---
license: apache-2.0
dataset_info:
  features:
  - name: stimulus_id
    dtype: int64
  - name: image
    dtype: image
  - name: angle
    dtype: string
  - name: proximity
    dtype: string
  - name: n_candidates
    dtype: int64
  - name: candidates
    dtype: string
  - name: actor
    dtype: string
  - name: label
    dtype: string
  - name: annotator
    dtype: string
  - name: list_id
    dtype: int64
  - name: site
    dtype: string
  - name: head_target
    dtype: string
  - name: path
    dtype: string
  - name: face_x_min
    dtype: float64
  - name: face_y_min
    dtype: float64
  - name: face_x_max
    dtype: float64
  - name: face_y_max
    dtype: float64
  - name: option1_x
    dtype: float64
  - name: option1_y
    dtype: float64
  - name: option2_x
    dtype: float64
  - name: option2_y
    dtype: float64
  - name: option3_x
    dtype: float64
  - name: option3_y
    dtype: float64
  - name: option4_x
    dtype: float64
  - name: option4_y
    dtype: float64
  splits:
  - name: train
    num_bytes: 496067566
    num_examples: 2273
  download_size: 497192807
  dataset_size: 496067566
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

## Vision-Language Models Mistake Head Orientation for Gaze Direction: Nonverbal Conversation Cues

Accepted by ACL 2026
- Project Webpage: https://zoryzhang.github.io/gaze/
- arXiv: https://arxiv.org/abs/2506.05412
- OSF Preregistration: https://osf.io/v8kdf
- Analysis Code: https://github.com/zoryzhang/referential-gaze

`Site0` maps to pilot stimuli; `SiteB` and `SiteH` map to two physical locations for main study stimuli. 
Only the latter controlled head orientation.

If you find our work helpful for your research, please cite as follows :)
```
@inproceedings{vlmGaze2026,
  title={Vision-Language Models Mistake Head Orientation for Gaze Direction: Nonverbal Conversation Cues},
  author={Zhang, Zory and Feng, Pinyuan and Wang, Bingyang and Zhao, Tianwei and Yu, Suyang and Gao, Qingying and Deng, Hokin and Ma, Ziqiao and Li, Yijiang and Luo, Dezhi},
  booktitle={Findings of the Association for Computational Linguistics: ACL 2026},
  publisher={Association for Computational Linguistics},
  year={2026},
  url={https://arxiv.org/abs/2506.05412},
}
```