There is little doubt that the twin challenges of language universals and language diversity are critical for understanding the architecture of the language faculty, its domain-specificity and evolutionary origins. Despite their crucial import, these questions remain unaddressed in most existing psycholinguistic research. Evans & Levinson (E&L) should be commended for reminding the cognitive science community of its outstanding intellectual debt in this area. Nonetheless, E&L's own conclusion – that the hypothesis of universal grammar is false – does not follow from the evidence they present. Here, I specifically consider E&L's analysis of phonological universals – the role of syntactic and semantic universals falls beyond the scope of this commentary.

In its bare minimum, the hypothesis of Universal Grammar (UG) states that the brains of all speakers represent a shared set of grammatical constraints. Although this hypothesis is often associated with the claims that UG constraints are innate, and domain- and species-specific, these additional claims are not logically linked to the basic hypothesis of grammatical universals. E&L appear to reject all four claims on the grounds that language typology exhibits no absolute, exceptionless regularities. Typological universals, however, are distinct from grammatical universals, and the link between them is complex. Grammatical universals – the object of cognitive inquiry – are mental representations (I-language), whereas typological universals are statistical generalizations concerning external linguistic outputs (E-language). Such outputs are shaped by multiple factors, of which putative grammatical universals are only one force – the restrictions on perception, motor control, conceptual structure and memory, coupled with cultural and social factors, are equally strong determinants.

Consider, for example, the typological prevalence of CV syllables (discussed by E&L). One theory of UG, Optimality Theory (Prince & Smolensky 1993/2004), attributes this fact to a universal, but violable, well-formedness constraint that requires all syllables to begin with an onset. Such a constraint, however, does not guarantee that CV syllables are most frequent typologically (typological frequency is also determined by extra-grammatical factors); nor does it preclude the existence of onsetless syllables (e.g., V – such syllables can be protected by other constraints enforcing faithfulness to grammatical inputs). Instead, the Onset constraint predicts that no grammatical process will actively transform syllables with an onset to onsetless ones (de Lacy, Reference de Lacy and Parker2008). Whether the case of Arrernte (cited by E&L) counters this prediction is debatable (Berry Reference Berry1998; Smith Reference Smith2005), but typological frequency alone clearly cannot decide this matter.

Although this conclusion calls for a more careful interpretation of the typological diversity, it does not render the UG hypothesis unfalsifiable: optimality theory asserts that universal well-formedness constraints are active in the grammars of all speakers, irrespective of whether the relevant structures are present or absent in their linguistic experience. This strong hypothesis has sparked a productive research program that uses experimental tools to test the role of grammatical language universals – an enterprise that has unfortunately gone unnoticed by E&L. The available findings suggest that speakers are sensitive to putatively UG restrictions unattested in their language while ignoring other regularities that are equally motivated on statistical and phonetic grounds (Becker et al., submitted; Davidson Reference Davidson2006; Hayes et al., submitted; Moreton Reference Moreton2008; Wilson Reference Wilson2006).

Consider, for example, the restrictions on onset clusters (e.g., bl in blocks). It is well known that onsets such as bl are typologically more frequent than lb, and languages that tolerate syllables like lba tend to allow bla. This fact is attributed to sonority – a scalar property that correlates with the intensity of consonants: least sonorous (softest) on the scale are stops (e.g., b,d), followed by nasals (e.g., n) and liquids (e.g., l). Accordingly, bla rises in sonority, whereas lba manifests a sonority fall. The typological preference for onsets like bl is captured by a scalar UG constraint that favors onsets with large sonority distances (e.g., bl>bn>bd>lb, where>indicates preference; Clements Reference Clements, Kingston and Beckman1990; Smolensky Reference Smolensky, Smolensky and Legendre2006).

Although sonority restrictions are widely documented, the typological evidence reflects only implicational tendencies, and many languages manifest outright reversals (e.g., Russian allows sonority falls, e.g., lb). Such observations might lead E&L to conclude that sonority restrictions are not grammatical universals, but rather, artifacts of modality-specific acoustic and articulatory preferences. However, sonority-based restrictions have been documented in sign languages (Corina Reference Corina, Ziolkowski, Noske and Deaton1990; Sandler & Lillo-Martin Reference Sandler and Lillo-Martin2006), and recent experimental work suggests that they are active in the brains of individual speakers even when the relevant structures are absent in their language. English speakers, for example, favor syllables that rise in sonority (e.g., bnif) compared to sonority plateaus (e.g., bdif), which, in turn, are preferred to sonority falls (e.g., lbif), and this preference shapes the perception of these syllables (Berent Reference Berent2008; Berent et al. Reference Berent, Steriade, Lennertz and Vaknin2007): the worst-formed onsets of falling sonority (e.g., lbif) are more likely to be misperceived (as lebif) compared to sonority plateaus (e.g., bdif), which, in turn, are misperceived relative to sonority rises (e.g., bnif). The misperception of ill-formed onsets is not due to an inability to extract their surface form from the acoustic input (e.g., it obtains with printed materials; Berent et al. Reference Berent, Lennertz, Smolensky and Vaknin-Nusbaum2009); nor is it explained by the statistical properties of English – similar results have been reported among speakers of Korean, a language that arguably lacks onset clusters altogether (Berent et al. Reference Berent, Lennertz, Jun, Moreno and Smolensky2008). Instead, these systematic misperceptions imply broad grammatical knowledge that triggers the active recoding of ill-formed structures. The convergence of sonority preferences across phonological systems, both spoken and signed (Sandler Reference Sandler1993), is consistent with a domain-specific phonological mechanism.

Nonetheless, the present results cannot determine whether phonology preferences are, in fact, universal or innate. Markedness (i.e., well-formedness) hierarchies, such as sonority, could vary in detail due to both predictable grammatical processes (e.g., conflation; de Lacy Reference de Lacy2006) and variation in fine-grained phonetic properties that could inform their inference (Hayes & Steriade Reference Hayes, Steriade, Hayes, Kirchner and Steriade2004). Whether phonological markedness hierarchies are experience-independent or learned is unknown, and there is vanishingly little information on their domain- and species-specificity. Far from being untestable, however, these questions call for a comparative cross-linguistic research program that combines typological, formal, and experimental methods. The emerging field of experimental phonology demonstrates the viability of this approach in evaluating the UG hypothesis.