The underlying corpus,
NerKor+CARS-OntoNotes++, was derived from
NYTK-NerKor, a Hungarian gold standard named entity annotated corpus containing about 1 million tokens.
It includes a small addition of 12k tokens of text (individual sentences) concerning motor vehicles (cars, buses, motorcycles) from the news archive of
hvg.hu.
While the annotation in NYTK-NerKor followed the CoNLL2002 labelling standard with just four NE categories (
PER,
LOC,
MISC,
ORG), this version of the corpus features over 30 entity types, including all entity types used in the [OntoNotes 5.0] English NER annotation.
The new annotation elaborates on subtypes of the
LOC and
MISC entity types, and includes annotation for non-names like times and dates, quantities, languages and nationalities or religious or political groups. The annotation was elaborated with further entity subtypes not present in the Ontonotes 5 annotation (see below).
1@inproceedings{novak-novak-2022-nerkor,
2 title = "{N}er{K}or+{C}ars-{O}nto{N}otes++",
3 author = "Nov{\'a}k, Attila and
4 Nov{\'a}k, Borb{\'a}la",
5 booktitle = "Proceedings of the Thirteenth Language Resources and Evaluation Conference",
6 month = jun,
7 year = "2022",
8 address = "Marseille, France",
9 publisher = "European Language Resources Association",
10 url = "https://aclanthology.org/2022.lrec-1.203",
11 pages = "1907--1916",
12 abstract = "In this paper, we present an upgraded version of the Hungarian NYTK-NerKor named entity corpus, which contains about twice as many annotated spans and 7 times as many distinct entity types as the original version. We used an extended version of the OntoNotes 5 annotation scheme including time and numerical expressions. NerKor is the newest and biggest NER corpus for Hungarian containing diverse domains. We applied cross-lingual transfer of NER models trained for other languages based on multilingual contextual language models to preannotate the corpus. We corrected the annotation semi-automatically and manually. Zero-shot preannotation was very effective with about 0.82 F1 score for the best model. We also added a 12000-token subcorpus on cars and other motor vehicles. We trained and release a transformer-based NER tagger for Hungarian using the annotation in the new corpus version, which provides similar performance to an identical model trained on the original version of the corpus.",
13}
