From Signal to Symbol: The Evolution of Language

Preface

This book has had a long gestation. One of us (Sterelny) has been working for the best part of fifteen years primarily on the evolution of human social life and on the capacities that make that life possible. As that work progressed, he became increasingly oppressed by the realization that a fully serious treatment of language was inescapable. He devoted a chapter of The Evolved Apprentice to the evolution of human communication in general, arguing there that the specific issue of deception as a threat to stable communication has been oversold, once one had an account of the stability of human cooperation in general, and once one takes fully into account the difficulty of project-managing deception in many-many interactions with much communication being about the here and now. However, with the exception of a defense of a gesture-first picture of the emergence of language, neither that chapter nor associated work burrowed deep into the specifics of language and its proximate ancestors.

So while Sterelny continued to build a detailed conception of human social evolution, language-specific issues remained somewhat on the back burner until Planer arrived in 2015 and we began to collaborate, with our picture of the book itself taking shape in 2018. Writing the book—the actual generation of text—has been thoroughly collaborative and approximately equal. We have both worked on revising the whole text, with Sterelny having written the first draft of the initial chapters and Planer of the later chapters. Intellectually, it is fair to say that the general conception of human evolution that structures the overall argument derives more from Sterelny, whereas the language-specific elements, and especially the ideas relating to cognitive neuroscience and the evolution of structure, owe more to Planer. Given that, he should be regarded as first author of the monograph. We will now turn to a brief picture of the book as a whole.

Language is foundational to human cognitive and social life. Hence, no adequate evolutionary account of the uniqueness of sapiens can escape the challenge of language evolution. As many have noted, this challenge is exceedingly difficult, in part because the gap between animal signaling and human linguistic communication is so glaringly large, and in part because evidence of the antecedents of language is elusive and indirect. Nevertheless, a growing number of researchers have begun to tackle this formidable challenge (whom we join here), and we think with valuable results. So we will begin by locating our own ideas against this cohort.

In the last decade or so, in addition to a new journal and various edited collections, we have seen important and ambitious books on the evolution of language. These include: Hurford (2007), Tomasello (2008), Bickerton (2014), Hurford (2011), Progovac (2015), Scott-Phillips (2015), Berwick and Chomsky (2016), Everett (2017). One major divide among these researchers is that between those defending a thoroughly gradualist model of language evolution and those who think language appeared much more suddenly (e.g., Berwick and Chomsky 2016). We figure firmly in the first camp, but that is still a diverse group. In our view, any adequate theory of language evolution must satisfy two important constraints. It must identify a plausible evolutionary trajectory from great-apelike communicative abilities to those of modern humans, where each step along the way is small, cumulative, and adaptive (or at least not maladaptive: there might be some role for drift). In other words, it must offer a lineage explanation (Calcott 2008) for language. Second, that account must be embedded within a larger account of human social, technological, and economic evolution, one enjoying independent evidential support from paleobiology, archaeology, and evolutionary biology. We have quite rich evidence of how and where ancient humans lived and moved. Information about their lifeways—about the resources they consumed, the environments they could exploit, the risks they could ameliorate, the territories over which they moved—constrains accounts of their cognitive, social, and communicative capacities (Sterelny 2016a). It is difficult to construct a coherent account that is both incremental and embedded, and any theory satisfying these requirements is no mere just so story.

Another major dividing line concerns the role that theory of mind played in the evolution of language. For some researchers (e.g., Tomasello 2008; Scott-Phillips 2015), the evolution of language is largely explained by a grade shift in theory-of-mind capacities. More specifically, on this view, the creation of language followed more or less inexorably once our ancestors were able to act with and recognize (notoriously complex) Gricean communicative intentions. The opposite extreme is also to be found in the literature, with some researchers understanding uniquely human theory-of-mind capacities to be the product rather than the cause of language. These researchers tend to be more concerned with the unique structural properties of language—in particular, its syntax and the basis of syntax in the human brain. As will become clear in the chapters that follow, here we very much occupy a middle position. In short: yes, an increase in theory of mind was important, but not necessarily for the reasons that those in the first camp typically cite. Moreover, an increase in those capacities was not the only critical intrinsic cognitive change that made language as we now know it possible. One way in which our view differs from others is that in general we are skeptical of single-factor, crucial-adaptation accounts of the evolution of language, just as we are of the emergence of the human mind and human social life more generally. Human language, and the cognitive and social capacities that support it, differ from other animal communication systems, including those of great apes, in a number of important ways. These include: the open-ended structural complexity of utterances (and here syntax is indeed critical); its diachronic flexibility (any competent user can coin new words and expressions); the great variety of social and communicative functions it can serve; its expressive richness, and the contextual sensitivity of that richness; the extraordinary speed, reliability, and agility of multiperson conversational interaction; its power as a tool of social learning and organizer of thought, for the division of linguistic labor makes it possible for us to think things because we can say them. We doubt that any of these features of language, with their supporting features of mind, are privileged: the single crucial difference from which all else follows. Instead, we think the evolution of language began from a mosaic of initially smaller differences from baseline capacities, and their interaction drove the evolution of communicative systems that became increasingly unlike any seen elsewhere in animal life.

