Entity Linking over Nested Named Entities for Russian

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4458–4466 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4458 Entity Linking over Nested Named Entities for Russian Natalia Loukachevitch1, Pavel Braslavski2,3, Vladimir Ivanov4, Tatiana Batura5,6, Suresh Manandhar9, Artem Shelmanov7,1, Elena Tutubalina3,8 1Lomonosov Moscow State University, Moscow, Russia 2Ural Federal University, Yekaterinburg, Russia 3HSE University, Moscow, Russia 4Innopolis University, Innopolis, Russia 5Novosibirsk State University, Novosibirsk, Russia 6Ershov Institute of Informatics Systems, Novosibirsk, Russia 7Artificial Intelligence Research Institute, Moscow, Russia 8Sber AI, Moscow, Russia 9Madan Bhandari University of Science and Technology Development Board, Nepal louk nat@mail.ru, pbras@yandex.ru, nomemm@gmail.com, tatiana.v.batura@gmail.com, suresh.manandhar@mbustb.edu.np, shelmanov@airi.net, tutubalinaev@gmail.com Abstract In this paper, we describe entity linking annotation over nested named entities in the recently released Russian NEREL dataset for information extraction. The NEREL collection (Loukachevitch et al., 2021) is currently the largest Russian dataset annotated with entities and relations. The paper describes the main design principles behind NEREL’s entity linking annotation, provides its statistics, and reports evaluation results for several entity linking baselines. To date, 38,152 entity mentions in 933 documents are linked to Wikidata. The NEREL dataset is publicly available: https://github.com/nerel-ds/NEREL. Keywords: information extraction, entity linking, nested named entities, Russian language 1. Introduction Entity linking (EL) is a popular NLP task, where a system needs to link a named entity to a concept in a knowledge base such as Wikidata (Vrandeˇci´c and Kr¨otzsch, 2014), DBpedia (Auer et al., 2007), or Freebase (Bollacker et al., 2008). Entity linking is a crucial step in automatic knowl- edge base construction, it helps disambiguate entity men- tions in text documents and enrich them with external in- formation (Sevgili et al., 2020; Oliveira et al., 2021). Entity linking in the general domain is mainly implemented for named entities. Current studies in named entity ex- traction comprise creation of datasets with so called nested named entities, when an entity can be annotated within an- other named entity (Benikova et al., 2014; Ringland et al., 2019). In contrast to corpora with “flat” named entities, nested named entity datasets provide possibilities for fine- grained relations between entities and links to knowledge bases. For example, Lomonosov Moscow State University (Wikidata ID: Q13164) contains further entities – Moscow (the capital of Russia – Q649) and Mikhail Lomonosov (the founder of the Moscow State University – Q58720). In the example in Figure 1a, “Famous “Diary of Anne Frank” remains in Amsterdam”, the entities Diary of Anne Frank, Anne Frank, and Amsterdam help disambiguate each other. As Figure 1b shows, the word Metro in the sentence “Football team of England will play on Tuesday, writes Metro” refers to a UK newspaper and can be resolved correctly by considering internal entity England. In this paper, we present entity linking annotation over nested named entities in the recently developed Russian dataset NEREL. The NEREL collection (Loukachevitch et al., 2021) is currently the largest Russian dataset annotated with entities and relations, when compared to the existing Russian datasets (Gordeev et al., 2020; Trofimov, …

Content truncated to 500 words