RESEARCH

Research questions and objectives

Research questions

RQ1

Divergence from clinical definitions

How do mental health terms diverge from their psychiatric definitions as they enter everyday discourse?

RQ2

Variation across groups

How does perception and usage of mental-health terminology vary across different groups (e.g., laypeople, academic psychologists, people with lived experience)?

RQ3

Interpretation by language models

To what extent can LLMs distinguish between clinical and colloquial uses of mental-health terminology and accurately infer users’ intent to avoid safety-relevant misinterpretations?

Objectives

O1

MindShift benchmark

Create a multilingual, human-annotated benchmark capturing semantic proximity and affective dimensions across clinical and everyday contexts.

O2

Human perception

Characterize how people interpret the same terms differently, including perceived severity, stigma, misleading uses and distress signals.

O3

Computational evaluation

Evaluate how language technologies represent denotational and connotational variation, using human judgments as reference.

O4

Guardrails

Translate findings into practical evaluation tools and design guidance for safer, context-sensitive human–AI interaction.