In much of the work on human life, biology is contrasted with culture, with supposedly biological features of our lives and bodies contrasted with cultural ones. Language is an ideal case for undermining that pernicious dichotomy: language is both pervasively biological (supported by adaptations based on genetic changes, though often ones not specific to language) and also shaped by cultural processes on time scales from tens of thousands of years to minutes, as a witty new term catches on. Over the long duration of human deep time, there has almost certainly been a directional shift in which more of human communicative activity, and especially human vocalization, has been bought under top-down control: words are not much like snorts or grunts. But this should not be seen as a transition from biology to culture. Our eating is under top-down control, allowing what, when, and with whom we eat to be profoundly influenced by cultural learning. How, what, and with whom we eat is of great social significance. But that we have to eat, and what we can eat, are matters of morphology and physiology, though in this case a morphology and physiology that have been shaped by the cultural invention in the deep past of cooking. Eating is thus strongly biocultural. Laughter is another clear example of a biocultural trait, though it interestingly displays limited top-down control: genuine laughter is involuntary, and yet the triggers for laughter are profoundly influenced by culture. Language is typical rather than atypical in having a biocultural character.

One of the lessons of evolutionary and ecological theory is that life is messy, and this includes human life and deep human history. Biological kinds are fuzzy, without sharp boundaries; biological processes are typically driven by multiple interacting causal factors, and changes in those processes are often changes in relative importance. While formal models rightly idealize away much of this complexity, their results need to be interpreted in ways that reflect it. Our narrative is built around this recognition of multiple factors, with change in their relative importance over time. Our approach differs from others in a second way, one that reflects our disciplinary origins. Our argument is grounded in deep history rather than linguistics. Many of the books on language evolution to date have a shared approach that reflects the fact that they have mostly been written by linguists. These works tend to begin with a favored conception of language as it now is and work backward, attempting to identify the origins of language through the lens of their theoretical framework. Linguistic data are their empirical bedrock. For example, Progovac’s monograph embraces a specific theory of the syntax of human language in general, and develops an account of how such a syntax could be assembled incrementally. Since we do not commit to any specific account of the character of natural language syntax, our account could in principle complement hers. For we bring an alternative strategy: we attempt to identify the lifeways of ancient hominins in enough detail to sketch the cognitive and communicative capacities on which those lifeways depended. These lifeways changed over time, and in favorable cases we have sufficient information to outline the capacities and needs implied by these changed lifeways. So, we aim to sketch a plausible and empirically constrained sequence of changing communicative lives, beginning with lifeways not much more collaborative or complex than living great apes, and ending with lives whose social complexity and expressive needs approximate those of living humans. To the best of our knowledge, this is the only monograph-length model of the evolution of language organized around and embedded in a specific, reasonably detailed, empirically constrained view of changing hominin lifeways. Indeed, perhaps the most important difference between the approach exemplified in this book and those mentioned above is the extent to which our conception of the evolution of language is embedded within an archaeological context, and a more general account of the evolution of human sociality. Everett, for example, defends a view of the incremental emergence of language which is in some ways broadly similar to ours, but which places the establishment of something close to full language much earlier in time. We see no independent archaeological support for the existence of the very impressive cognitive capacities human language requires a million or more years ago. Likewise, Bickerton’s final foray into this problem was an attempt to solve Wallace’s problem: the supposed fact that humans are smarter than they need to be (and this includes our language abilities) to cope with the problems posed by our physical and biological environment. Wallace overlooked the fact that we have to be smart enough to cope with one another, and that this can generate positive feedback, ratcheting up our cognitive capacities. Bickerton, in our view, developed a solution in search of a problem.

