Though Evans & Levinson (E&L) cite us as foils, we agree with many of their points: that the documentation of linguistic diversity is important for cognitive science; that linguists have been too casual in assuming universals and formulating a defensible theory of universal grammar; that the language faculty is a product of gene-culture coevolution; that languages are historical compromises among competing desiderata; and that cross-linguistic generalizations are unlikely to be exceptionless. We do, however, endorse a version of the Universal Grammar (UG) hypothesis: that the human brain is equipped with circuitry, partly specific to language, that makes language acquisition possible and that constrains human languages to a characteristic design.
E&L do not make it easy to evaluate their position. Their claim that linguistic variation is “extraordinary,” “fundamental,” “amazing,” and “remarkable” is not just unfalsifiably vague but also myopic, focusing on differences while taking commonalities for granted. Any survey that fails to consider the larger design space for conceivable languages is simply unequipped to specify how “remarkable” the actual diversity of languages is. Consider these hypothetical languages:
Abba specifies grammatical relations not in terms of agent, patient, theme, location, goal, and so on, but in terms of evolutionarily significant relationships: predator-prey, eater-food, enemy-ally, permissible-impermissible sexual partners, and so on. All other semantic relations are metaphorical extensions of these.
Bacca resembles the two-word stage in children's language development. Speakers productively combine words to express basic semantic relations like recurrence, absence, and agent-patient, but no utterance contains more than two morphemes. Listeners use the context to disambiguate them.
The speakers of Cadda have no productive capacity: they draw on a huge lexicon of one-word holophrases and memorized formulas and idioms. New combinations are occasionally coined by a shaman or borrowed from neighboring groups.
The grammar of Daffa corresponds to quantificational logic, distinguishing only predicates and arguments, and using morphological markers for quantifiers, parentheses, and variables.
Fagga is a “rational” language spoken in a utopian commune. It lacks arbitrary signs larger than a phoneme, allowing the meaning of any word to be deduced from its phonology.
Gahha is a musical language, which uses melodic motifs for words, major and minor keys for polarity, patterns of tension and relaxation for modalities, and so on.
These and countless other conceivable languages are not obviously incompatible with cognitive limitations or communicative limitations (unless stipulated post hoc), and fall well outside the envelope reviewed by E&L.
Indeed, E&L concede an enormous stratum of universals in endorsing an innate core of Hockett-syle “design features”: discreteness, arbitrariness, productivity, duality of patterning, auditory-vocal specialization, multiple levels of structure between sound and meaning (including morphosyntax and semantics), and linearity combined with nonlinear structure (constituency and dependency). But if this set of universal mechanisms were laid out explicitly, as computational machinery consistent with the details of actual languages, we doubt that it would differ “starkly” from the kind of Universal Grammar that the two of us have posited (Jackendoff Reference Jackendoff1997; Reference Jackendoff2002; Pinker & Jackendoff Reference Pinker and Jackendoff2005; Culicover & Jackendoff Reference Culicover and Jackendoff2005).
E&L's focus on differences of detail, moreover, leads them to exaggerate differences in underlying design. Whether a language's inventory of phonemes or verbs is large or small, whether grammatical agreement plays a role that is indispensable or vestigial, whether a noun-verb difference is obvious to a linguist in every language or subtle in a few, whether grammatical combination is florid inside phrases but rudimentary inside words, or vice versa, whether embeddings are found in a high or a low percentage of phrases – none of these pertains to underlying computational systems, just the extent to which different parts of it are deployed (which we agree depends on the vagaries of history).
The force of many of E&L's examples depends more on having a splitter's (rather than a lumper's) temperament than on the existence of “extraordinary” variation. English, for example, has phenomena similar to Chinese classifiers (e.g., a piece of paper, a stick of wood), Athabaskan verb distinctions (among locative verbs; Levin Reference Levin1993; Pinker Reference Pinker1989; Reference Pinker2007), ideophones (response cries such as yum, splat, hubba-hubba, pow!; Goffman Reference Goffman1978), and geocentric spatial terms (e.g., north, upstream, crosstown; Li & Gleitman Reference Li and Gleitman2002). There are differences, to be sure, but it is unparsimonious to insist that they lack any common psycholinguistic mechanisms.
E&L try to minimize their own commitment to Universal Grammar in two ways. First, they try to trivialize it as “true by definition,” claiming that any system lacking these features would not be called a “language.” But that is both factually false (the word “language” is commonly applied to bee dance, ape signing, computer programming protocols, etc.) and theoretically misleading. The proposition that all human societies have “language” according to that definition is neither circular nor trivial.
Second, the authors note that these features are functional, as if that suffices to explain how children acquire them. But functionally useful structures do not materialize by magic. Even after they have developed historically as part of a tacit convention within a community of speakers, a child still has to acquire them, including the myriad grammatical details which do not perceptibly contribute to that child's ability to communicate but are simply part of the target language (semi-regular inflectional paradigms, arbitrary grammatical categories, complex co-occurrence restrictions, etc.). This requires sufficiently powerful learning abilities, which have evolved as the “gene” side of the “gene-culture coevolution” we all embrace.
The construct “Universal Grammar” (UG), as we have invoked it, refers to these learning abilities. UG is emphatically not a compendium of language universals (though it has often been taken to be such). UG, in our conception, is a toolkit. This implies that it does not, as E&L claim, contain all possibilities for all languages: one builds structures with tools, not from tools. We have suggested that UG may be a set of “attractors” for grammars (Culicover Reference Culicover1999; Culicover & Jackendoff Reference Culicover and Jackendoff2005; Jackendoff Reference Jackendoff2002) – strong biases on the child's learning that produce statistical tendencies in grammars, but that can be overcome by exceptional input, a possibility that E&L endorse.
These learning mechanisms are a black box for E&L, who suggest that they can be borrowed off the shelf from other domains of cognition. But they do not cite any analysis showing that some independently motivated set of cognitive processes actually suffices to acquire the detailed and complex structures mastered by children. In practice, attempts to model specific phenomena of language development using generic information-processing machinery (notably, connectionist networks) habitually smuggle innate linguistic structure in the back door in the form of unheralded design features and kluges (Marcus Reference Marcus2001; Pinker & Ullman Reference Pinker and Ullman2002a; Reference Pinker and Ullman2002b; Reference Pinker and Ullman2003).
For all these disagreements, we applaud E&L's call for a more rigorous study of linguistic variation and universals. When combined with rigorous specification of the psychological mechanisms that make language acquisition and use possible, we predict that the resulting psycholinguistic universals that give rise to the diversity of human language will be far from “unprofound.”
