Past-life Memories and Foreign Languages: An Exploration of Xenoglossy in Cases of the Reincarnation Type

Cozzolino, P. J., Pehlivanova, M., & Tucker, J. B. (2025). Past-life Memories and Foreign Languages: An Exploration of Xenoglossy in Cases of the Reincarnation Type. Journal of Anomalous Experience and Cognition, 5(2), 111–135. https://doi.org/10.31156/jaex.27464

AI Assessment

A database study from the University of Virginia’s reincarnation-research group examining xenoglossy, the claim that a child recalling a past life can speak a language they had no ordinary way of learning. Comparing 40 xenoglossy cases with 872 others in the CORT database (1959 to 2020), the authors report that xenoglossy was not linked to ordinary explanations (early speech, parental education) but was linked to emotion during recall, an intentionally violent mode of death, a desire to return to the previous family, and overall case strength. The study has an unusually honest feature, the authors disclose their own prior belief scores, and it applies a Bonferroni correction and directly tests mundane explanations. But it rests on a proprietary, non-public database and on subjective variables coded by a team that believes in the phenomenon, with a small number of xenoglossy cases (as few as 21 in the multivariate model). This audit reports what the study found and how; it takes no position on whether reincarnation occurs.

Provenance

DOI. 10.31156/jaex.27464 · Journal of Anomalous Experience and Cognition 2025, 5(2), 111 to 135. Open access under CC-BY. (The DOI printed inside the article itself contains a typo, 10.311356; the resolvable identifier for this journal is 10.31156.)

Study type. An archival analysis of an existing case database, using logistic regression. It is observational, not experimental.

Authors. Philip J. Cozzolino, Marieta Pehlivanova, and Jim B. Tucker, of the Division of Perceptual Studies, University of Virginia School of Medicine, the group that continues Ian Stevenson’s reincarnation-case research.

Data source. The Cases of the Reincarnation Type (CORT) database, a proprietary collection held at the University of Virginia and not publicly available, which limits independent replication.

Source basis. Every figure below is taken from the article’s own Abstract, Methods, and Results.

What the paper reports

Cases of the reincarnation type are children, usually aged 2 to 7, who spontaneously describe what appear to be memories of a previous life, investigated with a forensic protocol that seeks first-hand evidence for and against the claims.¹ Xenoglossy, speaking a foreign language the child should not know, is one of the more extraordinary features sometimes reported. The authors test a novel emotion-trauma hypothesis: that xenoglossy is tied to emotionally intense, violent, or distressing past-life memories, rather than to mundane factors like precocious verbal ability or educated parents.

Xenoglossy was not associated with variables related to an ordinary explanation. The emotion-trauma hypothesis was supported, in that xenoglossy was associated with participants’ emotionality, desires to return to their purported previous family, claiming to have died as a result of intentional/violent means, and having a stronger case.

How it was run

• Sample. 40 cases coded as exhibiting xenoglossy versus 872 that did not (912 total), drawn from the CORT database and spanning 1959 to 2020.
• Predictors. Two “ordinary explanation” variables (age at first coherent speech, average parental education) and four hypothesis variables (degree of emotion during recall, intentionally violent mode of death, desire to return to the previous family, and an adjusted strength-of-case score with its geographic-distance component removed).
• Analysis. Six univariate binary logistic regressions predicting xenoglossy, with a Bonferroni-corrected significance threshold of p = .008 (0.05 divided by 6), followed by a multivariate model combining the individually significant predictors.
• Belief disclosure. The three authors reported their own a-priori belief in the phenomenon on a 5-point scale (4, 4, and 3), an explicit acknowledgment of their priors that is rare in this or any literature.

Results, as reported

Values are reproduced from the article’s Results section. The pattern supports the emotion-trauma hypothesis over the ordinary-explanation variables, but the effects rest on few xenoglossy cases (40 in the univariate tests, 21 in the multivariate model), and in that multivariate model the mode-of-death result (p = .035) does not meet the study’s own Bonferroni threshold of .008.

