Input Matters: Evaluating Input Structure’s Impact on LLM Summaries of Sports Play-by-Play

Barkavi Sundararajan, Somayajulu Sripada, Ehud Reiter · 2025

Read Input Matters: Evaluating Input Structure’s Impact on LLM Summaries of Sports Play-by-Play by Barkavi Sundararajan, Somayajulu Sripada, Ehud Reiter (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-21034.txt.

Description

Barkavi Sundararajan leads two co-authors on how input structure changes factual errors in language model summaries of NBA play by play data. Sports reporting is accuracy critical, and generated recaps that invent plays erode trust fast.

Over thirty three hundred factual errors were hand annotated across one hundred eighty summaries from two large models, fed the same games as rows, as JSON, or as unstructured text. Structured JSON input cut error rates by roughly two thirds against unstructured input for both models, and row format also helped substantially. Input formatting, often an afterthought, emerges as a major lever on factual reliability.

Catalog record

CREATORBarkavi Sundararajan, Somayajulu Sripada, Ehud Reiter
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-21034

Browse the full library · About the Tower of Babel