A rough heuristic can be deployed in comparative approaches to learning: If a behavior is fixed, it derives from the organism (e.g., early genetic or epigenetic processes); if it is moderately variable, it is learned, although possibly with constraints; if it is highly flexible, it is acquired by highly context-sensitive learning processes. Language may be the quintessentially flexible behavior. So, why have classical approaches focused so heavily on what might lie within the organism?

Evans & Levinson (E&L) offer a remarkable gift to cognitive science. In their sweeping synthesis of typological linguistics and their daring criticism of prevailing orthodoxy, they push us to focus on the flexibility and prowess of the language capacity, if there even is such a thing anymore.

Chomsky proposed Universal Grammar (UG) to ease the learning of language; yet, thanks to E&L, we see that in many cases even basic assumptions like syllabification or word order might set children learning certain languages down the wrong path. Although it is possible to soften UG by positing limits on the form of languages (as opposed to features that languages must have), even this maneuver may fail. Swedish, for example, violates the hallowed “size principle” in phonology: using two features to represent a binary contrast for which one would suffice (Helgason & Ringen Reference Helgason and Ringen2008). Given the linguistic data described by E&L, we suspect that a UG based on limits will fare no better than one based on necessary features.

E&L, by questioning whether anything about language is universal (and therefore could be properly deemed innate), compel us to consider a flexible learning process that is capable of giving rise to a great deal of emergent complexity. So, a language learner may rely on a rich, redundant set of cues, using seemingly irrelevant speaker information to acquire phonology (Rost & McMurray Reference Rost and McMurray2009) and seemingly irrelevant phonological detail to acquire syntax (Christiansen et al. Reference Christiansen, Onnis and Hockema2009), all in the context of a reinforcement environment that surpasses that of all other species in complexity.

Therefore, why do only humans acquire language? Hold-outs might argue that disparities in basic learning processes among species (Newport et al. Reference Newport, Hauser, Spaepen and Aslin2004) are responsible. And, indeed, there is longstanding and compelling evidence that animals differ in basic, although perhaps parametric ways (Bitterman Reference Bitterman1975). Yet, other than Skinner (Reference Skinner1957), no other learning theorist has attempted to apply basic learning processes to language acquisition as a whole (although connectionist approaches have begun to address this piecemeal; e.g., Christiansen & Chater Reference Christiansen and Chater2001).

One might also observe that basic cognitive capacities like categorization (Lazareva & Wasserman Reference Lazareva, Wasserman and Menzel2008), same-different discrimination (Cook & Wasserman Reference Cook, Wasserman, Wasserman and Zentall2006), sequence learning (Gentner et al. Reference Gentner, Fenn, Margoliash and Nusbaum2006), and expectation generation and evaluation (Keller & Hahnloser Reference Keller and Hahnloser2009) can be seen in various forms among diverse species. Yet, exhibiting many of these complex forms of cognition, all necessary for language, may be insufficient to move animals to a level of communication performance comparable to humans.

So where does that leave us? When confronted with such diversity of possible language inputs, one might first look for even more powerful learning mechanisms. But, perhaps language is not the product of a single mechanism. There is tremendous diversity of learning and cognition among animals (Wasserman & Zentall Reference Wasserman and Zentall2006). Finding one simple mechanism across many different species is highly improbable (Bitterman Reference Bitterman1975). Given the diversity of human language so richly detailed by E&L, perhaps the parallel holds across humans.

This possibility would seem at first blush to undermine the view of humans sharing fundamental cognitive capacities. Yet, perhaps that is not so. Language may not be a monolithic structure to be internalized. Rather, language may be an ensemble of skilled behaviors that are shaped by a variety of forces and are sensitive to any available source of information. In this light, humans (much like evolution) assemble language with a variety of learning mechanisms and sources of information, this assembly being guided by the particularities of the language they are learning.

Learning even something as restricted as the phonology of English may demand mechanisms as diverse as unsupervised clustering (McMurray et al. Reference McMurray, Aslin and Toscano2009), dimensional weighting or cue neutralization (Rost & McMurray Reference Rost and McMurray2009), anticipatory learning of transitions (Elman Reference Elman1990; Saffran et al. Reference Saffran, Aslin and Newport1996), top-down feedback from the lexicon (Thiessen Reference Thiessen2007), and nonspecific social contingencies (Goldstein et al. Reference Goldstein, King and West2003). Perhaps none of them is necessary for mastery of all languages; but most of them are necessary for any given language. Still, none of these learning mechanisms could operate without a rich information source consisting of overlapping categories, information buried in the relative variation in particular cues, sequential relationships between sounds, and vertical relationships between sound and meaning. These relationships are provided by the language and the culture. Although this may appear circular, as is so often the case in evolution, it is merely evidence of coevolving systems. Language is not a structure to be internalized; rather, it is an extremely complex assembly of behaviors to be developed.

Is language a thing? Or is it a skill? If it is a skill, we should not be studying “structure” at all, but rather, studying moment-by-moment behaviors. Evolution does not select for genes, it selects for phenotypes, or rather, for the developmental processes that give rise to them (Lickliter & Honeycutt Reference Lickliter and Honeycutt2003). Similarly, in language, the fundamental unit is not a piece of structure – it is an ongoing behavior, the learning processes that give rise to it, and the cultural factors that make it useful. So what is language? It is a behavior … that develops … in a culture … that develops …

The evidence and argument of E&L have freed developmental theorists from having to explain the abstract and arcane of narrow linguistic theories; language can be virtually anything. The authors have moved us to a more productive consideration of what is central to language: how language is used and acquired, and what flexible and emergent mechanisms are needed to accomplish this remarkable behavior.