This study explores the challenges and potential of AI translation in capturing the nuances of inter-Korean sensory language, with a focus on onomatopoeia and mimetic words. While onomatopoeia and mimetic words represent a highly developed linguistic category in Korean, AI translation systems frequently demonstrate a phenomenon where these unique language features are systematically erased or significantly diminished during the translation process. To address these limitations, the study proposes a two-way interactive learning model for AI translation. This approach incorporates deep learning methodologies to improve the frequency and contextual accuracy of sensory expressions in translations. The research also examines North Korean translation strategies for sensory language, which could provide valuable insights for enhancing AI models' handling of Korean onomatopoeia and mimetic words1. The study emphasizes the need for AI translation to evolve beyond word-for-word substitution and advocates for a framework that respects the cultural and linguistic specificity of each language. By integrating both human and machine translation methodologies, this research aims to develop more sophisticated AI translation tools capable of accurately conveying the richness of Korean sensory language.