The LM Playschool Workshop accepts submissions to two tracks: the Paper-only Track and the LM Playschool Challenge Track. Submissions may be archival or non-archival, as specified below. All submissions should follow the ARR submission guidelines and use the official ACL style template.
Paper-only Track
The Paper-only Track welcomes original research, work in progress, position papers, and recently published papers on language models that learn, adapt, or improve through situated, conversational, collaborative, task-oriented, or multi-turn interaction.
Authors may submit either:
- Short papers of up to 4 pages of content; or
- Long papers of up to 8 pages of content.
Paper-only submissions may be archival or non-archival. Recently published papers may only be submitted as non-archival submissions and should clearly indicate the venue and status of the original publication.
➡️ The page to submit to the Paper-only Track is on OpenReview.
LM Playschool Challenge Track
The Challenge Track is for technical reports describing systems submitted to the LM Playschool Challenge. Challenge papers must be at least 3 pages long and may use up to 5 pages of content, excluding references.
Challenge papers should provide enough detail for reviewers and future researchers to understand, evaluate, and reproduce the submitted system. In particular, they should include:
- A precise description of the methodology, including the learning algorithm, prompting or training procedure, model adaptation strategy, and any post-processing or inference-time mechanisms.
- A full experimental setup, including the base model, model size, dataset sizes, train/dev splits, number of training steps or epochs, batch size, optimisation details, hyperparameters, random seeds, decoding parameters, and compute budget.
- A complete description of the training data, including its source, size, composition, filtering, preprocessing, synthetic data generation procedures (if any), and any use of dialogue-game transcripts, playpen data, external datasets, or model-generated data.
- Training and validation results, including intermediate development results where available, final validation performance, and any relevant ablations or comparisons to the base model.
For the camera-ready version, challenge papers should provide a description of the submitted model and reproducibility information: the exact model checkpoint used for final evaluation, any deviations from the public evaluation setup, and links to code, model cards, training cards, configuration files, and model checkpoints where possible.
➡️ The page to submit your technical report to the Challenge Track is on OpenReview.
General Requirements
All submissions should be self-contained. Appendices and supplementary materials may be included, but reviewers are not required to consult them, so all central claims, methods, and results should appear in the main paper.
Review versions should be anonymised unless the submission is explicitly non-archival and based on already published work. Accepted archival papers will appear in the workshop proceedings. At least one author of each accepted paper must present the work at the workshop.
Submissions must follow the ARR policies on publication ethics, citation and comparison, multiple submission, use of writing assistance, limitations, ethical considerations, and responsible NLP research.