paper

How reliable are speech and language markers?

Before a speech marker can be clinically useful, it has to be reliable. A new paper in Computer Speech & Language, “What is the retest reliability of computationally extractable speech and language markers?”, shows why this matters:

  • acoustic-prosodic markers were relatively stable across repeated assessments,
  • many individual semantic and syntactic features were much less so,
  • semantic features looked more promising at the level of latent clusters than as single markers.

For language-based markers in psychiatry, clinical usefulness is not just about signal strength or prediction performance, but also about test-retest reliability.

The work comes from the LAMBDA study, an interdisciplinary collaboration across psychiatry, linguistics and cognitive neuroscience at the University of Cologne. Congratulations to Derya Çokal, Martin Villalba, Annkathrin Böke, Klaus von Heusinger, Wolfram Hinzen, Kai Vogeley, Philipp Homan, Rui He, Claudio Flores Palominos and the whole team.

Read the paper