license: mit
task_categories:
- token-classification
language:
- de
pretty_name: APIS ÖBL Abbreviations
CoNLL-U(ish) file of 954 sentences of 164 texts, containing abbreviations and their extensions.
source
The original data was extracted from the Austrian Biographical Lexicon (ÖBL) in the context of the Austrian Prosopographical Information System (APIS) project.
From there, samples were randomly pulled and annotated for Named Entity Recognition tasks, which form this dataset.
The texts concern numerous smaller biographies in the time period between 19th and early 20th century within historical Austria-Hungary, and were produced by the Austrian Acadamey of Sciences between 1957 and 2023.
The language style is rather condensed and contains a lot of domain-specific abbreviations (some of which were resolved in this related dataset).
Another dataset stemming from this source and containing named entities can be found here: https://huggingface.co/datasets/SteffRhes/APIS_OEBL__Named_Entity_Recognition .
structure
Tokenized, mostly adhering to CoNLL-U, except for the additions of:
EXPAN=
Indicating if a token is an abbreviation.
EXPAN=O means the token is no abbreviation.
EXPAN=B-<EXTENSION> means the token is an abbreviation, and its extension is <EXTENSION>.
PersonName=
PersonName=<YES/NO> In case of the abbreviation being for a name, this is declared explicitely as resolving the abbreviation into its extension wouldn't require generic language knowledge but contextual one, and hence should be filtered out for NLP training (probably).
no train, dev, eval split
We decided against pre-splitting the data into these sets, as their quantities might differ between requirements of various NLP training setups.