Sports commentary improves the audience’s engagement while delivering a real-time description and analysis of sporting events. However, sometimes the fast-paced nature leads to occasional linguistic errors which includes grammatical inconsistencies, lexical inaccuracies, and discourse level ambiguities. This study will categorize these errors and evaluate various NLP models for detection and processing. The data set of 100 h of transcribed football basketball and tennis commentary was preprocessed and annotated. Several NLP rule-based models such as languageTool and Hunspell, machine learning models such as SpaCy and Stanford NLP, and deep learning models such as AraBERT and GPT-4 Fine-Tuned we’re all assessed based on their precision, recall, F1-score, and real time visibility. GPT 4 achieved the highest F1-score with 91% but has high computational costs which limits its real time applicability. Rule-based models; on the other hand, though faster, they have struggled with domain specific errors. The results suggest that a hybrid NLP approach combining both rule-based and AI-driven techniques can optimize both accuracy and efficiency. This research will contribute to sports broadcasting AI driven analytics and real time speech processing, which in turn will improve automated commentary and speech to text applications. Future work should focus on real time optimization, multimodal learning, and dataset expansion to other spoken discourse domains.