Frontmatter

EMNLP 2021

Introduction

The 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) marked a significant milestone in the evolution of the field. As the volume of research in Natural Language Processing (NLP) continues to grow exponentially, the community has had to adapt its publication models to ensure that high-quality research is disseminated efficiently without being constrained by the physical or temporal limits of a traditional conference schedule. This paper summarizes the organizational framework and outcomes of the Findings of EMNLP 2021, a volume comprising 419 peer-reviewed papers published alongside the main conference proceedings.

The Challenge of the "Publication Explosion"

The primary driver behind the creation of the Findings series is the massive growth of the NLP field. For the 2021 cycle, EMNLP received a staggering 3,600 submissions. Historically, main-track conferences have faced a "space crunch," where many technically sound and well-executed papers were rejected simply because there were not enough presentation slots. This created a "reject-and-resubmit" cycle that delayed the publication of important work and increased the burden on the reviewer pool.

The Findings series was established to address three core objectives:

  1. Timely Publication: Accelerating the speed at which high-quality work reaches the community.
  2. Reviewer Efficiency: Leveraging the existing review infrastructure to identify a broader range of publishable research beyond the top 20%.
  3. Recognition of Solid Work: Providing a "full publication" status for research that meets the quality bar but may lack the broad appeal or "flashiness" typically required for a main-stage plenary session.

Methodology: The Editorial Framework

The Findings of EMNLP 2021 utilized a dual-track review system. Papers were evaluated using the same criteria and the same Program Committee as the main EMNLP conference. Authors of highly-rated papers that were not selected for the main conference were invited to publish in the Findings volume.

Notably, this cycle served as a pilot for the ACL Rolling Review (ARR), integrating five papers from this new, continuous review process. This move signaled a shift toward a journal-style model where the review of research is decoupled from the scheduling of specific events. To ensure these papers remained visible, authors were required to submit pre-recorded video presentations (6 minutes for long papers, 3 minutes for short) to be hosted on the Underline platform.

Key Metrics and Community Impact

The scale of the Findings volume highlights its importance to the ecosystem:

  • Total Submissions: 3,600 papers.
  • Findings Volume: 419 papers (300 long, 119 short).
  • Aggregate Acceptance Rate: When combining the main conference and Findings, the aggregate acceptance rate was 34.9%.

This aggregate rate provides a more realistic view of the "publishable" research being produced in the field, contrasting with the hyper-competitive ~20% rates of traditional main tracks. The community's adoption of this model has been strong; only 26 authors declined the invitation to publish in Findings, suggesting that researchers value the "full publication" status and its impact on academic and professional career trajectories.

Ethical Standards and Quality Control

Despite the increased volume, the Program Chairs maintained a rigorous focus on research ethics. A dedicated Ethics Committee, chaired by Margot Mieskes and Chris Potts, was employed to review papers. Even papers that were "conditionally accepted" were required to meet the field's evolving standards for responsible AI, data usage, and potential societal impact before final inclusion in the volume.

Limitations and Future Directions

While the Findings model successfully prevented 419 papers from being lost to the rejection cycle, challenges remain regarding visibility. Because these papers do not have dedicated presentation slots in the main conference, there is a risk of lower engagement and citation counts.

To mitigate this, EMNLP 2021 integrated approximately 100 Findings papers into specialized workshops for live academic discussion. Future iterations aim to further refine the integration with the ACL Rolling Review and find better ways to ensure that excellent, widely-read work within Findings receives the recognition it deserves. The ultimate goal is to move toward a future where the quality of the research, rather than the constraints of a physical venue, dictates its place in the permanent record of the ACL Anthology.

Summary updated

BibTeX
@inproceedings{2021frontmatter,
  title = {{Frontmatter}},
  booktitle = {Conference on Empirical Methods in Natural Language Processing},
  year = {2021},
}
Figure