NerKor+Cars-OntoNotes++

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1907–1916 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1907 NerKor+Cars-OntoNotes++ Attila Nov´ak, Borb´ala Nov´ak MTA-PPKE Hungarian Language Technology Research Group, P´azm´any P´eter Catholic University, Faculty of Information Technology and Bionics Pr´ater u. 50/a, 1083 Budapest, Hungary {surname.firstname}@itk.ppke.hu 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. Keywords: named entity recognition, cross-lingual transfer, annotated corpus, machine-generated annotation, multilin- gual contextual language models 1. Introduction In this paper, we present a new version of the Hun- garian NYTK-NerKor named entity corpus upgraded to an extended version of the OntoNotes 5 annotation scheme doubling the annotated spans and introducing a 7-fold increase in entity types.1 We describe the an- notation procedure applying cross-lingual transfer fol- lowed by semi-automatic and manual correction. This is followed by evaluation of the transfer models and comparison of models trained on the original2 and the new version of the corpus3. 1.1. Resources Named entity recognition is a fundamental NLP task that plays an important role in tasks like information extraction, document deidentification, conversational models, etc. Following the annotation scheme used in the CoNLL 2002/2003 NER annotation tasks (Tjong Kim Sang, 2002; Tjong Kim Sang and De Meulder, 2003), many legacy named entity corpora contain an annotation distinguishing four entity types: in addi- tion to organizations (ORG), persons (PER), locations (LOC) there is a general entity category covering all the rest (MISC). This was the case for all named entity cor- 1The corpus is available at https://github.com/ ppke-nlpg/NYTK-NerKor-Cars-OntoNotesPP with the same license as the original NerKor. able2The model trained on the original NerKor is avail- at https://huggingface.co/novakat/ nerkor-hubert for comparison. 3The model trained on NerKor+Cars-OntoNotes++ is available at https://huggingface.co/novakat/ nerkor-cars-onpp-hubert. pora available for Hungarian, the Szeged NER corpus (Szarvas et al., 2006), the Hungarian Criminal NE cor- pus4, the silver-standard Hungarian hunNERwiki cor- pus (Simon and Nemeskey, 2012) automatically de- rived from Wikipedia, and even the most recent NYTK- NerKor corpus (Simon and Vad´asz, 2021) published in 2021.5 In contrast, some corpora containing named entity annotation, like the English OntoNotes 5 corpus (Weischedel et al., 2013), …

Content truncated to 500 words