Home

Translation Workflow Wiki

Welcome to the Translation Workflow documentation! This project provides tools for translating Tibetan Buddhist texts using LLM-powered semantic analysis.

---

📚 Table of Contents

Tibetan Translator Modules

The main translation pipeline with two versions - commentary-based (v1) and semantic-analysis-based (v2).

Page	Description
Tibetan-Translator-Comparison	Key differences between v1 and v2
Tibetan-Translator-V1	Commentary-based translation workflow
Tibetan-Translator-V2	UCCA + Word-by-Word + Summary based workflow
Translator-Prompts	All prompt templates for translation
Translator-Models	Pydantic models for translation

UCCA Generator Module

The UCCA (Universal Conceptual Cognitive Annotation) Generator creates semantic graph representations of Tibetan Buddhist texts.

Page	Description
UCCA-Generator-Overview	Architecture and module overview
Data-Models	Pydantic models (UCCANode, UCCAGraph, State, etc.)
UCCA-Agent-Workflow	LangGraph iterative generation-evaluation workflow
Prompts	Key prompt templates for generation and evaluation
API-Reference	Function and class documentation

---

🚀 Quick Start

Using Tibetan Translator (v1 - Commentary-based)

from tibetan_translator import optimizer_workflow

state = {
    "source": "བོད་ཀྱི་བསྡུད་མ་འདུག",
    "sanskrit": "Sanskrit text",
    "commentary1": "Commentary 1...",
    "commentary2": "Commentary 2...",
    "commentary3": "Commentary 3...",
    "language": "English",
    # ... other fields
}

result = optimizer_workflow.invoke(state)
print(result["translation"][-1])

Using Tibetan Translator V2 (Semantic-analysis-based)

from tibetan_translator_v2 import optimizer_workflow

state = {
    "source": "བོད་ཀྱི་བསྡུད་མ་འདུག",
    "sanskrit": "Sanskrit text",
    "ucca": "UCCA graph analysis...",
    "word_by_word": "Word-by-word translation...",
    "multilevel_summary": "Multi-level summary...",
    "language": "English",
    # ... other fields
}

result = optimizer_workflow.invoke(state)
print(result["translation"][-1])

Using UCCA Generator

from ucca_generator.ucca_generator import run_ucca_generation_pipeline

result = run_ucca_generation_pipeline("མི་རྟག་པ་སྒོམ་པ་ལ་དགོས་པ་ཅི་ཡོད།")

if result['ucca_graph']:
    print(f"Root ID: {result['ucca_graph'].root_id}")
    print(f"Nodes: {len(result['ucca_graph'].nodes)}")

---

🏗️ Project Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                         Translation Workflow Project                         │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                              │
│  ┌─────────────────────┐     ┌─────────────────────┐                        │
│  │ tibetan_translator  │     │ tibetan_translator  │                        │
│  │        (v1)         │     │        (v2)         │                        │
│  │                     │     │                     │                        │
│  │ Commentary-based    │     │ Semantic-based      │                        │
│  │ • 3 commentaries    │     │ • UCCA graphs       │                        │
│  │ • Traditional       │     │ • Word-by-word      │                        │
│  │                     │     │ • Multi-level sum   │                        │
│  └─────────┬───────────┘     └──────────┬──────────┘                        │
│            │                            │                                    │
│            │         ┌──────────────────┘                                    │
│            │         │                                                       │
│            │         ▼                                                       │
│            │  ┌─────────────────────┐                                       │
│            │  │   ucca_generator    │◄──────────────────────────────────┐   │
│            │  │                     │                                    │   │
│            │  │ • UCCA parsing      │     ┌─────────────────────┐       │   │
│            │  │ • Graph generation  │     │ word_by_word_       │       │   │
│            │  │ • Evaluation loop   │     │    generator        │       │   │
│            │  └─────────────────────┘     └─────────────────────┘       │   │
│            │                                         │                   │   │
│            └─────────────────────────────────────────┴───────────────────┘   │
│                                                                              │
└─────────────────────────────────────────────────────────────────────────────┘

Module Relationships

Module	Purpose	Dependencies
tibetan_translator (v1)	Commentary-based translation	Standalone
tibetan_translator_v2	Semantic-analysis translation	Uses outputs from ucca_generator, word_by_word_generator
ucca_generator	UCCA semantic graph generation	Standalone
word_by_word_generator	Word-by-word translations	Standalone

---

🔧 Dependencies

• langchain-anthropic - Claude LLM integration
• langgraph - Workflow orchestration
• pydantic - Data validation and models

---

📁 Project Structure

translation-workflow/
├── tibetan_translator/          # V1: Commentary-based translation
│   ├── cli.py                   # Command-line interface
│   ├── config.py                # Configuration settings
│   ├── models.py                # Pydantic models
│   ├── prompts.py               # Prompt templates
│   ├── utils.py                 # Utilities and LLM setup
│   ├── workflow.py              # LangGraph workflow
│   └── processors/              # Processing functions
│       ├── commentary.py        # Commentary translation
│       ├── translation.py       # Translation generation
│       ├── evaluation.py        # Quality evaluation
│       ├── formatting.py        # Format checking
│       └── glossary.py          # Glossary extraction
│
├── tibetan_translator_v2/       # V2: Semantic-analysis based
│   └── (same structure as v1)
│
├── ucca_generator/              # UCCA graph generation
│   ├── models.py
│   ├── workflow.py
│   ├── ucca_agent.py
│   └── ucca_generator.py
│
└── word_by_word_generator/      # Word-by-word translation
    └── generator.py

---

🔄 Version Comparison at a Glance

Aspect	tibetan_translator (v1)	tibetan_translator_v2
Input Source	Raw commentaries	Pre-processed semantic data
Processing	Translates commentaries in parallel	Loads UCCA/word-by-word/summary
Philosophy	Commentary-driven	Semantic-analysis-driven
Dependencies	Standalone	Requires ucca_generator output
Best For	Traditional scholarly texts	Detailed linguistic analysis

See Tibetan-Translator-Comparison for detailed differences.