# Potato Featured at EMNLP 2022

Source: https://www.potatoannotator.com/blog/potato-emnlp-2022

> **Note:** This post describes the original Potato system presented at EMNLP 2022. Potato has since grown to include AI-powered annotation, multimedia support, and 30+ annotation types. See the [current documentation](/docs) for up-to-date information.

Our paper introducing Potato was accepted at EMNLP 2022. It's the result of a few years of work trying to make data annotation less painful for researchers, and we're glad to see it in the program.

## The Paper

**"Potato: The Portable Text Annotation Tool"** lays out the design, architecture, and capabilities of Potato. The argument is simple: if you can specify an annotation interface in a config file, you don't have to write code to get a good one.

### Key Contributions

1. **Configuration-First Design**: Complex annotation interfaces can be specified entirely through YAML, so most projects need no custom code.

2. **Flexible Annotation Schemes**: Potato supports radio buttons, checkboxes, spans, Likert scales, and more, and you can combine them for more involved tasks.

3. **Built-in Quality Control**: The paper covers how we manage annotator quality through training phases, attention checks, and inter-annotator agreement monitoring.

4. **Crowdsourcing Integration**: Potato connects to Prolific and Amazon Mechanical Turk for large-scale studies.

## Citing Potato

If you use Potato in your research, please cite our paper:

```bibtex
@inproceedings{pei2022potato,
  title={Potato: The Portable Text Annotation Tool},
  author={Pei, Jiaxin and Anber, Aparna and Jurgens, David},
  booktitle={Proceedings of the 2022 Conference on Empirical Methods
             in Natural Language Processing: System Demonstrations},
  pages={327--337},
  year={2022}
}
```

## Why We Built Potato

Potato came out of our own frustration as NLP researchers. Every new annotation project seemed to leave us with bad options: learn a heavy platform full of features we didn't need, build an interface from scratch, or settle for a worse annotation experience because the tool couldn't do what we wanted.

So we built something simple to start (minutes, not days), flexible enough for whatever task we threw at it, easy to run without much infrastructure, and built around academic workflows and reproducibility.

## Impact and Adoption

Research groups have used Potato for sentiment and emotion work, named entity recognition and relation extraction, content moderation and toxicity detection, argument mining and stance detection, and clinical and biomedical text mining, among others.

## Looking Forward

The EMNLP paper was the starting point. Since then we've added image and audio annotation, AI-powered annotation assistance, active learning, and better collaboration features.

We keep developing based on what people ask for. If you have a feature request or an idea, open an issue on our [GitHub repository](https://github.com/davidjurgens/potato).

## Acknowledgments

Thanks to the annotators, beta testers, and early adopters who shaped Potato, to the University of Michigan for supporting the research, and to the EMNLP reviewers for their feedback.

---

*Read the full paper on [ACL Anthology](https://aclanthology.org/2022.emnlp-demos.33/) or watch our [demo video](https://www.youtube.com/watch?v=potato-demo).*