Enhanced Entity Annotations for Multilingual Corpora

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3732–3740 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3732 Enhanced Entity Annotations for Multilingual Corpora Michael Strobl, Amine Trabelsi, Osmar Zaiane University of Alberta, Lakehead University, University of Alberta mstrobl@ualberta.ca, atrabels@lakeheadu.ca, zaiane@ualberta.ca Abstract Modern approaches in Natural Language Processing (NLP) require, ideally, large amounts of labelled data for model training. However, new language resources, for example, for Named Entity Recognition (NER), Co-reference Resolution (CR), Entity Linking (EL) and Relation Extraction (RE), naming a few of the most popular tasks in NLP, have always been challenging to create since manual text annotations can be very time-consuming to acquire. While there may be an acceptable amount of labelled data available for some of these tasks in one language, there may be a lack of datasets in another. WEXEA is a tool to exhaustively annotate entities in the English Wikipedia. Guidelines for editors of Wikipedia articles result, on the one hand, in only a few annotations through hyperlinks, but on the other hand, make it easier to exhaustively annotate the rest of these articles with entities than starting from scratch. We propose the following main improvements to WEXEA: Creating multi-lingual corpora, improved entity annotations using a proven NER system, annotating dates and times. A short evaluation of the annotation quality of WEXEA is added. Keywords: Wikipedia, Entity Annotations, Distant Supervision 1. Introduction Understanding natural language and storing informa- tion in a structured way at human level has been an is- sue for decades, e.g. see the SHRDLU program (Wino- grad, 1972), which was able to conduct a dialogue in English with a human to move blocks by the com- puter in a virtual box. More recent approaches often involve the population of a Knowledge Graph (KG), as described in (Zhang et al., 2016), which explicitly (or implicitly for end-to-end approaches) involves the fol- lowing sub-tasks: 1. Named Entity Recognition (NER): The detec- tion of Named Entities (NE). The most com- mon dataset is derived from news data (Sang and De Meulder, 2003) and aims to detect four en- tity types: PERSON, LOCATION, ORGANIZA- TION, MISC. 2. Co-reference Resolution (CR): CR aims to de- tect co-reference clusters of entities mentioned throughout a document, not necessarily NEs. 3. Entity Linking (EL): EL links NEs to an ex- isting entity in a Knowledge Base (KB), e.g. Wikipedia1. 4. Relation Extraction (RE): RE systems are able to detect relations, typically, between NEs and their co-references recognized by NER and CR sys- tems. All of these tasks ideally require large amounts of data on which to train models, which is typically limited, es- pecially if languages other than English are of interest. Wikipedia, as-is, is already able to provide datasets in the following ways: 1https://www.wikipedia.org/ 1. NER: If a class-article-mapping is available, e.g. from (Ghaddar and Langlais, 2017), existing an- notations in Wikipedia can be tagged with their corresponding type and used for NER models to be trained on. …

Content truncated to 500 words