This repository contains the source code for TwoSA (Talmud Word Similarity Analysis) — a web-based tool that uses machine learning techniques to analyze word similarities in the Babylonian and Jerusalem Talmuds.
TwoSA allows researchers to compare how specific tractates use a given word by computing word embeddings (using BEREL) and clustering them based on similarity scores. It maps phrases into a multidimensional matrix, computes distances between occurrences, and sorts them into clusters using K-Means and Hierarchical Clustering algorithms. The results are visualized, enabling researchers to explore patterns in Talmudic language usage.
- Pull the Docker Image:
docker pull ghcr.io/songkai-z/twosa:latest
- Run the Docker Container
docker run -p 80:80 ghcr.io/songkai-z/twosa:latest
- Access the Tool
Open your browser and visit:
http://localhost:80
If you want to run the Streamlit app directly without Docker:
pip install -r requirements.txt
streamlit run app.py├── Bavli_talmud_texts.csv # Babylonian Talmud texts
├── Yerushalmi.csv # Jerusalem Talmud texts
├── Dockerfile # Docker configuration for building and running the app
├── LICENSE # Licensing information (MIT License)
├── app.py # Main application logic (Streamlit-based)
├── requirements.txt # Python dependencies
The development has been supported by Brown University and the Center for Digital Scholarship at the Brown University Library. The texts have been downloaded from Sefaria and further refined by Michael Sperling.
- Avi Shmidman, Joshua Guedalia, Shaltiel Shmidman, Cheyn Shmuel Shmidman, Eli Handel, Moshe Koppel, "Introducing BEREL: BERT Embeddings for Rabbinic-Encoded Language", Aug 2022 [arXiv:2208.01875]
This project is licensed under the MIT License – see the LICENSE file for details.