# What You Can't See Is What You Learn: Slot-Selective Evidence Masking Favors Compositional Generalization in Shared-Genome Language-Model Societies

arXiv: https://arxiv.org/abs/2608.20054 · HTML: https://mlsanity.com/paper/2608.20054/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 27 of its 35 references that are in the map.
- First posted on arXiv 20 Aug 2026.

## Builds on (most cited first, 27 of 27)

- [Communication-Efficient Learning of Deep Networks from Decentralized Data](https://mlsanity.com/paper/1602.05629/) (2016)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation](https://mlsanity.com/paper/2308.08155/) (2023)
- [Improving Factuality and Reasoning in Language Models through Multiagent Debate](https://mlsanity.com/paper/2305.14325/) (2023)
- [CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society](https://mlsanity.com/paper/2303.17760/) (2023)
- [Causal Abstractions of Neural Networks](https://mlsanity.com/paper/2106.02997/) (2021)
- [Recurrent Independent Mechanisms](https://mlsanity.com/paper/1909.10893/) (2019)
- [Finding Alignments Between Interpretable Causal Variables and Distributed Neural Representations](https://mlsanity.com/paper/2303.02536/) (2023)
- [Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability](https://mlsanity.com/paper/2301.04709/) (2023)
- [Multi-Agent Cooperation and the Emergence of (Natural) Language](https://mlsanity.com/paper/1612.07182/) (2016)
- [Compositionality and Generalization in Emergent Languages](https://mlsanity.com/paper/2004.09124/) (2020)
- [Coordination Among Neural Modules Through a Shared Global Workspace](https://mlsanity.com/paper/2103.01197/) (2021)
- [Capacity, Bandwidth, and Compositionality in Emergent Language Learning](https://mlsanity.com/paper/1910.11424/) (2019)
- [Emergent Communication: Generalization and Overfitting in Lewis Games](https://mlsanity.com/paper/2209.15342/) (2022)
- [How Modular Should Neural Module Networks Be for Systematic Generalization?](https://mlsanity.com/paper/2106.08170/) (2021)
- [Scaling can lead to compositional generalization](https://mlsanity.com/paper/2507.07207/) (2025)
- [Dynamics of specialization in neural modules under resource constraints](https://mlsanity.com/paper/2106.02626/) (2021)
- [Do Latent Channels Actually Communicate? A Causal Audit of Latent Multi-Agent LLM](https://mlsanity.com/paper/2607.26773/) (2026)
- [StateBridge: Training-free Hidden-state Alignment for Latent Communication in LLM Multi-Agent Systems](https://mlsanity.com/paper/2608.13317/) (2026)
- [AC-VLA: Robust Out-of-Distribution Action Execution via Compositional Learning](https://mlsanity.com/paper/2607.15714/) (2026)
- [When Does Latent Communication Pay? A Causal Audit of Relayed KV Caches in Multi-Agent LLMs](https://mlsanity.com/paper/2608.04893/) (2026)
- [XBridge: Entity-Grounded Latent Bridge for Heterogeneous LLM Communication](https://mlsanity.com/paper/2608.11676/) (2026)
- [Input Pathways Shape Few-Shot, Not Zero-Shot, Binding in Tiny Transformers: A Fully-Enumerable Study](https://mlsanity.com/paper/2607.04926/) (2026)
- [Hidden APIs in Language Models: Discovering Reusable Causal Interfaces from Forked Futures](https://mlsanity.com/paper/2607.27617/) (2026)
- [Emergent Compositional Communication for Latent World Properties](https://mlsanity.com/paper/2604.03266/) (2026)
- [Making Transformers Logical](https://mlsanity.com/paper/2608.00418/) (2026)
- [Information Abundance Paradox: Long-Context Training Undermines Parametric Knowledge](https://mlsanity.com/paper/2608.12218/) (2026)

## Built on by (most cited first, 1 of 1)

- [What You Can't See Is Still What You Learn: A Preregistered Sixty-Society Confirmation That Evidence Masking Drives Compositional Generalization](https://mlsanity.com/paper/2609.17637/) (2026)

## Abstract (arXiv)

Multi-module neural systems often expose every module to the full input. We test whether a slot-selective evidence-masking regime -- restricting each module to its own evidence span -- changes which solutions gradient-based training discovers. Four-cell societies share one frozen pretrained language model and one low-rank adapter, communicating only through two model-width continuous vectors in a fixed relay. On a prospectively sealed natural-language function-composition task, we train ten matched restricted/global pairs identical except for the attention mask. Restricted-visibility societies outperform their globally visible twins by at least 20 percentage points at both depths in 9 of 10 pairs, with median paired advantages of 0.7648 and 0.6050. Cutting communication reduces every restricted society to chance, and in a post hoc collision-stratified analysis the depth-three advantage remains 0.558 on programs whose complete affine map never appeared in training. In six post hoc-selected restricted societies, packet interventions on correctly answered held-out episodes are consistent with approximately value-indexed relay states; the sole high-performing global model also requires communication, but its same-value packets are not interchangeable across episodes. Thus restricted visibility is not necessary for composition. Under the tested seeds, streams, task world, and training budget, the masking regime strongly shifted which solutions training discovered: a post hoc mask crossover finds both arms mask-native. Because the restricted mask both blocks foreign evidence and implicitly identifies each cell's assigned slot, attribution to evidence visibility alone awaits a role-marked control. The preregistered battery nevertheless formally fails because restricted-arm median depth-three accuracy is 0.6988, below the 0.70 floor; an earlier qualification cohort yielded 0/10 complete passes.