Book announced at https://linguistlist.org/issues/32-3531

Title: Reflections on language evolution
Subtitle: From minimalism to pluralism
Series Title: Conceptual Foundations of Language Science
Publication Year: 2021

Publisher: Language Science Press
http://langsci-press.org
Book URL: https://langsci-press.org/catalog/book/142

Author(s): Cedric Boeckx

Reviewer: Conor Houghton

SUMMARY

Cedric Boeckx’s monograph Reflections on language evolution: From minimalism to pluralism can be read as a useful reminder of the need for more diversity in our approach to linguistic research. However, its central thesis is more ambitious and controversial: it seeks to reformulate the field by suggesting that linguistics should be centered on the problem of how our language ability evolved. It also suggests that this language ability is best studied as a collection of individual skills and that the method best suited to this problem is paleo-genetic evidence.

EVALUATION

Somewhere in our evolutionary history we started to use language. Reflections on language evolution: From minimalism to pluralism [1] identifies this as the central problem of linguistics. Moreover, in Boeckx’s view, this problem is not about a single advance or a particular skill which evolution stumbled upon, language is an accumulation of many parts:
“I find it more useful to think of our language capacity as a collection of (generic) cognitive biases put to the task of acquiring and using “an art” as Darwin defined our language instinct.”
Each element, each grain in this heap of “cognitive biases”, has an evolutionary story and each can be studied using its own subfield while remaining directed towards the central question of language evolution, referred to in the monograph as the Darwin Problem. Largely, the subfield considered is paleogenetic-linguistics. Paleogenetic-linguistics took off with the discovery of FOXP2, once called the grammar gene though recent research has revealed it is not specific to grammar. Boeckx describes the discovery as
“arguably one of the most significant achievements in the language sciences in the past twenty five years”
and suggests that it demonstrates that genetic evidence can play a significant role in solving the Darwin Problem.

It is interesting to consider the paleogenetic-linguistics of FOXP2 as it shows both what is fascinating and, ultimately, what is frustrating about this approach. FOXP2 is highly expressed in brain regions critical for language [2]. The human version of the gene is shared with Neanderthals, [3], whereas other primates have different versions. This is interesting because the Neanderthals might have had language. However, we are not really sure. Since Boeckx believes in the heap theory of language, he is able to take an intermediate view and has suggested that there is evidence for vocal learning in Neanderthals, but not necessarily evidence for a full language [4]. The problem is that we are left unsure if FOXP2 is evidence for Neanderthal linguistic ability, or that the evidence for vocal learning in Neanderthals tells us something about FOXP2.

Generally, the debate about language evolution is a debate about the nature of language. Chomsky has suggested that the evolution of language was discrete, sudden, and serendipitous [5]. This arises from Chomsky’s belief that language is rooted in a human capacity for recursion emerging from a single, highly-advantageous mutation; what we might call the monolith theory, in contrast to Boeckx’s heap theory. The monolith theory leads to odd views on evolution: Chomsky is skeptical of standard selectionist and gradualist evolutionary models for language. His opposition to selectionism has been broadly criticized, most famously in Pinker and Bloom [6]. The debate about gradualism is more subtle. An adaptation can create its own landscape: some changes can act like traversing a narrow ridge that leads to a new set of steep slopes along which the species rapidly ascends. A seemingly small change can open up new opportunities, including, in the case of language, opportunities created by increased learning and cooperation [7, 8]. If this is the case it renders the distinction between the monolith and heap theories less distinct: language could be a set of skills that accumulated rapidly following a single key adaptation. Indeed, my own view is that our language ability represents the unfurling of a single event: the internalization of utterance into a tool for thought, as suggested in [9].

