BUILDING LINGUISTIC CORPUS FOR VERBAL LIE DETECTION: ADVANCING CRIMINAL INVESTIGATIONS THROUGH NEURO-LINGUISTIC TECHNIQUES

Abstract

Lying is a global phenomenon that everyone encounters in everyday life, as it significantly affects communication. This study aims to develop a mini specialized speech-to-text corpus for deception detection in criminal investigations. Furthermore, it assesses the use of neuro-linguistic programming techniques in Pakistani courts, as well as their effectiveness in detecting deception through technologies such as machine learning, Python, and natural language processing. This research also identifies the linguistic markers of trickery to support decision-making in the Pakistani legal system.  A mixed research approach is employed to achieve these objectives, incorporating exploratory research and quasi-experimental research. The linguistic corpus is collected from primary and secondary sources, including the accused, suspects or criminals, spectators, complainants, and witnesses who are questioned by police officers, lawyers, or inspecting journalists, selected using convenience and purposive sampling techniques. The primary source consists of data from the session court, where legal proceedings are included for observation of neurolinguistic technique, while the secondary source is obtained from public legal content on YouTube, such as the ‘Pukar’ media program. The corpus contains 10,360 words from legal cases related to murder (Section 302) and violence (Section 376), featuring both male and female speakers. The result shows that truthful sentences are simple in structure, while liars create complex sentences during speaking. The result concluded that the neurolinguistic technique is an effective method to detect lies, as it gives the required results within a limited time. Furthermore, the analysis shows that direct elicitation, emotional trigger, accusation/challenge, and accusation/pressure on the brain are the best techniques for disgorge of truth. Whereas role, or framing of identity, Wh-questions, and direct elicitation of NLP techniques are best for lie detection.