Machine translation of anonymized documents with human-in-the-loop

Machine translation of anonymized documents with human-in-the-loop Konstantinos Chatzitheodorou M.ª ´Angeles Garc´ıa Escriv`a Carmen Grau Lacal Pangeanic Av. Corts Valencianes, 26 Bloque 5, 46015 Valencia {k.chatzitheodorou, ma.garcia, c.grau}@pangeanic.com Abstract In this paper, we introduce a workflow that utilizes human-in-the-loop for post- editing anonymized texts, with the aim of reconciling the competing needs of data privacy and data quality. By combining the strengths of machine translation and human post-editing, our methodology fa- cilitates the efficient and effective transla- tion of anonymized texts, while ensuring the confidentiality of sensitive informa- tion. Our experimental results validate that this approach is capable of providing all necessary information to the transla- tors for producing high-quality translations effectively. Overall, our workflow offers a promising solution for organizations seeking to achieve both data privacy and data quality in their translation processes. 1 Introduction Almost five years ago, the European Union, setting a milestone for data protection, enforced the General Data Protection Regulation (GDPR). Private and public organizations were required to remove sensitive content from public distribution involving European citizens under this legisla- tion (European Parliament and Council of the European Union, 2016). Text may need to be anonymized before it is translated to protect sensitive or confidential information. Text anonymization is a critical step in protecting sensitive or confidential information before machine translation (MT). Anonymization © 2023 The authors. This article is licensed under a Creative Commons 4.0 licence, no derivative works, attribution, CC- BY-ND. involves removing or disguising personally iden- tifiable information or other sensitive data in a text to protect the privacy and confidentiality of individuals or organizations mentioned in the text (Pil´an et al., 2022). Anonymization is particularly important in sit- uations where the translated text may be viewed by individuals who are not authorized to access the sensitive information contained in the original text. For example, in the case of medical records or legal documents, it may be necessary to remove personally identifiable information to protect patient or client privacy (Papadopoulou et al., 2022). Moreover, post-editing machine-translated texts is often required to ensure that the translation accurately conveys the intended meaning and tone of the original text. A human-in-the-loop workflow for post-editing machine-translated texts can improve the quality of the final translation by leveraging the strengths of both human and MT (Lee et al., 2021). By anonymizing the text before translation and utilizing post-editing workflows, the confiden- tiality and privacy of sensitive information can be maintained, while allowing the text to be effectively translated and used for its intended purpose. Furthermore, MT incorporates the factor of speed, meaning that post-editing is faster than translating from scratch. In this paper, we propose a human-in-the- loop workflow for post-editing machine-translated documents that have been anonymized to protect sensitive information. The proposed workflow leverages the strengths of both humans and MT to improve the quality of the final translation while ensuring that the privacy and confidentiality of sensitive information are maintained. 2 Challenges of translating anonymized texts Translating an anonymized text from one language to another …

Content truncated to 500 words