Agent-created languages are morphologically productive, coining never-seen valid forms
EliasEskin · x · 2026-09-03
Analysis of the emergent languages shows they are compositional and productive: agents correctly produce and parse new forms licensed by the language's morphosyntax that were never observed in their history. The morphosyntax sometimes diverges from English—e.g., contrastive capitalization (likely from per-character cost pressure) and different word order.
Related event: GlossoGen Platform Systematically Studies Emergent Agent Languages(12 posts)→
More from Research
- Apodex 1.1 Released: Training Agents to Deliver Verifiable Complete Work — rohanpaul_ai · 2026-09-04
- 'The craftsmanship of training very deep models is lost in the LLM age' — _arohan_ · 2026-09-04
- DeepLoop debuts to make looped transformers stable and scalable, echoing 48Lx2 frontier model rumors — burny_tech · 2026-09-04
- Bespoke Labs post-trains Inkling with SFT+GRPO, gaining 57pp on repo-level coding and 40% token efficiency — AlexGDimakis · 2026-09-04
- Your Agent's Bottleneck Isn't the Model — It's Your Tests — aparnadhinak · 2026-09-04
- Prof. Tom Yeh Publishes Kimi 3 Seminar Tracing Attention from Vanilla to Kimi Delta Attention — ProfTomYeh · 2026-09-04