Reference Summary

Reference summaries, crucial for evaluating text summarization models, are the focus of ongoing research aiming to improve their quality, generation, and use in evaluation. Current work explores diverse approaches, including contrastive learning for data-efficient cross-lingual summarization, the impact of source modality (speech vs. transcript) on annotation quality, and the development of more robust evaluation metrics using large language models (LLMs) to better align with human judgment. These advancements are vital for enhancing the accuracy and fairness of automated summarization systems across various domains, from news articles to scientific papers and medical records, ultimately improving information access and analysis.