dyslexia-friendly-text-simplifier

🧠 Dyslexia-Friendly Text Simplifier

This model is a fine-tuned BART model specifically designed to simplify educational content for students with dyslexia. It transforms complex academic text into clear, accessible language while maintaining educational value and accuracy.

🎯 Model Description

• Base Model: facebook/bart-base
• Training Date: 2025-07-18T19:11:01.378252
• Training Samples: 506 samples
• Subjects: Mathematics, Science, English, Computer Science
• Grade Levels: Grades 5, 6, 7, 9, 10, 11, 12
• Optimization: MacBook MPS/Apple Silicon optimized

🚀 Quick Start

from transformers import BartForConditionalGeneration, BartTokenizer
import torch

# Load model and tokenizer
model = BartForConditionalGeneration.from_pretrained("elvisbakunzi/dyslexia-friendly-text-simplifier")
tokenizer = BartTokenizer.from_pretrained("elvisbakunzi/dyslexia-friendly-text-simplifier")

# Auto-detect device (MPS for Apple Silicon, CUDA for GPU, CPU fallback)
device = torch.device("mps") if torch.backends.mps.is_available() else torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
model.to(device)

def simplify_text(text, max_length=256):
    inputs = tokenizer(text, max_length=512, truncation=True, return_tensors='pt').to(device)
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_length=max_length,
            num_beams=3,
            length_penalty=0.8,
            early_stopping=True,
            no_repeat_ngram_size=2
        )
    
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Example usage
complex_text = "The derivative represents the instantaneous rate of change of a function."
simplified = simplify_text(complex_text)
print(simplified)
# Output: "The derivative shows the immediate rate of change of a function."

📊 Performance Metrics

• Average Grade Level Reduction: 2-4 Flesch-Kincaid grade levels
• Word Reduction: 15-25% average reduction in word count
• Processing Speed: ~0.2-0.5 seconds per text (on MPS/GPU)
• Device Support: CPU, CUDA, MPS (Apple Silicon)
• Training Samples: 430 training samples
• Validation Samples: 76 validation samples

🎓 Training Details

Dataset

• Total Samples: 506 educational texts
• Subjects: Mathematics, Science, English, Computer Science
• Grade Distribution: Balanced across grades 5-12
• Content Types: Textbook passages, educational materials, academic explanations

Training Configuration

• Base Model: facebook/bart-base
• Epochs: 3
• Batch Size: 4 (optimized for MacBook)
• Learning Rate: 3e-5
• Optimization: MPS (Metal Performance Shaders) for Apple Silicon
• Evaluation Strategy: Per epoch
• Early Stopping: Enabled

📝 Example Transformations

Mathematics

Input: "To solve linear equations of the form ax + b = c, where a, b, and c are constants, we must isolate the variable x through systematic algebraic manipulation."

Output: "To solve equations like ax + b = c, we need to get x by itself. We do this by doing opposite operations."

Science

Input: "Cellular organelles perform specialized functions essential for maintaining cellular homeostasis and metabolic processes."

Output: "Cell parts do special jobs to keep cells healthy and balanced."

English Literature

Input: "The protagonist's psychological transformation throughout the narrative demonstrates the author's sophisticated understanding of human nature."

Output: "The main character changes during the story. This shows the author understands how people think and feel."

🎯 Dyslexia-Friendly Features

• ✅ Simplified Vocabulary: Complex academic terms → Common everyday words
• ✅ Shorter Sentences: Reduced cognitive load with concise statements
• ✅ Clear Structure: Logical flow and organization
• ✅ Consistent Formatting: Predictable text patterns
• ✅ Educational Accuracy: Maintains scientific and academic correctness

🔧 API Integration

Flask API Example

from flask import Flask, request, jsonify
from transformers import BartForConditionalGeneration, BartTokenizer

app = Flask(__name__)
model = BartForConditionalGeneration.from_pretrained("elvisbakunzi/dyslexia-friendly-text-simplifier")
tokenizer = BartTokenizer.from_pretrained("elvisbakunzi/dyslexia-friendly-text-simplifier")

@app.route('/simplify', methods=['POST'])
def simplify():
    data = request.get_json()
    text = data['text']
    # ... simplification logic
    return jsonify({"simplified": simplified_text})

Production Deployment

# For production use with caching and optimization
import torch
from functools import lru_cache

class DyslexiaFriendlySimplifier:
    def __init__(self):
        self.model = BartForConditionalGeneration.from_pretrained("elvisbakunzi/dyslexia-friendly-text-simplifier")
        self.tokenizer = BartTokenizer.from_pretrained("elvisbakunzi/dyslexia-friendly-text-simplifier")
        self.device = torch.device("mps") if torch.backends.mps.is_available() else torch.device("cpu")
        self.model.to(self.device)
        self.model.eval()
    
    @lru_cache(maxsize=100)
    def simplify(self, text):
        # Implementation with caching for repeated requests
        pass

📈 Use Cases

• Educational Platforms: Integrate into learning management systems
• Reading Assistance Tools: Help students with dyslexia access complex content
• Content Adaptation: Automatically simplify educational materials
• Accessibility Services: Make academic content more inclusive
• Study Aids: Generate simplified versions of textbook content

🔬 Evaluation

The model was evaluated on dyslexia-specific metrics:

• Flesch-Kincaid Grade Level: Consistent 2-4 grade reduction
• Reading Ease: Significant improvement in accessibility scores
• Word Complexity: Reduction in syllable count and word length
• Sentence Structure: Shorter, clearer sentence construction
• Educational Accuracy: Maintained factual correctness across subjects

🤝 Contributing

If you use this model and have suggestions for improvement:

1. Open an issue on the model repository
2. Share your use case and results
3. Contribute additional training data for specific subjects

📄 License

Apache 2.0 - Feel free to use in educational and commercial applications.

🔗 Citation

If you use this model in your research or application, please cite:

@misc{dyslexia-friendly-text-simplifier-2025,
  title={Dyslexia-Friendly Text Simplifier: BART-based Educational Content Simplification},
  author={Elvis Guy Bakunzi},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/elvisbakunzi/dyslexia-friendly-text-simplifier}
}

📞 Contact

For questions, issues, or collaboration opportunities, please reach out through the model repository discussions.

---

This model was developed to make educational content more accessible for students with dyslexia and other reading differences. Every student deserves access to quality education.