Efficient NLP Policy Document

Our Recommendations

To address concerns related to the growing computational demands of NLP research, we suggest three measures for the ACL community:

  1. Encourage authors and reviewers to ensure better alignment between experiments and hypotheses, requiring justifications for how experiments support the claims made.
  2. Provide incentives for researchers to release trained models of various sizes (e.g., visible branding of papers in conference proceedings).
  3. Introduce dedicated efficiency tracks in *ACL conferences, motivating researchers to focus on achieving efficiency gains.

The policy was officially adopted by the ACL in 2022.


    title = "Efficient NLP Policy Document",
    author = {Yuki Arase and Phil Blunsom and Mona Diab and Jesse Dodge and Iryna Gurevych and Percy Liang and Colin Raffel and Andreas Rücklé and Roy Schwartz and Noah A. Smith and Emma Strubell and Yue Zhang},
    url = "https://www.aclweb.org/portal/sites/default/files/Efficient%20NLP%20policy%20document%20full%20document.pdf"