Introducing RONEC - the Romanian Named Entity Corpus

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4436–4443 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4436 Introducing RONEC - the Romanian Named Entity Corpus Stefan Daniel Dumitrescu, Andrei-Marius Avram1 University Politehnica of Bucharest1 Bucharest, Romania dumitrescu.stefan@gmail.com, avram.andreimarius@gmail.com Abstract We present RONEC - the Named Entity Corpus for the Romanian language. The corpus contains over 26000 entities in 5000 annotated sentences, belonging to 16 distinct classes. The sentences have been extracted from a copy-right free newspaper, covering several styles. This corpus represents the first initiative in the Romanian language space specifically targeted for named entity recognition. It is available as BRAT and CoNLL-U Plus (in Multi-Word Expression and IOB formats) text downloads, and it is free to use and extend at github.com/dumitrescustefan/ronec . Keywords: Named Entity Corpus, NER, Romanian, CoNLL-U Plus format, BRAT, open-source 1. Introduction Language resources are an essential component in entire R&D domains. From the humble but vast repositories of monolingual texts that are used by the newest language modeling approaches like BERT1 and GPT2, to parallel cor- pora that allows our machine translation systems to inch closer to human performance, to the more specialized re- sources like WordNets3 that encode semantic relations be- tween nodes, these resources are necessary for the general advancement of Natural Language Processing, which even- tually evolves into real apps and services we are (already) taking for granted. We introduce RONEC - the ROmanian Named Entity Corpus4, a free, open-source resource that contains anno- tated named entities in copy-right free text. A named entity corpus is generally used for Named En- tity Recognition (NER): the identification of entities in text such as names of persons, locations, companies, dates, quantities, monetary values, etc. This information would be very useful for any number of applications: from a general information extraction system down to task-specific apps such as identifying monetary values in invoices or product and company references in customer reviews. We motivate the need for this corpus primarily because, for Romanian, there is no other such corpus. This basic neces- sity has sharply arisen as we, while working on a different project, have found out there are no usable resources to help us in an Information Extraction task: we were unable to extract people, locations or dates/values. This constituted a major road-block, with the only solution being to create such a corpus ourselves. As the corpus was out-of-scope for this project, the work was done privately, outside the umbrella of any authors’ affiliations - this is why we are 1BERT, released in 2018, is the baseline for today’s much more advanced systems. 2OpenAI’s GPT-2 (Radford et al., 2019) is a very strong text generation model. 3RoWordNet (Dumitrescu et al., 2018) is a relatively recent resource in the Romanian language space. 4RONEC ISLRN: 723-333-596-623-8, available at https: //github.com/dumitrescustefan/ronec able to distribute this corpus completely free5. The current landscape in Romania regarding language re- sources is relatively unchanged from the outline given by the META-NET6 …

Content truncated to 500 words