BabyLM Challenge Returns for Round 4 at EMNLP 2026 with New Multilingual Track
LChoshen · x · 2026-10-22
The BabyLM Challenge returns for its 4th edition as a shared task and workshop at EMNLP 2026, keeping its core goal: sample-efficient pretraining on developmentally plausible, human-scale data budgets.
- Data: detoxified 100M-word Strict and 10M-word Strict-Small datasets, hosted on Hugging Face.
- New track: a dedicated MultiLingual track based on BabyBabelLM, evaluating English, Dutch, and Chinese; Multimodal and Interaction are no longer standalone tracks and are folded into Strict/Strict-Small (image-text pairs and teacher-model feedback allowed).
- Evaluation: the 2026 pipeline and official leaderboard are live; submissions via ACL Rolling Review or OpenReview.
Key venue for teams researching low-data pretraining and language acquisition.
More from Research
- Recursia hits 28-53% zero-shot on LIBERO-Pro Long, far ahead of ASPIRE and π0.5 — jiqizhixin · 2026-10-22
- New Paper Uses 288M Job Postings to Measure Where AI Bottlenecks Raise Work Value — soumitrashukla9 · 2026-10-22
- Python array API standard unifies NumPy-like operations across libraries — GaelVaroquaux · 2026-10-22
- scikit-learn's CPU-bound dependencies have become a decade-old bottleneck — GaelVaroquaux · 2026-10-22
- JHU Paper Shows Aggregate Benchmark Scores Can Overstate Model Competence — DanielKhashabi · 2026-10-22
- Megakernel framework claims trillion-parameter pretraining in a single kernel — basedjensen · 2026-10-22