PhiloBERTA: A Transformer-Based Cross-Lingual Analysis of Greek and Latin Lexicon

Overview

PHILOBERTA is a specialized transformer-based framework designed to analyze and quantify semantic relationships between Ancient Greek and Latin philosophical terms. By leveraging modern language models and classical linguistics, our approach bridges the gap between ancient languages and contemporary NLP techniques.

The project demonstrates remarkable effectiveness in:

• Identifying semantic parallels between etymologically related terms across languages
• Maintaining robust performance despite variations in textual context and genre
• Providing quantitative insights into classical philosophical terminology

Key Results

Our latest analysis shows:

• 81.4% similarity scores for etymologically related pairs (±0.3%)
• 78.0% baseline similarity for control pairs (±2.3%)
• Statistically significant differentiation (p = 0.012, t = 3.219)
• Strong performance across major philosophical concepts:
  • ἐπιστήμη-scientia: 0.820
  • δικαιοσύνη-iustitia: 0.814
  • ἀλήθεια-veritas: 0.814

Features

Core Components

• Multilingual transformer model trained on classical texts
• Cross-lingual term alignment system
• Statistical significance testing framework

Analysis Tools

• Automated term extraction and alignment
• Cross-lingual similarity metrics
• Configurable context window analysis

Visualization Tools

• Term relationship networks
• Semantic space projections
• Comparative analysis plots

System Architecture

graph TD
    A[Input Terms] --> B[Text Processor]
    B --> C[Embedding Generator]
    B --> D[Context Extractor]
    
    C --> E[Cross-Lingual Aligner]
    D --> E
    
    E --> F[Similarity Analyzer]
    F --> G[Statistical Validator]
    
    G --> H[Results & Visualizations]
    
    subgraph Data Sources
        I[Greek Texts] --> B
        J[Latin Texts] --> B
    end
    
    subgraph Analysis Pipeline
        B
        C
        D
        E
        F
        G
    end

Loading

Installation

git clone https://github.com/rumiallbert/philoberta.git
cd philoberta
pip install -r requirements.txt

Requirements

• Python 3.8+
• PyTorch 2.0+
• Transformers
• CLTK (Classical Language Toolkit)
• scikit-learn
• pandas
• matplotlib
• seaborn

Usage

from philoberta.model import PHILOBERTA

# Initialize the model
model = PHILOBERTA()  # Uses default bert-base-multilingual-cased

# To analyze terms with context
greek_contexts = ["Context 1 with ψυχή", "Context 2 with ψυχή"]
latin_contexts = ["Context 1 with anima", "Context 2 with anima"]

similarity = model.cross_similarity(
    greek_term="ψυχή",
    latin_term="anima",
    greek_contexts=greek_contexts,
    latin_contexts=latin_contexts
)

# Generate visualizations
model.plot_similarity_distribution(similarity)

Citation

@misc{allbert2025philobertatransformerbasedcrosslingualanalysis,
      title={PhiloBERTA: A Transformer-Based Cross-Lingual Analysis of Greek and Latin Lexicons}, 
      author={Rumi A. Allbert and Makai L. Allbert},
      year={2025},
      eprint={2503.05265},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2503.05265}, 
}

License

This project is licensed under the MIT License - see the LICENSE file for details.