Resource of Wikipedias in 31 Languages Categorized into Fine-Grained Named Entities

Proceedings of the 29th International Conference on Computational Linguistics, pages 3769–3777 October 12–17, 2022. 3769 Abstract This paper describes a resource of Wikipedias in 31 languages categorized into Extended Named Entity (ENE), which has 219 fine-grained NE categories. We first categorized 920K Japanese Wikipedia pages according to the ENE scheme using machine learning, followed by manual validation. We then organized a shared task of Wikipedia categorization into 30 languages. The training data were provided by Japanese categorization and the language links, and the task was to categorize the Wikipedia pages into 30 languages, with no language links from Japanese Wikipedia (20M pages in total). Thirteen groups with 24 systems participated in the 2020 and 2021 tasks, sharing their outputs for resource-building. The Japanese categorization accuracy was 98.5%, and the best performance among the 30 languages ranges from 80 to 93 in F- measure. Using ensemble learning, we created outputs with an average F-measure of 86.8, which is 1.7 better than the best single systems. The total size of the resource is 32.5M pages, including the training data. We call this resource creation scheme “Resource by Collaborative Contribution (RbCC)”. We also constructed structuring tasks (attribute extraction and link prediction) using RbCC under our ongoing project, “SHINRA”. 1 Introduction Wikipedia consists of a large volume of entities that are significant resources for the knowledge base (KB) used in many Natural Language Processing (NLP) applications, including Question Answering, Information Extraction, and so on. To maximize the use of such a KB, the information in Wikipedia has to be categorized and structured in a consistent manner for machines to perform inference, reasoning, and many other purposes. The current categorization and structure of Wikipedia and other KB derived from it, such as DBpedia, YAGO, and Wikidata, are extremely noisy for NLP applications. This noise is inherent to Wikipedia categories and structures because they are created by multiple independent crowd workers using a bottom-up approach. There are no consistent rules exists for most parts of Wikipedia. As a result, instead of the cumbersome Wikipedia categories and structures, we must rely on a well-defined ontology. Extended Named Entity (ENE) (ENE homepage) is one such ontology for named entities (NEs). ENE version 8 has 219 hierarchical categories, and a set of attributes is defined for each category. Our final goal is to transform Wikipedia information into the structure of ENE so that machines can use the rich information in Wikipedia. This study reports a resource of Wikipedias in 31 languages categorized into fine-grained named entity (ENE) categories. Categorization is the first task in creating a structured KB. After developing the techniques of automatic categorization are developed, we can structure the contents of Wikipedia pages in each category. We worked on 31 languages, rather than a few major languages. We created resources for practical NLP applications in various languages, instead of platforming a few major languages. 2 Extended Named Entity To construct a useful KB for NLP applications, a well-structured ontology is essential and must be designed in a …

Content truncated to 500 words