Adjectives and deception: A view from linguistic theory

Abstract

This study addresses the challenge of deceptive opinion spam, a growing concern for e-commerce and consumer trust. Building on established psychological theories of deception and focusing on hotel reviews, we expand current approaches by incorporating a Radical Construction Grammar (RCG; Croft, 1990, 1991, 2001, 2022) perspective on adjectives. Traditional part-of-speech taggers define adjectives largely through morphological and syntactic criteria, lumping property modifiers together with property predicates. Based on Croft’s more refined framework, we suggest that the cognitive load associated with property words used attributively (e.g., the white door) is higher than in predicative positions (e.g., the door is white). We analyse a subset of the Deceptive Opinion Spam Corpus (DOSC) and find attributive property words to be significantly more frequent in truthful reviews, whereas predicative forms show no variation. This distinction proved more effective than a traditional POS-tagger based definition of adjectives in separating authentic from fake reviews. The manual coding required for the RCG-based approach was resource-intensive, but even modest accuracy gains could be crucial in high-stakes scenarios. Future work should investigate whether a Croftian approach can be operationalised through automated taggers and whether these findings extend to other deceptive contexts. The paper highlights the benefit of a more theoretically grounded view of linguistic categories in forensic settings. A truly interdisciplinary effort that draws on advanced linguistic theory as much as on psychological theories of deception, and operationalises the approach computationally, thus promises to yield efficient and more effective deception detection systems.