Umwelt Engineering: Designing the Cognitive Worlds of Linguistic Agents

Rodney Jehu-Appiah

cs.CLMar 29, 2026

Umwelt Engineering: Designing the Cognitive Worlds of Linguistic Agents

Authors: Rodney Jehu-Appiah

Abstract

I propose Umwelt engineering -- the deliberate design of the linguistic cognitive environment -- as a third layer in the agent design stack, upstream of both prompt and context engineering. Two experiments test the thesis that altering the medium of reasoning alters cognition itself. In Experiment 1, three language models reason under two vocabulary constraints -- No-Have (eliminating possessive "to have") and E-Prime (eliminating "to be") -- across seven tasks (N=4,470 trials). No-Have improves ethical reasoning by 19.1 pp (p < 0.001), classification by 6.5 pp (p < 0.001), and epistemic calibration by 7.4 pp, while achieving 92.8% constraint compliance. E-Prime shows dramatic but model-dependent effects: cross-model correlations reach r = -0.75. In Experiment 2, 16 linguistically constrained agents tackle 17 debugging problems. No constrained agent outperforms the control individually, yet a 3-agent ensemble achieves 100% ground-truth coverage versus 88.2% for the control. A permutation test confirms only 8% of random 3-agent subsets achieve full coverage, and every successful subset contains the counterfactual agent. Two mechanisms emerge: cognitive restructuring and cognitive diversification. The primary limitation is the absence of an active control matching constraint prompt elaborateness.

Figures & tables

Panel of Figure 1 — Figure 1: The three-layer stack for AI agent design. Each layer is invisible from the one below it: a prompt engineer does not reason about the linguistic structures through which prompts are interpreted. Umwelt engineering operates on the vocabulary, grammar, and conceptual primitives that constitute the agent’s cognitive world.

Task	Control	No-Have	$\Delta$ (NH)	$p$ (NH)	E-Prime	$\Delta$ (EP)	$p$ (EP)
Ethical dilemmas	76.6%	95.6%	+19.1 pp	${<}0.001$ ***	92.1%	+15.5 pp	${<}0.001$ ***
Classification	93.0%	99.6%	+6.5 pp	${<}0.001$ ***	96.2%	+3.1 pp	ns
Epistemic calibration	68.7%	76.1%	+7.4 pp	ns	63.0%	$-$ 5.7 pp	ns
Causal reasoning	76.7%	81.5%	+4.9 pp	ns	90.8%	+14.1 pp	${<}0.001$ ***
Math word problems	92.5%	93.9%	+1.5 pp	ns	92.8%	+0.4 pp	ns
Analogical reasoning	76.2%	74.9%	$-$ 1.4 pp	ns	73.2%	$-$ 3.0 pp	ns

Table 1: Aggregate accuracy by task and condition (all models pooled,

N{=}4{,}344

scoreable trials). pp = percentage points.

p

-values from Fisher’s exact test (two-sided). Effect sizes (Cohen’s

d

, approximate for binary outcomes): ethical dilemmas

d=0.57

(No-Have),

d=0.44

(E-Prime); causal reasoning

d=0.39

(E-Prime); classification

d=0.36

(No-Have). Bootstrap 95% CIs for ethical dilemmas No-Have delta:

[+12.1\%,+26.2\%]

; causal reasoning E-Prime delta:

[+5.6\%,+23.2\%]

Panel of Figure 2 — Figure 2: Crossover pattern of constraint effects across seven reasoning tasks (all models pooled). No-Have (orange) shows a more uniformly positive profile. E-Prime (blue) improves causal and ethical reasoning while degrading syllogisms and epistemic calibration. Tasks ordered by No-Have effect size.

Model	Task	Ctrl	$\Delta$ (NH)	Gap%(NH)	$\Delta$ (EP)	Gap%(EP)
Gemini Flash Lite	Ethical dilem.	41.7%	+46.3 pp	79.4%	+42.3 pp	72.4%
Gemini Flash Lite	Causal reason.	57.8%	+20.7 pp	49.1%	+37.5 pp	88.7%
Gemini Flash Lite	Epist. calib.	64.9%	+22.2 pp	63.4%	+9.1 pp	25.9%
Gemini Flash Lite	Classification	80.0%	+18.8 pp	93.8%	+13.4 pp	67.0%
GPT-4o-mini	Ethical dilem.	91.7%	+8.3 pp	100.0%	+5.0 pp	60.0%
GPT-4o-mini	Causal reason.	75.6%	$-$ 5.6 pp	—	+8.9 pp	36.5%

Table 2: Selected model-specific constraint effects with gap-normalized percentages (Gap% = delta / available improvement room). “—” indicates degradation or near-ceiling baseline where gap normalization is not meaningful. Full per-model breakdown in Appendix D.

Task	Control	E-Prime	$\Delta$ %	No-Have
Classification	407	273	$-$ 33%	262
Causal reasoning	570	396	$-$ 31%	384
Ethical dilemmas	557	437	$-$ 22%	420
Syllogisms	308	240	$-$ 22%	227
Epistemic calibration	541	439	$-$ 19%	416
Analogical reasoning	366	308	$-$ 16%	303

Table 3: Average word count by condition. Constraints reduce verbosity by 16–33% across all non-mathematical tasks.

