First Call for Abstracts for a Special Issue Preserving the Privacy of Language Data: Emerging Technologies, Evolving Standards and Beyond at the Language Resources and Evaluation Journal

Guest editors and contact information:

Elena Volodina, University of Gothenburg, Sweden, ( elena dot volodina at svenska dot gu dot se )
Maria Irena Szawerna, University of Gothenburg, Sweden, ( maria dot szawerna at gu dot se )

Call for submissions

In the age of GDPR and heightened awareness about data privacy, incorporating de-identification techniques when working with language data has become increasingly important. In academic contexts, de-identification is employed e.g. when using examples in publications (Heaton, 2022; Wang et al., 2024) or in dataset collection and sharing (Megyesi et al., 2018; Eder et al., 2019). It is also relevant for public uses, such as redacting legal documents (Cabrera-Diego and Gheewala, 2024) or varied industry applications (Gardiner et al., 2024; Hou et al., 2025). Manual de-identification is a costly and time-consuming process, which is why various automatic methods for detecting and replacing personal information, ranging from rule-based systems to Large Language Model (LLM) approaches, have been developed in recent years (Deußer et al., 2025). While rule-based approaches allow for processing the data locally, reducing the risk of the data falling into the wrong hands, they lack the ability to take semantic context and world knowledge into account; LLMs, on the other hand, can handle these challenges to a certain degree, but present ethical concerns when it comes to data leakage, environmental costs, or over-relying on encoded social and gender biases (Bender et al., 2021; Volodina et al. 2025, Szawerna and Suchardt, 2026).
This special issue will cover several aspects related to the problem of using technology for effective and ethical de-identification of linguistic data, including (but not limited to) the following topics:

Detection and classification of personal information (PI): Automatic identification of PI in text, speech, and multimodal data; context-dependent and indirect indicators of identity. Which information you are replacing, including how and why this is done.
Replacement and transformation of PI: Context-sensitive pseudonymization and anonymization methods; substitution, masking, obfuscation; maintaining coherence across discourse and modalities.
Utility and bias after de-identification: Effects of de-identification on downstream task performance, linguistic research validity, readability, and bias amplification or reduction.
Approaches to evaluation and adversarial testing: Metrics and frameworks for assessing de-identification quality; adversarial re-identification attempts; robustness and failure-mode analysis.
Dataset creation for de-identification research: Methodological, ethical, and annotation-related considerations in building corpora for training or evaluating de-identification systems.
Low-resource scenarios: Techniques for de-identification in settings with limited data, scarce annotations, or underrepresented languages; transfer and multilingual approaches.
Speech-specific challenges: Removing speaker identity cues in audio; voice anonymization; cross-modal leakage between text, transcripts, and acoustic features.
Cross-disciplinary applications and challenges: Integrating de-identification techniques into real-world workflows in areas such as linguistics, social sciences, digital humanities, healthcare, and other private- or public-sector data environments.

Submission information

Abstracts should be around 500 words excluding references and statements. The submission link can be found here. Authors of accepted abstracts will be notified by the end of November and invited to submit a full article by 1st March 2027.
Full Articles should follow the LRE journal guidelines. Each submission should include an AI disclosure statement. The submission link will be shared later.

The full call can be found https://mormor-karl.github.io/events/LREJ-special-issue/