We said above that we are not in the hunt for a crucial breakthrough, a new adaptation that took our ancestors into a new adaptive zone. In our view, rudimentary forms of all or most of the elements on which language depends are found in the great apes and were part of the equipment of the earliest species in our lineage. The elements include: theory of mind (as mentioned above); executive control, providing the ability to plan and execute complex and precise action sequences; the ability to notice and parse action sequences performed by others; but also prosocial motivations of various kinds, since conversation is typically a form of cooperation. This view of language evolution gives a prominent role to intermediate language forms, such as early protolanguage, enriched protolanguage, and early language, and to the flexibility and adaptability of hominin minds, even ancient hominin minds. By recognizing these intermediate language forms and the intermediate minds supporting them, we can begin to close the gap between great ape communication systems and language without positing evolutionary miracles along the way; we can, in other words, begin to glimpse a lineage explanation for language.

But as we also said, any adequate account of the evolution of language must be embedded within a broader frame of hominin evolution. We argue that the broader context in which language evolution played out is usefully conceived of in terms of two cooperation revolutions. The first of these connects great-apelike social behavior with that of early Homo. It marks the transition from the rather limited forms of cooperation characteristic of chimpanzees to a regime of obligate mutualistic cooperation (Tomasello, Melis, et al. 2012; Tomasello 2014). This period saw an initial upgrade in several cognitive abilities critical to more sophisticated forms of communication—enhanced social learning, enhanced theory of mind, enhanced memory, future-directed action, and more. The second revolution connects the already quite complex lifeways of late erectus and Homo heidelbergensis with that of modern humans. This revolution involved the appearance of regular, delayed-return forms of cooperation, such as giving a person a tool today in exchange for some meat next week (Sterelny 2014). Life is even more complex if your generosity with your tools is rewarded by help from a third party, recognizing your reputation as a reliable and helpful member of your community. We argue that the shift to delayed-return cooperation introduced a range of new social challenges. It is at this stage that tracking the reputation of third parties via gossip or its equivalent became essential to the stability of cooperation, and this drove the need for more complex communicative technologies. If gossip matters, and if it matters that gossip is accurate, those who exchange gossip need to be able to specify what happens off stage—who did (or failed to do) what to whom—precisely and unambiguously.

This shift to a more efficient but more complex and challenging form of cooperation within hominin bands or residential groups probably partly overlapped in time with an expansion of the social and spatial scale of cooperation, as residential groups became more open to one another, with freer movement in and out, and as they became networked into larger communities. This too led to important benefits (information exchange, a larger reproductive market, demographic buffering). But once again it made the social and communicative landscape more complex (Sterelny 2021).

In telling our particular story, we will draw upon some familiar ideas. Among them: (i) that gesture played a key role in launching human language evolution (Donald 2001; Tomasello 2008; Corballis 2011; Sterelny 2012b); (ii) that stone toolmaking drove the evolution of human syntactic abilities, and plausibly other cognitive abilities linked to language (Stout 2011; Stout and Chaminade 2012; Planer 2017b); (iii) that large-game hunting and the control of fire drove increases in communicative complexity by intensifying demands on human cooperation and coordination (Pickering 2013); (iv) that the evolution of singing in our line played an important role in preadapting humans for vocal language (Gamble, Gowlett, and Dunbar 2014; Killin 2017a; Killin 2017b), with the increased control of fire providing an ecological context selecting for the transition to the vocal channel; (v) that the human release from proximity was facilitated by the evolution of complex kin terminologies and other linguistically enabled ways of keeping track of relationships between people (Gamble 2013; Planer 2020a). In developing our conception of the evolution of language, we bring these ideas together into a novel, coherent package, while also updating and extending them in ways supported by more recent paleontological, archaeological, phylogenetic, and genetic evidence. Or so we hope!

