NorNE: Annotating Named Entities for Norwegian

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 4547–4556 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 4547 NorNE: Annotating Named Entities for Norwegian Fredrik Jørgensen,† Tobias Aasmoe,‡ Anne-Stine Ruud Husevåg,♦ Lilja Øvrelid,‡ Erik Velldal‡ Schibsted Media Group,† Oslo Metropolitan University,♦ University of Oslo‡ fredrik.jorgensen@schibsted.com,† annesh@oslomet.no,♦ {tobiaaa,liljao,erikve}@ifi.uio.no‡ Abstract This paper presents NorNE, a manually annotated corpus of named entities which extends the annotation of the existing Norwegian Dependency Treebank. Comprising both of the official standards of written Norwegian (Bokmål and Nynorsk), the corpus contains around 600,000 tokens and annotates a rich set of entity types including persons, organizations, locations, geo-political entities, products, and events, in addition to a class corresponding to nominals derived from names. We here present details on the annota- tion effort, guidelines, inter-annotator agreement and an experimental analysis of the corpus using a neural sequence labeling architecture. Keywords: Named Entity Recognition, corpus, annotation, neural sequence labeling 1. Introduction This paper documents the efforts of creating the first pub- licly available dataset for named entity recognition (NER) for Norwegian, dubbed NorNE.1 The dataset adds named entity annotations on top of the Norwegian Dependency Treebank (NDT) (Solberg et al., 2014), containing manu- ally annotated syntactic and morphological information for both varieties of written Norwegian – Bokmål and Nynorsk – comprising roughly 300,000 tokens of each. The corpus contains mostly news texts (around 85% of the corpus), but also other types of texts, such as government reports, parlia- ment transcripts and blogs. The treebank has following its release also been converted to the Universal Dependencies standard (Øvrelid and Hohle, 2016; Velldal et al., 2017). We correspondingly distribute the annotations of NorNE in two versions, mirroring both the original NDT and the UD- version. The annotations in NorNE include a rich set of entity types. In short, they comprise the following (more details are given in the guideline discussion in Section 3): ◦ Person (PER): Named real or fictional characters. ◦ Organization (ORG): Any collection of people, such as firms, music groups, political parties etc. ◦ Location (LOC): Geographical places and facilities. ◦ Geo-political entity (GPE): Geographical regions de- fined by political and/or social groups, additionally sub-categorized as either: • GPE with a locative sense (GPE_LOC), • GPE with an organization sense (GPE_ORG). ◦ Product (PROD): Artificially produced entities, in- cluding abstract entities such as radio shows, program- ming languages, ideas, etc. ◦ Event (EVT): Festivals, weather phenomena, etc. ◦ Derived (DRV): Nominals that are derived from a name, but not a named entity in themselves. 1https://github.com/ltgoslo/norne/ In addition to discussing the annotation process and guide- lines, we also provide an exploratory analysis of the re- sulting dataset through a series of experiments using state- of-the-art neural architectures for named entity recognition (combining a character-level CNN and a word-level BiL- STM, feeding into a CRF inference layer). The purpose of the experimental section is to provide some preliminary baseline results while simultaneously validating the con- sistency and usability of the annotations, …

Content truncated to 500 words