<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
        <title>Journal of Open Humanities Data Feed</title>
        <link>https://reference-global.com/journal/UPJJOHD</link>
        <description>Reference Global RSS Feed for Journal of Open Humanities Data</description>
        <lastBuildDate>Mon, 26 Oct 2026 12:31:03 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <image>
            <title>Journal of Open Humanities Data Feed</title>
            <url>https://sciendo-parsed.s3.eu-central-1.amazonaws.com/ubiquity/68406cf3da18af458e768440/cover-image.jpg</url>
            <link>https://reference-global.com/journal/UPJJOHD</link>
        </image>
        <copyright>All rights reserved 2026, Ubiquity Press</copyright>
        <atom:link href="https://reference-global.com/2/rss/journal/UPJJOHD" rel="self" type="application/rss+xml"/>
        <item>
            <title><![CDATA[Evaluating UDPipe 2 and Stanza for Morphosyntactic Annotation of Late Latin: A Pilot Study on Sixth-Century Texts]]></title>
            <link>https://reference-global.com/article/10.5334/johd.630</link>
            <guid isPermaLink="false">/article/10.5334/johd.630</guid>
            <pubDate>Mon, 26 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[To date, annotated Late Latin treebanks and natural language processing (NLP) models specifically trained on this linguistic variety remain extremely limited. In this low-resource setting, the present study investigates which combination of annotation tool and model currently available for Latin provides the most reliable basis for developing a Late Latin treebank. To address this question, a manually annotated gold standard is created, comprising four sixth-century texts. The creation of this reusable dataset constitutes the first contribution of the study, as it provides a benchmark for the evaluation of existing and future NLP models on Late Latin. The evaluation focuses on part-of-speech tagging and dependency parsing, comparing two tools, UDPipe 2 and Stanza, across their available pretrained models. The results show that UDPipe 2 generally outperforms Stanza on the corpus examined, with the winning submission of EvaLatin 2024 achieving the best overall performance among the models tested. These findings suggest that UDPipe with the EvaLatin 2024 model provides a suitable starting point for a semi-automatic workflow for the creation of new Late Latin treebanks, in which automatic pre-annotation is followed by expert manual revision.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Irish and the Ordnance Survey Letters – Dialects, Accent Features, and Levels of Irish]]></title>
            <link>https://reference-global.com/article/10.5334/johd.619</link>
            <guid isPermaLink="false">/article/10.5334/johd.619</guid>
            <pubDate>Wed, 21 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[In the 1830s and 1840s, scholars employed by the Topographical Section of the Ordnance Survey of Ireland were sent across Ireland to collect place names. The scholars decided as well to collect information about Irish speakers and the Irish they spoke. As Irish-speakers themselves, the scholars in the Topographical Section had a unique opportunity to document the Irish spoken. This included dialectal variations and the levels of Irish in a townland/parish, which they recorded in the Letters they wrote back to the Section Headquarters in Dublin. This small dataset records their findings from close reading the Letters. Hosted on Zenodo, this dataset is of interest to those involved in historical linguistics, dialectology, and nineteenth-century Irish Studies.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Aligning the Odes: A Character-Level Parallel Dataset for the Anhui University and Fùyáng Bamboo-Slip Manuscripts]]></title>
            <link>https://reference-global.com/article/10.5334/johd.644</link>
            <guid isPermaLink="false">/article/10.5334/johd.644</guid>
            <pubDate>Mon, 19 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[This dataset aligns graph-level transcriptions of three witnesses to the Book of Odes: the received Máoshī 毛詩, the Warring States so-called ‘Anhui University’ (Āndà 安大) bamboo manuscript, and the Western Han Fùyáng 阜陽 bamboo manuscript. The release represents each witness in Unicode JSON, uses Ideographic Description Sequences for graphs without suitable encoded characters, and records explicit one-to-one, one-to-many, and many-to-many correspondences. Separate mappings connect transcription positions to physical locations in the source editions and those locations to images of the manuscript characters. JSON Schemas, integrity tests, a static read-only viewer, and reviewed crop recipes support validation and reproducibility. The corpus makes manuscript variation available both as structured data and through a visual interface in which readers can inspect aligned passages, graph images, and their positions on reconstructed manuscript strips. Its format can accommodate later manuscript publications and competing transcriptions of the same physical witness. It turns manuscript and transcription variation into an inspectable, reusable research infrastructure for early Chinese textual transmission, palaeography, historical phonology, and digital philology.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Construction of Consensus-Labeled Data for the Study of Emotions in Korean Full-Length Novels]]></title>
