Skip to content

Leichte Sprache API

Welcome to the Developer Documentation

The Leichte Sprache API is a powerful German text analysis and generation service that helps make content accessible to people with cognitive disabilities by checking compliance with German accessibility writing rules.

What is this project about?

This API provides automated analysis and transformation of German texts according to Leichte Sprache (Simple Language) standards - a regulated form of German designed to be easily understood by people with learning difficulties, cognitive disabilities, or limited language skills.

Key Features

  • 18 specialized rules covering syntax, lexical, stylistic, and technical aspects
  • Real-time violation detection with detailed feedback
  • Annotated text output highlighting problematic areas
  • Comprehensive statistics and improvement suggestions
  • LLM-powered transformation to Simple Language
  • Iterative optimization with escalation strategies
  • Multiple provider support (OpenAI, Google, Ollama, Mistral, Anthropic)
  • Quality scoring with HIX readability index and faithfulness metrics
  • REST API with OpenAPI/Swagger documentation
  • Docker deployment ready for production
  • Comprehensive test suite with 100+ test cases
  • Dynamic rule engine for easy extensibility

Who is it for?

Primary Users

  • Government agencies ensuring accessibility compliance
  • Content creators writing for diverse audiences
  • Educational institutions adapting learning materials
  • Healthcare providers making information accessible
  • Legal services simplifying complex documents

Developers

  • API integrators building accessible content tools
  • ML engineers working on text simplification
  • Accessibility experts implementing compliance solutions
  • Researchers studying text complexity and comprehension

What problems does it solve?

Accessibility Challenge

In Austria, Leichte Sprache is increasingly required for public information under accessibility legislation, including the Austrian Web-Zugänglichkeits-Gesetz (Web Accessibility Act) and UN Convention on the Rights of Persons with Disabilities. Manual compliance checking is time-consuming and requires specialized expertise.

Before: Manual Process

  • ❌ Time-intensive manual review by experts
  • ❌ Inconsistent rule application
  • ❌ Limited scalability for large content volumes
  • ❌ High costs for specialized reviewers

After: Automated Solution

  • ✅ Instant analysis of any German text
  • ✅ Consistent rule application across all content
  • ✅ Scalable processing from single paragraphs to entire documents
  • ✅ Cost-effective automated pre-screening and generation

How does it work?

graph TD
    A[Input Text] --> B[spaCy NLP Processing]
    B --> C[Rule Engine]
    C --> D[18 Specialized Rules]
    D --> E[Violation Detection]
    E --> F[Annotated Output]

    A --> G[LLM Generation Request]
    G --> H[Iterative Optimization]
    H --> I[Violation Feedback Loop]
    I --> J[Simple Language Output]

    style A fill:#e1f5fe
    style F fill:#e8f5e8
    style J fill:#e8f5e8

Technical Architecture

  1. NLP Foundation: Uses German spaCy model (de_core_news_lg) for linguistic analysis
  2. Rule Engine: 18 dynamically-loaded rule modules covering all aspects of Leichte Sprache
  3. ML Components: BERT-based models for complex tasks like abbreviation detection and complexity scoring
  4. LLM Integration: Multi-provider support with intelligent escalation strategies
  5. REST API: FastAPI with comprehensive validation and error handling

Rule Categories

Category Rules Examples
Syntax Sentence length, subordinate clauses, passive voice Max 15 words per sentence
Lexical Foreign words, compound words, complexity scoring "Administration" → "Verwaltung"
Stylistic Negations, idioms, personal pronouns Avoid double negatives
Technical Numbers, abbreviations, punctuation, grammar "z.B." → "zum Beispiel"

Quick Start

Get started with the API in minutes:

# 1. Clone and setup
git clone https://github.com/acolono/leichte-sprache.git
cd leichte-sprache
uv venv && source .venv/bin/activate
uv sync

# 2. Download required models
uv run python -m spacy download de_core_news_lg

# 3. Start the API
python api_main.py

# 4. Test analysis
curl -X POST "http://localhost:8000/analyse" \
  -H "Content-Type: application/json" \
  -d '{"text": "Die komplexe Verwaltung organisiert eine außerordentliche Veranstaltung."}'

Docker Alternative

docker run -p 8000:8000 leichte-sprache-api

Next Steps

  • Getting Started

    Set up your development environment and run your first analysis

    Quick Start Guide

  • API Reference

    Explore all endpoints, parameters, and response formats

    API Documentation

  • Architecture

    Understand the system design and rule engine architecture

    System Overview

  • Contributing

    Learn how to add rules, improve the system, and contribute code

    Developer Guide

Community and Support

  • Issues: Report bugs and feature requests on GitHub Issues
  • Discussions: Join the community on GitHub Discussions
  • Documentation: This comprehensive guide covers all aspects of development and usage

Made with ❤️ for accessible communication