The architecture of police interviews: a speech unit-level analysis of communicative intent and topic in false and presumed true confessions
Year: 2026
Type: dissertation
Abstract: This dissertation examined whether police interview transcripts containing false confessions differ linguistically from those containing presumed true confessions. Previous research has shown that false confessions can be as detailed as presumed true confessions but less is known about whether they differ in broader language patterns. To address this gap, this research analyzed police interviews at the micro speech-unit level, considering both communicative intent (i.e., how they talk) and topic (i.e., what they talk about). The sample included 20 false and 20 presumed true confessions in police interview transcripts from the United States. Large language models were trained to segment speaking turns into speech units (i.e., Segmentation), and analyze communicative intents (i.e., Continent) and topics (i.e., Content). The resulting annotations were analyzed using generalized linear mixed-effects models (GLMMs) and cluster-bootstrap contrasts. The findings showed differences by confession type and speakers. First, police officers and false confessions followed a more constrained interview style. Across communicative intent and topic, police officers produced more clarifications and closed-ended questions, and more often introduced topics related to actions, people, and objects. They also showed higher content density than suspects, especially in false confessions. Second, presumed true confessions showed more affective and expressive content. And third, suspects' contributions in false confessions were shorter and more dependent on prior speech units. Overall, the findings suggest that false and presumed true confessions differed beyond case-relevant detail. Although these findings do not yet have diagnostic application, they may inform research on interview profiles associated with false confessions.
Author: Blanco-Velasco, Guadalupe
Language: English
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7208488769
