---
language:
- en
- ja
pretty_name: NVFCorpus
size_categories:
- 10B<n<100B
---
# NVFCorpus

ver. 0.9.2

The Nonverbal Function Corpus (NVFCorpus) is a multimodal conversation dataset developed to investigate the multifunctional aspects of nonverbal behaviors of interlocutors in multiparty conversations. It centers on four-person consensus-building dialogues conducted among six mixed-gender groups. Each group comprises Japanese participants, with each individual engaging in four 10-minute sessions conducted in both face-to-face and remote settings. The total length of conversation sessions is 264 minutes. For the face-to-face sessions (125 minutes), the corpus includes manually annotated function labels for over 120 nonexclusive functions. These functions encompass head movements, facial expressions, gaze, and listener vocal backchannels (referred to as Aizuchi). Three external annotators conducted function labeling at the frame level.

#  Content

1) [Video](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/video) 
2) [Audio](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/audio)
3) [Text transcript](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/text_transcript)
4) [Questionnaire](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/questionnaire)
5) [Function label](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/function_label)
6) [Other label](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/other_label)
7) [OpenFace files](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/tree/main/openface_files)

The filename convention: 

G[x]S[y][environenmnet].[type].P[z].[extension] denotes the x-th group’s y-th session’s z-th participant’s [type] of data, where x = 1, …, 6, y = 1, …, 4, z = 1, …, 4. The environenmnet flag indicates that F denotes the face-to-face sessions and Z denotes the remote sessions (using Zoom). 


Further details are available in the following reference.

# Citation

If you use any of the resources provided on this page in any of your publications we ask you to cite the following paper.

Kazuhiro Otsuka, Takumi Nishihira, Yuya Nishimura, and Issa Tamura, "NVFCorpus: A Corpus for Analyzing Multimodal and Multifunctional Nonverbal Behavior in Multiparty Conversations," In Proceedings of ACM 28th ACM International Conference on Multimodal Interaction (ICMI ’26), October 05–09, 2026, DOI: 10.1145/3776574.3831172

Bibtex: <br>
@inproceedings{NVFCorpus,<br>
author = {Otsuka, Kazuhiro and Nishihira, Takumi and Nishimura, Yuya and Tamura, Issa},<br>
title = {NVFCorpus: A Corpus for Analyzing Multimodal and Multifunctional Nonverbal Behavior in Multiparty Conversations},<br>
booktitle = {Proc. 28th Int. Conf. Multimodal Interaction (ICMI'26)},<br>
year = {2026},<br>
doi = {10.1145/3776574.3831172}<br>
}


# Copyright

See [license.txt](https://huggingface.co/datasets/Otsuka-Lab/NVFCorpus/blob/main/License.txt)


# Note

This corpus is under construction. The paper will be presented in ICMI2026 in October 2026. 

# Version history

Ver. 0.9.1: September, 2026. the construction-phase pre-release of the corpus before ICMI2026. 

CAUSION!!: Octorber, 2026. We found that Openface data for G4S2Z's P1, P3, and P4 were wrong files. We will correct them soom.

Ver. 0.9.2: October, 23st, 2026: We corrected the Openface data for G4S2Z's P1, P3, and P4, which were wrong files. We replaced the files with the correct ones. 

# Contact
 
Kazuhiro Otsuka, Ph.D.  (email: otsuka@ieee.org)
Yokohama National University