To the editors:

Over the past twenty or so years, detailed comparative studies have revealed that important qualities observed in human languages are also present, in some form or another, in the communication systems of other species. Although nonhuman primates have been one obvious locus of interest, especially given their evolutionary closeness to our own species, relevant findings have also emerged from studies involving birds, mongoose, and even bacteria.¹

One especially interesting finding concerns the calls of chestnut-crowned babblers (Pomatostomus ruficeps).² The calls are comprised of individual elements that are perceptibly distinct within calls, perceptibly equivalent across calls, and convey no relevant information on their own. But when used in combination, these elements form constructions that do convey relevant information. This is the collection of qualities that characterize duality of patterning—also known as double articulation—a property many linguists have identified as foundational.³ Here, the elements are phonemes: individual sounds, such as /k/ and /b/. Like the elements of the calls of chestnut-crowned babblers, phonemes are perceptibly distinct within utterances, perceptibly equivalent across utterances, and convey no relevant information on their own. When they are used in combination they form constructions, known as morphemes, that do convey relevant information. /kju:b/ (‘cube’) is meaningful in English, even though /k/, /j/, /u:/ and /b/ are not.⁴ The discovery of the same collection of properties in the calls of a songbird is an important and intriguing development.

In his essay, Riny Huybregts argues that this discovery does not constitute duality of patterning proper, because for chestnut-crowned babblers the generation of meaningful units from meaningless elements is not truly open-ended. “In contrast to birdcall communication,” Huybregts writes, “the normal use of human language is creative and unbounded.” True enough, but the similarities are striking. And in any case, the systematic study of combinatorics in nonhuman species has only just begun and further discoveries are likely. Whatever the similarities and differences in this case, the question now and for the future must be, how should linguists interpret such discoveries and accommodate them in linguistic theory?

For the last few decades, evolutionary approaches to language have argued for a kind of Necker cube switch. The Necker cube is a familiar optical illusion that was first published in 1832 by Swiss crystallographer Louis Albert Necker. The illusion arises from the question of which panel represents the front of the cube. Two alternate perspectives are possible, and with a little bit of concentration the viewer can freely switch between them.

In the present discussion, the question can be restated as follows: are language users adapted to languages, or are languages adapted to language users? Famously, Noam Chomsky argued that because all human newborns succeed at the task of language learning—aphasias notwithstanding—despite the happenstance and limited data provided by their everyday environment, there must be some preexisting cognitive constraints on language form.⁵ These constraints structure the child’s learning and define the limits of language, which is otherwise unbounded. Steven Pinker, among others, tied this thesis to the insight that human cognition, like everything in biology, is subject to adaptation by natural selection; and in doing so he effectively articulated one perspective on the Necker cube. Human minds are adapted, via biological evolution, to be able to acquire and use languages.⁶

Yet this can all be switched around. From the other perspective, languages adapt to language users. Consider a surreal language in which, say, tense morphemes, such as English “-ed,” must be used at both the beginning and the end of the verb, and should also be phonologically reversed depending on the time of day—such that the past of, say, “want” was either “edwanted,” “dewanted,” “edwantde,” or “dewantde,” depending on whether it is morning, afternoon, evening, or night. Unwieldy constructions like this are more likely than others to be either dropped by the community, or adjusted and changed in the direction of greater practicality. Fields such as sociolinguistics, language acquisition, and language contact describe many real world cases. The perspective has now switched: language users are out there in the world, and languages adapt, via cultural evolution, to fit the needs and limitations of those users. This perspective on the Necker cube has become increasingly widespread and influential over the past two decades.⁷

Should we prefer one perspective over the other? With the Necker cube itself neither view can be said to be wrong in any objective sense, but in the case of languages there are asymmetries between the two perspectives which might provide good reasons to prefer one over the other. In particular, computational models suggest that the speed of cultural evolution effectively prevents the biological evolution of specific cognitive constraints on language form.⁸ If languages readily adapt to the needs and constraints of language users, then how could biological evolution ever cause language users to adapt to language form? How could that even happen, if cultural evolution does all the work first?

This Necker cube switch in turn suggests a new way to interpret the findings of combinatorics in other species. Perhaps what is being revealed by these studies is that the combinatorial apparatus employed in languages is nascent in many species—basic combinatorics are present even in bacterial communication⁹—and, as such, is available to be co-opted by any culturally evolving communication system, be it human languages, bird vocalizations, or any other. If so, duality of patterning and other recurrent features of the world’s languages are best seen not as pre-given foundations for language, but rather as emergent constructions that recur across languages simply because they are an effective means of doing what their users need them to, which is to be useful in communication and learnable by newborns.

