Semantic Register Compression in Multi-Agent LLM Cascades
Quick Answer
The study identifies 'semantic register compression' in multi-agent LLM systems, quantifying a 41.7% reduction in label separability during critical evaluations across tasks like political fact-checking and sentiment analysis.
Quick Take
The study identifies 'semantic register compression' in LLM systems, quantifying a 41.7% reduction in label separability during critical evaluations across tasks like political fact-checking and sentiment analysis. This phenomenon poses risks in decision-making processes, emphasizing the need for safety evaluations in high-stakes applications.
Key Points
- Semantic register compression reduces label separability by 41.7% in multi-agent LLM systems.
- Evaluations in political fact-checking, sentiment analysis, and medical triage reveal consistent compression effects.
- Five architectural variants isolate semantic transformation as the main driver of compression.
- A credibility-seeking variant minimizes geometric compression while shifting outputs toward mostly-true.
- Prompt-level regression accounts for 78% of variance in compression across tasks.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Multi-agent LLM systems commonly decompose complex tasks into specialized roles. However, this modularity introduces a representational risk: when intermediate agents transform text across linguistic registers, they can systematically compress the semantic distinctions needed for accurate downstream decisions. We term this phenomenon semantic register compression and characterize it as an observable failure mode in multi-agent cascades. Using a three-agent pipeline (Collector-Evaluator-Decider), we quantify compression via inter-label separation in sentence-transformer embedding space. Across political fact-checking (LIAR), sentiment analysis (SST-5), and medical triage (Triagegeist), critical evaluation consistently reduces label separability by 41.7% at the Evaluator stage, while identity passthrough preserves it nearly fully. Five architectural variants causally isolate oriented semantic transformation as the primary driver. A credibility-seeking variant produces minimal geometric compression yet shifts outputs toward mostly-true, demonstrating that transformation valence controls the direction of distributional collapse independently of compression magnitude. Compression generalizes across the three domains with varying intensity: 41.7% in fact-checking, 27.2% in sentiment, and 20.0% in triage. Prompt-level regression explains 78% of the variance, with operational constraints associated with lower compression. These results demonstrate that semantic register compression is a measurable and generalizable phenomenon in multi-agent LLM systems, with implications for safety evaluation in high-stakes domains.
| Comments: | 15 pages, 2 figures, 4 tables |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.14119 [cs.CL] |
| (or arXiv:2607.14119v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.14119 arXiv-issued DOI via DataCite |
Submission history
From: Manuele Tele Junior Fernandez [view email]
[v1]
Tue, 12 May 2026 01:11:12 UTC (36 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Letting the Data Speak: Extracting Keywords from Crowdsourced Collections with AI
The study evaluates three NLP approaches—Named Entity Recognition, Keyword Extraction, and Topic Modelling—using the Their Finest Hour Online Archive to automate keyword extraction from crowdsourced WWII collections. Findings suggest that while NLP methods show promise, no single approach is sufficient, and ethical considerations in automated keyword extraction are crucial for responsible stewardship.