            <link>https://reference-global.com/article/10.5334/johd.626</link>
            <guid isPermaLink="false">/article/10.5334/johd.626</guid>
            <pubDate>Fri, 16 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[This study proposes a consensus-labeling methodology for constructing emotion data for Korean full-length novels by integrating multiple annotations, model predictions, and researcher verification. Emotion interpretation in literary texts may vary across researchers, while sentence-level emotion prediction models may not fully capture emotions that depend on broader narrative context. First, for 5,282 sentences from Sohyeonseongrok, one set of deep learning model predictions and two sets of human annotations were combined. Provisional labels were derived through majority rule and then subjected to researcher verification based on surrounding narrative context. Next, eight emotion prediction models trained under different training-data configurations were applied to 15,426 sentences from Sossi-Samdaerok. When five or more models predicted the same emotion, the shared label was treated as a provisional model consensus and subsequently reviewed by the researcher. Model comparisons showed that predictions varied according to training-data configuration, while the model trained on the Sohyeonseongrok consensus-labeled data achieved agreement comparable to a model trained on a substantially larger dataset formed by concatenating three annotation datasets. These findings suggest that consensus labeling can provide practical shared reference labels for computational analysis while preserving information about differing annotation perspectives, rather than establishing a single objectively correct emotion label. The resulting datasets are publicly available on Zenodo under a CC-BY 4.0 license for reuse in Digital Humanities and NLP research on Korean classical literature.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Editorial. Wikidata Across the Humanities: Datasets, Methodologies, Reuse]]></title>
            <link>https://reference-global.com/article/10.5334/johd.624</link>
            <guid isPermaLink="false">/article/10.5334/johd.624</guid>
            <pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[In this editorial, we present Wikidata Across the Humanities: Datasets, Methodologies, Reuse, a special collection dedicated to the publication of discussion and data papers that explore the use of Wikidata in the digital humanities. With 45 articles, it is by far the largest special collection in the Journal of Open Humanities Data. The collection deliberately spans a wide range of disciplines, from history and linguistics to archaeology and museology, which demonstrate the growing prominence and maturity of Wikidata as a knowledge base that is increasingly used not only in the humanities and social sciences, but also in the cultural heritage sector and GLAM institutions. The success of this collection reflects not only the adoption and dissemination of free and open knowledge, but also the research community’s interest in integrating methods and tools that enable the interoperability of existing datasets – an issue that has become increasingly important in recent years. We hope this collection will encourage further researchers to contribute to Wikidata and similar collaborative platforms, not as a peripheral activity, but as an integral part of how humanities data is built and shared.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[A Dataset for the Study of Jezebel in Late Antique Christian and Jewish Literature]]></title>
            <link>https://reference-global.com/article/10.5334/johd.635</link>
            <guid isPermaLink="false">/article/10.5334/johd.635</guid>
            <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[This dataset includes all references to Jezebel in late ancient literature available in digital corpora in Syriac, Latin, Greek, Hebrew, and Jewish Literary Aramaic before 850 CE. The dataset presents excerpts annotated for collocations, metadata, rhetorical, intertextual and thematic properties in TSV files in Zenodo. It is useful to anyone studying the reception of biblical women or Jezebel in particular, as well as the use of the Bible and Biblical Storyworld in late antiquity. It also serves as a dataset for the study of late ancient rhetoric across languages and religious divides through the prism of one wicked woman.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Ibero-Romance Handwriting in Arabic Script (IRHAS): A First Step to Manuscript Text Recognition in Spanish Ajami (Aljamiado)]]></title>
            <link>https://reference-global.com/article/10.5334/johd.646</link>
            <guid isPermaLink="false">/article/10.5334/johd.646</guid>
