• Roberta Pireddu KU Leuven

KU Leuven

The growing accessibility of cultural heritage data, together with the large‑scale digitization of collections, has transformed how heritage materials are discovered and used. While this expansion offers clear benefits, it also introduces new challenges: collections are often published online without sufficient curatorial oversight, which can unintentionally reinforce discriminatory narratives and hinder genuine inclusion in cultural and digital spaces. This underscores the need for heritage institutions to ensure that metadata is responsibly curated or at least that potential issues are clearly identified and flagged. In response to this, digital technologies and AI are currently being explored for the identification and flagging of potentially biased or harmful metadata in large‑scale digital collections. 

These issues were recently addressed in the context of DE-BIAS, a two-year project (2023–2025) co-funded under the Digital Europe Programme (DIGITAL) of the European Union and grounded on the collaboration between 11 European GLAM institutions. The project developed co‑creative, technology‑driven strategies to support the contextualization and identification of bias in heritage collections, and to encourage cultural heritage institutions to describe digital collections more respectfully and inclusively. It implemented a capacity‑building strategy that combined co‑creation methodologies in workshops and events with tools for co-detecting bias in heritage collections. The project also developed an AI‑powered tool which has been integrated into the Europeana portal, that, by leveraging a multilingual vocabulary -also produced within the project- allows to automatically flag and provide contextual information about potentially contentious terminology in heritage metadata. 

Building on the outcomes of the DE‑BIAS project, this study investigates how explainable AI and data‑visualization techniques can be adopted to support GLAM institutions in systematically detecting issues within their collections’ metadata. By using as a case study photographic collections from the Wellcome Collection and Photoconsortium, this research draws on the iCANDID infrastructure to analyse collections metadata and to inform the development of a diagnostic dashboard designed to assess collections’ metadata quality and identify areas requiring remediation. The objective is to provide cultural heritage institutions with tools that support systematic evaluation and diagnosing of their collections by transforming metadata into visual structures that highlight discrepancies or imbalances in the descriptive patterns. This process aims at facilitating the interpretation and identification of potential biases or descriptive imbalances, such as limited richness, incomplete records, multilingual inconsistencies, and representational gaps.

• William Thorne University of Sheffield

University of Sheffield

Cultural heritage institutions hold vast collections, and the effort to digitally catalogue them for research and public access has been enormous. Yet the backlog continues to grow. Many existing records were created before current standards and technologies matured, resulting in inconsistent, incomplete, or error-prone metadata. Language models offer a route to addressing this at scale but the typical paradigm of large, compute-heavy models is ill-suited for a sector with limited hardware budgets and strict obligations around copyright and data privacy. Institutions often cannot share collection data with third-party services and rarely have the infrastructure to even run models that are referred to as “small” in language modelling research. The result is a transformative technology, placed tantalisingly out of reach of a sector that stands to gain enormously from its application.

This paper investigates how language models can be practically deployed within Galleries, Libraries Archives and Museums (GLAM) digitisation and data enrichment workflows, in partnership with the National Gallery. Rather than pursuing scale, it promotes the fine-tuning of smaller models for specific tasks on institutional data. The work identifies several pivotal tasks involved in producing quality data for heritage research and public dissemination where this proves especially valuable: correcting errors introduced during digitisation, standardising inconsistent metadata, extracting structured information from unstructured text and disambiguating references to people, places, and concepts within collections records.

To keep deployment feasible under real institutional constraints, the research makes use of contemporary efficiency techniques such as quantisation and Low-Rank Adaptation (LoRA). By sharing a single base model and attaching lightweight, task-specific adapters, multiple specialised models can be served without substantially increasing memory requirements.

Fine-tuning demands training data that many institutional tasks lack. In such cases, synthetic data generation becomes a practical necessity. The research examines strategies for producing synthetic data of a higher quality, that are more challenging and more representative of the characteristics present in heritage collections. In particular, it investigates the counterintuitive notion of introducing noise into evaluation datasets to better emulate the reality of typical source material.

