NUTSHELL: A Dataset for Abstract Generation from Scientific Talks
This addresses the problem of limited data for researchers working on automated abstract generation from scientific talks, though it is incremental as it provides a new dataset rather than a novel method.
The paper tackles the lack of large-scale datasets for Speech-to-Abstract Generation (SAG) by introducing NUTSHELL, a multimodal dataset of ACL conference talks paired with abstracts, and establishes strong baselines to highlight challenges and benefits of training on this dataset.
Scientific communication is receiving increasing attention in natural language processing, especially to help researches access, summarize, and generate content. One emerging application in this area is Speech-to-Abstract Generation (SAG), which aims to automatically generate abstracts from recorded scientific presentations. SAG enables researchers to efficiently engage with conference talks, but progress has been limited by a lack of large-scale datasets. To address this gap, we introduce NUTSHELL, a novel multimodal dataset of *ACL conference talks paired with their corresponding abstracts. We establish strong baselines for SAG and evaluate the quality of generated abstracts using both automatic metrics and human judgments. Our results highlight the challenges of SAG and demonstrate the benefits of training on NUTSHELL. By releasing NUTSHELL under an open license (CC-BY 4.0), we aim to advance research in SAG and foster the development of improved models and evaluation methods.