WikiBio: a Semantic Resource for the Intersectional Analysis of Biographical Events

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 12370–12384 July 9-14, 2023 ©2023 Association for Computational Linguistics WikiBio: a Semantic Resource for the Intersectional Analysis of Biographical Events Marco Antonio Stranisci , Rossana Damiano , Enrico Mensa Viviana Patti , Daniele Paolo Radicioni , Tommaso Caselli Dipartimento di Informatica, Università degli Studi di Torino, Italy CLCG, University of Groningen aequa-tech, Turin, Italy marcoantonio.stranisci@unito.it Abstract Biographical event detection is a relevant task for the exploration and comparison of the ways in which people’s lives are told and represented. In this sense, it may support several applica- tions in digital humanities and in works aimed at exploring bias about minoritized groups. De- spite that, there are no corpora and models specifically designed for this task. In this paper we fill this gap by presenting a new corpus an- notated for biographical event detection. The corpus, which includes 20 Wikipedia biogra- phies, was compared with five existing corpora to train a model for the biographical event de- tection task. The model was able to detect all mentions of the target-entity in a biography with an F-score of 0.808 and the entity-related events with an F-score of 0.859. Finally, the model was used for performing an analysis of biases about women and non-Western people in Wikipedia biographies. 1 Introduction Detecting biographical events from unstructured data is a relevant task to explore and compare bias in representations of individuals. In recent years, the interest in this topic has been favored by studies about social biases on allegedly objective public archives such as Wikipedia. Sun and Peng (2021) developed a resource for investigating gender bias on Wikipedia biographies showing that personal life events tend to be more frequent in female ca- reer sections than in those of men. Lucy et al. (2022) developed BERT-based contextualized em- beddings for exploring representations of women on Wikipedia and Reddit. The detection of biographical events has been addressed with complementary approaches by dif- ferent research communities. Projects in Digital Humanities have focused mostly on representa- tional aspects, delivering ontologies and knowledge graphs for the collection and study of biographi- cal events (Tuominen et al., 2018; Fokkens et al., 2017; Plum et al., 2019; Krieger, 2014). When it comes to NLP, the focus has been mainly on developing models for the detection and classifica- tion of events (Rospocher et al., 2016; Gottschalk and Demidova, 2018). Few are the works that di- rectly target biographies and focus on identifying biographical events with varied approaches (super- vised and unsupervised) across different datasets (e.g., Wikipedia vs. newspaper articles), making their comparison impossible (Bamman and Smith, 2014; Russo et al., 2015; Menini et al., 2017). Al- though not directly targeting biographies, some works focused on the identification of entity-related sequences of events (Chambers and Jurafsky, 2008) and entity-based storylines (Chambers and Juraf- sky, 2009; Minard et al., 2015; Vossen et al., 2016). Despite the above mentioned variety of ap- proaches to biographical event detection, there are …

Content truncated to 500 words