ROUGE: Recall-Oriented Understudy for Gisting Evaluation Metric for Summarization

ROUGE (Recall-Oriented Understudy for Gisting Evaluation) is a widely used metric for evaluating automatic text summarization, based on n-gram overlap with reference summaries. Though effective for extractive methods, its surface-level matching limits its ability to assess paraphrasing, semantic equivalence, and factuality-particularly in the context of modern large language models (LLMs) such as GPT-4, Claude, and PaLM 2. This chapter reviews the ROUGE metric, its mathematical foundations, and its variants, while highlighting strengths and common use cases. We critically examine its limitations for evaluating abstractive generation and survey alternatives such as BERTScore, MoverScore, BLEURT, QAEval, and FactScore. While ROUGE remains useful for legacy benchmarking and coverage estimation, we argue that hybrid evaluation frameworks combining lexical, semantic, and factual dimensions are necessary to better align with human judgment in the era of LLMs.