A T5-BASED TRANSFORMER MODEL FOR INTERPRETABLE BIDIRECTIONAL TRANSLATION OF INFORMAL GEN-Z LANGUAGE

Developments in natural language processing (NLP) have highlight the growing need to tackle informal digital
language. However, understanding and translating the rapidly changing Gen-Z slang is difficult. The increase
in use of slang during online communication has created a language gap between younger users and those less
familiar with new expressions, especially millennials and non-social media users. This research focuses on a
two-way translation approach using a fine-tuned Text-to-Text Transfer Transformer (T5) model trained on an
expanded English–Gen-Z parallel dataset. This proposed framework treats slang translation as a single text
generation task. This helps the model capture style variations while keeping the original meaning intact, even

with limitations like limited data and changing language patterns. The findings show that the Transformer-
based method improves translation accuracy and clarity compared to traditional rule-based normalization

methods. The model achieves noticeable gains in translation quality measured by BLEU scores and semantic
similarity metrics. Lexical mapping analysis also shows consistent slang transformation patterns that make
understanding easier. These results suggest that fine-tuned Transformer models offer an impactful way to
process informal language and reduce communication gap between different generations. This approach has
practical applications in social media communication, language accessibility tools, and NLP systems that need

to handle dynamic slang expressions. This will help many people who are not familiar with the new fast-
growing slangs can now understand and adapt with these slangs