Humanist Discussion Group, Vol. 39, No. 375.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org




        Date: 2026-03-17 11:56:15+00:00
        From: Matthias Hadi Benabdellah <matthias.h.bh@gmail.com>
        Subject: HTR evaluation for Syriac manuscripts (Transkribus vs Kraken)

Dear colleagues,

I've published an evaluation of HTR tools for Syriac Estrangelo
manuscripts, comparing Transkribus and Kraken on an 8th-century manuscript
(Codex Ambrosianus).

Key findings:
- Transkribus HD: 21.71% CER
- Kraken HD (Sophro Mhiro models): 28.10% CER
- High-resolution imaging are critical for both tools
- Vision LLMs (GPT-4, Claude) are unsuitable as of now due to content
hallucinations

Recent update: Manual segmentation correction in eScriptorium can achieve
~6% CER (analysis by Daniel Stökl Ben Ezra, EPHE-PSL).

Full post:
https://etcbc.nl/methodology/htr-for-syriac-manuscripts-an-assessment-as-of-
march-2026/


Matthias Hadi Benabdellah
PhD Candidate (University of Lorraine – Vrije Universiteit Amsterdam)
Lecturer – Semitic linguistics, Syriac, Arabic, Aramaic, Old Testament
(University Paris 3, Catholic Institutes of Paris & Toulouse, University of
Lorraine)


_______________________________________________
Unsubscribe at: http://dhhumanist.org/Restricted
List posts to: humanist@dhhumanist.org
List info and archives at at: http://dhhumanist.org
Listmember interface at: http://dhhumanist.org/Restricted/
Subscribe at: http://dhhumanist.org/membership_form.php