., 2022, Icelandic Gigaword Corpus (IGC-2022) - unannotated version, CLARIN-IS, http://hdl.handle.net/20.500.12537/253.