Developmental sensitive periods suggest architectural shaping. Hannon and Trehub (2005) found that brief exposure restores 12-month-olds' but not adults' sensitivity to non-native meters. Subsequent work confirmed that while adults improve on non-native meters behaviorally, they show persistent differences in neural responses: Hannon, Soley, and Ullal (2012) found that American listeners, despite behavioral improvement on Turkish meters, showed different error patterns than Turkish listeners, suggesting different underlying representations rather than mere parameter differences (Hannon, Soley, & Ullal, 2012, 543-548). The pattern of behavioral improvement without representational normalization is what constitution would predict.
Statistical learning creates categories, not just weightings. While Saffran et al. (1996) demonstrated universal statistical learning mechanisms, subsequent work shows that early statistical learning creates categorical boundaries that persist despite later exposure. Werker and Tees' (1984) phonemic narrowing research provides a template: infants lose sensitivity to non-native phonemic contrasts not merely through disuse but through active reorganization of perceptual space (Werker & Tees, 1984, 49-54). Rhythmic categories may undergo analogous reorganization.
Ethnomusicological evidence documents qualitative differences. McGraw (2008) reports that foreign students studying Balinese gamelan describe not merely learning new rhythms but experiencing time differently, with phenomenological reports consistent with architectural rather than parametric change. However, such reports require systematic collection and validation before serving as strong evidence. The McGraw study, in any case, offers only preliminary results.
Adult plasticity is substantial. Musicians successfully learn non-native rhythmic systems such as Balkan meters, West African polyrhythm, Indian tala. While difficulty varies, acquisition is possible. Constitution would predict harder limits than typically observed.
Neural mechanisms appear shared. Basal ganglia involvement in timing is documented across populations studied (Grahn & Brett, 2007). Oscillatory entrainment occurs universally. The mechanisms appear identical; only the inputs differ, consistent with universal statistical learning over variable input.
Parsimony favors modulation. One universal system with variable parameters is simpler than multiple constituted architectures. The infant narrowing data might reflect efficiency differences (early learning is faster) rather than architectural differences (only early learning is possible).
Neural mechanisms appear shared. Basal ganglia involvement in timing is documented across populations studied (Grahn & Brett, 2007). Oscillatory entrainment occurs universally. The mechanisms appear identical; only the inputs differ, consistent with universal statistical learning over variable input.
Most importantly, the Jacoby 7:2:3 finding permits multiple interpretations. This category could reflect culturally constituted architecture absent elsewhere, or simply statistical learning from a specific repertoire producing a category that could form in any population given equivalent exposure. Distinguishing these requires training studies not yet conducted.
No one wins, as Always
The evidence genuinely underdetermines theory choice. Both positions accommodate most findings through auxiliary hypotheses. What would decisively favor one position?
For constitution, a training study where adults achieve behavioral parity with native performers but show persistent implicit processing differences (MMN, phase-locking) exceeding pre-registered thresholds.
For modulation, evidence that adult training produces implicit processing signatures statistically indistinguishable from developmental exposure, given sufficient training duration.
For modulation, evidence that adult training produces implicit processing signatures statistically indistinguishable from developmental exposure, given sufficient training duration.