Evans & Levinson (E&L) claim that diversity, rather than uniformity, is the fundamental fact of language relevant for cognitive scientists. This is not the conclusion that modern linguistics arrived at over the last half century: careful testing of specific universalist hypotheses permitted in many cases the discovery of precise and stable patterns of invariance.
First, consider locality. In the late 1970s, the observation that wh-phrases can be extracted from indirect questions in Italian led to abandoning the hypothesis of a universal Wh Island Constraint (WhIC), and to the formulation of a parametrized version of the relevant Universal Grammar (UG) principle, Subjacency (Rizzi Reference Rizzi, Dubisson, Lightfoot and Morin1978). The attempt to systematically check the empirical validity of the model led to the discovery of argument/adjunct asymmetries. Wh-adjuncts (why, how, …) are strictly non-extractable even in languages which are permissive with argument extraction:
(1)
a. Quale problema non sai come risolvere?
“Which problem don't you know how to solve?”
b. *Come non sai quale problema risolvere?
“How don't you know which problem to solve?”
The critical cross-linguistic observation came from the study of wh-in-situ languages (with wh-elements remaining in clause-internal position). Huang(Reference Huang1982) discovered that in Chinese, the argument-adjunct asymmetry (1) holds at the interpretive level. Sentence (2) can mean (3)a, with interpretive extraction of the argument, but not (3)b, with interpretive extraction of the adjunct:
(2)
Akiu xiang.zhidao [women weishenme jiegu-le shei] ne?
Akiu want.know [we why fire-Prf who] Qwh
(3)
a. “Who is the person x such that Akiu wonders [why we fired person x]?”
b. “*What is the reason x such that Akiu wonders [whom we fired for reason x]?”
Huang's conclusion was that in wh-in-situ languages the wh-elements move abstractly to the front, and the abstract movement in (2) is sensitive to the same fundamental locality constraint as the overt movement in (1) (see also, Tsai Reference Tsai1994). A huge literature was then devoted to understanding the nature of the asymmetry in (1) to (3) (see Cinque Reference Cinque1990; Pesetsky Reference Pesetsky2000; Starke Reference Starke2001, among others), the generality of adjunct non-extractability (Lasnik & Saito Reference Lasnik and Saito1992), and the parametrization involved in argument extraction. So, the refutation of an unqualified WhIC gave rise to a deeper cross-linguistic generalization, which was soon extended to a much larger class of Weak Islands (Szabolcsi Reference Szabolcsi, Everaert and van Riemsdijk2006), and deeply influenced the development of the theory of locality (Chomsky Reference Chomsky1995, Reference Chomsky, Martin, Michaels and Uriagereka2000; Rizzi Reference Rizzi1990).
Here and elsewhere, variation is not haphazard but gives rise to precise cross-linguistic patterns, with some logical possibilities systematically unattested: we don't find the “mirror image” of Italian or Chinese, that is, of a language allowing (overt or covert) extraction of adjuncts and barring extraction of arguments. If the non-universality of the WhIC had been interpreted as showing that anything goes in extraction across languages, linguists would have continued to collect catalogues of extraction facts for individual languages, with little hope of ever uncovering systematic generalizations in the patterns of variation.
In this case, the discovery of invariance required considerable abstractness and theoretical sophistication. Much of the typological tradition discussed by E&L has chosen to stick to an extremely impoverished, non-abstract descriptive apparatus, a legitimate methodological decision, but one which severely restricts the possibility of capturing forms of invariance. In spite of these limitations, it is far from true that the typological literature globally supports an “anything goes” position. In fact, the typological findings can be an important source of inspiration for more structured and abstract approaches. Consider Cinque's (2005) discussion of Greenberg's (1963a) Universal 20, a generalization essentially confirmed by 40 years of typological investigations: in languages in which N appears at the end of the NP, the order of certain modifiers is rigid (Demonstrative – Numeral – Adjective – Noun); in languages in which N is initial, significant ordering variation is observed. Cinque offers an explanation of this pattern through the assumption that N movement is the fundamental engine determining reordering. If N moves, it can determine different orders on the basis of familiar parameters on movement. If N remains in the final position, the order is invariably the basic one, plausibly determined by compositional semantics. Again, only a small subset of the logical possibilities is attested, and precise patterns of variation hold, which are amenable to plausible computational principles.
Finally, consider the theory of binding, constraining referential dependencies (E&L, sect. 3, para. 8). There is some limited variation in the locality conditions for the principles holding for anaphors and pronominals, expressible through restrictive and highly falsifiable parametric models (Manzini & Wexler Reference Manzini and Wexler1986). Clearly, we are very far from a picture of indefinite variability. Moreover, there is no variation at all in non-coreference effects: no language seems to allow coreference between a pronoun and a NP when the pronoun c-commands the NP (*He isaid that John iwas sick: Lasnik Reference Lasnik1989), a property which children are sensitive to as early as testing is feasible (Crain Reference Crain1991). Furthermore, referential dependencies are systematically regulated by the hierarchical relation c-command (Reinhart Reference Reinhart1976), not by other imaginable relations. In sum, we find highly restricted variation, or strict uniformity. The same kinds of considerations would apply to many other domains discussed by E&L (constituent structure, the status of subjects, etc.).
I certainly agree that language variation is of great interest for the cognitive sciences. Much promising research already focuses on that. Brain-imaging studies try to determine how the brain learns formal rules falling within or outside the range of attested cross-linguistic variation (Moro Reference Moro2008; Tettamanti et al. Reference Tettamanti, Alkadhi, Moro, Perani, Kollias and Weniger2002); acquisition studies uncover systematic “errors” in language development which correspond to structures possible in other adult languages, thus suggesting that children explore UG-constrained grammatical options which are not supported by experience (Rizzi Reference Rizzi, Deen, Nomura, Schulz and Schwartz2006; Thornton Reference Thornton2008); and so on. The cognitive sciences should not overlook the results of half a century of linguistic research which has seriously attempted to identify the limits of variation: it is simply not true that languages can vary indefinitely. This fundamental fact is of critical importance for addressing central questions of acquisition and processing, questions on the expression of the linguistic capacities in the brain, and on the domain specificity or generality of the relevant cognitive constraints. The astounding observational diversity of living organisms has not deflected biologists from the search for the fundamental common features of life. There is no reason why linguists and cognitive scientists should give up the theory-guided search for deep invariants and the ultimate factors of linguistic variation.
