Corpus Inscriptionum Armenicarum Electronicum
The inscription notices of the ten volumes of the Corpus of Armenian Inscriptions, each linked to the page of the printed edition it was read from.
Editorial signs are ignored, so ՍՈՒՐԲ also finds Ս[ՈՒՐ]Բ. Or browse all notices.
Volumes
| Vol. | Title | Editor(s) | Year | Notices |
|---|
| Ref. | Notice |
|---|
Select a notice.
Data
Export
Sources
Fields
Licence and citation
The data are released under the Creative Commons Attribution-ShareAlike 4.0 licence. Corrections arrive continuously: when citing a download, give the date and the correction commit shown above. How to cite
Corpus Inscriptionum Armenicarum Electronicum
Project
The Corpus Inscriptionum Armenicarum Electronicum (CIAE) is a digital version of the Divan hay vimagrutʻyan (Դիվան հայ վիմագրության, Corpus of Armenian Inscriptions), the series compiled by the Institute of Archaeology and Ethnography of the National Academy of Sciences of Armenia since 1966 (see Bibliography).
Led by Calfa in collaboration with Inalco (ANR project DALiH), the CIAE has three aims:
- to make the information in the printed volumes easy to search and query;
- to provide a lasting platform for documenting Armenian epigraphic heritage, on which records can be updated and enriched over time;
- to provide silver data in Old and Middle Armenian for training language models (for instance, to decipher damaged inscriptions) and for exploring methods in the digital humanities. This aim is pursued jointly with the MESROP project, funded by DISTAM.
The CIAE rests above all on the remarkable work of compilation and description carried out over the decades in the Divan hay vimagrutʻyan. Any use of the digital corpus must cite the original volume as well as the digital version consulted (see the citation on each record, and How to cite below). The raw data are available on GitHub and through the export module.
Team
Method and data quality
The corpus was extracted and structured with a pipeline developed by Calfa that combines a vision-language model, for the OCR and semantic parsing of each record, with a YOLO detector that locates the inscription drawings. The pipeline is being documented for publication, and its formal evaluation is under way. A qualitative assessment by our experts has nonetheless judged the output good enough to be searched and used now. Most errors concern editorial signs.
The platform is also a proofreading tool. Records can be corrected in several ways, from reassigning drawings to fixing typos, and every proposed correction is checked by our experts (see Proofreading).
This section will be updated as the evaluation progresses.
Editorial signs
Transcriptions are diplomatic and reproduce the signs of the printed edition. Their exact use varies between volumes; each volume's table of conventional signs is authoritative.
Bibliography
The Corpus of Armenian Inscriptions
Further reading
- Barkhudarian, Makar. Artsʻakh. Baku, 1895.
- Greenwood, Tim. “A Corpus of Early Medieval Armenian Inscriptions.” Dumbarton Oaks Papers 58 (2004): 27–91.
- Elliott, Tom, et al. EpiDoc Guidelines. epidoc.stoa.org
- Vidal-Gorène, Chahan, and Aliénor Decours-Perez. “Detecting and Deciphering Damaged Medieval Armenian Inscriptions Using YOLO and Vision Transformers.” In Document Analysis and Recognition – ICDAR 2024 Workshops, edited by H. Mouchère and A. Zhu, 22–36. Lecture Notes in Computer Science 14936. Cham: Springer, 2024. doi:10.1007/978-3-031-70642-4_2
How to cite
Cite the printed edition first, then the digital record with its address and the date you consulted it:
The data are released under CC BY-SA 4.0.
Research projects that build on the corpus, each with its own team and funding.
The Artsakh corpus
This project is building the first digital corpus of Armenian inscriptions from Artsakh. Some 1,071 historical and architectural monuments are recorded in the region, with inscriptions ranging from the early Middle Ages to the 18th century (Divan V, 1982). They have received little scholarly attention, owing to their number, the shortage of specialists and restricted access, while the fragility of the monuments makes their documentation urgent.
Its first task is to check, as a priority, the records of volume V in this digital corpus. It then extends them from photographs and drawings, starting with the inscriptions of Martakert. Each inscription receives two aligned layers: a diplomatic transcription that keeps lineation, abbreviations and graphemic detail, and a normalized text that expands abbreviations and regularizes spelling for search and language processing. Metadata follow a schema inspired by EpiDoc (location, monument and material, dating, script, donor or scribe, state of preservation). The project builds on a pilot corpus for the Shirak region (Master's thesis, Inalco, 2022–2023), and its results will be released openly.
- Project leader
- Period
- 2026
- Partners
MESROP
Machine learning for Epigraphic Study and Recognition Of the Armenian Patrimony. MESROP is an ongoing project that explores what computer vision and language models can bring to Armenian epigraphy. It takes up the third aim of the CIAE.
Much of the Armenian epigraphic heritage is at risk. Inscriptions stand in conflict zones, weathering wears them away, and fewer and fewer readers can make sense of abbreviated, damaged texts in Old Armenian. Many now survive only in photographs, drawings or printed editions.
MESROP asks how far machines can support that reading: by locating and transcribing inscriptions on photographs and 3D models, by proposing plausible readings for lost or worn passages, and by tracking the physical degradation of monuments over time. The structured records of the Divan published here serve as training data, and the 3D models produced by TUMO tie each text back to its monument. The project also aims at an open, reproducible approach that can serve other non-Latin scripts; its models and datasets will be released openly.
- Project leader
- Status
- Ongoing
- Partners
-
Publications
Other Calfa heritage projects
Proofreading
How corrections work
- EditOpen a notice and press Edit. Changes stay in this browser.
- SendReview the change and open it as a GitHub issue. Nothing is sent until you press Create on GitHub.
- ReviewedAn automatic check validates the format. Corrections from outside the team are then reviewed by an editor against the scan before they are published under your name. The machine reading is kept.