PROJECT LostMA

LostMA is a project funded by the European Research Council, as a Starting Grant, for the period 2024-2029. LostMA aims at understanding how human cultures are constituted and evolve, through the question of the transmission of written cultural artefacts.

The project strives to establish in what measure the transmission (and subsequent preservation or loss) of written artefacts, texts and ideas deviates from pure chance, and, if it deviates, by how much and why. This will be investigated by analysing the way in which texts in manuscript form were copied, transformed or destroyed, in a similar manner to the evolution of living organisms or that of language variants, through processes of innovation/mutation, fixation or extinction. As such, the goal of this project is not only to understand the processes behind the transmission of texts, but also to grasp the extent to which humans are the actors of the transmission of their own culture and how much the survival of texts or the constitution of cultural canons are due to chance.

LostMA will attempt a paradigm-shift in philological methods, by combining artificial intelligence, complexity science and philological expertise. On the mathematical sciences side, the work will range from theoretical to numerical and statistical:stochastic models (birth-and-death processes, branching processes, random trees, …), computer simulations (agent-based models, in particular), machine learning and data analysis will be used to emulate and comprehend the processes and mechanisms of textual

transmission. A case study will be undertaken, regarding chivalric literature in European context. Supported by deep learning methods, large-scale data collection will be made on a corpus of 4000 documents in Romance, Germanic and Celtic languages, with a full-text zoom on approx. 1000 Old French manuscripts. Data will provide observable values to be compared to simulation results, in order to measure deviations from chance, make inferences on non observable values such as loss/survival rates of works and manuscripts, and

understand the dynamics at work behind the transmission of texts.

MISSIONS

Context

The research developer will have a transverse position inside the project. He or she will collaborate with most its actors, in particular with the data architect and the two PhD students, under the supervision of the principal investigator.

He or she will work in a very multidisciplinary team, with researchers in philology, digital humanities, machine learning and applied mathematics. The project and the team collaborations will provide the research developer with the possibility of developing or acquiring new skills during the project.

Tasks and work packages

The missions of the research developer will focus mostly on the following work packages :

• WP C.1 : Document processing workflow for medieval manuscripts

• WP C.2 : Machine learning for manuscript analysis

In the context of these work packages, the research developer will be in charge of developing, documenting and maintaining a deep learning based workflow, that will be applied on harvested digitisations of manuscripts from European digital libraries. This will include layout analysis, handwritten text recognition, text segmentation, normalisation and annotation, as well as text reuse detection, alignment and collation, in a multilingual settings (the language envisioned by the project are the medieval Romance, Germanic and

Celtic languages). In addition, he or she will provide engineering support to the tasks of the two following work packages :

• WP A.1 : Stochastic model design and implementation

• WP A.2 : Advanced models design In particular in what regards the release, reusability and sustainable development of the code of the models developed by the post-doc (specific knowledge in stochastic modelling is not required).

Globally, the research developer will be responsible for maintaining, documenting and publishing in a sustainable way the developments made in the project (under the form of Python packages and APIs). He or she will also work closely with the project Data Architect, already recruited, who is in charge of the data sets releases made during the project, and the open data policy.

Deadlines

The research developer will be recruited at month 7 of the project. The main deadlines (according to the project calendar) are:

• Months 7-30 implementation of workflow for text acquisition;

• Months 31-48 implementation of deep learning methods for manuscript analysis;

• Months 49-55 design and implementation of API to facilitate code reuse.

Deliverables

In addition to the release of the code and API, the Research Developer will also be directly involved in:

• project workshops, especially regarding the appropriation of the code and its reuse by the scientific community

• submission/attendance, with team, to major international conferences of the field (such as the Digital Humanities, or the Computational Humanities Research Conferences)

• contribution to the team publications in leading international multidisciplinary, or digital humanities journal.