Dateline: April, 2022, Issue 3

Can false confessions be distinguished substantively or linguistically from true confessions?

False confessions are not rare. The National Registrar of Exonerations has reported that of the 2,400 exonerees in their database, 292 (12%) had falsely confessed. The Innocence Project has helped exonerate 375 individuals incarcerated for murders and rapes through post-conviction DNA testing, of whom 28% falsely confessed.

False confessions also are powerfully persuasive evidence of guilt. False confessions are potent even when the interrogation is coercive, the decision-makers are trial judges, the confessors are juveniles, the confessions are reported secondhand by a motivated informant, and the confessions are contradicted by DNA and other evidence (see, for review, Rizzelli et al., 2021).

False confessions cannot be distinguished from true confessions based on information only a perpetrator would know. Almost all false confessions contain accurate crime details not in the public domain. For example:

While substantive content has been unable to differentiate false confessions from true confessions, the grammar and linguistic cues used by confessors have been found by researchers to be able to distinguish false from true confessions.

Rizzelli and colleagues (2021) compared 37 confessions that have been proven false (obtained from the Innocence Project, DNA Exoneration Database, etc.) to 98 confessions that are presumed true (obtained from FBI files housed at John Jay College of Criminal Justice). Two computer programs were used to analyze the confessions: a psychological language analysis program called Linguistic Inquiry and Word Count and a corpus analysis program called Antconc. Proven false confessions were found to differ linguistically from presumed true confessions.

Proven false confessions had a higher incidence of third person plural pronouns (e.g., they, them, their), impersonal pronouns (e.g., this, those, it), assent words (e.g., okay, yeah) and nonfluencies (e.g., hm, oh, ahh), and their language was more formal, logical and hierarchical (as opposed to informal and narrative). Presumed true confessions had a higher incidence of words of six or more letters, first person pronouns and possessives (e.g., I, me, mine), auxiliary verbs (e.g., be, do, have, will, shall, would, should, can, could, may, might, must, ought), conjunctions (e.g., but, and, because) and adjectives.

The three linguistic categories accounting for the most difference between proven false and presumed true confessions were impersonal pronouns, personal pronouns, and conjunctions. A high frequency of impersonal pronouns was associated with confessions proven false, while a high frequency of personal pronouns and conjunctions were associated with confessions presumed true.

Using only the incidence of impersonal pronouns, personal pronouns and conjunctions occurring in a confession, Rizelli and colleagues conducted two tests to determine if they could predict whether specific confessions were true or false confessions. The incidence of impersonal pronouns, personal pronouns and conjunctions were able to correctly classify (as presumed true or proven false) 37 out of 50 (74%) confessions in one test, and 20 of 24 (83%) confessions in a second test.

The context of utterances in confessions also differed between proven false and presumed true confessions. Proven false confessions used the verb "to know" more often to express a lack of a confessor's knowledge (e.g., "I don't know", "I didn't know") than did presumed true confessions. When uttering "I do not/don't remember/recall", 49% of the time in proven false confessions this sentence was ended at that point, with no further details provided; in presumed true confessions, only 19% of the time did those words terminate the sentence and in the remaining 81% of the time, the confessor explained what was not remembered (e.g., "I don't remember what he was saying", "I don't remember if I had blood on them"). Both "I guess" and "I mean" were more frequently used in proven false than presumed true confessions, adding a tentative aspect to the narrative.

In sum, substantive details only a perpetrator would know cannot distinguish true from false confessions: 95% of false confessions contain accurate crime facts known only to the police. The grammar and language of confessions can distinguish a true from a false confession, with the frequency of impersonal pronouns, conjunctions and personal pronouns able to predict accurately whether a confession is true or false 75% or more of the time. False confessions contain more impersonal pronouns, such as "it", "that", and "what"; more conjunctions such as "and", "then" and "but"; and fewer personal pronouns, such as "I", "he" and "me".

Source Appleby, S.C., Hasel, L.E. & Kassin, S.M. (2013). Police-induced confessions: An empirical analysis of their content and impact. Psychology, Crime & Law, 19(2), pp. 111-128.

Source Appleby, S.C. & Petrillo, J.T. (2015). Confessions as Hollywood Productions: A Content Analysis Comparison of Corroborated and False Confessions. Paper presented at the annual meeting of the American Psychology-Law Society, San Diego.

Source Garrett, B.L. (2010). The substance of false confessions. Stanford Law Review, 62(4), pp. 1051-1118.

Source Garrett, B.L. (2015). Contaminated confessions revisited. Virginia Law Review, 101(2), pp. 395-454.

Source Rizzelli, L., Kassin, S.M. & Gales, T. (2021). The language of criminal confessions: A corpus analysis of confessions presumed true vs. proven false. The Wrongful Conviction Law Review, 2(3), pp. 205-225.