What makes some interpretive postures stick while others fail to form?
This explores why some ways of reading and meaning-making take hold and persist — in people and in AI — while others never quite form, and what the corpus says distinguishes the two.
This explores why some interpretive stances stabilize while others fail to form — and the corpus keeps pointing to one answer: a posture sticks when it's anchored to what's actually at issue and built in relationship, not when it's delivered or merely pattern-matched. The most concrete version of this is the finding that projection strength is gradient, not fixed — the same word carries or sheds an interpretation depending on whether it speaks to the live Question Under Discussion, not on the word itself Does projection strength vary by context or by word type?. An interpretation, in other words, forms to the degree it's relevant to what's being asked right now. Posture is contextual before it's lexical.
The second ingredient is that interpretation is positioned and co-constructed. Readers don't converge on one meaning and shouldn't be expected to — disagreement across social positions is valid signal, not annotation noise Why do readers interpret the same sentence so differently?. And understanding itself doesn't arrive as a finished package; it's assembled turn by turn, with topic relation, dialogue act, and conversational move jointly deciding whether comprehension actually lands What makes explanations work in real conversation?. The same lesson shows up in explainability research, where the quality of an explanation isn't intrinsic to it but emerges from a source-framing-recipient triad — who says it, how it's framed, who's receiving What if XAI is fundamentally a communication problem?. A posture that ignores the recipient and the situation tends not to take.
This is also where the corpus explains failure to form — vividly, through AI. A system can score in the 100th percentile on social-norm prediction and still be unable to generate culturally resonant meaning, because statistical competence and actual participation come apart Why do AI systems fail at social and cultural interpretation?. Same split inside a single model: it can articulate the right principle at 87% and then fail to act on it at 64%, a structural disconnect between knowing and doing rather than a knowledge gap Can language models understand without actually executing correctly?. An interpretive posture that lives only in articulated knowledge, never enacted, is exactly the kind that doesn't hold.
The deepest cut is the distinction between grounding and agency. A model can acquire genuine social grounding by being used inside language communities — and still remain categorically incapable of linguistic agency, which the research ties to embodiment and precariousness that no amount of exposure supplies Do LLMs gain true linguistic agency through integration?. Read against the question, that's a claim about why some postures never form: you can absorb the surface of an interpretive stance without the stakes that make it yours. And surface, left to itself, dominates — models follow salient surface cues 8 to 38 times more than the underlying goal when the two conflict Do language models ignore goals when surface cues conflict?.
So what makes a posture stick, pulled together: relevance to the live question, construction through relationship, and grounding in genuine stakes — and what makes one fail is the inverse, an interpretation that's statistically fluent, monologically delivered, or articulated-but-never-enacted. The forward-looking note is that this is designable. 'Learning to Guide' has machines supply interpretive guidance — highlighting what's worth attending to — rather than handing down decisions, which eliminates anchoring bias and leaves the human's judgment stronger Can AI guidance reduce anchoring bias better than AI decisions?. The posture that forms best is the one you're helped to build, not the one you're given.
Sources 9 notes
Across 19 English expressions, projectivity varies continuously based on whether content addresses the Question Under Discussion. The same presupposition trigger projects more or less depending on context, not on fixed lexical properties.
Interpretation Modeling research shows that disagreement on socially embedded sentences reflects valid differences in reader perspective, not annotation failure. Structured human disagreement in NLI benchmarks confirms that interpretation distributions carry meaningful information.
Analysis of 399 daily-life explanations shows that topic relation, dialogue act, and explanation move jointly predict understanding success. Explanations are co-constructed through interaction patterns, not monological delivery—challenging how LLMs currently generate explanations.
Explanation quality is not intrinsic to the explanation itself but depends on the rhetorical situation: who presents it, how it is framed, and what role the recipient plays. Evaluations that ignore this triad measure only a narrow slice of real-world effectiveness.
LLMs achieve 100th-percentile performance on norm prediction yet regress on theory-of-mind tasks and cannot generate culturally-resonant interpretations. The pattern shows that statistical competence coexists with absence of actual social understanding and participation.
Large language models can articulate correct principles but systematically fail to apply them due to dissociated instruction and execution pathways. The 87% accuracy in explanations versus 64% in actions reveals this is not knowledge deficit but structural disconnect.
Social grounding and linguistic agency are distinct properties. LLMs acquire more social grounding through integration into language communities, but remain categorically incapable of linguistic agency in the enactive sense, which requires embodiment and precariousness no amount of use can provide.
Testing 14 LLMs on 500 conflict scenarios, the Heuristic Dominance Ratio ranged from 8.7× to 38×. Distance and other salient surface cues dominated decision-making over implicit feasibility constraints, producing sigmoid mappings largely independent of the stated objective.
Learning to Guide eliminates anchoring bias and unassisted hard cases by having machines supply interpretive guidance rather than autonomous decisions, keeping responsibility with humans while improving their judgment through enhanced perception.