This project implements a deep learning model for translating English sentences into Persian using an Encoder-Decoder architecture. The Encoder-Decoder model has gained popularity in machine translation tasks due to its effectiveness in capturing semantic information from input sequences and generating corresponding output sequences.
- Encoder-Decoder Architecture: Utilizes an encoder to process input English sentences and a decoder to generate corresponding Persian translations.
- Sequence-to-Sequence Learning: Implements sequence-to-sequence learning paradigm, enabling the model to handle variable-length input and output sequences.
- Attention Mechanism: Incorporates attention mechanism to help the decoder focus on relevant parts of the input sequence during the translation process, enhancing the model's translation quality.
- Bidirectional LSTM: Employs bidirectional Long Short-Term Memory (LSTM) units in the encoder to capture both past and future context of input sequences, enabling better representation learning.
- Beam Search Decoding: Implements beam search decoding technique to explore multiple translation hypotheses and generate more accurate translations.
- Training and Inference Modes: Supports both training mode for training the model on a dataset of English-Persian sentence pairs, and inference mode for translating English sentences into Persian.
This project is licensed under the MIT License - see the LICENSE file for details.
- This project is inspired by the Encoder-Decoder architecture and attention mechanism proposed in the literature.
- Parts of the code are adapted from open-source repositories and research papers.