Challenge Submission
Submission of models and results via HuggingFace and evaluation pipeline; must adhere to 10-epoch training limit.
NLP/CORE Unranked
BabyLM 2026
EMNLP 2026 Workshop: BabyLM ARR
BabyLM 2026 is a shared task and workshop co-located with EMNLP 2026 that promotes sample-efficient language model pretraining using developmentally plausible, human-scale data budgets. It aims to democratize LM research by focusing on low-resource training scenarios inspired by child language acquisition, with updated datasets and a new MultiLingual track emphasizing English, Dutch, and Chinese.
Key deadlines
Verified September 9, 2026Full paper
August 3, 2026
AoE
Workshop timeline
Submission and decisions
Full paperKey deadline
August 3, 2026 · AoE
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Workshop Paper
Paper submission following CoNLL template (4-8 pages) with hyperparameter form; not required to follow challenge compute limits.
ARR Commitment
Commitment of papers already submitted to ACL Rolling Review with reviews, to be considered for the BabyLM workshop.
Research areas in scope
01
Challenge Tracks
Strict (100M-word detoxified dataset)Strict-Small (10M-word detoxified dataset)MultiLingual (BabyBabelLM, 100M-token mixture focusing on English, Dutch, and Chinese)
Policies worth checking twice
- Challenge submissions must not exceed 10 epochs over training data; workshop papers are exempt.
- Submissions require a HuggingFace model link and results file generated via collect_results.py.
- Workshop and Challenge submissions must use separate OpenReview portals; Challenge participants must not submit to the Workshop portal.
- ARR submissions are only for papers already reviewed by ACL Rolling Review.
- Papers must follow the CoNLL template with 4-8 pages and include a hyperparameter form.
- MultiModal and Interaction tracks are no longer standalone; image-text data and teacher feedback are permitted within Strict/Strict-Small tracks.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026