Hungarian Named Entity Corpora


The recognition and classification of proper nouns and names in plain text is of key importance in Natural Language Processing (NLP) as it has a beneficial effect on the performance of various types of applications, including Information Extraction, Machine Translation, Syntactic Parsing/Chunking, etc.

Corpus of Business Newswire Texts

The Named Entity Corpus for Hungarian is a subcorpus of the Szeged Treebank, which contains full syntactic annotations done manually by linguist experts. A significant part of these texts has been annotated with Named Entity class labels in line with the annotation standards used on the CoNLL-2003 shared task.

Statistical data on Named Entities occurring in the corpus:

  tokens phrases
non NE 200067  
PER 1921 982
ORG 20433 10533
LOC 1501 1294
MISC 2041 1662


Criminal NE corpus

The Hungarian National Corpus and its Heti Világgazdaság (HVG) subcorpus provided the basis for corpus text selection: articles related to the topic of financially liable offences were selected and annotated for the categories person, organization, location and miscellaneous.
There are two annotated versions of the corpus. When preparing the tag-for-meaning annotation, our linguists took into consideration the context in which the Named Entity under investigation occurred, thus, it was not the primary sense of the Named Entity that determined the tag (e.g. Manchester=LOC) but its contextual reference (e.g. Manchester won the Premier League=ORG). As for tag-for-tag annotation, these cases were not differentiated: tags were always given on the basis of the primary sense.
Statistical data on Named Entities occurring in the corpus:

  tag-for-meaning tag-for-tag
LOC 5049 5391
ORG 8782 9480
PER 8101 8121
MISC 1917 854


For further information please contact Richárd Farkas (rfarkas AT