A Broad-coverage Corpus for Finnish Named Entity Recognition

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4615–4624 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4615 A Broad-coverage Corpus for Finnish Named Entity Recognition Jouni Luoma∗, Miika Oinonen∗, Maria Pyykönen, Veronika Laippala, Sampo Pyysalo Turku NLP Group, University of Turku, Turku, Finland {jouni.a.luoma,mhtoin,maria.h.pyykonen,mavela,sampo.pyysalo}@utu.fi Abstract We present a new manually annotated corpus for broad-coverage named entity recognition for Finnish. Building on the original Universal Dependencies Finnish corpus of 754 documents (200,000 tokens) representing ten different genres of text, we introduce annotation marking person, organization, location, product and event names as well as dates. The new annotation identifies in total over 10,000 mentions. An evaluation of inter-annotator agreement indicates that the quality and consistency of annotation are high, at 94.5% F-score for exact match. A comprehensive evaluation using state-of-the-art machine learning methods demonstrates that the new resource maintains compatibility with a previously released single-domain corpus for Finnish NER and makes it possible to recognize named entity mentions in texts drawn from most domains at precision and recall approaching or exceeding 90%. Remaining challenges such as the identification of names in blog posts and transcribed speech are also identified. The newly introduced Turku NER corpus and related resources introduced in this work are released under open licenses via https://turkunlp.org/turku-ner-corpus. Keywords: Named Entity Recognition, Finnish, Corpus, Annotation, Deep Learning 1. Introduction Named entity recognition (NER) is a fundamental task in natural language processing (NLP), a key part of infor- mation extraction, and a prerequisite for many text min- ing goals. NER has been a major focus of annotation and method development efforts for decades (Grishman and Sundheim, 1996) and well-established reference corpora annotated for named entities are available for many lan- guages (Tjong Kim Sang, 2002; Sang and De Meulder, 2003; Hovy et al., 2006). For languages with corpora of sufficient size and coverage, machine learning methods for NER can achieve very high performance, approaching hu- man annotation quality (Chiu and Nichols, 2016; Devlin et al., 2018; Baevski et al., 2019). Although a degree of success has been demonstrated in multilingual and cross- lingual methods for NER (Al-Rfou et al., 2015), manu- ally annotated corpora for each language remain required to fully realize the benefits of the most recent advances in NER methodology, and many lower-resourced languages still lack broad-coverage corpora of sufficient size and qual- ity to train NER methods comparable to those available for high-resource languages. In this paper, we focus on Finnish, a Uralic language spo- ken (nearly exclusively) by approx. 5 million people of the 5.5 million Finnish population. Alongside with the mi- nority language Swedish, Finnish is the official language of the country. Finnish has a long and strong tradition of linguistic research (Setälä, 1880; Hakulinen et al., 2004) and a number of modern language resources and NLP tools are available for the language, such as manually annotated treebanks (Haverinen et al., 2014) and morphological and syntactic analyzers (Pirinen, 2015; Kanerva et al., 2018). However, resources for …

Content truncated to 500 words