Conference item
Multimodal large language model-assisted metadata extraction from historical concert programmes (1872–1928)
- Abstract:
- Historical concert programmes are widely recognised as valuable sources for musicological historiography; yet, although many collections are increasingly becoming available online, identifying relevant programmes within larger archival holdings remains difficult. When present, descriptive metadata is often provided only at the collection level rather than for individual items. Even then, the absence of granular metadata at the level of individual programmes continues to preclude systematic analysis of their contents at scale. This paper reports the results of an experiment, testing whether general-purpose multimodal large language models (MLLMs) can assist in preparing semi-structured concert programme metadata as a first-pass metadata record ready for subsequent expert verification and reconciliation. Using a sample of 100 programmes (1872–1928) from three Oxford student music societies held at the Bodleian Libraries, we implement a lightweight workflow comprising low-cost image capture using everyday consumer hardware, minimal preprocessing, schema-constrained JSON output, and repeated sampling using a simple consensus strategy of extracted metadata fields. We compare MLLM outputs against a manually curated reference dataset using Levenshtein distance as a character-level error measure, treating extraction correctness primarily in terms of sufficient intelligibility and data consistency for future reconciliation. We find that MLLM output quality depends strongly on model choice and on expert-controlled design decisions (including hierarchical JSON schema definition and prompt specification). We conclude with practical recommendations for institutions and researchers who wish to treat MLLMs as assistive components in ephemera metadata creation workflows, while retaining expert authority over final metadata records.
- Publication status:
- Published
- Peer review status:
- Peer reviewed
Actions
Access Document
- Files:
-
-
(Preview, Version of record, pdf, 4.6MB, Terms of use)
-
- Publisher copy:
- 10.1145/3815723.3815726
Authors
+ Leverhulme Trust
More from this funder
- Funder identifier:
- https://ror.org/012mzw131
- Grant:
- RPG-2024-080
- Publisher:
- Association of Computing Machinery
- Host title:
- Proceedings of the 13th International Conference on Digital Libraries for Musicology
- Pages:
- 19-28
- Publication date:
- 2026-07-01
- Event title:
- DLfM '26: Proceedings of the 13th International Conference on Digital Libraries for Musicology
- Event series:
- 13th International Conference on Digital Libraries for Musicology (DLfM 2026)
- Event location:
- Thessaloniki, Greece
- Event website:
- https://doi.org/10.1145/3815723.3815726
- Event start date:
- 2026-07-02
- Event end date:
- 2026-07-02
- DOI:
- ISBN:
- 9798400723698
- Language:
-
English
- Keywords:
- Pubs id:
-
2442309
- Local pid:
-
pubs:2442309
- Deposit date:
-
2026-07-06
- ARK identifier:
Terms of use
- Copyright holder:
- Eck and Page
- Copyright date:
- 2026
- Rights statement:
- © 2026 Copyright held by the owner/author(s). This work is licensed under a Creative Commons Attribution 4.0 International License.
- Licence:
- CC Attribution (CC BY)
If you are the owner of this record, you can report an update to it here: Report update to this record