Measuring Individual Differences in Meaning: The Supervised Semantic Differential

The Supervised Semantic Differential (SSD) is a new method for measuring differences in how people understand the same concept using their own language, instead of questionnaire results. SSD builds a “personal concept vector” for each participant by aggregating the word embeddings of the local linguistic contexts (±3-word windows) around a small, targeted lexicon (e.g., “climate,” “vaccination”) in their open-ended writing. These vectors are then reduced with PCA and regressed onto questionnaire scores; the resulting regression weights are back-projected to recover a single semantic gradient in embedding space that captures how meaning shifts as the psychological construct changes. This gradient is interpreted through nearest-neighbor retrieval and clustering, producing human-readable contrasts in meaning. We applied SSD to seven corpora (Politics, Climate, Vaccination, Trust, Scientists, Health, Immigration) collected from 1,736 Polish adults, each of whom wrote short essays and completed a corresponding attitude item (e.g., perceived political influence, readiness to change lifestyle for the climate, trust in vaccines). Six of seven regressions showed reliable associations between language use and questionnaire scores (adjusted R² ≈ .03–.12), with Immigration showing no coherent gradient. SSD revealed theoretically meaningful semantic contrasts (e.g., politics as empowered civic participation vs. politics as corrupt and irrelevant; climate change as urgent emotional distress vs. abstract historical process). Finally, SSD-derived scores showed convergent patterns with demographics and information-seeking behavior that broadly paralleled those of the questionnaires, indicating that SSD recovers psychologically interpretable individual differences. Simulations showed that SSD holds perfect statistical power even in samples lower than 100 datapoints. Taken together, SSD joins qualitative meaning with quantitative measurement thereby opening a new avenue for psychological research in which language itself becomes an interpretable, testable operationalization of psychological constructs rather than just their narrative residue.