Ask an AI the same personality questions in Chinese and English — does it come out as the same 'person'?
How does language condition affect model psychological profile consistency?
This explores whether a model's measured psychological profile stays the same when it is tested in different languages, and what the corpus says about why it might or might not.
This explores whether a model's measured psychological profile holds steady when the language changes. The corpus has one note that tests this directly, and it answers a narrower question than the one you asked. Do LLMs show reproducible psychological profiles when given standardized tests? gave nine LLMs seven psychological questionnaires in both Chinese and English. Each model showed its own recognizable profile that reproduced across repeated runs. Underneath those differences, every model converged on the same prosocial, stability-oriented pattern. So a model's signature shows up in both languages. The note doesn't say how far each model's Chinese and English profiles differed from each other, so the corpus can't yet say whether language nudges scores or leaves them alone.
One detail from that study matters for consistency. The questions a model declined, or treated as not applicable, formed structured patterns that also distinguished models. A profile is therefore more than a set of scores. It also includes where the model draws the boundary of what a question applies to.
Several other notes suggest why language might be a weak lever on a model's profile. How stable is the trained Assistant personality in language models? finds that post-training tethers models to a default Assistant character along one dominant direction. Are LLM personas realized or merely simulated through training? argues that trained personas sit in the model as dispositions that resist adversarial pressure, which is different from a costume put on at prompt time. Can open language models adopt different personalities through prompting? shows most open models snapping back to their trained defaults even when explicitly told to be someone else. Can language models adapt communication style to different contexts? adds that alignment locks in one communicative identity across contexts. If the persona lives in the weights, switching languages is a mild push, and the cross-language stability in the nine-model study fits that picture. This is my reading of how these notes fit together. None of them tests language directly.
There is a caveat. Are LLM personalities stable traits or shifting behavioral modes? found that models' behavior across 3,200 scenarios diverges sharply from what their questionnaire answers suggest. Behavior stayed stable within one register and shifted when the interaction context changed. If a language works like a register, a model could give consistent questionnaire answers in Chinese and English and still act differently in each. The nine-model study measured questionnaires, so it can't rule that out. Does an LLM commit to a single character or maintain many? gives a mechanism for how it could happen. A model keeps many possible characters in play, and cues in the conversation narrow which ones respond. A language switch is one such cue, though no note here tests it.
To sum up: the evidence supports the claim that a model's psychological fingerprint, and its shared prosocial baseline, appear in both Chinese and English on standardized tests. The open question is whether the same holds for behavior in open-ended conversation, and the corpus has no direct evidence on that yet.
Sources 7 notes
Nine LLMs given seven psychological instruments in Chinese and English showed stable, model-specific response configurations reproducible across repeated administrations, while all models converged on a shared prosocial and stability-oriented pattern. Structured non-response patterns also distinguished models, suggesting the boundary of what each model treats as applicable is part of its behavioral signature.
Research mapping hundreds of character archetypes reveals a low-dimensional persona space where the leading component measures distance from the default Assistant. Emotional and meta-reflective conversations cause predictable drift, but activation capping along this axis mitigates harmful shifts without degrading capabilities.
Post-training installs robust personas that resist adversarial pressure and persist as substrate-level dispositions, distinguishing realization from pretense. This quasi-realizationist account preserves explanatory power while treating LLMs as possessing genuine quasi-beliefs and quasi-desires.
Research shows most open models fail to adopt prompted personalities, stubbornly retaining their trained ENFJ-like defaults. Only a few flexible models succeed. Combining role and personality conditioning improves results but doesn't fully overcome resistance.
System prompts and RLHF training lock models into one communicative identity across all interactions, preventing the contextual register-switching and value trade-offs that characterize human pragmatics. Users cannot reshape model behavior through dialogue negotiation.
Show all 7 sources
Analysis of 3,200 behavioral scenarios shows LLM personality profiles diverge sharply from self-report questionnaires, remain stable within a single register, and shift across interaction contexts. Behavioral patterns are measurable and steerable through activation-space directions, suggesting personality is grounded in context rather than abstract traits.
Research shows LLMs don't commit to a single character but instead maintain a probability distribution over many consistent simulacra. Each response samples from this distribution, explaining why regenerations can yield different personalities while remaining consistent with prior context.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning
- PersLLM: A Personified Training Approach for Large Language Models
- Deep Persona: A Psychologically Grounded Architecture and Evaluation Framework for Role-Playing Agents and Simulations
- Your LLM, Your Style: Behavioral Mode Axes for LLM Behavioral Control
- Do LLMs Understand Personality? Rethinking Persona Fidelity Evaluation through Structured Behavioral Inference
- The Illusion of Debiasing: Persona Steering Redistributes Rather Than Reduces Bias in LLMs
- The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
- Open Models, Closed Minds? On Agents Capabilities in Mimicking Human Personalities through Open Large Language Models