Hayashi · Environmental health and preventive medicine 2004 · cross-sectional questionnaire survey · n=1395

Factors affecting the appearance of 'twin language': An original language naturally developing within twin pairs.

Cited 9 times in the scientific literature.

Level 4 - case-series / case-control

Cross-sectional survey (by design analogy, Oxford CEBM Level 4)

PubMed 21432318 · doi:10.1265/ehpm.9.103 · record verified 2026-08-26

What was done

A mailed questionnaire survey was administered to 2,733 mothers of twins to determine factors associated with the emergence of 'twin language' (a unique language understandable only within the pair). Completed questionnaires were returned by 1,395 mothers. The survey collected data on birth weight, age at first spoken word, physical resemblance ('as alike as two peas in a pod'), household members, and non-verbal play. Multivariate logistic regression adjusting for twin age was used to analyze predictors.

What was found

Twin language was reported in 598 of 1,395 pairs (42.9%). Occurrence rates were 48.4% in exactly alike female pairs (140/289), 47.6% in exactly alike male pairs (129/271), 40.4% in not-alike male pairs (105/260), 39.3% in not-alike female pairs (106/270), and 38.4% in opposite-sex pairs (112/292). Multivariate logistic regression controlling for age showed that twin language was significantly more frequent in twin pairs who were exactly alike, engaged in non-verbal play, and had fewer older siblings. Specific numerical odds ratios, confidence intervals, and p-values were not provided in the abstract.

Why it matters

This study provides observational evidence on the frequency of autonomous twin language and links its development to closer physical resemblance/potential monozygosity, close dyadic non-verbal interaction, and reduced linguistic input from older siblings.

Limits

The study relied entirely on subjective maternal recall and questionnaire responses without direct linguistic observation or genetic zygosity testing. The response rate was 51.0% (1,395/2,733), introducing potential selection bias. The abstract omits effect sizes, confidence intervals, and exact p-values for the multivariate model.

Cited by