The difficulty is that the Darwin Problem is an historical question and so much of the evidence is historical, relying on artefacts rather than experimental study. Boeckx suggests that more direct, experimental evidence can be found if we are willing to imagine a linguistic equivalent of the recapitulation principle, known by the slogan ontogeny recapitulates phylogeny. We are usually scrupulous in distinguishing the Darwin Problem from what could be called the Schleicher Problem, the question of what rules govern the evolution of the languages we speak. However, in Boeckx’s view, the two problems are related and the evolution of languages from generation to generation recapitulates the evolution of our language ability: the Schleicher Problem restates the Darwin Problem! In practice, the approach Boeckx has in mind is the pioneering work by the language evolution group in Edinburgh. In their experiments, languages are evolved in the laboratory by asking participants to communicate using made-up words, they then pass on the makeshift languages to a new set of participants and this is repeated, creating a kind of generation-to-generation evolution of language [10, 11].

If we accept the heap theory, these laboratory-evolved languages may exploit the individual elements of evolution in turn, recapitulating evolution. However, while participant-based iterated-learning experiments do elegantly tease apart different aspects of language, it will be difficult to integrate insights from iterated learning with paleo-genetics because it seems difficult to decide if elements of language identified through participant studies correspond directly to specific interpretable genetic changes discovered by comparisons between our genes and those of other hominins.

There is a broader challenge. When we study an emergent phenomenon it is common to equate importance with a hierarchy of causality; to assume that since genetics contains the template for the brain and the brain creates language, then genetics must be the best level for the study of language. However, the most accessible and explanatory structure in an emergent phenomenon can arise at any level and, while it is tempting to privilege genetics because it is so low in the hierarchy, it is going to be extremely difficult to answer questions about language from the perspective of genetics because there are so many steps in between. Boeckx frames the issue differently. By centering the Darwin Problem and defining linguistics as the study of the evolutionary history of language, he suggests that paleo-genetic evidence is among the only concrete evidence available. In this view, emphasizing genetics does not reflect a bias toward the lowest scale of inquiry; it is a way to match the method to the subject. Thus, first: linguistics is the study of evolution and, second: evolution is best studied using genetics.

Paleogenetic-linguistics is without doubt a valuable subdiscipline in linguistics; the question is whether the Darwin Problem and, by association, paleogenetic-linguistics, should be the core preoccupation of linguistics. Since the Chomskyan revolution linguistics has been centered on uncovering the abstract structures of language. In Chomsky’s original formulation, language is the expression of an innate, biologically-driven predisposition in the brain to recognize and generate particular syntactic forms [12, 13]. This innate expectation allows us to learn language as children and means that language has a universal structure.

We are still, in reality, untangling Chomsky’s radical reformulation of linguistics. This book is a brave attempt to perform this task in a radical manner by proposing we move on and reinvent the mission and methods of linguistics: by proposing the Darwin Problem as the central concern of linguistics, by asserting that language is an accumulation of skills, and by promoting paleo-genetics as the pre-eminent tool of the field. This seems premature. The questions posed by the Chomskyan revolution remain unresolved, either positively or negatively. The unanswered questions include the existence of a language acquisition device, the relationship that this putative device has to universal properties of language, the arbitrariness of these properties and the utility of trying to formalize them using a generative grammar. While some of the original ideas may no longer be as useful as they once were, the insight that identified language structure, universal properties, and language acquisition as the key preoccupations of linguistics remains important.

In fact, Boeckx seems to object to the wrong part of Chomskyan linguistics. He suggests that the Chomskyan programme has succeeded in describing language through formal grammars, referring to this as an “underappreciated consensus”, while rejecting the Chomskyan focus on language as the subject of linguistics. We are still unsure how to formulate language to match our internal processing. Transformational grammars are an elegant way to study formal systems like computer code where multiple levels of recursion are common and the rules governing recursion must be strictly enforced. However, they are an inefficient and unparsimonious approach to human language in the everyday performance of communication. For example, they require multiple caveats and a duplication of categories to capture the rules about agreement that languages seem to carelessly add to their structure. While it is possible, though fiddly, to write the grammars of real languages as transformational grammars, this has not revealed a single universal transformational grammar. Such a universal grammar would need to specify, through a clear set of principled adjustments, all the grammars found in human languages while excluding grammars that are unattested.

