- Abdul Rafay (rafaya@bu.edu)
- Mehmet Bora Sarioglu (sarioglu@bu.edu)
Korean online communities have developed a unique form of phonetic text coding for discreet communication. This project aims to develop an automated decoder for this coded Korean text. We will create a dataset of coded and decoded text, train a model to recognize transformation patterns, and deploy a functional tool for real-time decoding. This tool will bridge the communication gap, improving accessibility and understanding of online discourse.
Koreans online often use a phonetic text coding system to communicate candidly, sometimes obscuring messages. By subtly altering word phonetics, users convey meaning while maintaining a degree of privacy. This practice is common in online reviews, social media, and forums. While effective for native speakers, this coded language presents challenges for automated systems and non-Koreans. This project develops a tool to systematically decode such text into standard Korean.
Our primary goal is a reliable and efficient system for decoding phonetically altered Korean text with high accuracy. Key objectives include:
- Dataset Creation: Construct a comprehensive dataset of coded Korean text and their standard Korean translations. We will explore online forums, reviews, and social media, considering ethical implications and anonymization. We may also explore generating synthetic data using encoding tools and leveraging existing Korean language corpora like KoNLPy and CLIcK.
- Model Development: Train a robust NLP model to recognize and reverse phonetic transformations. We will investigate architectures like sequence-to-sequence models or transformer networks, optimizing for decoding accuracy and performance.
- Tool Development and Deployment: Develop an intuitive tool for real-time decoding, deployed as a web-based application for broader accessibility.
- Evaluation and Refinement: Rigorous evaluation will be conducted using the created dataset and user feedback. We will compare model output with original text to measure accuracy and iteratively refine the model and tool.
The decoding tool will have a simple interface. Users input coded Korean text, and the decoded standard Korean is displayed. Planned features include:
- Real-time decoding: Immediate feedback as the user types or pastes text.
- Example usage: Showcasing examples of coded and decoded text.
- Potential for integration: Exploring integration with other platforms.
This project has significant potential impact:
- Non-native Speakers: Facilitating understanding of online discourse.
- Researchers: Providing a valuable resource for studying Korean text phonetic transformations.
- NLP Developers: Offering a solution to a niche problem allowing for further applications like sentiment analysis and machine translation.
- Businesses: Enabling better understanding of customer feedback in the case of encoded Korean text.
This project addresses a unique linguistic challenge. The resulting tool may have an impact on Korean online communication, linguistic research, or other NLP applications such as sentiment analysis. We are confident that our approach will lead to a successful project.