Lies in Conversation: An Examination of Deception Using Automated Linguistic Analysis - eScholarship
Bibliographic record
Abstract
Lies in Conversation: An Examination of Deception Using Automated Linguistic Analysis Jeffrey T. Hancock (jeff.hancock@cornell.edu) Department of Communication, Cornell University 320 Kennedy Hall, Ithaca, NY 14850 USA Lauren E. Curry (lec26@cornell.edu) Department of Communication, Cornell University 320 Kennedy Hall, Ithaca, NY 14850 USA Saurabh Goorha (sg278@cornell.edu) Department of Communication, Cornell University 320 Kennedy Hall, Ithaca, NY 14850 USA Michael T. Woodworth (mwoodwor@dal.ca) Department of Psychology, Dalhousie University Life Sciences Center, Halifax, NS, B3K 1L4, Canada suggests that liars tend to make less sense and tell less plausible stories (e.g., making discrepant and ambivalent statements), among other verbal characteristics (for review, see DePaulo, Lindsay, Malone, Mulenbruck, Charlton, & Cooper, 2003). The present study employs automated linguistic analysis, in which a computer program is used to analyze the linguistic properties of texts, to examine the verbal content of deceptive and truthful conversations. As Pennebaker, Mehl, and Niederhoffer (2003) note, words used in daily interactions reveal both psychological and social aspects of peoples’ worlds. Certain words and parts of speech can be markers of emotional, psychological, and cognitive states. Given that deceiving others likely involves changes in emotional or psychological states, linguistic cues detected using automated techniques may indicate lying in conversation. Abstract The present study investigated changes in both the sender’s and the receiver’s linguistic style across truthful and deceptive dyadic communication. A computer-based analysis of 242 transcripts revealed that senders used more words overall, increased references to others, and used more sense-based descriptions (e.g., seeing, touching) when lying as compared to telling the truth. Receivers naive to the deception manipulation produced more words and sense terms, and asked more questions with shorter sentences when they were being lied to than when they were being told the truth. These findings are discussed in terms of their implications for linguistic style matching. Introduction Maxims such as “honesty is the best policy” and “let the truth be told” reinforce the notion that telling the truth is the best way to communicate. When telling everyday lies, then, deceivers must be careful to assume a position of sincerity in order to make their partners believe them and avoid being viewed in a negative light. In fact, this feat might not be very difficult to accomplish. Previous research suggests that it is quite difficult to catch a liar as deception detection rates in many experiments are not much better than chance (Vrij, In general, there are three methods for trying to detect deceit. The first method focuses on vocalic and physical nonverbal behaviors (e.g., movements, smiles, voice pitch, speech rate, stuttering, and eye gaze) (Vrij, 2000). The second method involves measuring physiological responses with various technologies, such as polygraph machines (Vrij, Edward, Roberts, & Bull, 2000). The third method is concerned with the content of what is said (e.g., verbal behavior, as well as a study of linguistic properties of liars’ texts). For example, previous research Linguistic Indicators of Deception A review of the relatively small literature concerned with automated linguistic analyses of deception indicates that, to date, at least four main types of linguistic cues have been associated with deception: 1) word counts 2) pronoun usage, 3) words pertaining to feelings and the senses, and 4) exclusive terms (Burgoon, Buller, Floyd, & Grandpre, 1996; Burgoon, Bliar, Qin, & Nunamaker, 2003; Newman, Pennebaker, Berry, & Richards, 2003; Pennebaker et al., Consider first differences in word counts across deceitful and truthful messages. Previous studies have found that senders offer fewer details when lying than when telling the truth (Burgoon et al., 2003; DePaulo et al., 2003; Vrij, 2000). Senders may offer fewer details because they are less familiar with what they are discussing, or because they are trying to avoid providing details that may be inconsistent
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.004 |
| Science and technology studies | 0.000 | 0.002 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".