README.md

---
library_name: transformers
tags:
  - english-to-braille
  - braille translation
  - accessibility
  - educational content
  - text summarization
license: mit
datasets:
  - ccdv/arxiv-summarization
  - xsum
  - cnn_dailymail
language:
  - en
base_model:
  - facebook/bart-large-cnn
---

## Model Details

### Model Description

<!-- Provide a longer summary of what this model is. -->

The English-to-Braille Translator combines advanced natural language processing with a custom conversion algorithm. In the first stage, the model uses a pre-trained and fine-tuned version of the Facebook BART model (facebook/bart-large-cnn) to create abstractive summaries of educational materials drawn from datasets such as ccdv/arxiv-summarization, xsum, and cnn_dailymail.

In the second stage, the generated summary is converted into Braille. Instead of a neural translation approach, the system employs a handcrafted dictionary-based mapping mechanism. This mapping converts each English character—and, where applicable, certain contractions and abbreviations—into their corresponding Braille Unicode representations. Multiple versions are supported (including a baseline, an advanced context-aware variant, and our custom implementation) and are evaluated using metrics such as character accuracy, word-level precision/recall, and edit distance.


- **Developed by:** Srimeenakshi K S
- **Model type:** English-to-Braille Translation and Summarization
- **Language(s) (NLP):** English
- **License:** MIT License
- **Finetuned from model:** facebook/bart-large-cnn


## Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

### Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
This model can be used as a standalone tool for converting English texts into Braille. Simply input your educational document, and the model will (1) generate a concise summary and (2) translate the summary into Braille characters using the mapping dictionary.


### Downstream Use

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
The model is ideal for incorporation in accessibility pipelines – for instance, as a backend service for e-learning platforms, libraries, or digital accessibility applications that aim to provide visually impaired users with Braille-compatible summaries of long educational documents.


### Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
This model is specifically designed for educational content and might not perform well on texts that require nuanced or domain-specific translations beyond the scope of its dictionary. Its dictionary-based conversion approach does not account for context beyond a basic character and common contraction mapping; therefore, it should not be deployed for highly technical documents without additional validation.


## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->
While the summarization component is built on a well-established BART model, the Braille conversion relies on a fixed dictionary. This mapping approach may struggle with ambiguous punctuation, special formatting, or non-standard abbreviations. Users should be aware that:
 - The summarization output might occasionally omit vital context.
 - The dictionary mapping, while effective for most cases, is inherently limited and could misrepresent characters where multiple mappings exist.
 - Evaluation metrics indicate strong performance overall, but edge cases (especially with highly technical jargon) may require manual review.


### Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Deploy the model in contexts where the educational content adheres to a standard vocabulary and formatting. For critical applications, supplement automated outputs with human verification, particularly where accuracy in Braille representation is imperative.


## How to Get Started with the Model

Use the code below to get started with the model.

```
from transformers import pipeline

# Step 1: Summarize the English text using the fine-tuned BART model
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
english_summary = summarizer("Your long educational text goes here.", max_length=200, truncation=True)[0]['summary_text']

# Step 2: Convert the summary to Braille using the custom dictionary mapping
from your_custom_braille_module import braille_to_text_map, braille_to_text  # Ensure you import your conversion functions

# (For an English-to-Braille conversion, you might invert the mapping)
def text_to_braille(text, mapping):
    # Invert the mapping (note: for a complete solution, handle duplicate values and contractions appropriately)
    inverted = {v: k for k, v in mapping.items()}
    braille = ''.join(inverted.get(char, char) for char in text.lower())
    return braille

mapping = braille_to_text_map()
braille_summary = text_to_braille(english_summary, mapping)
print("English Summary:", english_summary)
print("Braille Summary:", braille_summary)
```

## Training Details

### Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

The summarization component of this model was fine-tuned on a mix of educational and general summarization datasets:

 - [ccdv/arxiv-summarization](https://huggingface.co/datasets/ccdv/arxiv-summarization)
 - [xsum](https://huggingface.co/datasets/EdinburghNLP/xsum)
 - [cnn_dailymail](https://huggingface.co/datasets/abisee/cnn_dailymail)

The Braille translation itself does not involve training but instead relies on a manually curated mapping between English characters (and common contractions) and Braille Unicode characters.


## Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->


#### Preprocessing

 - **Text Summarization:** Standard preprocessing steps such as tokenization, truncation, and padding were employed to prepare texts for BART.
 - **Braille Conversion:** The mapping was manually constructed using expert knowledge of Braille representations, with additional additions for common contractions.
 
#### Training Hyperparameters (for the summarization model)

- **Epochs:** 3

- **Batch size:** 4

- **Learning rate:** 5e-5

- **Precision:** fp16 mixed precision

## Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

### Testing Data, Factors & Metrics

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->

The summarization quality was evaluated on validation splits from xsum and cnn_dailymail, while the Braille conversion was compared against baseline conversions on a set of educational excerpts.


#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

 - Character-level accuracy

 - Word-level precision, recall, and F1 scores

 - Edit distance

 - Overall text similarity

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

Evaluation of the Braille translation is based on:

 - Character Accuracy

 - Word Precision, Recall, and F1 Score

 - Edit Distance (Levenshtein distance)

 - Text Similarity

   
### Results

In evaluations:

 - Our custom Braille model showed high character accuracy (above 90%) on average.

 - Word-level F1 scores and edit distances indicate that the advanced mapping variant performs comparably to context-aware corrections (improving simulated accuracy by approximately 10% in controlled tests).


#### Summary

The combined pipeline delivers robust summarization and effective Braille translation for standard educational texts. However, performance may vary on content with unconventional formatting or specialized vocabulary.



## Model Examination

<!-- Relevant interpretability work for the model goes here -->

The evaluation includes detailed comparisons of three Braille conversion methods:

 - **Our Custom Braille Model:** Uses full mapping with contractions.

 - **Baseline Braille Translator:** Uses a simplified mapping.

 - **Advanced Braille Translator:** Incorporates context-aware simulation for slight correction improvements.

Further interpretability work can analyze how minor changes in the mapping affect overall accuracy and readability, especially for borderline cases in character conversion.


## Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

- **Hardware Type:** NVIDIA GeForce RTX 4050
- **Hours used:** 3 hours for fine-tuning

## Technical Specifications

### Model Architecture and Objective

 - **Architecture:** Sequence-to-sequence transformer (BART) for summarization, followed by a custom rule-based English-to-Braille mapping.

 - **Objective:** Generate accessible Braille summaries from long-form educational texts.


### Compute Infrastructure

#### Hardware

- **GPU:** NVIDIA GeForce RTX 4050
- **RAM:** 16GB

#### Software

- **Framework:** PyTorch
- **Library Version**: Hugging Face Transformers version 4.44.2
- **Additional Libraries:** nltk, datasets, rouge, wandb, and scikit-learn for evaluation

## Citation

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**

@model{srimeenakshiks2025eng2braille,
  title={English-to-Braille Translator for Educational Content},
  author={Srimeenakshi K S},
  year={2025},
  publisher={Hugging Face}
}



**APA:**

Srimeenakshi K S. (2025). English-to-Braille Translator for Educational Content. Hugging Face.
## Glossary

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

 - **Abstractive Summarization:** The process of generating a concise summary that captures the essence of an input document using natural language generation techniques.

 - **Braille Translation:** The conversion of written text into Braille, typically represented using Unicode Braille patterns.

 - **Levenshtein Distance:** A metric for measuring the difference between two strings by counting the number of single-character edits required to change one string into the other.

## Model Card Authors

- **Author:** Srimeenakshi K S