Henry Wang, Saman Sarraf, Arbi Tamrazian
This paper proposes a novel end-to-end solution to convert tabular data into natural-sounding sports narratives. The solution leveraged large language models (LLM) such as T5 and natural language generation techniques such as back translation and paraphrasing. The solution improves the readability of the narratives by 13% compared to the baseline rule-based template solution. The solution can also be easily scaled to include new statistics and expanded to other sports domains for future business needs.