In his essay, Huybregts effectively makes the same assertion that many do: that language makes unbounded expression possible. But here too things are, in fact, the exact other way around. The human capacity for truly unbounded communication makes languages possible as a means of communication and interaction.¹⁰ After all, linguistic communication is but one example of open-ended, unbounded human communication: there are also points, nods, and winks; spontaneous and improvised gestures; exaggerated displays of enthusiasm; and countless others. All can be used in myriad and unbounded ways.

Human communication is different. It is different not because of some special or distinctive feature of grammar, but because only humans naturally communicate by showing and reading intentions.¹¹ This is a highly cooperative, open-ended means of interaction that might be evolvable only in the sorts of social ecologies that characterize our species.¹² Duality of patterning—not to mention all the other recurrent features of natural languages—is an exploitation of the open-endedness of this means of interaction. The constituent parts of grammar are best seen as epistemic tools that allow us to show and read our intentions more effectively than we otherwise could; and which have evolved, culturally, to fit the various biases and needs of language users. Animal linguistics is revealing the extent to which we share these prior dispositions with other species.

Thom Scott-Phillips

Riny Huybregts replies:

The evolutionary linguist Thom Scott-Phillips offers a Necker cube perspective for a different approach to language. Necker cubes yield optical illusions of perceived orientation in space that do not exist in the real world. By contrast, the distinct biolinguistic and sociolinguistic perspectives on language are as real as any other research methodology and exist independently of one another. From a biological evolutionary perspective, language users may be said to “adapt to language form,” as Scott-Phillips writes. Biology would effectively guide language learning, thus avoiding the poverty of stimulus problem. From a cultural evolutionary perspective, languages may be said to “adapt to language users.” Only languages that are sufficiently expressive—or “useful,” as he writes—and learnable by newborns would survive.¹³ However, external language change and evolution of internal language are different and not mutually exclusive. Consequently, Scott-Phillips’s further suggestion that a cultural evolutionary perspective is to be preferred to the biological perspective is doubly mistaken. Both perspectives coexist, but the internal language system takes priority over its communicative use. Use of X function cannot exist ahead of X structure. While lungs are essential for breathing, breathing did not select for lungs in vertebrates. In fact, radically different lungs evolved in birds, who have flow-through ventilation, and mammals, who use reciprocating ventilation.¹⁴ Still, humans may apply different modes to use their biologically evolved respiratory system, such as abdominal breathing or chest breathing. This example would be a matter of cultural evolution in Scott-Phillips’s view.

Perspectives

The Necker cube is an illusion in the real world. The different perspectives of language sketched are real but affect different aspects of language. Languages are varied, complex along different dimensions, and change constantly. But the variety, complexity, and mutability are plausibly narrowly localized in lexicon and externalization, the systems and operations that yield the phonology, morphology, morpho-syntax, and word order of individual languages. The term cultural evolution would be a misleading label for historical language change, and does not yield any specific insight in the matter. There is no controversy here except that there is no generative, clear, explicit, formal theory of cultural evolution yet to elucidate the evolution of language use. In this sense, ancillary parts of language may readily adapt to the needs and constraints of language users, though not to their brains, the biologically constrained internal systems dedicated to language.

Externalization is external to the brain-internal generative language system, its basic property (BP), which yields an unbounded array of structured expressions for potential use in speech or sign.¹⁵ This capacity for unbounded language is genetically endowed in humans specifically. Its computational procedure, a biologically evolved property, is apparently unique, applying universally within the species. As far as can be determined, there has not been any significant mutation of this biological property since humans left Africa some 75,000 years ago. The BP seems to be a conservative property of the human brain inaccessible to cultural whims. In contrast, external language is volatile and prone to change.¹⁶

Did human brains adapt to acquire and use language? Prior to the human-specific mutations that led to the BP, there was no hierarchically structured language that could have selected for the BP. Rather it is the other way around. Structure dependence is a property of a biologically evolved BP, a crucial property distinguishing human language from nonhuman animal communication systems.¹⁷ Consequently, evolution of language is essentially a question of the evolution of the systems and operations that led to the BP. How were they genetically introduced into the organism? How were they reorganized to meet the conditions of physical law? How did natural selection, if in effect at all, reduce computational variety to binary Merge?¹⁸ These are the relevant questions about the evolution of language.

