Bidimensional Leaderboards: Generate and Evaluate Language Hand in Hand

Introduction

We propose a generalization of leaderboards, bidimensional leaderboards (Billboards), that simultaneously drives progress in language generation tasks and their evaluation. We accept two types of submissions:

• Generator developers submit output text. A Billboard computes all metric scores.
• Metric developers submit an executable program. A Billboard computes correlations with the human judgments, updates the ensemble metric, and measures how much it overrates machine over human generations.

Anonymous submissions are allowed!!

Submit

Submission guides and examples are available here.

Scoring Results

Scoring results for all past public submissions are available here. We have generator-name||metric-name.csv files from the Cartesian product between the generators and metrics: each contains instance-level scores.

Citations

Bidimesional Leaderboards

@inproceedings{kasai2022billboard,
  author    = {Jungo Kasai and
               Keisuke Sakaguchi and
               Ronan Le Bras and
               Lavinia Dunagan and
               Jacob Morrison and
               Alexander R. Fabbri and
               Yejin Choi and
               Noah A. Smith},
  title     = {Bidimensional Leaderboards: Generate and Evaluate Language Hand in
               Hand},
  year      = {2022},
  url       = {https://arxiv.org/abs/2112.04139},
  booktitle={Proc.\ of NAACL},
}

MSCOCO Captioning Evaluations and THumB 1.0 Protocol

@inproceedings{kasai2021thumb,
    title   = {Transparent Human Evaluation for Image Captioning},
    author  = {Jungo Kasai and Keisuke Sakaguchi and Lavinia Dunagan and Jacob Morrison and Ronan Le Bras and Yejin Choi and Noah A. Smith},
    year    = {2022},
    booktitle = {Proc.\ of NAACL},
    url     = {https://arxiv.org/abs/2111.08940}, 
}

CNNDM Summarization Evaluations

@article{fabbri2021summeval,
    title   = {{SummEval}: Re-evaluating Summarization Evaluation},
    author  = {Fabbri, Alexander R and Kry{\'s}ci{\'n}ski, Wojciech and McCann, Bryan and Xiong, Caiming and Socher, Richard and Radev, Dragomir},
    journal = {TACL},
    year    = {2021},
    url     = {https://arxiv.org/abs/2007.12626},
}

WMT20 ZH-EN/EN-DE Machine Translation Evaluations

@misc{freitag2021experts,
      title={Experts, Errors, and Context: A Large-Scale Study of Human Evaluation for Machine Translation}, 
      author={Markus Freitag and George Foster and David Grangier and Viresh Ratnakar and Qijun Tan and Wolfgang Macherey},
      year={2021},
      url={https://arxiv.org/abs/2104.14478},
}