PixT3: Pixel-based Table-To-Text Generation

This repository contains code and datasets for the ACL 2024 paper PixT3: Pixel-based Table-To-Text Generation.

We release PixT3 model checkpoints for the TControl, LControl, and OpenE settings as well as ToTTo, Controlled Logic2Text, and SLC pretraining datasets alongside their corresponding rendered tables for each setting. This repository also contains the code to train and evaluate these models.

Datasets

Download the ready-to-use datasets in Files and versions.

Model checkpoints

Download model checkpoints in Files and versions.

Model names:

• PixT3 (TControl): pixt3_tcontrol
• PixT3 (LControl): pixt3_lcontrol
• PixT3 (OpenE): pixt3_opene
• PixT3 (SLC): pixt3_slc This is the model pretrained with the Structure Learning Curriculum. It mainly serves as initialization checkpoint for PixT3 (LControl) and PixT3 (OpenE).

Reference

If you find this project useful, please cite it using the following format

@inproceedings{alonso-etal-2024-pixt3,
    title = "{P}ix{T}3: Pixel-based Table-To-Text Generation",
    author = "Alonso, I{\~n}igo  and
      Agirre, Eneko  and
      Lapata, Mirella",
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.364",
    pages = "6721--6736",
}