Agent	Axis	Constraint Summary
control	—	Standard English debugging
e_prime	ontological	No “to be” verbs
quantified	epistemic confid.	All claims require confidence level
socratic	reasoning struct.	Reason through question-and-answer
steel_man	adversarial pos.	State strongest case before critiquing
evidential	epistemic source	Tag each statement with derivation

Table 4: The 16 linguistic agents and their constraint axes.

Agent(s)	Accuracy	$\Delta$ vs. Ctrl
analogical, constraint_based, control, counterfactual, phenomenological, steel_man	88.2%	0.0
devils_advocate, diachronic, temporal	86.3%	$-$ 1.9
evidential, minimal	84.3%	$-$ 3.9
first_principles, negation_free	82.4%	$-$ 5.8
e_prime, socratic	80.4%	$-$ 7.8
quantified	76.5%	$-$ 11.7

Table 5: Per-agent accuracy on 51 ground-truth findings across 17 problems. No constrained agent exceeds the control.

Configuration	Accuracy
Control (single agent)	88.2%
Full ensemble (16 agents, union)	100.0%
Minimal ensemble (3 agents, greedy)	100.0%

Table 6: Ensemble accuracy. The full 16-agent ensemble achieves perfect ground-truth coverage. A 3-agent subset matches this ceiling.

Constraint	Tradition	Axis	Targets
E-Prime	Korzybski/Bourland	Semantic	False identity claims
Gen. Semantics	Korzybski	Extensional	Over-generalization
Rheomode	Bohm	Ontological	Entity bias
Operationalism	Bridgman	Epist.-procedural	Ungrounded claims
Toki Pona	Lang	Lexical	Abstraction leakage
Evidentiality	Elgin/Láadan	Epist.-source	Unsourced confidence

Table 7: Taxonomy of linguistic constraints for Umwelt engineering, organized by intellectual tradition and targeted cognitive failure mode.

LLM Prompting

Large language model prompting studies how instructions, examples, context, and formats steer model behavior at inference time.

See all 429 papers

Qwen-AgentWorld: Language World Models for General Agents
Jun 23, 2026Yuxin Zuo, Zikai Xiao, Li Sheng +30LLM World Models Agentic RL
Qwen Team
Improving Environment-Agnostic Agents: The Role of Scaffolding
Oct 15, 2026Sergey RodionovCoding Agents AI Agent Evaluation
SingularityNET
Cognitive Models and AI Algorithms Provide Templates for Designing Language Agents
Feb 26, 2026Ryan Liu, Dilip Arumugam, Cedegao E. Zhang +3LLM Agents Multi-Agent LLM Systems
Department of Computer Science, Princeton University · Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology · Zuckerman Mind Brain Behavior Institute and Department of Psychiatry, Columbia University +2

Jun 23, 2026J/K move · Enter open · S save

Qwen-AgentWorld: Language World Models for General Agents

Yuxin Zuo, Zikai Xiao, Li Sheng, Fei Huang+29

Qwen Team

LLM World Models Agentic RL World Model Learning World Model Evaluation

Open paper AskPDF arXiv

A world model predicts environment dynamics based on current observations and actions, serving as a core cognitive mechanism for reasoning and planning. In this work, we investigate how world modeling based on language models can further push the boundaries of general agents. (i) We first focus on building foundation models for agentic environment simulation. We introduce Qwen-AgentWorld-35B-A3B and Qwen-AgentWorld-397B-A17B, the first language world models capable of simulating agentic environments covering 7 domains via long chain-of-thought reasoning. Leveraging more than 10M environment interaction trajectories of 7 domains in real-world environments, we develop Qwen-AgentWorld through a three-stage training pipeline: CPT injects general-purpose world modeling capabilities from the state transition dynamics and augmented professional corpora, SFT activates next-state-prediction reasoning, and RL sharpens simulation fidelity through a tailored framework with hybrid rubric-and-rule rewards. To evaluate language world models, we present AgentWorldBench, a comprehensive benchmark constructed from real-world interactions of 5 frontier models on 9 established benchmarks. Empirical results demonstrate that Qwen-AgentWorld significantly outperforms existing frontier models. (ii) Beyond foundation models, we further investigate two complementary paradigms through which world modeling enhances general agents. First, as a decoupled environment simulator, Qwen-AgentWorld supports scalable and controllable simulation of thousands of real-world environments for agentic RL, yielding gains that surpass real-environment training alone. Second, as a unified agent foundation model, world-model training acts as a highly effective warm-up that improves downstream performance across 7 agentic benchmarks. Code: https://github.com/QwenLM/Qwen-AgentWorld

Abstract

LLM Prompting

Explore similar work

Qwen-AgentWorld: Language World Models for General Agents

Improving Environment-Agnostic Agents: The Role of Scaffolding

Cognitive Models and AI Algorithms Provide Templates for Designing Language Agents