From a distance, the structures of the world's languages do look gloriously diverse and endlessly varied. But since when is it sound strategy to take diversity at “face value”? All other sciences have progressed precisely by taking nothing at face value – diversity included. Evans & Levinson (E&L) claim, in effect, that linguistics is different from all other fields. If they are right, the search for deeper laws behind linguistic structure is a fool's errand, and languages are just as inexplicably diverse as they seem at first glance. It is thus surprising that E&L's article contains no discussion of the actual research to which they supposedly object. Instead, their article offers only (1) a parade of capsule descriptions of phenomena from the world's languages, coupled with (2) blanket assertions that each phenomenon falls outside the scope of all (non-vacuous) general theories of linguistic structure. The argument must therefore rest on two premises: (1) that the capsule descriptions of phenomena are correct, and (2) that these phenomena fall so obviously beyond the pale of research on Universal Grammar (UG) that no argument is necessary. I believe there are reasons for caution about the first supposition and that the second is simply wrong.

Confidence in the accuracy of E&L's examples is undermined at the outset by a comparison of E&L's claims with their sources. For example, as a demonstration that “semantic systems may carve the world at quite different joints” (sect. 2, para. 2), E&L cite Mithun (Reference Mithun1999, p. 81) for an “unexpected number” marker (sect. 2.2.5, para. 3) in Jemez (misidentified by E&L as Kiowa). It is supposedly because “two” is an expected number of legs but not stones that the marker means “any quantity but two” when added to “leg” but “exactly two” when added to “stone.” According to Mithun, however, the actual determinant is the Noun-Class to which the noun belongs, a partly arbitrary classification (like Indo-European gender). In Jemez, “nose” belongs to the same class as “leg,” and “arm” belongs to the same class as “stone.” Consequently, if the number marker truly indicates how Jemez speakers “carve the world,” “one” must be an unexpected quantity of noses, but an expected quantity for arms (cf. Harbour Reference Harbour2006). Similarly, although E&L use Mundari ideophones to exemplify a “major word class” whose very existence “dilutes the plausibility of the innatist UG position” (sect. 2.2.4, para. 10), their source, Osada (Reference Osada1992, pp. 140–44), actually does not discuss ideophone syntax at all. Mayan positionals are mentioned for the same reason – but although they do constitute a distinct category of root (used to derive words belonging to standard classes), they do not appear to constitute a syntactically distinguishable word class – at least not in Mam (England Reference England1983, p. 78), Tzotzil (Haviland Reference Haviland1994, p. 700), Chol (Coon & Preminger Reference Coon, Preminger, Avelino, Coon and Norcliffe2009), or Tzeltal (Brown 1994, p. 754; cited by E&L).

But even if E&L's sources had been quoted perfectly, there is a deeper danger in arguments founded on capsule descriptions. Because linguistic puzzles are complex, even the best descriptions are incomplete. Fundamental generalizations often remain undiscovered for years, until a creative researcher suddenly asks a question whose answer unlocks one of the language's secrets. And how does one find such questions? As it happens, among the most productive question-generating devices is the very idea that E&L deride: that languages share structural properties, often masked by one of their differences. Again and again, investigations guided by this possibility have altered our basic picture of the world's languages.

Compare E&L's assertion that “syntactic constituency … is not a universal feature of languages” (sect. 5, para. 8), supposedly supported by the existence of languages with “free word order,” with the alternative – that such languages do not fundamentally differ from others in basic constituent structure, but merely allow these constituents to be reordered more freely. Although E&L attribute the appeal of this alternative to mere anglophone prejudice, its actual appeal rests on a body of hard-won results, achieved by researchers pursuing lines of inquiry that no one previously thought to pursue.

One recent example is Cable's (2007; 2008) investigation of wh-questions in Tlingit, a language of Alaska with considerable word order freedom. In languages with more rigid word order, there is a well-known split in how they form wh-questions. In languages like English and Italian, interrogative wh-phrases move to a dedicated position near the left periphery of the clause, precedable only by discourse topics (cf. This book, who would read it?; Rizzi Reference Rizzi and Haegeman1997). In languages like Chinese and Korean, by contrast, there is no dedicated position for wh-phrases. Instead, they appear wherever their noninterrogative counterparts would:

1. (1)

   1. (a) English

      Who did John see?

   2. (b) Chinese (Huang Reference Huang1982)

      Zhangsan kajian-le shei

      Zhangsan see-asp who

At first glance, Tlingit appears to pattern with Chinese rather than English in lacking a dedicated wh-position, as expected if free word order entails absence of constituent structure:

1. (2) Tlingit (Cable Reference Cable2007; Reference Cable and Matthewson2008)

   1. a. Yá x'úx' aadóoch sá kwgwatóow?

      this book who-erg Q he.will.read.it

      “Who will read this book?”

   2. b. Aadóoch sá yá x'úx' kwgwatóow?

      who-erg Q this book he.will.read.it

      (= a)

   3. c. Kéet  daa sá  axá?

      killer.whale what Q  he.eats.it

      “What do killer whales eat?”

On the basis of his fieldwork and analysis of Tlingit texts, Cable showed that, contrary to appearances, Tlingit wh-phrases do occupy a dedicated wh-position – in fact, a position syntactically and semantically identical to its counterpart in languages like English. Specifically, Cable discovered that phrases to the left of an interrogative wh-phrase – “this book” in (2a) and “killer whale” in (2c) – are always discourse topics, and if a wh-phrase appears to the right of the verb, the result is not interpretable as a wh-question. These findings are explained if the interrogative phrase in both Tlingit and English must occupy the same position at the left periphery, precedable only by a discourse topic – that is, if this aspect of Tlingit clause structure resembles English, despite first impressions. Cable's findings constitute a discovery, not anglophone prejudice. Earlier work on Tlingit, though excellent, had failed to spot the evidence for a wh-position. Cable's discoveries were made precisely because his investigation was informed by what linguists had learned about wh-questions in other languages.

Remarkably, the same type of discovery has been reported by Legate (Reference Legate and Pica2001; Reference Legate2002, building on Laughren Reference Laughren and Amberber2002) for Warlpiri, the free word order language par excellence. Legate not only found the same dedicated wh-position in Warlpiri (precedable by topics) but also showed that the full array of standard tests for phrase structure in languages like English yield the same results for Warlpiri – important findings directly attributable to the very research program that E&L consider bankrupt.

Questions like those addressed by Legate and Cable must be asked for every item in E&L's linguistic Wunderkammer. Are they truly as “different” as they seem? Sometimes, of course, the answer might be “yes.” But how will we ever distinguish real differences from those that merely reflect our ignorance if we grant – even for an instant – E&L's dictum that apparent structural differences must just “be accepted for what they are” (sect. 1, para. 1)?