NUTSHELL: A Dataset for Abstract Generation from Scientific Talks

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.

More informatation can be found in our paper NUTSHELL: A Dataset for Abstract Generation from Scientific Talks.

Dataset Splits

Split	Number of Examples
train	4000
dev	885
test	1431

Dataset Fields

Field	Type	Description
video_path	string	The video URL to the ACL talk.
audio
	- array	A numpy.ndarray representing the audio signal.
	- sampling_rate	The sampling rate of the audio.
sr	int	The sampling rate of the audio.
abstract	string	The abstract of the ACL paper corresponding to the talk.
language	string	The language of the videos and audios: English.
split	string	The data split to which the entry belongs, such as "train," "dev," or "test."
duration	float	The duration of the video/audio content in seconds.
conference	string	The name of the conference associated with the dataset entry.
year	string	The year of the conference.

Citation

@inproceedings{zufle-etal-2025-nutshell,
    title = "{NUTSHELL}: A Dataset for Abstract Generation from Scientific Talks",
    author = {Z{\"u}fle, Maike  and
      Papi, Sara  and
      Savoldi, Beatrice  and
      Gaido, Marco  and
      Bentivogli, Luisa  and
      Niehues, Jan},
    editor = "Salesky, Elizabeth  and
      Federico, Marcello  and
      Anastasopoulos, Antonis",
    booktitle = "Proceedings of the 22nd International Conference on Spoken Language Translation (IWSLT 2025)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria (in-person and online)",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.iwslt-1.2/",
    doi = "10.18653/v1/2025.iwslt-1.2",
    pages = "19--32",
    ISBN = "979-8-89176-272-5",
    abstract = "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."
}