The Harmony Research Group

We love words and their ability to communicate who we are, how we feel, and what we think.

That is why we analyse language from text, audio and video recordings, to gain psychological and clinical insights into how people experience health and illness.

Working at the intersection of psychology, psychiatry and computer science, we study both mental and physical health, and aim to improve research and clinical practice by developing, adapting and validating AI-driven methods.

Probed language-based assessments

Over the past decade, we have developed probed language-based assessments, which let people describe their thoughts, feelings and symptoms in their own words. Language is a natural way to communicate complex states of mind and body, and analysing these descriptions with AI produces highly valid and reliable assessments. Using robust and open methods, we have demonstrated that these assessments can be:

Valid and reliable: Accurate and consistent, including when compared with clinical diagnoses.
Descriptive: Capturing the richness of personal experiences – providing insights beyond rating scale scores.
Predictive: Providing insights into future behaviours and health outcomes, such as healthcare use.

We have used probed language-based assessments to study, among other things:

Mental health, including depression, anxiety, stress, PTSD, and suicide and self-harm risk
Physical health, including how people describe their physical health and how this relates to medical diagnoses and healthcare costs
Well-being, including experienced well-being, harmony in life and satisfaction with life
Personality, including implicit motives

Illustrations of well-being, depression and personality

Software for Social Sciences

We develop and maintain an ecosystem of open-source R-packages designed to facilitate the use of state-of-the-art AI methods and large language models in social sciences:

talk transforms voice recordings into text, audio features, or embeddings optimized for capturing psychological dimensions in speech.
text converts written text into word embeddings and includes machine learning techniques optimized for social sciences to analyze these embeddings.
topics visualizes language patterns into topics to generate psychological insights.

talk and text provide access to Large Language Models from Hugging Face.

Logos of the talk, text and topics R-packages

The L-BAM Library

The Language-Based Assessment Model (L-BAM) Library is an open collection of pre-trained models that turn people’s own descriptions into scores for psychological constructs, with models for mental and physical health, personality, well-being, implicit motives and more.

Every model comes with documentation of its training data, performance and development process, so you can judge whether it suits your data before relying on it. The models are downloaded and applied directly in R with the text package:

library(text)
textPredict(
  model_info = "mentalhealth_interview_mxbai23_kjell2025",
  texts = "Lately I feel low most days, I sleep badly and I worry about everything."
)

Browse the library · Read the paper

Courses & Workshops

Coursera specialization · online

The Mental Health Clinician and AI

Three online courses for students and professionals in psychology, medicine and other clinical disciplines on how to navigate AI in mental health care: understanding and evaluating AI for mental health, AI applications in clinical practice, and implementation, safety and the future of mental health care.

View on Coursera

SMaRT Workshops · online

Analysing Human Language using R

A hands-on workshop on the R Language Analysis Suite: from audio recordings and text to transcriptions, L-BAM scores, visualisations of the language patterns behind them, and publication-ready reporting. No prior experience with NLP or machine learning is required.

Workshop details & registration

Funding

Our research is made possible by support from:

AI Lund