Update README.md

README.md

@@ -1,73 +1,101 @@
-# **BLEU Score Comparison for English-to-Japanese Translations**
+# **BLEU and chrF Score Evaluation for English-to-Japanese Translations**
 
 ## **Overview**
-This project demonstrates the calculation and visualization of BLEU scores for English-to-Japanese translations. The BLEU scores evaluate the performance of two different models: an **LSTM-based model** and a **Seq2Seq model**, based on their ability to translate input sentences into Japanese.
+This project evaluates the performance of two translation models:
+1. **LSTM-based Model**
+2. **Seq2Seq Model**
 
-## **Models Evaluated**
-1. **LSTM-based Model**:
-   - A simpler model that predicts translations based on a sequential structure.
-   - Tends to perform moderately well but lacks sophistication in handling complex language patterns.
+The evaluation is based on two standard metrics:
+- **BLEU Score**: Measures n-gram precision with a penalty for shorter translations.
+- **chrF Score**: Measures character-level n-gram precision and recall with a focus on fluency.
 
-2. **Seq2Seq Model**:
-   - A more advanced model designed for sequence-to-sequence tasks.
-   - Expected to perform better due to its ability to learn complex patterns and context.
+The dataset contains translations from **English to Japanese**, where both the reference (ground truth) and predicted translations are evaluated.
 
-## **Key Features**
-- Calculates BLEU scores using the **SacreBLEU** library.
-- Visualizes BLEU scores as a bar chart for easy comparison.
-- Saves the BLEU scores to a CSV file for further analysis.
+---
 
-## **Implementation**
-### **Steps in the Code**:
-1. **Dataset Preparation**:
-   - The dataset contains English sentences and their corresponding Japanese translations (used as references).
-   - Predictions from both LSTM and Seq2Seq models are compared against these references.
+## **Project Structure**
+- **Code**: Contains Python scripts for computing BLEU and chrF scores using the `sacrebleu` library.
+- **Input Data**:
+  - Reference translations (ground truth in Japanese).
+  - Predictions generated by LSTM and Seq2Seq models.
+- **Output**:
+  - BLEU and chrF scores for each model.
+  - Visualizations of the comparison as bar charts.
+  - Results saved to `.csv` files.
 
-2. **BLEU Score Calculation**:
-   - BLEU scores are computed using SacreBLEU to quantify the overlap between the model predictions and the ground truth references.
+---
 
-3. **Visualization**:
-   - BLEU scores are visualized using a bar chart to provide an intuitive comparison of model performance.
+## **Evaluation Steps**
 
-4. **Saving Results**:
-   - The BLEU scores for both models are saved to a CSV file named `bleu_scores_english_to_japanese.csv`.
+### **1. BLEU Score Evaluation**
+The **BLEU** metric evaluates n-gram matches between model predictions and reference translations. Higher scores indicate better translation quality.
+
+**Process**:
+1. Compute BLEU scores using `sacrebleu.corpus_bleu`.
+2. Compare scores for LSTM and Seq2Seq models.
+3. Save results to `bleu_scores.csv`.
+4. Visualize the results with a bar chart.
+
+**Example BLEU Results**:
+| Model   | BLEU Score |
+|---------|------------|
+| LSTM    | 60.45      |
+| Seq2Seq | 85.78      |
+
+---
+
+### **2. chrF Score Evaluation**
+The **chrF** metric evaluates character-level n-gram precision and recall, making it more sensitive to fluency and grammatical correctness.
+
+**Process**:
+1. Compute chrF scores using `sacrebleu.corpus_chrf`.
+2. Compare scores for LSTM and Seq2Seq models.
+3. Save results to `chrf_scores_updated.csv`.
+4. Visualize the results with a bar chart.
+
+**Example chrF Results**:
+| Model   | chrF Score |
+|---------|------------|
+| LSTM    | 72.36      |
+| Seq2Seq | 93.12      |
+
+---
 
 ## **Files**
-- `main.py`: The primary Python script containing the code for BLEU score calculation, visualization, and saving results.
-- `bleu_scores.csv`: Output file containing the BLEU scores for both models.
 
-## **Requirements**
-### **Dependencies**:
+- **`bleu_scores.csv`**: Contains BLEU scores for LSTM and Seq2Seq models.
+- **`chrf_scores_updated.csv`**: Contains chrF scores for LSTM and Seq2Seq models.
+- **Python Script**: Computes BLEU and chrF scores, generates visualizations, and saves results.
+
+---
+
+## **Dependencies**
 - Python 3.x
-- Libraries:
-  - `sacrebleu`
-  - `matplotlib`
-  - `csv`
+- `sacrebleu`: Library for computing BLEU and chrF scores.
+- `matplotlib`: For plotting visualizations.
+- `csv`: To save results as `.csv` files.
 
-To install the required dependencies, run:
+Install dependencies using:
 ```bash
 pip install sacrebleu matplotlib
 ```
 
-## **Usage**
-1. Clone this repository and navigate to the project directory.
-2. Run the script:
-   ```bash
-   python main.py
-   ```
-3. View the BLEU scores printed in the console and the generated bar chart.
-4. Check the `bleu_scores_english.csv` file for the saved results.
-
-## **Results**
-- The BLEU scores for both models are displayed in the console and visualized in the bar chart.
-- Example output:
-  ```
-  BLEU Score Comparison (English-to-Japanese):
-  LSTM Model BLEU Score: 45.32
-  Seq2Seq Model BLEU Score: 70.25
-  BLEU scores have been saved to bleu_scores.csv
-  ```
-
-## **Acknowledgments**
-This project uses the SacreBLEU library for BLEU score calculation and Matplotlib for visualization.
+---
+
+## **How to Run**
+1. Results will be saved as `.csv` files, and bar charts will be displayed.
+
+---
+
+## **Visualization**
+Both BLEU and chrF results are displayed as bar charts for easy comparison:
+- **X-axis**: Models (LSTM, Seq2Seq).
+- **Y-axis**: Scores (BLEU or chrF).
+- Each chart highlights the comparative performance of the models.
+
+---
+
+## **Conclusion**
+- **Seq2Seq Model**: Achieves higher BLEU and chrF scores, demonstrating better translation accuracy and fluency.
+- **LSTM Model**: Performs adequately but lags behind Seq2Seq in both metrics.