BibTeX

@inProceedings{ilinykh-szawerna-2025-i-349004,
	title        = {“I Need More Context and an English Translation”: Analysing How LLMs identify Personal Information in Komi, Polish, and English},
	abstract     = {In this paper we present a pilot study and a qualitative analysis of the errors made by three large language models (LLMs) prompted to identify personal information (PI) in texts written in languages with varying resource availability: Komi (extremely low), Polish (medium), and English (high). Our analysis shows that LLMs perform better in detection of PI when provided with JSON-eliciting prompts. We also conjecture that the rich morphology and inflectionality of languages like Komi and Polish might affect the models’ performance. The small-scale parallel dataset of text that we introduce here can be used as a starting point in developing benchmarks for evaluation of PI detection with longer textual contexts and LLMs. },
	booktitle    = {Proceedings of the Third Workshop on Resources and Representations for Under-Resourced Languages and Domains (RESOURCEFUL-2025), March 2, 2025, Tallinn, Estonia / Špela Arhar Holdt, Nikolai Ilinykh, Barbara Scalvini, Micaella Bruton, Iben Nyholm Debess, Crina Madalina Tudor (eds.)},
	author       = {Ilinykh, Nikolai and Szawerna, Maria Irena},
	year         = {2025},
	publisher    = {University of Tartu Library},
	address      = {Tartu, Estonia},
	ISBN         = {978-9908-53-121-2},
	pages        = {165–178},
}

@inProceedings{szawerna-etal-2025-devils-348547,
	title        = {The Devil’s in the Details: the Detailedness of Classes Influences Personal Information Detection and Labeling},
	abstract     = {In this paper, we experiment with the effect of different levels of detailedness or granularity—understood as i) the number of classes, and ii) the classes’ semantic depth in the sense of hypernym and hyponym relations — of the annotation of Personally Identifiable Information (PII) on automatic detection and labeling of such information. We fine-tune a Swedish BERT model on a corpus of Swedish learner essays annotated with a total of six PII tagsets at varying levels of granularity. We also investigate whether the presence of grammatical and lexical correction annotation in the tokens and class prevalence have an effect on predictions. We observe that the fewer total categories there are, the better the overall results are, but having a more diverse annotation facilitates fewer misclassifications for tokens containing  correction annotation. We also note that the classes’ internal diversity has an effect on labeling. We conclude from the results that while labeling based on the detailed annotation is difficult because of the number of classes, it is likely that models trained on such annotation rely more on the semantic content captured by contextual word embeddings rather than just the form of the tokens, making them more robust against nonstandard language.},
	booktitle    = {Proceedings of the Joint 25th Nordic Conference on Computational Linguistics and 11th Baltic Conference on Human Language Technologies (NoDaLiDa/Baltic-HLT 2025), March 3–4, 2025 Tallinn, Estonia) / Richard Johansson and Sara Stymne (eds.)},
	author       = {Szawerna, Maria Irena and Dobnik, Simon and Muñoz Sánchez, Ricardo and Volodina, Elena},
	year         = {2025},
	publisher    = {University of Tartu Library},
	address      = {Tartu, Estonia},
	ISBN         = {978-9908-53-109-0},
	pages        = { 697–708},
}