            <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[The Ibero-Romance Handwriting in Arabic Script (IRHAS) dataset provides ground-truth data for Automatic Text Recognition (ATR) of Spanish Ajami or Aljamiado (Spanish written in Arabic script) manuscripts. Comprising 22 extracts from 19 Mudejar and Morisco codices from the late 15th to the early 17th centuries (663 pages, 10,825 lines, 550,299 characters in total), the corpus was curated via eScriptorium and controlled using Choco-Mufin. Stored and documented on Zenodo under an open-access licence, IRHAS enables the fine-tuning of generic ATR models, achieving an average Character Accuracy Rate of 92.54%, which will facilitate the production of scholarly editions and, eventually, text mining and analysis.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[A Benchmark Dilemma: Between Representativeness and Domain-Specificity in Visual Arts Domains]]></title>
            <link>https://reference-global.com/article/10.5334/johd.623</link>
            <guid isPermaLink="false">/article/10.5334/johd.623</guid>
            <pubDate>Tue, 13 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[In this paper we address the state of benchmarking in visual arts domains, the challenges associated with this task and we present a systematic benchmarking of transformer models for art historical classification tasks. With this, we link our benchmarking to a fundamental discourse of digital/computational art history, presenting a framework for an in-depth, but also practical implementation of machine learning methods in art historical questions. Finally, our outlook addresses the role of benchmarks and art historical research on a more general basis.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Picturing Stories: A Citizen-Science Annotated Dataset of Children’s Book Illustrations]]></title>
            <link>https://reference-global.com/article/10.5334/johd.592</link>
            <guid isPermaLink="false">/article/10.5334/johd.592</guid>
            <pubDate>Mon, 12 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[Illustrated children’s books have played a central role in the cultural transmission of narratives, values, and models of childhood for more than three centuries. Yet despite growing interest in computational approaches to cultural heritage and multimodal analysis, large-scale datasets focused on historical children’s book illustrations remain limited. We introduce Picturing Stories, a new citizen-science annotated dataset containing 44,043 illustrations drawn from a collection of 3,388 children’s books published between 1731 and 2011. The dataset combines materials from the Internet Archive’s Children’s Library and the International Children’s Digital Library, spanning historical and contemporary works across 37 languages. Annotations were produced through a 42-week Zooniverse project in which 10,647 volunteers contributed over 1.3 million total annotations, capturing core dimensions of visual narrative, including character presence, setting, objects, actions, emotional intensity, family relations, and caregiving behaviors, alongside a subset of ~27,000 images with two or more captions. This data can be used to study the history of children’s books and benchmark AI for the study of further datasets.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[A Structured Dataset of Payments, Places and Persons in the Cartulary of Santo Toribio de Liébana (790–1626)]]></title>
            <link>https://reference-global.com/article/10.5334/johd.633</link>
            <guid isPermaLink="false">/article/10.5334/johd.633</guid>
            <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[This paper presents a structured, open dataset derived from the Cartulario de Santo Toribio de Liébana, a monastic cartulary from northern Spain edited by Sánchez Belda (1948). The dataset converts 566 charters (790–1626) into five linked relational files: (i) 328 payment events; (ii) 531 payment items, retaining goods, quantities, units and legal context; (iii) a gazetteer of 603 place names with medieval variants, municipal attribution and, for 94 places, geographic coordinates; (iv) an index of 1,348 persons with their attested roles; and (v) a concordance between cartulary and edition numbering. The extraction pipeline, implemented in Python 3.12, is released with the data, alongside an audit trail recording penal-clause contamination and Optical Character Recognition (OCR) defects. The dataset supports research on medieval monetisation, diplomatic formulae, historical toponymy and the economic history of northern Iberia.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Studying the Ownership Structures of an Algiers’ Neighbourhood on the Eve of Its Demolition. The Creation of a Geographic Information System Using Historical Land Records from the 1930s]]></title>
            <link>https://reference-global.com/article/10.5334/johd.616</link>
            <guid isPermaLink="false">/article/10.5334/johd.616</guid>
            <pubDate>Wed, 07 Oct 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[The history of colonial cities often comes to a halt at the neighbourhood level: sources enabling analysis at the street level, let alone at the level of individual buildings, are extremely scarce. This is the case for Algeria and, in particular, for its capital, Algiers. Thus, the ownership structure is most often known only through statistical aggregations at the city-wide level.
