Kakaymi10/kanembu_English_translation

A translation model that uses RNNs with LSTM units and attention mechanisms to translate Kanembu text into English. # Project Name: Translation Model with Limited Dataset: English-Kanembu ## Table of Contents 1. Introduction 2. Dataset Creation and Preprocessing 3. Model Architecture and Design Choices 4. Training Process and Hyperparameters 5. Evaluation Metrics and Results 6. Insights and Potential Improvements --- ## Introduction This project focuses on building a simple translation model using a very small dataset of around 10 sentences. The model is tasked with translating short sentences from English to another language. Given the limited amount of data, the focus is on understanding the performance of the model, exploring its limitations, and identifying areas for potential improvements. --- ## Dataset Creation and Preprocessing The dataset is small, consisting of just 10 sentences. These sentences were manually gathered to serve as input-output pairs for the translation task. The dataset contains simple sentences such as: - **English:** "Abakar is looking for banana" - **Translation:** "Abakar banana mâi" ### Preprocessing Steps: 1. **Tokenization:** Each sentence was tokenized into words. 2. **Padding:** Since sentence lengths vary, padding was applied to ensure that all inputs are of equal length. 3. **Vocabulary:** A limited vocabulary was created from the dataset, including both the source and target languages. 4. **Encoding:** Each word in the sentences was mapped to an integer for feeding into the neural network. --- ## Model Architecture and Design Choices Given the simplicity of the task and the small dataset, the model architecture chosen was a basic sequence-to-sequence (Seq2Seq) model. - **Encoder:** The encoder takes the input English sentence and converts it into a hidden state representation using a series of LSTM layers. - **Decoder:** The decoder uses the hidden state from the encoder to generate the translated sentence, also using LSTM layers. - **Embedding Layer:** Both the encoder and decoder have embedding layers to represent words as dense ve …