# Portable Semantics, Private Dialects: Reuse and Negative Transfer in Latent Communication Between Language-Model Cells

arXiv: https://arxiv.org/abs/2609.11365 · HTML: https://mlsanity.com/paper/2609.11365/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 34 of its 42 references that are in the map.
- First posted on arXiv 10 Sep 2026.

## Builds on (most cited first, 30 of 34)

- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Linear Mode Connectivity and the Lottery Ticket Hypothesis](https://mlsanity.com/paper/1912.05671/) (2019)
- [Git Re-Basin: Merging Models modulo Permutation Symmetries](https://mlsanity.com/paper/2209.04836/) (2022)
- [Causal Abstractions of Neural Networks](https://mlsanity.com/paper/2106.02997/) (2021)
- [Fine-Tuning Pretrained Language Models: Weight Initializations, Data Orders, and Early Stopping](https://mlsanity.com/paper/2002.06305/) (2020)
- [The Primacy Bias in Deep Reinforcement Learning](https://mlsanity.com/paper/2205.07802/) (2022)
- [Understanding image representations by measuring their equivariance and equivalence](https://mlsanity.com/paper/1411.5908/) (2014)
- [Characterizing and Avoiding Negative Transfer](https://mlsanity.com/paper/1811.09751/) (2018)
- [Revisiting Model Stitching to Compare Neural Representations](https://mlsanity.com/paper/2106.07682/) (2021)
- ["Other-Play" for Zero-Shot Coordination](https://mlsanity.com/paper/2003.02979/) (2020)
- [Relative representations enable zero-shot latent space communication](https://mlsanity.com/paper/2209.15430/) (2022)
- [Critical Learning Periods in Deep Neural Networks](https://mlsanity.com/paper/1711.08856/) (2017)
- [A study on the plasticity of neural networks](https://mlsanity.com/paper/2106.00042/) (2021)
- [Maintaining Plasticity in Deep Continual Learning](https://mlsanity.com/paper/2306.13812/) (2023)
- [Investigating Transferability in Pretrained Language Models](https://mlsanity.com/paper/2004.14975/) (2020)
- [On the role of population heterogeneity in emergent communication](https://mlsanity.com/paper/2204.12982/) (2022)
- [Quasi-Equivalence Discovery for Zero-Shot Emergent Communication](https://mlsanity.com/paper/2103.08067/) (2021)
- [Natural Fingerprints of Large Language Models](https://mlsanity.com/paper/2504.14871/) (2025)
- [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)
- [A Negative Result on Cross-Model Activation Transfer in a Pythia Multi-Hop Setting](https://mlsanity.com/paper/2606.03280/) (2026)
- [Model Alignment Search](https://mlsanity.com/paper/2501.06164/) (2025)
- [Unsupervised Translation of Emergent Communication](https://mlsanity.com/paper/2502.07552/) (2025)
- [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)
- [Architecture-Dependent Causal Transfer of Activation States Across Large Language Models](https://mlsanity.com/paper/2608.16347/) (2026)
- [Frozen Memory Is Not Enough: Rethinking External Memory as Extraction](https://mlsanity.com/paper/2608.17050/) (2026)
- [Context-dependent communication under environmental constraints](https://mlsanity.com/paper/2305.05821/) (2023)
- [Bridging Functional and Representational Similarity via Usable Information](https://mlsanity.com/paper/2601.21568/) (2026)
- [Hidden APIs in Language Models: Discovering Reusable Causal Interfaces from Forked Futures](https://mlsanity.com/paper/2607.27617/) (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)

In shared-genome language-model societies, restricted evidence visibility favors reusable, value-indexed latent packet interfaces, whereas the sole high-performing globally visible model in the parent study learned an episode-entangled code. This companion study asks whether independently trained societies share one packet language, where strict zero-shot transfer fails, and whether inherited interface state helps or harms later learning. First, a leakage-controlled causal interoperability audit over all 30 ordered pairs of six independently trained restricted societies -- under sealed held-out structure and a preregistered raw/orthogonal/linear/nonlinear alignment ladder -- shows the six semantically similar interfaces do not form one raw language: one same-initialization pair is exactly interoperable in both directions, a second shows asymmetric partial compatibility, and all 26 cross-initialization directions fail every frozen alignment rung. Second, within the tested decomposition and a single sealed source formulation, a source-span control localizes strict zero-shot failure to interpretation and execution of the new operator instructions. Third, in a matched adaptation factorial, the globally trained communication interface acts as a severe negative-transfer prior: reinitializing only the packet reader, writer, and mouth raises final depth-three accuracy from 0.169 to 0.857. Fourth, across two restricted checkpoints and two independently frozen target streams each, inherited interfaces never exceeded fresh-interface controls by the preregistered 0.10 margin. All primary conclusions are bounded to a near-transfer 17-state setting; the negative-transfer factorial concerns one globally visible parent-cohort checkpoint, while an appendix adds a post hoc tagged-global twin case study.