Dirk Wulff: "Using embeddings to automate jingle--jangle detection and tackle taxonomic incommensurability"

  • Datum: 17 maj 2024, kl. 10.15–12.00
  • Plats: Blåsenhus, 12:128
  • Typ: Seminarium
  • Föreläsare: Senior Research Scientist Dirk Wulff, MPI for Human Development, Center for Adaptive Rationality, Berlin, Germany
  • Arrangör: Institutionen för Psykologi, Avdelningen för perceptions- och kognitionspsykologi
  • Kontaktperson: Håkan Nilsson

Allmänna seminariet

Title: Using embeddings to automate jingle--jangle detection and tackle taxonomic incommensurability

Abstract: Taxonomic incommensurability denotes the difficulty in comparing scientific theories due to incompatible use of concepts and operationalizations. We show that item, scale, and label embeddings—representations of psychometric items, scales, and construct labels in a vector space obtained from language models—can help tackle this problem in psychology. We analyze different data sets (e.g., International Personality Item Pool) spanning thousands of items and hundreds of scales and constructs, to show that embeddings can be used to predict empirical relations between measures, perform automated detection of jingle–jangle fallacies, and suggest more parsimonious taxonomies that eliminate a number of extant psychological con- structs. All in all, our work suggests that embeddings offer a useful tool to tackle taxonomic incommensurability in the sciences.

Keywords: large language models, embeddings, machine learning, psychology, personality