Historical Wine Labels of the German Mosel Region: Enabling Insights into Visual Cultural Heritage using Linked Open Data



Christof Schöch, Maria Hinzmann,
Veronica Wassermayr, Joëlle Weis, Achim Rettinger

Trier Center for Digital Humanities
Trier University, Germany

Digital Humanities Conference 2025
Universidade NOVA, Lisbon, Portugal

16 Jul 2025

Overview

  • 1 – Background
  • 2 – Historical Wine Labels
  • 4 – Wikibase
  • 5 – Image annotation
  • 6 – Conclusion

Background:
Wine Making in the German Mosel Region

Winemaking in the Moselle region

  • Practised since Roman times
  • Today, ~200 wine-making villages along the Mosel river
    • Beilstein
    • Bremm
    • Merl
    • Piesport

Wine cultivation around Trier University

Trier Center for Digital Humanities (TCDH)

  • Founded in 1998 at Trier University
  • Focus areas
    • Scholarly Digital Editions
    • Digital Lexicography
    • Research Software Engineering
    • Research Infrastructure
    • Computational Literary Studies
    • Digital Cultural Heritage
  • Wine label research as part of our LODinG initiative

Historical wine labels

What do these labels tell us about?

  • Document historical vineyard names
  • Reflect artistic periods
  • Show historical views (landscapes, villages)
  • Reflect legally required information
  • Show winemakers, traders and printers
  • Showcase wine marketing strategies

Amerine Collection (UC Davis)

Weinetikettensammler Database

Map of the German Wine Institute (DWI)

Multimodal LLM-based Annotation of Historical Images

  • Survey: Caffagni et al. (2024)
  • Application to the Sphaera Corpus: focus on illustrations; Büttner et al. (2022)
  • Application to the Documerica collection: prose captions that enhance search; Arnold and Tilton (2024)

The Wine Label Wikibase

Data model: major classes

  • Top-level classes
    • Organisation
    • Spatial Entity
    • Physical Object
  • Main classes of interest
    • Label
    • Vineyard
    • Wine estate

Descriptive vocabularies (examples)

  • For wine labels
    • grape varieties (Riesling, Gewürtztraminer, Elbling, Scheurebe, …)
    • quality markers (Auslese, Gutsabfüllung, naturrein, …)
    • depicted objects (coat of arms, vineyard, cat, …)
  • For vineyards
    • soil type (slate, schist, loam, sand, clay, …)
    • exposure (south, south-east, south-west, …)

External identifiers

State of play of the dataset

  • 1905 wine label entries
    (~ 1810–1960)
  • 1234 vineyards
    • 466 current vineyards
    • 768 historical vineyards
  • 641 wine estates
  • 200 villages and towns
  • 45 label printing companies
  • ~1600 depicted objects

Image annotation

Manual annotation of labels for depicted objects

  • grapes (Q881)
  • wine leaves (Q879)
  • wine branch (Q977)
  • eagle (Q883)
  • crown (Q911)
  • border (Q979)
  • ornaments (Q971)
  • sign (Q922)

Using multimodal LLMs for object identification

  • Challenge: not (only) prose, but use controlled vocabulary
  • Model generally remembers all identifiers (long list)
  • Big improvements from ChatGPT o3 to o4, especially in output structure
  • Gemma 3 with fairly easily parsable output (currently prepared model)

Using Ollama for object identification

  • Ollama for local access to open models (gemma3:27b)
  • Python interface to Ollama
  • System prompt with inventory of objects
  • Prompt: create lists of textual and visual elements
  • Iterate over label collection and save output
  • Compare model output with annotations created by humans

Using Ollama for object identification

Conclusion

Current state of affairs

  • Basic data model, reference entities, and controlled vocabularies are there
  • Almost 2000 labels are included in the Wikibase
  • First interesting queries are possible

Next steps

  • Continue digitising labels
  • Refine the data model (e.g. inventory of objects)
  • Scale up image annotation and evaluation experiments
  • Fine-tune models with annotated labels

Thank you!

Bonus slides

Validation using EntitySchema

Map of locations with labels

Using Ollama for object identification

References

References

Arnold, Taylor, and Lauren Tilton. 2024. “Explainable Search and Discovery of Visual Cultural Heritage Collections with Multimodal Large Language Models.” In CHR 2024: Computational Humanities Research Conference. Aarhus: CHR.
Büttner, Jochen, Julius Martinetz, Hassan El-Hajj, and Matteo Valleriani. 2022. “CorDeep and the Sacrobosco Dataset: Detection of Visual Elements in Historical Documents.” Journal of Imaging 8 (10): 285. https://doi.org/10.3390/jimaging8100285.
Caffagni, Davide, Federico Cocchi, Luca Barsellotti, Nicholas Moratelli, Sara Sarto, Lorenzo Baraldi, Lorenzo Baraldi, Marcella Cornia, and Rita Cucchiara. 2024. “The Revolution of Multimodal Large Language Models: A Survey.” arXiv. https://doi.org/10.48550/arXiv.2402.12451.
DWI. 1999. ABC Des Weinetiketts. Genuss Und Geschmack Auf Der Spur. Mainz: Deutsches Weininstitut GmbH.
Jones, Andrew. 1994. The Stories Behind the Labels: The History, Romance & Characters of the World of Wine and Drink. Bathurst, NSW: Crawford House.
Laufner, Richard, Wolfgang Binsfeld, Heinz Cüppers, and Rheinisches Landesmuseum Trier, eds. 1987. 2000 Jahre Weinkultur an Mosel-Saar-Ruwer: Denkmäler und Zeugnisse zur Geschichte von Weinanbau, Weinhandel, Weingenuß. Trier: Rheinisches Landesmuseum.
Matheus, Michael, ed. 2019. Weinkultur und Weingeschichte an Rhein, Nahe und Mosel. Mainzer Vorträge 22. Stuttgart: Franz Steiner Verlag.
Obis, E. 2018. “Wine Labels and Consumer Culture in the United States.” InMedia. https://doi.org/10.4000/inmedia.1029.
UC Davis Library. 2020. “Label This: Wine Label Transcription Project.” Label This: Wine Label Transcription Project. https://labelthis.library.ucdavis.edu/.
Weis, Joëlle, and Christof Schöch. 2024. “Vom Perler Hasenberg zur Lehmener Würzlay – Weinetiketten digital erschließen.” In Sammlungsforschung im digitalen Zeitalter, edited by Katharina Günther and Stefan Alschner, 19–28. Göttingen: Wallstein.