This repository contains a Neural Machine Translation (NMT) model designed to translate English sentences into Hindi. The model uses an encoder-decoder architecture with Long Short-Term Memory (LSTM) layers and is trained on the Hindi-English Truncated Corpus dataset.

Key Features:

• Data preprocessing for text normalization and tokenization.
• Custom batch generator for memory-efficient training.
• Encoder-Decoder architecture with embedding layers.
• Trained using the Keras deep learning framework.

• Overview
• Dataset
• Model Architecture
• Setup
• Training
• Model Checkpoints
• Contributing

The Hindi-English Truncated Corpus dataset is used for training the model. The dataset contains English and Hindi sentence pairs, sourced from TED talks. Only sentences with a maximum length of 20 words are used for training.

1. Convert all text to lowercase.
2. Remove special characters, numbers, and extra spaces.
3. Add START_ and _END tokens to Hindi sentences for better decoding.

The model is built using the Encoder-Decoder architecture with the following components:

• Encoder:
  • Embedding layer for English sentences.
  • LSTM layer to generate context vectors (hidden and cell states).
• Decoder:
  • Embedding layer for Hindi sentences.
  • LSTM layer initialized with encoder states.
  • Dense layer with a softmax activation for predicting the target words.

Ensure you have the following installed:

• Python 3.7 or later
• TensorFlow/Keras
• Numpy
• Pandas
• Matplotlib
• Seaborn

1. Clone the repository:

   git clone https://github.com/mbithesss/Language-Translation-with-Deep-Learning.git
   cd english-to-hindi-translation
   

2. Install the required libraries:

   pip install -r requirements.txt

The model is trained using the following parameters:

• Optimizer: RMSprop
• Loss Function: Categorical Crossentropy
• Batch Size: 128
• Epochs: 100

Checkpoints are saved after every epoch to ensure progress is not lost.

Checkpoints are stored in checkpoints.h5 directory. To load a checkpoint:

model.load_weights('/checkpoint.h5')

Contributions are welcome! Feel free to open issues or submit pull requests.

1. Fork the repository.
2. Create a new branch:

   git checkout -b feature-branch

3. Commit your changes and push them to your fork:

   git push origin feature-branch

4. Open a pull request.