Does thought influence language? Well, at times, ...

The extent to which language influences thought is an old and ongoing debate in cognitive science. I wish that the debate would show equal concern for letting thought influence language. In a recent article (Wall Street Journal), Lera Boroditsky (cool website) outlines some recent research in her field: it is imaginative and fascinating and there’s no doubt in my mind it makes substantial and significant contributions. However, I was struck by Boroditsky’s attempt to situate her work against the purported failure of Chomsky’s approach to language. Tying one’s work to one of the central figures of cognitive science adds caché, especially if one is overturning what that figure has said. But what the article presents reveals a fundamental misunderstanding of what language is and which aspects of it generative linguistics attempts to study.

Boroditsky writes: “Dr. Chomsky proposed that there is a universal grammar for all human languages essentially, that languages don’t really differ from one another in significant ways. And because languages didn’t differ from one another, the theory went, it made no sense to ask whether linguistic differences led to differences in thinking. // The search for linguistic universals yielded interesting data on languages, but after decades of work, not a single proposed universal has withstood scrutiny.” This not only misrepresents universal grammar and its investigation but also misconceives which aspects of language Chomsky et al’s versus Boroditsky’s et al’s research relates to.

Chomsky’s proposal is that languages do not significantly differ from each other in terms of narrow syntax, that is, in terms of syntactic algorithms and the set of primitives over which those algorithms operate. The kinds of universals that arise with this approach are statements like:

• All branching is maximally binary.
• Extraction is only possible from phase edges.
• Minimality of attraction is relativized to particular features.

What Boroditsky has in mind (I’m willing to bet) are universals of the form:

• Any language with head-internal relative clauses permits null subjects.
• Any language that is verb-initial has construct states.
• All languages have pronouns.

The difference between these two types of universals is massive: one concerns the deep infrastructure of language, the other, the superficial description of its output. (This misunderstanding also afflicts a much discussed paper by Evans and Levinson, BBS, and I wonder whether it represents a prevalent misapprehension.) However, to anyone who understands what generative linguistics is concerned with, it is wholly obvious that the entire research program of universal grammar could be successfully completed without there being a single universal of the latter (superficial) variety. Universals of output simply aren’t the universals that universal grammar is concerned with and unless Boroditsky knows something surprising about binary branching, long-distance extraction, and relativised minimality, her claim that no universals have “withstood scrutiny” “after decades of research” is simply untrue.

That said, although universal grammar does not require superficial universals, it should be emphasised that universals derived from the nature of the primitives over which syntax operates (rather than from the syntactic algorithms themselves) remain in good health “after decades of research”: feature systems with a dual still differentiate singular from plural, and those with a trial still differentiate singular, plural and dual; feature systems with a three-way person contrast still conflate first inclusive with first exclusive, not with second person; and there is still no choric-hearer or choric-speaker feature system. So, universal grammar doesn’t need superficial universals, but it has them anyway.

Yet, even if Boroditsky’s motivational critique of Chomsky is wrong, that hardly means her work is without motivation. The domains of study of the two approaches are substantially without overlap. Consider some of Boroditsky’s examples.

• English, Hebrew and Pormpuraaw speakers may arrange pictures differently to reflect chronological sequence, but this tell us nothing about how narrow syntax works, because representation of chronological sequence has nothing to do with binary branching, extraction, etc. The irrelevance of this data is underlined by Boroditsky’s own observation that the Hebrew versus English orderings reflect the directions in which the two languages are written: writing systems are means of representing the output of universal grammar, parametrized to a given language, after it has passed through the morphological, phonological and phonetic systems; they are very distant from the study of universal grammar itself.

• Lexical facts—such as whether Russian speakers, who lack a cover term for sky blue and navy blue, are more attuned to different shades of these colours than are English speakers to whom they are both blue—are also irrelevant to how syntax may construct long-distance dependencies or whether syntactic branching is binary. Much variation in the lexicon fails to impact on syntax: elks are nothing like elephants, but one struggles to find syntactic ramifications of this fact.

• English and Japanese both have constructions that identify and ignore agency, but their different rates of usage represents a sociological fact about language usage, not anything to do with differences in the narrow syntax. If your society generally expects you to impute agency when speaking whatever language it uses, then, it turns out that you are more attuned to agency. This is not overly surprising: when my lifestyle includes a lot of violin playing, my intolerance for others’ off-key playing increases. Speaking English is part of my lifestyle, so why should the set of distinctions that being a member of an English-speaking society routinely makes me make not be just like having perfect pitch? The propensity of English speakers to identify causes reflects a sociolinguistic fact, not something that universal grammar is concerned with.

It is irritating that such errors of logic and such misrepresentations of generative linguistics should have been made in so public a forum as the Wall Street Journal. What’s more worrying to me, however, is the persistent gap in understanding by other cognitive scientists as to what universal grammar is about, what evidence could falsify it, and how other research enterprises may pertain to language and yet be orthogonal to it.