With this general picture sketched, we now present a brief summary of the book’s chapters. In chapter 1, we develop our view of the constraints that a theory of the evolution of language should satisfy, building on the work of Laland (2017), and then commit to an incremental view of the evolution of language in which language evolved through a series of increasingly rich protolanguages. We then discuss skeptically one popular incremental conception of language evolution, the idea of progress through indexes and icons to true symbols. Models of the evolution of language have often appealed to the index-icon-symbol distinctions, suggesting that the puzzle of language is tied to the problem of explaining the emergence of the cognitive capacities and social environments that support symbolic communication (e.g., Deacon 1997). We argue that the index-icon-symbol distinctions are much less clean and transparent than has been supposed. We prefer the sender-receiver framework, much discussed in the recent literature, as our organizing conception of communication and its evolution. This setup chapter ends with our proposal about empirically constraining pictures of language evolution.

Chapter 2 begins the substantive project. We begin by using the resources of comparative biology and paleoanthropology to identify the cognitive and communicative baseline: the capacities of early hominins. That is one end of the pathway; in the second section, we give our characterization of the language mosaic, the cognitive and social abilities that support language. This is the other end of the pathway. The critical claim about this mosaic is that few, if any, of these capacities are unique to humans or to language. That matters. For if that claim is right, most, perhaps even all, are much-elaborated versions of capacities that existed in somewhat rudimentary forms in earlier hominins. Moreover, since these capacities feed into many skills, there can be a historical signal of their existence, for they can be manifest in behaviors like foraging and toolmaking, behaviors that do leave physical traces. Having identified the two bookends, this chapter elucidates the first advances of Pliocene hominins beyond great ape capacities, with the core idea being that a shift to obligate bipedal lifeways brought with it important cognitive and behavioral changes, including, in particular, changes to cooperation, coordination, and communication; in our view, via an expanded role for gesture.

Chapter 3 further articulates our view of the early stages of language evolution. This chapter takes as its principal question the transition from animal signals to words. Animal signals tend to be triggered by specific environmental states and linked to specific behavioral responses. They are in general tied to the here and now and tend to be produced unaccompanied by other signals. Moreover, the stock of signals belonging to an animal communication system tends to be fixed over time, whereas human lexicons are readily expandable. We explore how each of these constraints might have been lifted, yielding words or at least wordlike elements. These changes take place, in our view, against the background and in the aftermath of the first cooperation evolution. We estimate that some form of protolanguage using wordlike elements was very likely ancient, dating at least back to cooperative hunting, which is probably at least 1.7 million years old. As the hominin ecological repertoire expanded very gradually over time, we claim that the lexical stock expanded with it. So our crucial claim in this chapter is that erectine hominins, the hominins that appeared after about the first third of the Pleistocene, were probably equipped with quite rich protolanguages. These were protolanguages with structured signs, with displaced reference, and with enough flexibility to add new items over time. We aim to show that our information about those lifeways supports the view that these hominins were both capable of using a protolanguage and very likely needed a protolanguage, one with flexibility and displaced reference. We introduce the link between gesture and structured signs in chapter 3, but structure is the focus of chapters 4 and 5.

More specifically, chapter 4 addresses the origins and establishment of what we call composite signs—signs with less than basic syntax but which are an important precursor to basic syntax. Composite signs have parts that work together in some way to fix the meaning of the whole, but the order of the parts does not yet encode meaning. In developing this view, we respond to an important challenge raised by Thom Scott-Phillips (2015). He argues that the introduction and use of composite signs requires very rich cognitive capacities. We argue that a somewhat simpler cognitive skill set suffices, a view that supports a more incremental model of the emergence of structured signs.

Chapter 5 takes up the evolution of syntax proper. In particular, it addresses the evolution of hierarchical structure. Despite the attention hierarchical structure has received in the evolution-of-language literature (see, e.g., Stout 2011; Berwick and Chomsky 2016), exactly what such structure amounts to is far from clear. So one goal of this chapter is to offer an improved analysis of hierarchical structure and its role in communication. We then develop the idea that the computational machinery underpinning hierarchical structure evolved in the service of technological skill, the production of sophisticated stone tools in particular. (As our language here suggests, in this chapter, and throughout the book more generally, we commit to a broadly computational picture of cognition. We assume rather than defend this orientation in the book.) Toward this end, we consider evidence at the cutting edge of both lithic studies (pun intended!) and the neuroscience of complex, intentional action, and attempt to integrate the two. We suggest that, by the time composite tool production had become part of the human technical repertoire (between 500 and 250 kya), most if not all of the cognitive specializations for processing hierarchical structure were in place. So the central strategy of this chapter is to show that the cognitive capacities needed to master the impressive stone technologies of the erectine and especially heidelbergensian hominins (appearing at about 800 kya) could be, and probably were, co-opted for structured language processing. These tools could only be made through an accurate sequence of planned steps, and this skill could only be learned from others by being able to observe and recognize such sequences of planned steps. The advanced stone-working techniques of these hominins depended on hierarchically represented action plans, both in their execution and acquisition. In our view, likewise, this is the cognitive capacity on which producing and understanding hierarchically structured sentences ultimately depends.

