Published
Somewhere, Not Nowhere: First-Person Register, Epistemic Limitation, and Perspective Formation in Language Models
A model trained only on outside descriptions may learn facts without learning what it means to reason from a limited point of view, so this paper proposes first-person experiential narratives as a testable training signal for perspective.
Abstract
Language models learn about experience almost entirely from third-person description: text that reports what happened, what things are, and what people feel, written from outside the experience it describes. This paper asks whether a different training signal — synthetic first-person experiential narratives — can form better functional models of perspective: the situated, indexical, action-oriented point of view from which an agent perceives partially, acts under constraint, errs, and updates. We distinguish description from perspective and argue that third-person corpora, however large, may under-determine perspective-taking, because the point of view is precisely what objective description leaves out. We propose a corpus of first-person narratives structured by perceptual limitation, goal, obstacle, action, error, and revision, varied across agent types — including non-human umwelten — as a distinct pretraining or fine-tuning signal, together with a content-matched control that crosses register (first- vs third-person) with epistemic limitation (a marked, partial vantage vs an omniscient one), so that the grammatical voice and the represented vantage can be told apart rather than confounded. The claim is functional, not phenomenal: the corpus is not proposed to give a model experience or to make it conscious, but to test whether first-person register, represented epistemic limitation, or their interaction teach a measurable skill — perspective-taking, deictic reasoning, self/other distinction, agency attribution, and point-of-view-dependent decision — that an omniscient third-person rendering does not. We locate the proposal among synthetic-corpus, persona, egocentric, and embodied-experience work; differentiate it from first-person datasets built for interpretability or classification; and position it against the interaction-only prescription of the 'era of experience.' The predicted signature is a dissociation: register- or limitation-driven training improves perspective-dependent tasks without improving general factual recall. We give the dataset design, the evaluation, and the conditions under which the proposal is wrong.
In simple terms
The main idea: a view from somewhere
Most AI training text describes the world from the outside, as if the narrator can see everything. The paper calls this a view from nowhere.
Perspective is different. A perspective comes from somewhere: it has an I, a here, a limited view, a goal, a mistake, and a later correction.
Why knowing everything can be a problem
A third-person sentence can say, 'The animal did not see the food behind the leaf.' That gives the model the full answer from outside.
A first-person version says, in effect, 'The smell is strong, but I only see the leaf, so I leave. Later I learn the food was hidden behind it.' That teaches reasoning under limited information.
The experiment: change the lens
The paper proposes a large corpus of first-person experiential stories across many kinds of agents: humans, animals, insects, robots with broken sensors, or other limited viewpoints.
The design separates two factors: first-person voice and epistemic limitation. That way the test can tell whether the useful signal comes from saying 'I', from representing limited knowledge, or from both together.
The expected result
If the idea works, the model should improve on tasks that require perspective-taking, self/other distinction, agency attribution, and decisions under incomplete information.
It should not simply improve at general factual recall. That difference would show that perspective is a specific trainable skill, not just more knowledge.
Keywords
License
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International