The Bulgarian Event Corpus: Overview and Initial NER Experiments

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3491–3499 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3491 The Bulgarian Event Corpus: Overview and Initial NER Experiments Petya Osenova, Kiril Simov, Iva Marinova, Melania Berbatova Artificial Intelligence and Language Technology, Institute of Information and Communication Technologies Bulgarian Academy of Sciences Bulgaria petya@bultreebank.org, kivs@bultreebank.org, iva.marinova@identrics.net, melania.berbatova@gmail.com Abstract The paper describes the Bulgarian Event Corpus (BEC). The annotation scheme is based on CIDOC-CRM ontology and on the English Framenet, adjusted for our task. It includes two main layers: named entities and events with their roles. The corpus is multi-domain and mainly oriented towards Social Sciences and Humanities (SSH). It will be used for: extracting knowledge and making it available through the Bulgaria-centric Knowledge Graph; further developing an annotation scheme that handles multiple domains in SSH; training automatic modules for the most important knowledge-based tasks, such as domain-specific and nested NER, NEL, event detection and profiling. Initial experiments were conducted on standard NER task due to complexity of the dataset and the rich NE annotation scheme. The results are promising with respect to some labels and give insights on handling better other ones. These experiments serve also as error detection modules that would help us in scheme re-design. They are a basis for further and more complex tasks, such as nested NER, NEL and event detection. Keywords: Bulgarian event corpus, multidomain dataset, NER experiments, Social Sciences and Humanities 1. Introduction In recent years there is increasing interest in event- oriented applications within the NLP community for variety of tasks using different methods (like super- vised or unsupervised ones). Such applications are event detection and extraction, scheme induction, event profiling, among others. The successful extraction of events is not trivial but it became possible thanks to the advancements in technology (improved language mod- eling as well as cross-lingual and cross-domain meth- ods), on the one hand, and also thanks to the availability of a lot of hand-made, hand-induced or automatically produced data, on the other. In this paper we present our work on the creation of a Bulgarian Event Corpus (BEC) and the challenges be- hind that. We also report the results from the initial experiments with NER models trained on it. The main reason for the construction of the BEC cor- pus is to have appropriate data for training Named En- tity Recognition (NER), Named Entity Linking (NEL) and Event Recognition models. Such models would help us in the extraction of structured knowledge from domain texts in the area of Social Sciences and Human- ities (SSH). The extracted structured knowledge will be ultimately used for the creation of a Bulgarian-centric Knowledge Graph — see (Simov and Osenova, 2019). In the initial annotation of the corpus we concentrated on a rich set of Named Entities, on some general con- cepts and events that happen to be frequent in the texts from various genres and domains such as history and ethnography. Handling the typical NEs, …

Content truncated to 500 words