Chapter 6 takes up a crucial challenge for any view of language that begins with an expanded role for gesture. If protolanguages began as largely gestural systems, why and how did vocalization become so important? We meet that challenge through the idea of a firelight niche, and the changed social and physical environments that come with the control of fire. In developing this view, we adapt some ideas of Dunbar in identifying the selective forces that favored increased vocal control. In our view, selection for something like wordless singing and likely laughter led to improved vocal control. These behaviors helped to ease tensions and strengthen affiliative bonds as hominin social life became more complex and intense. With more vocal control available, the vocal channel offered various efficiencies, and we argue that those were particularly salient at the fireside, in the firelight niche. So we think a combination of increased vocal control (evolving independently of protolanguage) and the control of fire mediated the shift to a primarily vocal mode of communication.

Chapter 7 rounds out our story. It is organized around changes in human lifeways late in the Pleistocene, with the second cooperation revolution in full swing, perhaps beginning something like 150 kya (this claim is defended in detail in Sterelny 2014; Sterelny 2021). This second revolution both increased the demands on human communicative capacities and provided an environment in which cultural learning was more efficient. Jointly, these changes explain the transition from richer protolanguages to languages as we know them (though in our view there is no sharp boundary between the two). So this chapter presents a detailed account of the social tools and communicative capacities on which this second revolution depended. These include kinship terminologies, explicit norms, and the capacity to tell stories and report indirect speech. In our view, it is not until around this period that human lifeways began to closely resemble those of ethnographically known foragers in terms of the cooperation and coordination problems routinely faced and solved. In turn, we claim that it was probably not until around this period that fully modern language began to appear, evolving to meet the conflict flashpoints associated with intensified cooperation and coordination demands on humans.

While some might see this as a strikingly late date for the evolution of language, we see this as a natural consequence of taking the cultural-evolutionary character of language seriously. The genetic prerequisites for language are probably much more ancient, given the rather minor genetic differences between sapiens, Neanderthals, Denisovans, and hence, by inference, their immediate ancestor, the heidelbergensians (Dediu and Levinson 2013; Levinson and Dediu 2018). But we think the role of cultural evolution was of critical importance. Compare language with numerical cognition. It is one thing to have the genetic capacity to multiply 127 by 69; another to have that as part of one’s cognitive phenotype. In the middle Pleistocene, and in the lives of the heidelbergensians, there was little need for the full package of design features characteristic of modern language. Moreover its evolution may well have been prevented by demographic constraints. Our ancestors of 500 kya or even earlier may well have been language-ready—indeed, we think they probably were—but language readiness does not mean they were language-equipped.

Chapter 8 is self-assessment: we return to the success conditions identified in section 1.1, and evaluate our picture in the light of those criteria. While we certainly do not fully meet them, we claim to have sketched important parts of the map by offering empirically constrained and incremental models of an expandable lexicon, displaced reference, and the cognitive prerequisites of hierarchical syntax.