A major ‘renovation’ project in the highly central Marine district, involving demolition followed by reconstruction, required the expropriation of several hundred property owners in the 1930s. These expropriation procedures provided an opportunity to produce a plot map and prefectural decrees listing the owners by name. The map has been georeferenced and vectorised. The wealth of information contained in the prefectural decrees (the plot of land concerned, the owner’s name, address, gender, etc.) has been converted into a dataset. It is this dataset that is described and analysed here. It enables significant progress to be made in understanding the social history of Algiers following a century of colonisation.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[A Normalisation Layer for Old Turkic Text Editions in OCR-Based Workflows]]></title>
            <link>https://reference-global.com/article/10.5334/johd.627</link>
            <guid isPermaLink="false">/article/10.5334/johd.627</guid>
            <pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[For over a century, Old Turkic texts have been edited using divergent Latin-based transcription systems developed independently across German, Turkish, Russian, Japanese, French, and Chinese scholarly traditions. Because most of these editions survive only as printed volumes or PDFs, optical character recognition (OCR) is the primary route to digitisation, yet OCR does not resolve, and often amplifies, the transcriptional heterogeneity already embedded in the source editions. Even error-free character recognition would not yield interoperable text, since the incompatibility originates in editorial decisions rather than in the recognition process. This paper discusses a deterministic, representation-level normalisation framework that operates immediately after OCR extraction. The framework is additive, reversible, and never corrective. It preserves the raw OCR output unchanged in an orig layer and generates a canonical transcription in a reg layer using Text Encoding Initiative (TEI) P5 choice structures, with all philological interpretation frozen in static, edition-aware mapping tables rather than executed dynamically. Applied to 4,485 OCR-derived tokens from 13 editions published between 1919 and 2023, the approach allowed 94–97% of the tokens to be normalised using the defined mapping rules. By separating preservation from normalisation, it offers a reproducible basis for large-scale Old Turkic corpus construction and cross-edition search across different transcription traditions.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Digital Journeys: Access Modalities, Content &amp; Navigation Data from a Humanities and Social Sciences Open Digital Library]]></title>
            <link>https://reference-global.com/article/10.5334/johd.610</link>
            <guid isPermaLink="false">/article/10.5334/johd.610</guid>
            <pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[This dataset contains user navigation sessions along with metadata about the content accessed by users collected from OpenEdition, an open-access digital library for the humanities and social sciences. It was built from server logs that record page requests, timestamps, referral sources, device information, and search interactions, then enriched with geolocation and document metadata. Automated traffic was removed and sessions were reconstructed through a dedicated identification process. The dataset covers two comparable periods, December 2024 and December 2025, and is stored as Parquet and CSV files. It can support research on scholarly browsing behavior, search and recommendation systems, digital library design, and longitudinal studies of information seeking.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Manually Verified Polish Civil-Register Transcriptions and Weak Token Labels]]></title>
            <link>https://reference-global.com/article/10.5334/johd.628</link>
            <guid isPermaLink="false">/article/10.5334/johd.628</guid>
            <pubDate>Mon, 28 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[Polish civil registers are formulaic, but computational processing can be hard to inspect. This paper documents 100 manually checked TXT page transcriptions from Kraśnik with parallel PAGE-XML workflow exports, provenance, checksums, weak lexical categories, and a 480-row audit. Page-level files preserve layout and provenance; documentation separates descriptive labels from evaluation evidence. Scripts support classical and simulated quantum-kernel demonstrations. The resource aids teaching, error tracing, and evaluation design; purposive sampling, lexical repetition, absent scans, and non-contextual labels preclude representative, full-NER, or quantum-advantage claims.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Randomised Sentences and Word-Embeddings Generated During the National Corpus of Irish Project]]></title>
            <link>https://reference-global.com/article/10.5334/johd.617</link>
            <guid isPermaLink="false">/article/10.5334/johd.617</guid>
            <pubDate>Mon, 28 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[While collecting high-quality language data for the National Corpus of Irish project between 2022 and 2024 it became clear to the Gaois research group that there was a broader need for data in Irish-language NLP tasks and applied linguistic research.
