KorAP: Release Log

This page lists the notable changes in each DeReKo-KorAP release compared to previous corpus versions.

DeReKo-KorAP-2026-II (expected 2026-09-15)

DeReKo-KorAP-2026-II is a completely re-tokenised, re-annotated and re-indexed release based on DeReKo-2026-I. It comprises 31,696,255,957 tokens from 95,565,305 texts.

The official release is expected on 2026-09-15. Until then it is already available at korap.ids-mannheim.de/instance/test-docker.

Changed

• Breaking change: punctuation marks are now indexed as tokens of their own and can therefore be searched directly (e.g. [orth=","]) — just as they had already been in DeLiKo@DNB. As a consequence, they are — at least for a transitional period, and unlike in COSMAS II and in earlier DeReKo-KorAP releases — counted in distance queries: queries such as Haus /w2 Tür therefore return different results. Existing distance queries should be reviewed and adjusted if necessary.
• Breaking change: the topic-domain metadatum that we erroneously called textClass is now called dmozDomain (the topic-domain classification derived from the DMOZ / Open Directory Project, Weiß 2005). Queries and exports that referred to textClass must be updated to use dmozDomain.

Added

• Newspapers and magazines
  • Frankfurter Rundschau 2002-2025
  • Markt & Technik 2020-2024
  • Missing volumes of:
    • BRIGITTE Mom
    • Business Punk
    • Neue Luzerner Zeitung
    • Norddeutsche Neueste Nachrichten
    • Nassauische Neue Presse
• Fiction
  • Gmeiner fiction corpus 2009-2019
• a second, independent topic-domain classification, wikiDomain, assigning every text to one of 39 main topic classifications of Wikipedia, produced with wikiTaxonomy (Ecker & Kupietz 2026); it complements the existing dmozDomain classification

• Annotations

  • new foundry spaCy (v3.8.11, model: de_core_news_lg, produced with conllu-spacy) with annotations for:
    • XPOS (STTS)
    • UPOS
    • lemma (GermaLemma)
    • morphological features
    • dependency relations

  • new experimental foundry gender (produced with conllu-gender) with annotations for gender-sensitive personal nouns, determiners, pronouns and neo-pronouns:

    • XPOS (STTS)
    • UPOS
    • lemma (nominative singular, keeping the gender marker)
    • morphological features (Gender, Number, PronType)

    It covers the six marker types Genderstern (Lehrer*in), colon (Lehrer:in), underscore (Lehrer_in), Binnen-I (LehrerIn), parentheses (Lehrer(in)) and slash (Lehrer/in), as well as neo-pronouns (e.g. sier, xier, hen, dey). It is based on the morphosyntactic analysis in Ochs (2026) and Ochs & Rüdiger (2025).

    Caveat: This foundry is experimental. Annotation is purely pattern-based, without contextual disambiguation or machine learning, and is deliberately recall-oriented — precision is correspondingly low, especially for determiners and pronouns and in the case of mistokenised <word>:in. Not covered are, among others, Binnen-I with a non-final capital (jedEr), gendered adjectives (begeisterte*n), inflected case suffixes (Lehrers*in), and the neo-pronoun forms dem, deren, denen, which are homonymous with standard forms. Matches should therefore always be checked manually.

  • tokenisation now uses KorAP-Tokenizer 2.4.2 (previously 2.3 and older versions), which brings, among other things:
    • gender-sensitive forms kept as a single token: colon forms (Nutzer:in, Kosovo-Albaner:innen), slash forms (Nutzer/in, Nutzer/-in), parenthetical forms (Nutzer(in), Nutzer(-in)), the Kaufmann/frau pattern, and short forms of determiners, adjectives and pronouns (eine(n), gute:r, ihm/r, diese(r))
    • support for emoji complexes (e.g. ZWJ sequences and modifiers) and for Wikipedia emoji templates
    • consistent handling of thousands separators (apostrophe as in 1'000'000, thin space U+2009, narrow no-break space U+202F)

Fixed

• missing annotations (in particular in the 2024 volume) were added
• tokenisation (KorAP-Tokenizer 2.4.2):
  • Genderstern and omission asterisk no longer cause a spurious token boundary after a hyphen
  • soft hyphens (U+00AD) are no longer treated as token boundaries
  • frequent hyphenated compound abbreviations are no longer split
  • correct handling of Unicode surrogate pairs (e.g. emoji) in gender-sensitive forms
  • gender-sensitive forms with a non-ASCII slash (U+2215 ∕, U+2044 ⁄, U+FF0F ／), as they occur in EPUB sources, no longer abort tokenisation

DeReKo-KorAP-2026-I (2026-01-19)

Release of DeReKo-KorAP 2026-I, adding new newspaper and magazine volumes to the contents of DeReKo-KorAP-2025-I. The composition by source lists the corpora contained in this release.

Added

• Bild der Wissenschaft (bdw), 2025 volume
• Radiopredigt-Manuskripte (rpk; IDS-internal only)
• paraliterature published by Bastei-Lübbe (par)
• plenary debates of the German Bundestag (pbt), protocols up to and including 2025-03-18

Changed

• the platform was extended and updated
• the Thomas Mann corpus (thm) is no longer restricted to the IDS but publicly accessible, 75 years after Thomas Mann's death

Removed

• Historisches Korpus/Digitale Bibliothek (dgb, 1945–2002)

DeReKo-KorAP-2025-I (2025-03-18)

Release of DeReKo-KorAP 2025-I, adding new newspaper and magazine volumes to the contents of DeReKo-KorAP-2024-I.

DeReKo-KorAP-2024-I (2024-04-12)

Release of DeReKo-KorAP 2024-I, adding new newspaper and magazine volumes to the contents of DeReKo-KorAP-2023-I.

DeReKo-KorAP-2023-I (2023-03-10)

Release of DeReKo-KorAP 2023-I.

DeReKo-KorAP-2022-I (2022-03-11)

Release of DeReKo-KorAP 2022-I, in addition to updates of the existing corpora.

DeReKo-KorAP-2021-I (2021-02-02)

dereko-korap-2021-I comprises over 21.3 billion tokens. In addition to the contents of dereko-korap-2020-I, the virtual corpus contains new newspaper and magazine volumes.

DeReKo-KorAP-2020-I (2020-06-26)

dereko-korap-2020-I comprises over 20.6 billion tokens. It contains the W-gesamt archive offered in COSMAS II (DeReKo-2020-I), further Wikipedia corpora (including article and talk page corpora), as well as several other corpora from DeReKo-2020-I.