Open Science · Ancient Greek · NLP
An open-science initiative for freely accessible ancient Greek texts via a complete pipeline from scanned editions to structured TEI XML.
The Corpus Liberatum Linguæ Graecæ (CLLG) is a collaborative open-science project aimed at building a freely accessible, high-quality corpus of ancient Greek texts. It provides a complete pipeline from scanned edition images to structured TEI XML, serving the needs of philology, digital humanities, and natural language processing.
Access to ancient Greek texts is currently dominated by the Thesaurus Linguae Graecae (TLG), a proprietary resource that limits reproducible research. While existing open initiatives such as Perseus, First1KGreek, and the Patristic Text Archive (PTA) have made significant contributions, their coverage remains partial — in particular for late antique (Christian and non-Christian) and Byzantine texts. The CLLG aims to fill this gap with an open, sustainable alternative.
The project operates along three major axes:
The project operates within the framework established by French case law (TGI, Droz v. Classiques Garnier, 27 March 2014, confirmed on appeal 9 June 2017), which holds that the scholarly transcription of ancient texts does not constitute an original creative work protected by copyright, as the transcriber’s choices are governed by scientific method rather than personal expression.
This project is supported by the ANR (Agence Nationale de la Recherche) and carried out within the PIQ project. The project « Corpus Liberatum Linguae Graecae » was supported by the French National Research Agency (ANR) under the France 2030 grant reference number ANR-24-RRII-0002, operated by the Inria Quadrant Program.
The longer-term infrastructure goal is Biblissima Textes, a new component of Biblissima that will serve CLLG texts and other open corpora for ancient and medieval languages via the Distributed Text Services (DTS) API.
All data and tools are released under free licenses.
The Freely Editable and Distributable Corpus for ancient Greek, released on GitLab under a free license. Structured TEI XML, covering prose texts with canonical references.
View on GitLab →OCR and layout analysis models for ancient Greek and Latin texts.
Synthetic training dataset of ~185,000 pages generated from TEI/XML sources across 4,582 works with over 5,000 style combinations, plus real-world benchmarks of scanned critical editions.
Source code, tools, and supporting repositories.
Primary source code and corpus data hosted at Inria.
A ready-to-deploy DTS (Distributed Text Services) API server built on TEI structure, enabling standard programmatic access to corpus texts.
Validation tool for TEI XML files with CiteStructure encoding and DTS catalog registries. Allows automated testing and reference retrieval without human intervention.
JavaScript runtime for Kraken OCR/HTR models, running on Node.js and Electron without requiring Python at inference time. Includes segmentation, recognition, and a full end-to-end pipeline with WebGPU/CUDA support.
Research outputs from the CLLG project.
We investigate how visual language models perform on historical scholarly texts. We create a synthetic dataset of 185,000 pages from TEI/XML sources and a real-world benchmark of scanned critical editions. Qwen3VL-8B achieves 1.0% Character Error Rate on actual scans.
We examine how VLMs perform OCR on ancient Greek texts and find they can generate plausible but visually unsupported text, relying on language priors rather than actual visual input. Fluent output is not necessarily visually grounded.
A multidisciplinary team at ALMAnaCH, Inria Paris — combining Natural Language Processing and Computational Humanities.
