Publication

Abstract : This article introduces a new method for transforming incomplete and distorted children's stories into cohesive and contextually accurate narratives utilizing a Large Language Model (LLM). The suggested system tackles the difficulties of reconstructing fragmented material while maintaining narrative continuity, emotional tone, and age-appropriate language. The LLM, refined on a selected collection of children's literature, handles partially accessible or noise-affected inputs via a multi-step procedure that includes semantic context restoration, emotional coherence modification, and lexical simplification. In contrast to standard text-repair methods, the suggested approach focuses on maintaining narrative flow, cultural significance, and the retention of moral or educational purpose. The experimental assessment utilizing BLEU, ROUGE - 1,2,L, and METEOR metrics demonstrates steady enhancements compared to baseline models, with a +7.8% increase in BLEU, +9.2% in ROUGE - L, and +8.5% in METEOR. These enhancements indicate improved semantic consistency, thoroughness, and structural soundness of the revitalized narratives. The system showed fewer grammatical errors and less semantic drift compared to typical LLM outputs. Findings indicate that the method can consistently restore absent or damaged story parts while preserving the thematic and emotional core of the original. The approach has real-world uses in digital archiving, recovering educational materials, and assistive technologies aimed at preserving and revitalizing children's literature in environments with limited resources and noisy data.