The failure to find a universal transformational grammar led to the minimalist programme. This retains the focus on recursion while being silent on other aspects of language structure. It is exciting in its clarity but seems insufficient. Priming a child to use MERGE does not seem enough to constitute a language acquisition device: a minimal grammar does not appear a plausible key to the rapid and compulsive learning of language by children.

Since this monograph was published, there has been a sudden advance in the capabilities of Large Language Models (LLMs) offering a new opportunity to explore models of language [14]. Despite significant differences between brain and machine, LLMs demonstrate the power of statistical models operating on complex hierarchical representations of words and meaning, suggesting that this might also form at least a part of how the brain processes language. LLMs also demonstrate the need for a language acquisition device. Lacking the relevant biases, these algorithms require vast corpora of examples, far exceeding the exposure of infants to language during learning.

In summarizing Reflections on Language Evolution: From Minimalism to Pluralism, this review has oversimplified Boeckx’s more nuanced arguments and in an effort to give a clear picture of the book’s message I have also exaggerated its vehemence. Boeckx advocates for elevating the Darwin Problem within linguistics but he does not dismiss other objectives. However, it is useful to take his proposition at face value and ask how we should respond to the complex legacy of the Chomskyan revolution, how we should react to the remarkable advance in paleo-genetics and in machine learning, and how we should define the goals, methods and meaning of linguistics. While linguistics seems to be prone to a 70-year itch, a periodic need to radically reinvent itself, a reformulation away from a preoccupation with language itself does not seem to be required while the nature, and not just the origin, of our language ability is still an open question.

An expanded version of this review is available at doi.org/10.31234/osf.io/5vdbg

REFERENCES

[1] Cedric Boeckx. Reflections on Language Evolution: From Minimalism to Pluralism. Language Science Press, 2021.
[2] Cedric Boeckx. “The shape of the human language-ready brain”. Frontiers in Psychology, 2014, 5, 282.
[3] Wolfgang Enard et al. “Molecular evolution of FOXP2, a gene involved in speech and language”. Nature, 2002, 418(6900), 869–872.
[4] Pedro Tiago Martins, Maties Mari, and Cedric Boeckx. “SRGAP2 and the gradual evolution of the modern human language faculty”. Journal of Language Evolution, 2018, 3(1), 67–78.
[5] Noam Chomsky. Language and Mind. Harcourt Brace Jovanovich, 1972.
[6] Steven Pinker and Paul Bloom. “Natural language and natural selection”. Behavioral and Brain Sciences, 1990, 13(4), 707–727.
[7] Geoffrey E. Hinton and Steven J. Nowlan. “How learning can guide evolution”. Complex Systems, 1987, 1(3), 495–502.
[8] Conor Houghton. “Cooperation as well as learning: A commentary on ‘How learning can guide evolution’ by Hinton and Nowlan”. arXiv, 2024, arXiv:2409.15609.
[9] Jack Bunyan, Seth Bullock, and Conor Houghton. “An iterated learning model of language change that mixes supervised and unsupervised learning”. arXiv, 2024, arXiv:2405.20818.
[10] Simon Kirby. “Spontaneous evolution of linguistic structure: An iterated learning model of the emergence of regularity and irregularity”. IEEE Transactions on Evolutionary Computation, 2001, 5(2), 102–110.
[11] Simon Kirby, Kenny Smith, and Henry Brighton. “Iterated learning: A framework for the emergence of language”. Artificial Life, 2003, 9(4), 371–386.
[12] Noam Chomsky. Syntactic Structures. Mouton de Gruyter, 1957.
[13] Noam Chomsky. Aspects of the Theory of Syntax. MIT Press, 1965.
[14] Conor Houghton, Nina Kazanina, and Priyanka Sukumaran. “Beyond the limitations of any imaginable mechanism: Large language models and psycholinguistics”. Behavioral and Brain Sciences, 2023, 46, e395.

ABOUT THE REVIEWER

Conor Houghton is a computational cognitive scientist with a particular interest in linguistics, neurolinguistics and artificial intelligence; he is a member of the Intelligent Systems Laboratory in Bristol University and has recently been working on simple agent models of language change.