MariaYasmeen/NLP-Low-Resource-Paraphrase-detection

# Monolingual Paraphrase Detection - Low Resource Sindhi Lang at Sentence Level This project focuses on **sentence-level paraphrase detection for the Sindhi language**. The system classifies whether two Sindhi sentences convey the same meaning (paraphrase) or not (non-paraphrase). ## Overview Paraphrase detection is important for many NLP applications such as question answering, plagiarism detection, information retrieval, and text similarity systems. Most previous work focuses on English, while Sindhi remains a low-resource language. This project helps fill that gap by creating a dataset and building detection models. ## Objectives * Build a Sindhi paraphrase corpus * Train machine learning and deep learning models * Compare different feature representations * Develop a real-time prediction interface ## Dataset * Collected Sindhi sentence pairs from online news sources * Manually annotated into paraphrase and non-paraphrase * Preprocessed and cleaned the text data * Split into training and testing sets ## Methods Used The following techniques were implemented and compared: * N-gram features * FastText embeddings * Sentence Transformers * Feature fusion approach ## Model Training * Implemented in Python * Used Scikit-learn and HuggingFace Transformers * Experiments conducted on Google Colab * Evaluated using standard classification metrics ## Results * Feature fusion provided the best performance * Sentence Transformers showed strong semantic understanding * The system achieved reliable paraphrase classification for Sindhi text *(You can add your exact accuracy/F1 score here.)* ## Web Interface A simple web-based interface was developed where users can: * Input two Sindhi sentences * Get real-time paraphrase prediction ## How to Run 1. Clone the repository ``` git clone cd ``` 2. Install dependencies ``` pip install -r requirements.txt ``` 3. Run the model or notebook ## Tools & Technologies * Python * Google Colab * Scikit-learn * HuggingFa …