Program

Digital language resources for ANE/cuneiform (ORACC, CDLI, eBL) and classical languages (Perseus, PROIEL, Index Thomisticus). Linguistic annotation & NLP tasks. Types of NLP models: supervised / unsupervised. Evaluation process.
Teachers: Shai Gordin, Marco Passarotti

Aims & structure of the track. CNNs and transformers. Image classification, object detection, and segmentation. How to set up a demo with a pretrained model & Gradio.
Teacher: Morris Alper

10:00–11:00 Introductory lecture: What is network analysis? Examples from spatial, social, and lexical networks.
11:00–12:00 History of data visualization in ancient studies.
12:00–13:00 Layout of networks: graph layout and application of network algorithms in humanities and social sciences.
Teacher: Haleli Harel

Demo: supervised model (e.g. UDPipe). Setting up LLMs for the week. Collecting pre-defined linguistic questions from ANE/cuneiform & classical corpora. Student stream assignment: Guided (beginners) vs. Independent (advanced).
Teachers: Shai Gordin, Giulia Rambelli

Getting set up with Google Colab. Demo: image classification on OpenContext archaeological data with a pretrained model. Student stream assignment: Guided (beginners) vs. Independent (advanced).
Teacher: Morris Alper

Creating and editing network data formats, with sample data or students' own data.
Teacher: Haleli Harel

Word and contextual embeddings. Transformer architecture & attention mechanism. Practical implications for low-resource ancient languages.
Teacher: Rachele Sprugnoli

Classical through LLM-based approaches. Bounding box vs. end-to-end pipelines. Classical pipeline: line segmentation, dewarping. Frameworks: Transkribus, Kraken, Tesseract. Handwritten text recognition (HTR) with vision-language models.
Teacher: Morris Alper

10:00–11:00 Methods in network analysis: formats, nodes, edges, labels, and additional parameters.
11:00–12:00 Survey of tools for network analysis: spatial, social, and material data.
12:00–13:00 Survey of common tools and practices for building networks of language data.
Teacher: Haleli Harel

Zero-shot prompting on pre-defined ANE and classical linguistic questions. Supervised model(s) run in parallel where available. Recording results for comparison across days.
Teachers: Shai Gordin, Giulia Rambelli, Federica Iurescia

Fine-tuning OCR for Akkadian and Latin transcriptions. HTR for historical manuscripts using participants' data or provided samples.
Teacher: Morris Alper

Using tools with a given dataset or participants' data, with a pre-built script in a Colab notebook. Loading network data to network software (Gephilite, Vistorian, etc.).
Teacher: Haleli Harel

Knowledge injection: parametric (fine-tuning) vs. non-parametric (RAG). Prompt optimization. Prompt engineering: settings, few-shot, chain-of-thought (CoT).
Teachers: Eleonora Litta, Giovanni Moretti

Structure-from-Motion photogrammetry: principles and workflow. Collecting and processing image datasets of artefacts and sites. Introduction to 3D reconstruction pipelines.
Teacher: Steffen Bauer

10:00–11:00 Layout of networks — graph layout and application of network algorithms in humanities and social sciences: theoretical aspects.
11:00–13:00 Layout of networks — survey of examples.
Teacher: Haleli Harel

Few-shot prompting on ANE and classical linguistic questions. Chain-of-thought prompting. Supervised model(s) where available. Evaluation & comparison with Day 1 results.
Teachers: Shai Gordin, Giovanni Moretti, Roberta Leotta, Eleonora Litta

Hands-on: building a 3D model from participant image sets using SfM tools. Processing, cleaning, and evaluating photogrammetric reconstructions.
Teacher: Steffen Bauer

Editing layout and applying layout algorithms — testing with your dataset.
Teacher: Haleli Harel

Cuneiform NLP in practice: ORACC Parser, Babylonian Engine, ProtoSnap, AI for sign recognition. Prompt engineering: tree-of-thought, LLM workflows, API keys. Discussion: challenges of low-resource & script-diverse ancient languages.
Teachers: Shai Gordin, Marco Passarotti

Visualizing and annotating 3D models of artefacts and archaeological spaces. Tools and workflows for interactive 3D documentation. Discussion of annotation standards and data interoperability.
Teacher: Steffen Bauer

10:00–11:00 Dynamic Networks: how can we model the change of features over time or place — case studies.
11:00–13:00 Dynamic Networks: how can we model the change of features over time or place — methods.
Teacher: Haleli Harel

Tree-of-thought prompting on ANE and classical linguistic questions. Supervised model(s) where available. Evaluation & iterative refinement. Students begin preparing results presentations for Friday.
Teachers: Shai Gordin, Giovanni Moretti, Claudia Corbetta

Hands-on annotation of 3D models using participants' data. Experimenting with tools and annotation workflows; preparing results for Friday presentations.
Teacher: Steffen Bauer

Hackathon focus: How do we model change and motion with networks?
Teacher: Haleli Harel

Ancient Greek language resources & NLP tools (Perseus Digital Library, PROIEL, Ancient Greek Dependency Treebank). Student presentations: results on pre-defined linguistic questions — best methodology, strengths & weaknesses, open questions.
Teachers: Francesco Mambrini + all teachers present

CLIP and vision-language models. Open-vocabulary object detection and segmentation. MLLMs/VLMs for cultural heritage. Text-to-image for grounded tasks (e.g. segmentation). Cross-modal retrieval. Retrieval-augmented generation (RAG). Student presentations.
Teacher: Morris Alper

10:00–11:00 How do we understand a network: distribution, centrality, and trends across disciplines.
11:00–13:00 The future of network analysis: benefits and criticism of networks.
Teacher: Haleli Harel

Continued presentations if needed. Plenary discussion: state of the field, promising directions. Feedback session & certificates of completion. Informal social.
All teachers

Continued student presentations if needed. Plenary discussion: state of the field, promising directions in computer vision for cultural heritage. Feedback session & certificates of completion. Informal social.
Teacher: Morris Alper

Hackathon focus: How do we model change and motion with networks? Participants present their data, what they learned, and a potential modelling of change or motion.
Teacher: Haleli Harel