Overview

Provenance Data Provenance Research Data Provenance

Provenance

Sources: Glavic (2021: 3f.), Anderson (2024)

W3C Definition of Provenance

Source: W3C (2010)

Sources: Glavic (2021: 3f.), Simmhan, Plale and Gannon (2005: 1), Data Provenance Initiative (2024), Werder, Ramesh and Zhang (2022), Stein and Taentzer (2023), eResearch Alliance Universität Göttingen (2025)

Types of Provenance

Source: Herschel, Diestelkämper and Lahmar (2017: 882)

Provenance: any information describing the production process of an end product, which can be anything from a piece of digital data to a physical object.

Provenance meta-data: meta-data describing an arbitrary production process using an arbitrary data model and model of computation

Information system provenance: meta-data collected for an information-disseminating process that can be computed based on the input, the output, and the parameters of the process. (= processes producing digital data within information systems)

Workflow provenance: specializes information system provenance by further restricting the type of production processes to so-called workflows. (= directed graph where nodes represent arbitrary functions or modules [...] with some input, output, and parameters)

Data provenance: allows to track the processing of individual data items (e.g., tuples) at the “highest resolution,” i.e., the provenance itself is at the level of individual data items (and the operations they undergo). Collecting data provenance typically applies on structured data models and declarative query languages with clearly defined semantics of individual operators.

Source: Herschel, Diestelkämper and Lahmar (2017: 882f.)

Sources: highlighted terms from Herschel, Diestelkämper and Lahmar (2017: 882f.), Simmhan, Plale and Gannon (2005), IBM (2024) u.A.

FAIR data consortium

Source: FAIR Data Consortium

NFDI Sektion (Meta)daten, Terminologien, Provenienz

Representing Provenance

Source: Massari, Peroni, Tomasi and Heibi (2023), see also Sikos (2021), Sikos and Philp (2020)

Representing Gender

Why gender?

When it comes to managing gender data, common challenges and beliefs seem to be:

Certainly within the community of researchers working with cultural data, the desire to compare and aggregate diverse sources held together by a thin red thread of potential narrative cohesion, is only increasing. The KPLEX project (kplex-project.eu) is investigating these barriers to meaning-making. Our team has adopted a comparative, multidisciplinary, and multi-sectoral approach to this problem, focussing on key challenges to the knowledge creation capacity of cultural data such as the terms we use to speak about data in a cultural context, the manner in which data that are not digitised or shared become “hidden” from aggregation systems, the fact that data lacks the objectivity often ascribed to the term and the subtle ways in which data that are complex almost always become simplified before they can be aggregated.

Source: Edmond and Folan (2017)

Representation of gender in common standards

• Entity E76 Gender removed in 2001
• Gender modelled via P2 has type
• Alternatives discussed but complex, f.e. via gender assignment event [1]

• Standard approach is @sex attribute with values M, F, O (other), N (none)
• But any external standard can be used
• Alternative approaches exist but complex: f.e. via <trait type=“gender”> [2]

• Allowed values for gender in GND entities "male" and "female", without reference
• Other values allowed in free form in different field only if reference is provided

• Gender modelled via P21 sex or gender
• Used for animals and humans, conflates sex, gender identity and modality

Sources: [1] Andrews et al. (2024), [2] Flanders (2021)

Of course, not all is bad...

• Expansive linked data vocabulary for contemporary LGBTQ terms
• Not the only one, f.e. GSSO ontology

• Wikimedia-funded research project, reevaluated Wikidata model and recommended:
• Require references for gender statements
• Define standards for references
• Remove P21 sex or gender
• Separate gender identity and modality

...but it could be better.

Overview

Research Aims Research Questions

1. Take stock: How is gender represented in DH datasets? Is provenance information provided? How is it represented and communicated? Which standards, ontologies, vocabularies are used? Methods: Kuckartz' Qualitative Content Analysis; Kitchenham’s guidelines for Systematic Review (2004)
2. Identify issues: How may the quality of gender data in digital humanities projects be assessed? Is gender data FAIR? Is it interpretable, plausible and trustworthy? How does provenance information affect data quality?
3. Understand their origins: Why are certain standards and modelling approaches adopted/rejected? What are common requirements and constraints? Which types of provenance are tracked and how? What are typical data integration and reconciliation workflows? Methods: Interview stakeholders in DH projects: expert interviews; requirement elicitation techniques
4. Formulate recommendations: Which workflows and modelling approaches improve data quality in a linked data setting? How should provenance information be made available? Which types of provenance should be tracked when dealing with (historical) gender data? What strategies have proven successful in low resource settings? Output: Create repository with content analysis and interview data; publish workflows on SSH Open Marketplace

TL;DR

Digital humanities projects working with prosopographical data face a dilemma: historical gender data is inaccurate, messy, and mostly binary, but leaving it out means rendering gender as a social category of difference invisible. Approaches to tracking and representing the provenance of historical gender data vary and standardization is needed to improve interoperability and interpretability.

Preliminary Findings

1. Take stock: How is gender represented in DH datasets? Is provenance information provided? How is it represented and communicated? Which standards, ontologies, vocabularies are used? Methods: Kuckartz' Qualitative Content Analysis; Kitchenham’s guidelines for Systematic Review (2004)

Data collection

Python Scrapy used to scrape ADHO conference abstracts 2013-2023 from conference websites and repositories

XML-TEI 2013–2016, 2018, 2020, 2022, and 2023

Plaintext 2017

PDF to plaintext conversion 2019

Code available at: https://github.com/lipogg/dh-projects-scraper

Annotation

• Qualitative content analysis
• Mixed deductive-inductive category system development
• Selection variable funnel and content variables
• Quantitative evaluation

(Mayring, 2000)

Selection Variable Funnel

• Is the URL …
• online?
• a Digital Humanities resource?
• a dataset?
• a single dataset?
• unrestricted?
• And does it …
• contain personal data?
• contain gender data?

Inter-Coder Reliability Scores

Result interpretation based on Landis and Koch (1977)

Selection Variables

Gender expressions

Common non-gender expressions

Uncertainty Undetermined, Indeterminate, Contested

Inapplicability Not applicable

Unavailability Not provided, ?, Unknown

The “other” category Used for “other” genders, unknown values, animals and organizations

Gender expressions by structuredness of gender data

Standards used

Gender expressions by availability of provenance information

Provenance of gender data

Gender expressions by provenance

Limitations

• Global North overrepresented in DH conferences and conference abstracts
• Established projects favoured over small datasets
• Gender categories beyond the binary less easily enumerated for unstructured data

See e.g. Siddiqui (2023)

Outlook Timeline

1. Additional questions:
   • What is the language of the data / project?
   • What is the stage of the project (onset/ maintenance/ ...)
   • What types of provenance are represented and how?
   • How is provenance information communicated?
   Reinclude DOIs and Zenodo links? Incorporate Kitchenham's guidelines?

Literature