CITATION.cff

cff-version: 1.2.0

message: >-
  If you use MultiPII-X, please cite this dataset and acknowledge the
  relevant upstream resources listed below.

title: "MultiPII-X"
type: dataset

version: "1.0.0"
date-released: "2026-10-03"

doi: "10.57967/hf/10725"
url: "https://huggingface.co/datasets/Nan0NJ/multiPII-x"

authors:
  - given-names: Nenad
    family-names: Jakovchevski
    affiliation: "Jožef Stefan Institute, Department of Knowledge Technologies"

abstract: >-
  MultiPII-X is an extensible fine-grained dataset for personally identifiable
  information (PII), sensitive-entity recognition, and text anonymization.
  Version 1.0 contains Slovene data spanning multiple domains and source
  collections. The release contains 15,180 annotated texts with 89,812 entity
  annotations across 56 fine-grained entity types. Source material was
  translated, adapted, sampled, reannotated, or otherwise transformed under
  a unified annotation ontology. The canonical release preserves original text
  together with character-level entity spans and also provides derived
  representations for GLiNER and GLiNER2/GLiNER2.5.

keywords:
  - personally identifiable information
  - PII
  - anonymization
  - named entity recognition
  - NER
  - privacy
  - sensitive information
  - Slovene
  - Slovenian
  - low-resource language
  - multi-domain NLP
  - GLiNER

references:

  # --------------------------------------------------------------------------
  # TANAOS
  # --------------------------------------------------------------------------

  - type: data
    title: "Tanaos Text Anonymizer Training Dataset"
    authors:
      - name: "Tanaos"
    database: "Hugging Face Hub"
    entry: "tanaos/synthetic-text-anonymizer-dataset-v1"
    license: MIT
    scope: >-
      Source dataset for MultiPII-X. Selected synthetic English PII examples
      were adapted into Slovene and transformed to the MultiPII-X annotation
      ontology.

  # --------------------------------------------------------------------------
  # KURKOWSKI
  # --------------------------------------------------------------------------

  - type: data
    title: "Contextual Text Anonymizer Dataset"
    authors:
      - given-names: Michał
        family-names: Kurkowski
    year: 2025
    database: "Hugging Face Hub"
    entry: "kurkowski/synthetic-contextual-anonymizer-dataset"
    license: MIT
    scope: >-
      Source dataset for MultiPII-X. Contextual synthetic PII examples were
      adapted into Slovene and mapped/reannotated under the MultiPII-X
      fine-grained ontology.

  # --------------------------------------------------------------------------
  # GIGAFIDA 2.0
  # --------------------------------------------------------------------------

  - type: data
    title: "Corpus of Written Standard Slovene Gigafida 2.0"
    authors:
      - given-names: Simon
        family-names: Krek
      - given-names: Tomaž
        family-names: Erjavec
      - given-names: Andraž
        family-names: Repar
      - given-names: Jaka
        family-names: Čibej
      - given-names: Špela
        family-names: Arhar Holdt
      - given-names: Polona
        family-names: Gantar
      - given-names: Iztok
        family-names: Kosem
      - given-names: Marko
        family-names: Robnik-Šikonja
      - given-names: Nikola
        family-names: Ljubešić
      - given-names: Kaja
        family-names: Dobrovoljc
      - given-names: Cyprian
        family-names: Laskowski
      - given-names: Miha
        family-names: Grčar
      - given-names: Peter
        family-names: Holozan
      - given-names: Simon
        family-names: Šuster
      - given-names: Vojko
        family-names: Gorjanc
      - given-names: Marko
        family-names: Stabej
      - given-names: Nataša
        family-names: Logar
    date-released: "2019-06-13"
    database: "CLARIN.SI"
    entry: "hdl:11356/1320"
    scope: >-
      Source corpus for naturally occurring Slovene text in MultiPII-X.
      Selected passages were independently annotated under the PII ontology;
      the original Gigafida linguistic annotations are not treated as
      MultiPII-X PII annotations.

