Flawed methods, flawed results: Unpacking verbal lie detection research

Description

Much of the work on lie detection stems from the field of forensic psychology - it is largely experimental and quantitative in nature. The field works on the theoretical assumption that when someone lies, they exhibit distancing behaviour, they use additional cognitive effort, and they attempt to control their behaviour to stop the other person from detecting their lie. Many “cues to deception” have been identified - some verbal, some non-verbal - with varying degrees of reliability. Verbal cues include increased hedging, a reduction in use of first-person pronouns, and more fillers.

Such findings have been widely reported over the years, but recent work suggests they are based on faulty statistics: the result of low-powered studies (Luke, 2019), and prone to sociolinguistic variation (Gillings, 2020). Not only that, but the methods used to analyse such language data are inherently flawed. By far the most widely used tool in the field is LIWC, a tool which allegedly allows users to “analyze others’ language […] to understand their thoughts, feelings, and personality” (LIWC, 2022). The tool assigns each word to a category (based on a dictionary), then the researcher compares those frequencies across conditions and speakers. This work does not consider the context of these linguistic cues, with researchers instead relying on earlier theory to explain linguistic differences.

This talk will unpack and evaluate the field of deception detection from a corpus linguistic perspective. It will primarily focus on methodological issues, regarding both the tools used and researcher interpretation of results. It will suggest potential alternatives and report on a small corpus-based case study into pronoun variation across truthful and deceptive statements.

Period	30 Apr 2022
Event title	Austrian Association of University Teachers of English (AAUTE)
Event type	Unkonwn
Degree of Recognition	National