Anonymising the SAGT Speech Corpus and Treebank

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5557–5564 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5557 Anonymising the SAGT Speech Corpus and Treebank ¨Ozlem C¸ etino˘glu, Antje Schweitzer Institute for Natural Language Processing (IMS), University of Stuttgart {ozlem.cetinoglu,antje.schweitzer}@ims.uni-stuttgart.de Abstract Anonymisation, that is identifying and neutralising sensitive references, is a crucial part of dataset creation. In this paper, we describe the anonymisation process of a Turkish-German code-switching corpus, namely SAGT, which consists of speech data and a treebank that is built on its transcripts. We employed a selective pseudonymisation approach where we manually identified sensitive references to anonymise and replaced them with surrogate values on the treebank side. In addition to maintaining data privacy, our primary concerns in surrogate selection were keeping the integrity of code-switching properties, morphosyntactic annotation layers, and semantics. After the treebank anonymisation, we anonymised the speech data by mapping between the treebank sentences and audio transcripts with the help of Praat scripts. The treebank is publicly available for research purposes and the audio files can be obtained via an individual licence agreement. Keywords: code-switching, Turkish, German, anonymisation 1. Introduction Dataset sharing is important for replicability and so that other researchers can make use of existing data. When making datasets available, anonymisation is cru- cial both ethically and legally. Medlock (2006) defines anonymisation as “the task of identifying and neutral- ising sensitive references within a given document or set of documents.” Sensitive references could be direct such as a person’s name. They could also be indirect, in that they contribute to revealing identities when com- bined with other information. Example 1 demonstrates an indirect sensitive reference. As a consequence of anonymisation then, direct and in- direct references in a dataset have to be altered. How- ever it is also important to keep the data as intact as pos- sible so that anonymised data is still useful for research purposes. Our motivation here is to find a balance in anonymisation. Our work describes the anonymisation stages of a code-switching corpus, namely SAGT, which is a col- lection of interviews with Turkish-German bilinguals (C¸ etino˘glu and C¸ ¨oltekin, 2022). The interviews were collected as audio recordings and transcribed. The tran- scriptions were then turned into a treebank with various layers of annotation (C¸ etino˘glu and C¸ ¨oltekin, 2019). The interviews were conducted in an informal setting to help increase the frequency of code-switching. Com- mon conversation themes include studies, work, travel, future plans, and free time activities such as sports, books, and TV. Hence the data contains private infor- mation about participants. Anonymisation took place both on the resulting treebank side and on the audio files. We asked two main research questions during this process: (i) Which anonymisation strategy should we follow? (ii) How can we anonymise in one medium (i.e., text) and transfer it to the other one (i.e., audio). In data collection, processing, and maintenance we closely follow EU General Data Protection Regula- tion (GDPR)1. The …

Content truncated to 500 words