A Context-Aware Framework for Reducing Offensive Language into Polite Equivalents in English-to-Tamil Neural Machine Translation

In neural machine translation (NMT), it is crucial to effectively handle offensive or slur words; particularly in low-resource languages like Tamil, it is all the more important to translate English words and phrases and present them in a cultural and respectful manner. In the proposed approach, the use of a context-aware NMT framework detects and translates offensive English phrases and expensive English phrases and expressions into a dignified and polite manner. The conversion includes a combination of politeness-based on dictionary mapping and transformer-based paraphrasing. The accuracy of this approach is 95.4 %, and the quality, BLEU score, is 78.2. The presence of a confidence scoring mechanism produces accurate and reliable translations. The proposed approach applies well in moderation requirements for social media. Furthermore, the same approach can be used for AI-based customer support and content filtering. In low-resource languages like Tamil, the management and control of offensive languages and expressions is of utmost importance to achieve acceptable and respectful translations.)