Preventing Author Profiling through Zero-Shot Multilingual Back-Translation

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8687–8695 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 8687 Preventing Author Profiling through Zero-Shot Multilingual Back-Translation David Ifeoluwa Adelani, Miaoran Zhang, Xiaoyu Shen, Ali Davody, Thomas Kleinbauer, and Dietrich Klakow Spoken Language Systems Group, Saarland Informatics Campus, Saarland University, Germany {didelani,mzhang,xshen,adavody}@lsv.uni-saarland.de {thomas.kleinbauer,dietrich.klakow}@lsv.uni-saarland.de Abstract Documents as short as a single sentence may inadvertently reveal sensitive information about their authors, including e.g. their gen- der or ethnicity. Style transfer is an effective way of transforming texts in order to remove any information that enables author profiling. However, for a number of current state-of-the- art approaches the improved privacy is accom- panied by an undesirable drop in the down- stream utility of the transformed data. In this paper, we propose a simple, zero-shot way to effectively lower the risk of author pro- filing through multilingual back-translation us- ing off-the-shelf translation models. We com- pare our models with five representative text style transfer models on three datasets across different domains. Results from both an au- tomatic and a human evaluation show that our approach achieves the best overall perfor- mance while requiring no training data. We are able to lower the adversarial prediction of gen- der and race by up to 22% while retaining 95% of the original utility on downstream tasks. 1 Introduction Data collections of natural language utterances bear the risk of disclosing sensitive information about the recorded participants, including their gender, race, or political preferences. Unlike explicit men- tions of private information, like a user’s name or location (Tang et al., 2004; Adelani et al., 2020), such user traits are often encoded rather subtly in a user’s speaking or writing style. Nevertheless, they can be predicted with high accuracy by deep learning-based classifiers even when they are not obvious to humans (Elazar and Goldberg, 2018), enabling third-parties with access to the data sets to profile users without their knowledge. A common method to alleviate this problem is the application of an intermediate transformation step to remove sensitive information via text style transfer. While a number of different style transfer techniques exist (Shen et al., 2017; Fu et al., 2018; Madaan et al., 2020), they require large amounts of text data labeled with user trait information to perform well. Additional annotations need to be provided for every new user trait that the model is expected to handle, multiplying the associated costs and effort. Furthermore, the impact that such transformations can have on the utility of the re- sulting data is often overlooked. Conversely, we argue that the privacy-utility dichotomy should be at the heart of all research on this topic because it is fairly easy to consider one of the two but difficult to improve both at the same time. In this paper, we explore a simple yet effective zero-shot text transformation method based on mul- tilingual back-translation. Back-translation (BT) is an alternative approach without the prerequisites of labeled training data. Sensitive user traits can be significantly obfuscated …

Content truncated to 500 words