In his letter, Scott-Phillips wrongly conflates language with speech and communication. There is inner language without speech or communication, as when talking to oneself. For externalization, different sensorimotor modalities can be used, not just articulated speech. And externalized language can be used for communication as well as for other purposes. Conversely, communication can be nonlinguistic. It has been argued that the speed of cultural evolution would effectively preempt biological evolution of language form.¹⁹ But that doesn’t pose a dilemma. Scott-Phillips deals with externalized language, language change, and, in particular, the evolution of language use and communication—the topics of cultural evolution, if you will. Generative grammar deals with the underlying biological systems of internal language that yield linguistic products for possible use. Internal language and communication are different in kind: the first is biological and the second cultural. No selection for functional language, or communication, could give rise to an optimized computational system as used in internal language. It would be telling the story backward. The speech and communication of grown-up organisms cannot qualify or disqualify internal language and its initial state in growing organisms.²⁰

Explanation

Human language yields an unbounded array of hierarchically structured expressions and is essentially a recursive computational procedure.²¹ This basic property, also known as universal grammar (UG), is constrained by natural resource restrictions. The term UG, used in this way, is neither a grammar nor universal. It is the faculty of language, the human genetic endowment for language, the initial state that applies universally.²² It follows, for example, that the language Pirahã and its native speakers do not differ essentially from English and its native speakers. Both simply use the tools universally available in slightly different ways. Naturally, genes work within the texture of biophysical laws that “control the growing organism’s response to its genetic instructions.”²³ The genetics of language enables the human brain to organize the operations and structures that the laws of nature automatically provide. Language’s other secret is the third factor principles of computational efficiency that determine the simplest computational operation, binary set formation, as the universal response of the developing human organism to its genetic instructions.

Scott-Phillips’s proposal for a dual- or multiple-perspective approach to language is misplaced. There are not multiple answers to the BP. The problems these approaches are said to account for have nothing to do with structure dependence or the recursive procedure that generates hierarchical structure. Without the BP, there is no human language. While bacterial communication and human internal language have no intersecting properties, avian and nonhuman primate communication systems may share properties for externalization at the behavioral, cognitive, neural, and genomic levels. But these sensorimotor systems are external to the BP and ancillary to core language design, even in evolutionary terms.²⁴

Genuine explanations reach beyond descriptive adequacy, predicting what should be expected as well as what should never be—both the humanly possible and impossible languages. This property characterizes the enabling function of grammar, which, Chomsky writes, “provides options and systems for language that would have no reason to exist if language did not abide by the strong minimalist thesis.”²⁵ The simplest combinatorial operation, binary Merge, includes internal Merge,²⁶ predicting displacement and reconstruction effects to be naturally occurring phenomena of human language, as is confirmed. Moreover, constraints on extraction are grounded in general third factor principles of computational efficiency, such as phase impenetrability. The result is a genuine, minimalist, and purely formal account of some core traits of human language. Properties of cultural evolution, communicative function, or information structure have no role to play. They do not explain the BP or structure dependence, which is why and how language relies exclusively on structure and ignores word order altogether.²⁷ But it is primarily the nature of the BP, its enabling function, that is interesting here, not just its external manifestations.

Consider a simple linguistic experiment that manipulates structural positions of elements in exotic but expected constructions, such as parasitic gap (PG) and so-called knot sentences licensing extraction from relative clauses.²⁸ The enabling function of grammar precisely predicts these results under the proper structural conditions. Structure dependence is key to understanding.

Gaps may be parasitic on object gaps but never on subject gaps. The licensing asymmetry in these exotic constructions between subject gaps, e.g., *[He’s the sort of guy that –_SUBJ likes [every woman who knows –_PG ]], and non-subject gaps, e.g., [He’s the sort of guy that [every woman who knows –_PG likes –_OBJ ]], immediately follows from the enabling function of grammar.²⁹ This strong minimalist explanation replaces the earlier binding condition stipulation, which is also structure-dependent, that variables must not be bound by elements in argument position. Merge-based structure dependence is a crucial element in all of this. Enabling grammar leads to the expectation that these exotic cases occur—without specific stipulations—as possible extensions of non-exotic cases.

Another exotic case is extraction from relative clauses. Under specific conditions, selected violation of the complex NP constraint is allowed but only if the subject has been relativized. This is shown, for example, by This is a paper that there’s no one who can explain (cf. there’s no one who can explain it), contrasting with *This is a guy that there’s no problem which can’t explain (cf. there’s no problem which he can’t explain). Again, the subject versus object asymmetry falls out automatically from enabling grammar, with vacuous movement, an option that is universally available and a crucial element in the explanation. The empirical predictions extend to asymmetries within the parasitic gap clause itself, shown by the contrast between Here’s someone that everyone [ who [ — meets (him) ]] will like — immediately and Here’s someone that everything [ which [ *(he) breaks — ]] will upset — deeply. Here the object of “meets” cannot refer to “someone” when it is overt but must when covert. In contrast, “he” cannot be covert and cannot refer to “someone.” There is no explanatory account of such empirical facts in terms of information structure, relevance theory, intention reading, communicative efficiency, pragmatics, or cultural evolution. These exotic variants naturally fall out, without the need for stipulation, from non-exotic standard cases, assuming the strong minimalist thesis and restriction conditions on organic resources. Despite claims to the contrary, such as those made by Michael Tomasello,³⁰ the notion of UG is far from dead. Reports of its passing are greatly exaggerated.

