Autonomous Research & Recursive Self‑Improvement

The question is no longer speculative. Research agents now synthesize literature, propose hypotheses that have been taken into laboratory testing, and run closed research loops: proposing an idea, implementing it, running an experiment, evaluating the result, and choosing the next attempt. What this establishes is contested: it shows that generated hypotheses can be worth testing, not that novelty has been measured.

Self-improvement also reaches training: models generate practice data, rewards, curricula, and critiques through self-play, self-reward, and verifiable task generation, and agentic data systems improve the datasets they create. This broadens AI-for-AI to data curation, pretraining, midtraining, post-training, and systems optimization. Yet full recursive self-improvement remains unproven; evaluation must expose overfitting, hidden human labor, and reward tampering.

ARRSI fits AACL-IJCNLP because language is how research agents form hypotheses, retrieve multilingual evidence, call tools, critique claims, and explain results. The workshop brings together NLP, agentic systems, AI for Science, AI for AI, evaluation, and governance to establish definitions, benchmarks, reporting standards, and safety boundaries for systems that do not merely produce research-like text, but accumulate auditable evidence and use it to become better researchers.

Authors of accepted papers should submit their camera-ready revisions on OpenReview by November 9, 2026, 23:59 Beijing time (UTC+8).

Sign in using an author account, open your accepted paper, and select Camera Ready Revision. Upload the final non-anonymous PDF, provide the requested source files, and complete the author and presentation information on the form.

We plan to submit the final camera-ready papers for ACL Anthology proceedings. Authors must ensure that each paper is covered by the required conference registration.

ARRSI 2026 follows the ACL Anti-Harassment Policy.

We expect respectful participation in all workshop sessions and related online discussions. Harassment and hostile conduct are not acceptable.

Please also read the AACL-IJCNLP 2026 Code of Conduct.

This policy applies to all workshop participants, both in person and online. We will briefly remind participants of the policy and reporting routes during our opening remarks.