Grounding Text Embeddings in Stakeholder Associations
Provides a practical method for domain experts to evaluate whether embedding models capture the semantic distinctions that matter to them, addressing a critical validity issue in text analysis.
The paper introduces the Stakeholder Grounding Exercise to assess alignment between text embeddings and human expert associations, finding a substantial reliability gap (19-26 pp in Danish, 16 pp in English) that propagates to downstream clustering performance (Spearman ρ=0.9).
Text embeddings are widely used to analyse large corpora of complex texts. However, it is unclear whether the embeddings capture the same semantic distances as the human experts using them. Ensuring alignment between embedding representations and human intentions is essential for valid analyses. We present the Stakeholder Grounding Exercise, a method for making expert associations explicit and grounding embedding model results in human understanding. In our primary case study on Danish policy issues, we find that neural text embeddings are substantially less reliable than human experts (19-26 pp gap), and that this misalignment propagates to downstream clustering performance (Spearman $ρ=0.9$ between exercise ranking and cluster quality). A secondary study on US Federal AI use cases replicates the gap (16pp) in English, using a digital protocol and a different community of experts -- demonstrating that the gap is not an artefact of a single instrument or domain. The Stakeholder Grounding Exercise offers a practical method for assessing whether embedding models capture the semantic distinctions that matter most to domain experts.