Speech and Language Markers of Bipolar Disorder: Challenges and Opportunities.

backgroundClinicians aspire to predict the emergence of Bipolar Disorder (BD) in a timely manner. To accomplish this, markers reflecting mental states that can be gathered non-invasively and at large scale are needed. Here, we systematically evaluate evidence relating speech-based markers to mood states in BD.

methodsWe searched Medline and Google Scholar for all published studies in English up to February 2026 on the use of speech markers in BD. We undertook thematic analysis on abstracts using topic modeling and a qualitative gap analysis to identify potential opportunities for future research.

results43 out of 867 studies were included after screening. Topic analysis revealed an emerging focus on mapping mood states to automated speech features. Most studies focused on cross-sectional detection of bipolar mood states, or BD as a diagnosis, rather than the prediction of upcoming mood states. Speech features distinguished BD from schizophrenia, depression, and healthy controls. Manic states were characterized by quantifiable measures of pressured speech, derailment, grammatical errors, and word repetition; depressive states by an increased use of personal pronouns, reduced verbal fluency, and speech quantity. Overall, attempts to replicate observations were limited.

conclusionAcoustic and lexical-semantic markers vary with manic, psychotic, or depressive states. At present, the evidence is insufficient for clinical utility in relapse prediction, response monitoring, or diagnosing mixed episodes or state changes in BD. We recommend that future research leverages the growing capabilities of natural language processing through longitudinal and cross-linguistic studies to strengthen the evidence base and advance the clinical utility of speech markers for BD.