Deep learning-based end-to-end spoken language identification system for domain-mismatched scenario

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 741–752 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 741 “Beste Gr¨uße, Maria Meyer” — Pseudonymization of Privacy-Sensitive Information in Emails Elisabeth Eder1 Michael Wiegand1 Ulrike Krieg-Holz1 Udo Hahn2 1 Alpen-Adria-Universit¨at Klagenfurt, Klagenfurt, Austria 2 Friedrich-Schiller-Universit¨at Jena, Jena, Germany {elisabeth.eder | michael.wiegand | ulrike.krieg-holz}@aau.at udo.hahn@uni-jena.de Abstract The exploding amount of user-generated content has spurred NLP research to deal with documents from various digital media formats (tweets, chats, emails, etc.). Using these texts as language resources implies complying with legal data privacy regulations. To protect the personal data of individuals and preclude their identification, we employ pseudonymization. More precisely, we identify those text spans that carry information revealing an individual’s identity (e.g., names of persons, locations, phone numbers, or dates) and subsequently substitute them with synthetically generated surrogates. Based on CODE ALLTAG, a German-language email corpus, we address two tasks. The first task is to evaluate various architectures for the automatic recognition of privacy-sensitive entities in raw data. The second task examines the applicability of pseudonymized data as training data for such systems since models learned on original data cannot be published for reasons of privacy protection. As outputs of both tasks, we, first, generate a new pseudonymized version of CODE ALLTAG compliant with the legal requirements of the General Data Protection Regulation (GDPR). Second, we make accessible a tagger for recog- nizing privacy-sensitive information in German emails and similar text genres, which is trained on already pseudonymized data. Keywords: pseudonymization, data privacy, email corpus, German language resources, named entity recognition 1. Introduction With the rapidly increasing adoption of social media platforms, we observe an upsurge of digitally trans- mitted private communication and exploding volumes of so-called user-generated content (UGC). Respond- ing to this fundamental move in communication habits world-wide, digital (social) media communication has become a major focus of research in NLP. From a user perspective though, intentionally sharing personal opinions, stances and attitudes via social me- dia also leaves footprints behind that can be used for demographic profiling, further social grouping activi- ties based on the analysis of properties users (uninten- tionally) disclose in their digital discourse (Kosinski et al., 2013; Volkova et al., 2015). Despite the evi- dent relevance of privacy protection for digital media data, including the threat of re-identification of indi- vidual authors, this topic has long been neglected by mainstream NLP research. While it has always been of utmost importance for medical NLP (Meystre, 2015), it has received almost no attention in non-medical NLP domains for a long time (except for two early works by Rock (2001) and Medlock (2006)). In response to data privacy legislation (see, for example, two recent analy- ses of rules targeting privacy protection in data for the US (Mulligan et al., 2019) and the EU (Hoofnagle et al., 2019)), many more NLP studies nowadays address the protection of individual data privacy (Li et al., 2018; Coavoux et al., 2018; Elazar and Goldberg, …

Content truncated to 500 words