Finding the perfect match: the role of distributional learning in facilitating visual comparison performance

Abstract

Forensic feature-comparison practitioners (e.g. fingerprint or document examiners) conduct visual comparison tasks where they decide whether forensic samples are from the same or different sources. However, how they undertake this task is not yet well understood. The ability to learn which features are rare and diagnostic of a ‘match’ between two samples of evidence, and which are common and less diagnostic may assist with this task. This ability is known as distributional statistical learning. This thesis examines whether distributional learning facilitates visual comparison performance. Section 2 presents six experiments that develop and test a novel paradigm that examines distributional learning. Participants were able to learn distributional information from novel visual stimuli. Section 3 presents three experiments that explore the use of distributional learning to facilitate visual comparison decision-making when basic frequency information was available. Participants were not able to apply distributional knowledge to visual comparison decisions unless trained to do so. Section 4 presents two experiments that examine whether distributional learning facilitated visual comparison accuracy when joint probability information was available. Participants used their distributional learning to improve visual comparison performance and training improved their ability to do so. Section 5 presents three experiments that examine the distributional learning and visual comparison performance of forensic practitioners and novices. Practitioners’ distributional learning and visual comparison accuracy of novel stimuli did not differ to untrained novices. This suggests that practitioners’ use of distributional learning in a novel visual comparison task is equivalent to novices. Together, these results provide important insights about the role of distributional learning in visual comparison performance and could inform the development of training or selection tools for forensic feature-comparison practitioners.