HiTZ/PixT3

# 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](https://aclanthology.org/2024.acl-long.364/).

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",
}
```