This paper details specifications of the shareable by-products that were generated from some or all of these data; a 76,578,449-word dataset of randomised sentences and two different types of 100-dimension word embeddings that were generated from the 150-million words of high-quality Irish that was collected during the Corpas Náisiúnta na Gaeilge project. All datasets are shared as zipped TXT files.
This paper also explains the considerations for creating resources of this sort, some of which pertain to minority languages and smaller language communities, others to languages that have fewer language technologies.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[A Corpus of Keystroke Logging From Children and Adult Writing]]></title>
            <link>https://reference-global.com/article/10.5334/johd.609</link>
            <guid isPermaLink="false">/article/10.5334/johd.609</guid>
            <pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[We present a corpus of keystroke logs of writing sessions from children and adult French writers from which we extracted a set of data related to bursts of writing, their temporal characteristics and their textual function. The corpus includes the raw keystroke logging files (recording of text production by typing). The corpus can be used to further investigate the writing process across different constraints, each related to a specific cognitive process involved in writing (planning, translating and reviewing) for adults and exploring writing development in children in two genres of text (narratives and expository texts).]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Building the Parsed Corpus of Scottish Correspondence]]></title>
            <link>https://reference-global.com/article/10.5334/johd.622</link>
            <guid isPermaLink="false">/article/10.5334/johd.622</guid>
            <pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[The Parsed Corpus of Scottish Correspondence (PCSC) is the first syntactically parsed corpus of Scots, annotated according to the Penn Parsed Corpora of Historical English (PPCHE) framework (Kroch, 2020). Developed from the Helsinki Corpus of Scottish Correspondence, 1543–1747 (Meurman-Solin &amp; VARIENG, 2017), the corpus provides 270,553 words of manually corrected syntactic annotation together with rich metadata of, e.g., authors, recipients, chronology, geography, and scribal practice (e.g., whether the letter is autograph and what type of script is used). This paper discusses the methodological decisions involved in adapting an existing diplomatic corpus for syntactic annotation, including data preparation, automatic Part-of-Speech (PoS)-tagging, syntactic parsing, and manual correction. Particular attention is given to the challenges posed by non-standard and highly-variable orthography, editorial markup, as well as the absence of resources designed for Scots, and to the strategies adopted to facilitate annotation efficiency and consistency while preserving the linguistic characteristics of the source texts. Hence, beyond documenting the creation of the PCSC, the discussion highlights methodological considerations relevant to the development of parsed historical corpora more generally, particularly for under-resourced varieties. The paper outlines the annotation workflow and demonstrates how the resulting corpus supports reproducible syntactic searches using CorpusSearch2 (Randall, 2010; Randall et al., 2004). The corpus expands the range of PPCHE-type corpora available for comparative historical syntax and provides a resource for future quantitative research on the history of Scots.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Making Sense of Italian Legacy Resources: Integrating ItalWordNet and PAROLE-SIMPLE-CLIPS for Sense Annotation]]></title>
            <link>https://reference-global.com/article/10.5334/johd.532</link>
            <guid isPermaLink="false">/article/10.5334/johd.532</guid>