  # --------------------------------------------------------------------------
  # TELECOM CONVERSATION CORPUS
  # --------------------------------------------------------------------------

  - type: data
    title: "Telecom 200k Dataset"
    authors:
      - name: "eisenzopf"
    database: "Hugging Face Hub"
    entry: "eisenzopf/telecom-conversation-corpus"
    license: MIT
    scope: >-
      Source corpus for telecom-domain conversational material. Selected
      synthetic conversations were adapted into Slovene and annotated using
      the MultiPII-X ontology.

  # --------------------------------------------------------------------------
  # E3-JSI SYNTHETIC MULTI-PII DATASET
  # --------------------------------------------------------------------------

  - type: data
    title: "Synthetic Multi-Domain PII NER Dataset"
    authors:
      - name: "Department for Artificial Intelligence, Jožef Stefan Institute (E3-JSI)"
    year: 2024
    database: "Hugging Face Hub"
    entry: "E3-JSI/synthetic-multi-pii-ner-v1"
    doi: "10.57967/hf/3211"
    license: MIT
    scope: >-
      Source dataset for MultiPII-X. The Slovene subset was selected and
      reannotated using the expanded MultiPII-X entity ontology.

  # --------------------------------------------------------------------------
  # POVEJMO-VEMO-MED
  # --------------------------------------------------------------------------

  - type: data
    title: "Dataset of Slovene Medical Texts PoVeJMo-VeMo-Med 1.0"
    authors:
      - given-names: Miha
        family-names: Malenšek
      - given-names: Marko
        family-names: Bajec
    date-released: "2024-09-25"
    publisher:
      name: "Faculty of Computer and Information Science, University of Ljubljana"
    database: "CLARIN.SI"
    entry: "hdl:11356/1983"
    license: CC-BY-4.0
    scope: >-
      Source corpus for the medical domain. MultiPII-X uses sampled texts from
      the deduplicated Slovene medical corpus, which were newly annotated with
      the MultiPII-X PII, identifier, measurement, and health-related ontology.

  # --------------------------------------------------------------------------
  # COLESLAW
  # --------------------------------------------------------------------------

  - type: data
    title: "Collection of Slovenian Legal Texts COLESLAW 1.0"
    authors:
      - given-names: Miha
        family-names: Malenšek
      - given-names: Slavko
        family-names: Žitnik
      - given-names: Aleš
        family-names: Završnik
      - given-names: Saša
        family-names: Krajnc
      - given-names: Primož
        family-names: Križnar
      - given-names: Marko
        family-names: Bajec
    date-released: "2024-12-31"
    publisher:
      name: "Faculty of Computer and Information Science, University of Ljubljana"
    database: "CLARIN.SI"
    entry: "hdl:11356/2095"
    license: CC-BY-4.0
    scope: >-
      Source corpus for the legal domain. Selected Slovene legal passages were
      newly annotated using the MultiPII-X fine-grained PII and sensitive-entity
      ontology.

  # --------------------------------------------------------------------------
  # PRIOR SLOVENE PII BENCHMARK / ANNOTATION METHODOLOGY
  # --------------------------------------------------------------------------

  - type: conference-paper
    title: >-
      Privacy-Preserving NLP for a Low-Resource Language: Benchmarking
      LLM-Based Personally Identifiable Information Anonymization in Slovene
    authors:
      - given-names: Nenad
        family-names: Jakovchevski
      - given-names: Matej
        family-names: Martinc
    year: 2026
    conference:
      name: "17th International Conference of the CLEF Association (CLEF 2026)"
    collection-title: >-
      Experimental IR Meets Multilinguality, Multimodality, and Interaction
    start: 55
    end: 68
    publisher:
      name: "Springer"
    scope: >-
      Prior Slovene PII benchmark and annotation methodology on which part of
      the annotation conventions and earlier training-resource construction
      underlying MultiPII-X are based.