Eleven-dimension audit

Pre-registration

Not preregistered, but the study applies a pre-specified Bonferroni correction (p = .008) to its six univariate tests and states its emotion-trauma hypothesis in advance, which are partial substitutes. The exploratory latitude that remains is in the choice and construction of predictors (for example the geographic-distance-adjusted strength-of-case score).

Randomization

Not applicable: this is an archival analysis of existing cases with no assignment of conditions. Cases enter the database through the group’s field investigations, which is a form of selection to keep in mind.

Sensory leakage

Not applicable in the perceptual sense. The analogous validity threat here is coding bias, treated under Blinding: the outcome (xenoglossy) and several predictors (emotion, case strength) are judgments made by the research team from case files.

Blinding

This is the central methodological concern. Xenoglossy status, degree of emotion, and strength of case are coded from investigator records, and the paper does not establish that coders were blind to the hypothesis when doing so. Because the same group that investigates and believes in these cases also codes the variables, non-blind coding of subjective features could inflate the very associations reported.

Optional stopping

Not a concern: the analysis used a fixed historical database spanning 1959 to 2020 rather than accumulating data until a result appeared.

Outcome measure

The outcome is a binary xenoglossy classification. It is worth noting that the study analyzes cases already coded as xenoglossy; whether those involve genuine responsive foreign-language ability, as opposed to recitation or looser coding, is a prior classification the analysis takes as given rather than tests.

Effect size

The odds ratios are moderate for emotion (1.71) and mode of death (2.80) and small for case strength (1.05). With only 40 (then 21) xenoglossy cases, these estimates are imprecise, and small-sample logistic regression can produce unstable odds ratios.

Multiple comparisons

Handled better than most: six univariate tests are held to a Bonferroni threshold of p = .008. The multivariate follow-up is more permissive, and its mode-of-death effect (p = .035) would not survive the same correction, so that particular result should be read cautiously.

Internal replication

The multivariate model provides a partial internal check: emotion and mode of death survive the move from univariate to multivariate, while desire to return and case strength do not. That two of four predictors drop out is honest shrinkage and appropriately reported.

External replication

This is an original analysis, and the proprietary CORT database means no outside group can currently replicate it on the same data. External replication would require either access to this database or an independent collection of comparable cases.

Transparency

Strong in one unusual respect and weak in another. On the credit side, the authors disclose their own prior beliefs on a numeric scale, apply a multiplicity correction, and explicitly test (and reject) ordinary explanations, all marks of good faith. On the debit side, the underlying database is proprietary and cannot be shared, and the key variables are subjectively coded, so the analysis cannot be fully independently checked.

The adversarial record

• Non-public data, subjective coding. The strongest objection for a verifiability-first reader is that the CORT database is proprietary and the outcome and predictors are coded by a group that believes in the phenomenon, with no stated blinding. The reported associations cannot be independently verified, and coding expectancy is a live alternative to the emotion-trauma interpretation.
• Few cases. Forty xenoglossy cases in the univariate analysis and twenty-one in the multivariate model are small numbers for logistic regression; the odds ratios are correspondingly uncertain, and one multivariate result does not meet the study’s own significance threshold.
• Classification taken as given. The analysis presupposes that the “xenoglossy” cases genuinely involve unlearnable foreign speech. The long-standing skeptical critique of xenoglossy, that many claimed cases reduce to recitation, chance resemblance, or cued responses, sits upstream of this study rather than being tested by it.
• What is genuinely commendable. The a-priori belief disclosure is a model of transparency that proponent and skeptic studies alike rarely match; the Bonferroni correction is appropriate; and directly testing, and finding null, the ordinary-explanation variables is exactly the right adversarial move. Within the limits of a proprietary, investigator-coded database, this is a carefully and candidly reported analysis.