            <pubDate>Tue, 22 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[This paper discusses the creation of the Italian sense-annotated sub-corpus of a parallel corpus developed in the context of a large European initiative, and focuses on the reuse of legacy computational lexical resources to produce high-quality semantic annotation data for Italian. The work is situated within the current shift in Language Technology towards data quality, where expert-curated resources are increasingly valued for model evaluation, fine-tuning, and explainability. In the absence of an open, comprehensive digital dictionary for Italian, we adopted a bottom-up strategy based on the integration of two authoritative openly licensed legacy lexical databases, namely ItalWordNet and PAROLE-SIMPLE-CLIPS. These resources were semi-automatically merged and curated to construct a unified initial sense inventory, which formed the backbone of the manual sense annotation of a parallel Italian corpus carried out by expert linguists, and was subsequently refined and enriched as a direct outcome of the annotation process. Beyond the resulting dataset, the paper reflects on the challenges involved in merging heterogeneous legacy resources, the role of data reuse in countering digital obsolescence, their usefulness and limitations when applied to annotation tasks. In addition it reinforces reflections and best practices for sustainability, interoperability, and documentation, strongly advocated by Open Science and especially within large research infrastructures. Overall, the experience demonstrates how legacy lexical knowledge can be revitalized and made actionable for contemporary NLP and Digital Humanities research, provided consistent human work is dedicated to carry out or supervise the process. It also highlights the limitations of existing resources and the need for the continued creation of new, high-quality lexical data.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[Logainm Administrative Names]]></title>
            <link>https://reference-global.com/article/10.5334/johd.612</link>
            <guid isPermaLink="false">/article/10.5334/johd.612</guid>
            <pubDate>Mon, 21 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[In this paper we describe the Logainm Administrative Names dataset which constitutes a snapshot of toponymic data from the Placenames Database of Ireland (Logainm) and which is deposited as a collection in the Digital Repository of Ireland. Logainm is an ever-evolving database used to manage and disseminate the official Irish-language administrative placenames of Ireland, as established by the Placenames Branch of the Government of Ireland in consultation with the Placenames Committee (formerly the Placenames Commission). The dataset includes the names, categories and locations of c. 68,000 provinces, counties, administrative counties, cities, baronies, metropolitan districts, borough districts, municipal districts, towns, civil parishes, electoral districts and townlands. It will be of use to researchers and administrators, as well as place-based product developers.]]></description>
            <category>ARTICLE</category>
        </item>
        <item>
            <title><![CDATA[An Enriched Dataset of Radio Netherlands Worldwide Broadcast Documentation (1950–1966)]]></title>
            <link>https://reference-global.com/article/10.5334/johd.614</link>
            <guid isPermaLink="false">/article/10.5334/johd.614</guid>
            <pubDate>Wed, 16 Sep 2026 00:00:00 GMT</pubDate>
            <description><![CDATA[We present an enriched dataset of 226,359 pages from Radio Netherlands Worldwide (RNW) broadcast archives spanning 1950–1966 and November-December 1969 (of which 3,002 pages are from Nov–Dec 1969). RNW was the Netherlands’ international public broadcasting service, launched in 1947 to transmit Dutch news, culture, and perspectives to audiences abroad via shortwave radio. Source PDFs from the KITLV (Royal Netherlands Institute of Southeast Asian and Caribbean Studies) archive were transcribed using a vision-language model (Infinity-Parser-7B/Qwen2.5-VL), classified into six document types via fuzzy pattern matching, and enriched with broadcast dates and multilingual language distributions at both the page and line levels. The dataset also includes 72,212 music references extracted from music program pages, enriched with Discogs metadata. The resulting corpus enables computational analysis of Cold War-era international broadcasting, multilingual radio programming, and historical music distribution patterns.]]></description>
            <category>ARTICLE</category>
        </item>
    </channel>
</rss>