Combinatorial Communication

Unicellular bacteria communicate efficiently using quorum sensing, the bacterial cell-to-cell communication mechanisms that coordinate gene expression and depend on the synthesis of signal molecules that diffuse in and out of bacterial cells.³¹ Similarly, plants communicate symbiotically with arbuscular fungi through their root structure. In fact, symbiosis enabled early plants to colonize newly emerged land some 450 million years ago.³² The underlying mechanisms are quite clear, and the use of the term communicate in these instances would be metaphorical only.

But Scott-Phillips goes further than this and argues that communication in the bacteria P. aeruginosa is combinatorial, as it is in human language and the call systems of some nonhuman primates such as Campbell’s monkeys.³³ Two coexisting biochemical signals yield a gene expression effect that is different from the expressions that result from sensing each single signal separately, or no signal at all. It is suggested that there may be evolutionary implications for combinatorial communication in human language. But these are entirely different, unconnected systems—the molecular biology of these mechanisms is well-known and combinatorial communication has a specific biochemical interpretation. Form and function should be sharply distinguished here as well. Bacterial quorum sensing and human internal language are different organic systems with potential communicative functions that are species-specific. Once again, the superficial similarity is only suggestive; one metaphor is piled on another.³⁴

Incidentally, combinatorial or not, the combined signal molecules yield a novel response that is idiomatic rather than compositional, much like “kick” and “the bucket” combining to yield a non-compositional idiomatic expression. In this case, the expression means “die” instead of its non-idiomatic meaning, a compositional function of the meanings of verb and noun phrase and their syntactic organization, on a par with the semantics of “hit the pail.”

There are no evolutionary implications for the BP of human language in any of this. The terms communication, language, and combinatorial meaning are employed here broadly and loosely without explanatory content. But communication and language are different in kind. Not every communication system uses language as its medium or has the BP as its generative engine. At least, nonhuman animals and bacteria do not in their communication systems. The former may share properties of production and perception with humans; the latter do not. These potentially shared properties have relevance in externalization only. Significantly, structure dependence is absent in all nonhuman communication systems, but it has many phenotypic effects in all constructions and in all languages. The evolution of human language is essentially the evolution of the BP and the combinatory operation from which structure dependence follows.

For Morten Christiansen and Nick Chater, language is likewise easily learnable and usable because language has adapted to human brains.³⁵ Languages take on the form of systems that are “easy to learn” and are “sufficiently expressive.” Hence, the slogan “language as shaped by the brain.” But this leads to the question posed by Stevan Harnad of “why and how human brains acquired an evolved capacity to learn all and only UG-compliant languages in the first place.”³⁶ Again, it is rather the other way around. Humans are predisposed to automatically, reflexively learn only some languages, namely those languages that can be generated by their inborn generative engine, the BP, thus circumventing the poverty of stimulus problem. In fact, other languages that do not conform to the BP seem to be impossible for humans to learn, such as languages with only linear structure and string-based positional dependencies.³⁷ Objections to the BP raised by Simon Kirby and colleagues on the basis of their diffusion-chain experiments are similarly misconceived and can be countered analogously.³⁸

The term cultural evolution is misleading because languages do not evolve, only biological organisms do. Languages change but changes are restricted to their externalizations. The term may plausibly refer to elements external to the BP, in particular the use of spoken or signed language. In contrast, the BP of language, though widely contested, is a property of human biology, an organic system and subject to genetics and biophysical law. The capacity for this internal language emerged through mechanisms of biological evolution and, apparently, did not change further to yield any group differences among human populations. It is essentially a highly conserved system. In contrast, after its emergence in the biological world, the external use of the language system resulted in the variety, diversity, and complexity of the world’s languages. This is cultural evolution. Unsurprisingly then, what catches the eye is the changeability of externalized language. But what is basic are the hidden systems and operations of internal language, in particular the generative procedure that yields discrete and digital infinity in ways that are uniquely human and uniquely linguistic. From an initial state determined by the human genome, internal language grows in the developing organism. While structure and function are linked topics of inquiry, structure has epistemic priority over function, the former presupposing the latter. The study of language should at least attempt to explain evolution, development, and physiology of its basic property. Of course, language change and use of language remain legitimate topics of study that deal with choice of parameter values, properties of individually realized languages, and externalization to the sensorimotor systems. But, from a biological perspective of the human language capacity, they deal with ancillary aspects of language, peripheral to its BP. Hence, there is no culture in evolution, our answer to the question.