A further contribution addresses what is arguably the most significant barrier to responsible adoption of language models in the sector: the absence of domain-specific evaluation benchmarks. Without rigorous ways to measure whether a model actually performs well on heritage tasks, institutions have little basis for informed decision-making. This research constructs several benchmark datasets by leveraging the structure of existing collections metadata to derive labels, testing domain knowledge and performance across common sectoral tasks while leveraging the expense, effort and expert knowledge already invested into institutional collections.

As a primary case study, the research draws on real user search queries from the National Gallery's discovery endpoint to evaluate embedding models, providing a grounded measure of retrieval quality. This embedding evaluation is presented through an interactive dashboard, offering a more visual and accessible evaluation methodology.

Taken together, the result is a pragmatic approach to integrate language models into heritage workflows: efficient models that run within institutional constraints, methods for faithfully generating the training data those models need, and evaluation tools to assess whether they are actually working. The aim throughout is to move language models from a promising but inaccessible technology toward something heritage institutions can meaningfully adopt and critically assess.

• Hakeem O. Ogunmuyiwa University of Sheffield

University of Sheffield

The intention behind this topic/presentation is to seek an ethically plausible process of preserving a disappearing cultural practice and further provoke scholarly debates. This fading practice is known as (facial) tribal marks.

The inscription of facial marks, which are drawn vertically and/or horizontally on the cheek, is one of the ancient cultural practices linked to a number of communities in sub-Saharan Africa notably Nigeria, Togo, and the Benin Republic. Other communities in West Africa where tribal marks are traditionally worn include Hausa, Fulani (or Fula), Gobir, Mossi (Ghana/Burkina Faso), Dagomba/Dagbanba (Ghana), Sisala (Ghana), Bobo (Burkina Faso) and Kagoro (Soninke/Mali). They are traditional facial markings or scarification that were historically used for multiple purposes, which include identification, social stratification, and beauty enhancement, a piece of art. Tribal marks are traditionally made by cutting or scarring the skin with sharp instruments, partly with the belief that these marks would distinguish members of one family or community from another. The actual marks varied from one town, family and lineage to another, then worn like badges of honour with pride.

Although it dates back to as early as the 14th century, the donning of tribal marks became popular during the period 16th to 19th century where families were forcefully separated due to colonial activities such as transatlantic slave trade. It became a means of identification of the tribe or lineage individuals belonged to, especially when searching for a lost relation. It was also a powerful means of identification during inter-tribal wars. In addition, certain versions of the marks were also seen as facial beautification although many would nowadays argue against this.

With modernisation however, the purposes and benefits appear to be no more relevant. Indeed, modernisation has resulted into stigmatisation and discrimination of people with tribal marks. Its accompanying potential health risks has also been debated (Adisa et.al 2024; Afolabi 2023). In addition, the country’s Child Rights Act (CRA) of 2003 has banned all forms of child mutilation. All of this explains why the practice is fast disappearing.

Digitisation, a methodological provision often used in Digital Humanities, offers the opportunity to reactivate dying cultures through digitisation. Digitising such disappearing cultural practices can undoubtedly document the practice and widen access to archives and initiatives aimed at promoting decolonisation. While digitisation tools (e.g. Tropy, GIMP, Universal Viewer) can come handy for managing the images, a lot of challenges would need to be overcome. Although digitisation efforts of African cultural heritage resources are already in place, McGregor et.al (2025) have reported that the progress of edited archives and ethnographic collections that focused on Africa published in 2014 has been lopsided.

How can lived experiences support or substantiate the narratives around tribal marks? Emerging issues that require extensive exploration and appropriate methodological strategy are digital divide (lack of access to the right technology in many indigenous communities), colonial legacy, emotional connections of the locals and data privacy and ethics particularly with regards to intellectual property and controls of the digital outputs.