README.md

---
license: cc-by-sa-4.0
language:
- en
tags:
- MWE
- multiword expression
extra_gated_prompt: >-
  To prevent leaking the dataset to the training data of future models, it should not be distributed on the open web.
extra_gated_fields:
  I agree not to make the CoAM dataset, or any part of it, available on the open web: checkbox
---

CoAM is a dataset for multiword expression (MWE) identification, comprising 1.3K English sentences. The name stands for the Corpus of All-Type Multiword Expressions, where "types" refer to MWE categories such as Noun and Verb.

The construction process of CoAM and experiments using CoAM are described in [our paper](https://arxiv.org/abs/2412.18151), which is accepted to ACL 2025. Please consider citing it when you use the dataset.

```
@misc{ide2024coam,
      title={CoAM: Corpus of All-Type Multiword Expressions},
      author={Yusuke Ide and Joshua Tanner and Adam Nohejl and Jacob Hoffman and Justin Vasselli and Hidetaka Kamigaito and Taro Watanabe},
      year={2024},
      eprint={2412.18151},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.18151},
}
```

# Release notes

## v1.1 (Latest)

- Changed the spaCy model from `en_core_web_lg` to `en_core_web_trf`.
- Annotated sentences sourced from UD with MWE types.
- Manually corrected type tagging errors.
- Changed indexing from 1-based to 0-based.

## v1.0

- Initial version.
- Used for the v1 paper published on arXiv.