Psychological Constructs in Shared Semantic Space

A new computational framework proposes to map psychological constructs — from emotions to personality traits — as directional vectors inside a shared word-embedding space, allowing direct comparison of measurements that have historically been siloed in separate instruments and research traditions. The framework, detailed in a paper submitted 26 May 2026, uses a technique called Supervised Semantic Differential to estimate construct-specific semantic gradients from text-outcome associations and then projects them onto theoretically motivated reference axes [1][2]. As an initial test case, the authors employ the Valence, Arousal, and Dominance (VAD) affective coordinate system [1][2]. The approach first recovers interpretable VAD directions from English word-level affective norms. It then projects semantic gradients for 27 GoEmotions categories into the space, recovering the expected organization of emotions, especially along the valence and arousal dimensions [1][2]. The same procedure is applied to Big Five personality domains and facets derived from IPIP-NEO-300 item-factor associations; domain-level placements are broadly coherent, while facet-level results are more exploratory because they rely on sparse questionnaire text [1][2]. The work addresses a long-standing fragmentation problem in psychology. Constructs such as “anxiety” or “conscientiousness” are routinely measured with different questionnaires, in different labs, and on different populations, making it difficult to know whether researchers are studying the same underlying phenomenon. The proposed method leans on the distributional hypothesis — the idea that words with similar meanings appear in similar contexts — which also underpins older techniques such as latent semantic analysis [5]. Latent semantic analysis, patented in 1988, uses singular value decomposition to reduce a term-document matrix and compare documents by cosine similarity [5]. The new framework extends this lineage by orienting whole constructs, not just individual words, inside a continuous semantic space. The cognitive grounding for the work rests on the concept of semantic memory, the store of general world knowledge that includes word meanings, concepts, and facts [3]. Semantic memory is distinct from episodic memory — the recollection of personal experiences — and both are forms of explicit, declarative memory [3]. By anchoring psychological constructs in a shared semantic space derived from large text corpora, the framework effectively treats construct definitions as a form of culturally accumulated semantic knowledge that can be geometrically compared. The paper also touches on questions of self-knowledge, which psychology describes as the information an individual uses to answer “Who am I?” and “What am I like?” [4]. Self-knowledge is a component of the self-concept and includes the cognitive self — everything a person knows or thinks they know about themselves, from hair color to beliefs and values [4]. Personality inventories such as the IPIP-NEO-300 attempt to measure facets of that cognitive self, and the new framework’s ability to place those facets in a common space could, over time, help researchers assess whether different instruments are capturing the same self-relevant attributes. The authors caution that semantic placements must be assessed for stability and interpretability before the method can be widely adopted [1][2]. The paper’s submission file is 99 KB [1].