Cultural Heritage institutions increasingly publish their collections as data, enabling computational reuse and new forms of scholarly inquiry. Recent initiatives such as the Collections as data, International GLAM Labs, and CARE principles promote the publication of digital collections supporting computational and responsible use [1, 2, 3]. Metadata plays a crucial role as a way to describe the content [4, 5]. Within this context, Cultural Heritage collections are commonly described as multilingual, usually referring to the linguistic diversity of their content. Yet the multilingual character of the content is not necessarily reflected in the descriptive metadata, which may remain uneven, incomplete, or accessible only through a limited set of languages.

This work proposes a reproducible method to assess multilingual metadata in Cultural Heritage collections, focusing on how linguistic representation is structured, distributed, and curated in relation to the multilingual character of the underlying content [6, 7]. The approach treats each collection as a bounded metadata corpus and applies a transparent computational workflow to examine patterns of multilingual description, metadata field completeness, and access conditions across languages.

The approach proposed in this research works in four steps as seen in Figure 1. The first step consists of retrieving the data using APIs such as OAI-PMH and SPARQL. The second step is about defining the metrics to assess multilingual metadata using several approaches according to previous work [5, 7]. The following step corresponds to the execution of the code using cloud services. Finally, the last step consists of the analysis by means of charts, tables and textual documentation.

The analysis is applied to a sample of GLAM collections covering large and medium-sized institutions to exemplify different regimes of multilingual description, including federated collections, nationally curated heritage datasets, and historically multilingual holdings.

The results represent a snapshot of the collections at the time of extraction, reflecting the dynamic nature of Cultural Heritage infrastructures. Crucially, the reproducible design of the method allows the analysis to be re-executed over time in open and collaborative platforms such as Binder and EOSC Notebooks, enabling comparison across versions and supporting longitudinal assessment of metadata practices.

By framing multilingual metadata as an infrastructural property rather than a purely descriptive feature, this work contributes a reusable diagnostic approach for Cultural Heritage institutions seeking to understand and document linguistic inequities in their collections. It aligns the principles of Collections as Data with emerging concerns around metadata governance, transparency, and equity, and provides a practical framework for benchmarking multilingual metadata practices within the GLAM sector. Future work includes publishing the workflow in the Social Sciences and Humanities Open Marketplace and exploring its alignment with existing data infrastructures such as the European data space for cultural heritage and the European Cloud for Heritage Open Science (ECHOES-ECCCH).