‘Lyin' Ted’, ‘Crooked Hillary’, and ‘Deceptive Donald’: Language of Lies in the 2016 US Presidential Debates

Abstract

Language in the high-stakes 2016 US presidential primary campaign was contentious, filled with name-calling, personal attacks, and insults. Language in debates served at least three political functions: for image making, to imagine potential realities currently not in practice, and to disavow facts. In past research, the reality monitoring (RM) framework has discriminated accurately between truthful and deceptive accounts (~70% classification). Truthful accounts show greater sensory, time and space, and affective information, with little evidence of cognitive operations. An RM algorithm was used with Linguistic Inquiry and Word Count software to code candidates' language. RM scores were significantly higher in fact-checked truth statements than in lies, and debate language in the 2016 primaries was as deceptive as fact-checked lies. In a binary logistic regression model, one RM criterion, cognitive processes, predicted veracity using computerized RM, classifying 87% of fact-checked truth statements but only 28% of fact-checked lie statements (63% classification overall).

Detecting deceptive communication through linguistic concreteness

“…Numerous approaches aim to discriminate between truthful and deceptive texts, for example by automatically scoring texts on their richness of detail (Bond et al, 2017;Bond & Lee, 2005) or the proportion of named entities (Kleinberg, Mozes, Arntz, & Verschuere, 2017). Little is known about whether these approaches can be integrated through a common, underlying feature that may explain why a text is classified as truthful or deceptive.…”

“…Named entities (Kleinberg, Mozes, et al, 2017) Classifying words or phrases into, e.g. specific names, organizations or times Reality Monitoring (Bond et al, 2017;Bond & Lee, 2005) Increased perceptual, spatial and temporal details, decreased cognitive operations Linguistic specificity (Zhou et al, 2004) Increased perceptual details, first person pronouns, second person pronouns, third person pronouns Person index Personal pronouns, proper nouns and person names (ranging from abstract to concrete) Motion verbs (Newman et al, 2003) Concrete, simple motion verbs (i.e. concrete) versus complex evaluations and judgments (i.e.…”

“…Reviewing the literature from various fields (computational linguistics, legal psychology, cognitive psychology) we identified eight ways to operationalize concreteness: 1) crowd-sourced concreteness ratings (Brysbaert et al, 2014), 2) the Linguistic Category Model (Seih et al, 2017;Semin & Fiedler, 1988), 3) the Abstraction Index (Markowitz & Hancock, 2016), 4) the proportion of named entities (Kleinberg, Mozes, et al, 2017), 5) motion verbs (derived from Newman et al, 2003), 6) a Reality Monitoring score (Bond et al, 2017;Bond & Lee, 2005), 7) the Burgoon-Zhou model of specificity (Fuller, 2008;Zhou et al, 2004), and 8) a new person reference index (see Table 1). There is a reason to suspect that these eight operationalizations may not all function in the same way in the context of verbal deception detection.…”

“…There is a reason to suspect that these eight operationalizations may not all function in the same way in the context of verbal deception detection. For instance, some authors found no difference between truths and lies (e.g., "specificity" in Zhou et al, 2004), others found more concrete information in truths than in lies (e.g., "RM" in Bond et al, 2017; named entities in Kleinberg, Mozes, et al, 2017), and others found lies to be more concrete than truths (Newman et al, 2003). We systematically assessed their validity as a cue to deception on six different datasets of truthful and deceptive texts.…”

The first direct replication on using verbal credibility assessment for the detection of deceptive intentions

“…The LIWC counts how many words per input text belong to predefined psycholinguistic lexicon categories and has been used for verbal deception research before (e.g., Bond et al, 2017). For the current experiment, we used the categories “time” (e.g., “once” and “since”) and “space” (e.g., “above” and “outside”) each of which is standardized by the word count per statement and question type.…”