Benchmarking Large Language Models for News Summarization
This work addresses the problem of understanding and accurately evaluating LLMs for summarization, providing insights for researchers and practitioners, though it is incremental in benchmarking existing models.
The study investigated the factors behind large language models' success in news summarization, finding that instruction tuning, not model size, is key for zero-shot capability, and that using high-quality human references improves evaluation, with LLM summaries judged as on par with human-written ones.
Large language models (LLMs) have shown promise for automatic summarization but the reasons behind their successes are poorly understood. By conducting a human evaluation on ten LLMs across different pretraining methods, prompts, and model scales, we make two important observations. First, we find instruction tuning, and not model size, is the key to the LLM's zero-shot summarization capability. Second, existing studies have been limited by low-quality references, leading to underestimates of human performance and lower few-shot and finetuning performance. To better evaluate LLMs, we perform human evaluation over high-quality summaries we collect from freelance writers. Despite major stylistic differences such as the amount of paraphrasing, we find that LMM summaries are judged to be on par with human written summaries.