DavidkingMazimpaka/English-to-Kinyarwanda-Translation

# English to Kinyarwanda Translation Project ## Table of Contents - Introduction - Dataset Creation and Preprocessing - Model Architecture and Design Choices - Training Process and Hyperparameters - Evaluation Metrics and Results - Insights and Potential Improvements - Conclusion ## Introduction This project aims to build a translation model that translates text from English to Kinyarwanda. The project involves several key steps, including dataset preparation, model building, training, and evaluation. ## Dataset Creation and Preprocessing 1. **Dataset Sources:** - The dataset is created from three separate sources containing English-Kinyarwanda sentence pairs. - Each source was inspected for quality and consistency. 2. **Data Cleaning:** - Removed duplicate entries and irrelevant rows. - Handled missing values by removing rows with incomplete translations. 3. **Standardization:** - Converted all datasets to a common format (TSV) with consistent column names: `english` and `kinyarwanda`. 4. **Combining Datasets:** - The three datasets were merged into a single file using Pandas: ```python import pandas as pd dataset1 = pd.read_csv('dataset1.tsv', sep='\t') dataset2 = pd.read_csv('dataset2.tsv', sep='\t') dataset3 = pd.read_csv('dataset3.tsv', sep='\t') combined_dataset = pd.concat([dataset1, dataset2, dataset3], ignore_index=True) combined_dataset.drop_duplicates(inplace=True) combined_dataset.to_csv('combined_dataset.tsv', sep='\t', index=False) ``` 5. **Splitting the Dataset:** - The combined dataset was split into training (80%), validation (10%), and test sets (10%). ## Model Architecture and Design Choices - **Model Type:** Transformer-based architecture was chosen due to its effectiveness in handling sequence-to-sequence tasks. - **Layers:** The model consists of an encoder-decoder structure with the following specifications: - Number of layers: 6 - Hidden size: 512 - Number of attention heads: 8 - Dropout: 0.1 - **Embedding:** Used a shared embeddin …