We now turn to a pleasurable duty: acknowledging and thanking help along the way. It takes a village to raise a book, and we have been fortunate in our village. The local Australian National University (ANU) environment has been very supportive. That is true both of the School of Philosophy and of the interdisciplinary (though linguistics-heavy) Centre for the Dynamics of Language. In addition to housing, supporting, and tolerating us, these have been places where both authors tried out, individually and jointly, presentations that were early versions of these chapters (especially at assorted events organized through the Centre), and we have had a lot of helpful feedback: in particular from Carl Brusse, Nick Evans, Liz Irvine, David Kalkman, Anton Killin, Stephen Mann, Richard Moore, Ross Pain, Lauren Reed, and Matt Spike. These robust individuals read and responded to draft papers and suffered through presentations. We have also tried out various chunks at workshops in Australasia (in particular, at an evolution-of-language workshop organized at ANU; assorted CoEDLFests, the Wellington Empirical Philosophy workshops, and a series of archaeology-meets-philosophy workshops). For help at these, we would particularly like to thank Peter Godfrey-Smith and Peter Hiscock. They too responded to both presentations and drafts, and they have both contributed in essential ways to how we think about evolution, language, and theory-building in the historical sciences. In addition, four readers for the MIT Press read the whole manuscript, and their comments provoked many improvements. Two were nameless. The other two were Michael O’Brien and Richard Moore. We thank all four, but Richard Moore in particular provided an especially searching set of comments. Finally, we thank the Australian Research Grants Council for their support of this project in particular, and for their support of Sterelny’s projects over many years.

1 An Intractable Challenge?

1.1 The Scope of the Problem

Human language confronts the community of evolutionary theorists with a difficult but inescapable challenge. Inescapable, since language is manifestly central to human life, and so an account of the origins of human cognition and social life must include an account of the emergence of language. Moreover, that account must be evolutionary. That claim is not controversial, though there is debate about the extent to which the emergence of language was gradual and incremental (rather than abrupt), and about the relative roles of genetic and cultural evolution. For even if language was an invention, as writing clearly was, it was invented because of evolved features of hominin life, and invented using evolved features of hominin cognition. Moreover, even if language was an invention, almost certainly it was invented incrementally and cumulatively, exemplifying cultural if not genetic evolution (Heyes 2018). We will return to those debates shortly.

The challenge is difficult, because language is very different from other forms of communication. No other extant species uses a minimal or rudimentary version of language, and so it is difficult to exploit the resources of comparative biology to identify rudimentary forms of language, or the environmental features that select for its evolution. Difficult, also, because the explanatory target is in part a feature of the social environment, in part a feature of individual psychology. Evolutionary biologists distinguish between an organism’s genotype and its phenotype, the array of morphological, physiological, and behavioral traits influenced by its genes. Humans’ extraordinary large and complex brain is one such phenotypic trait; speaking English is not. Specific languages are created, maintained, and modified over time by the communities that use them. The authors of this book are both native English speakers, and in an important sense, in our respective childhoods, we joined English. We did not create it de novo from inherited resources; say, inherited instructions about how to build a language. In contrast, ordinary phenotypic traits are created de novo. We do not inherit mini-legs, or grab a pair of readymade legs from the environment. Our physical organs were built de novo as we developed from zygote to toddler. In contrast, our language was a persisting feature of our social worlds to which we adapted. But the capacity to speak and understand English is an individual trait, a feature of each of our (the authors’) realized phenotypes. Evolutionary theories of language sometimes take the social phenomenon as their primary target, sometimes the individual trait. An adequate theory needs to give an account of both, and of their interaction. A further difficulty is an ongoing and deep controversy about the nature of the explanatory target itself. There is much debate about the evolution of our large and complex brain. But at least there is consensus on what the brain is. Evolutionary theorists of language have no such luxury. The Chomskian idea that language is primarily an organ of thought, only secondarily exapted for communication (Berwick and Chomsky 2016), is just one example of a controversial claim about language of obvious relevance to theories of its evolution. In general, an account of how language evolved will typically be shaped by an account of what language is, and there is no consensus about that. Finally, and most obvious of all, our evidential basis is largely indirect. We have to infer the scope and limits of ancient hominins’ communicative and cognitive capacities from the traces left by the actions those capacities made possible. For having (or lacking) a language leaves no clear signal on the skeleton of a language user (and hence on the potential fossils of such users). It is true that speech depends on extraordinarily complex control of breath, tongue position, and mouth shape (Fitch 2010; Everett 2017, chapter 8), and there are fossil indicators of such control, as we shall see in chapter 6. But the capacity to produce complex streams of sound might reflect the ability to sing rather than speak, and language can be manifest in sign rather than sound. Evidence of breath control is important, but most evidence about ancient language is indirect.

Despite some skepticism, we think the case for an incremental conception of the evolution of language is overwhelming. For one thing, the cognitive capacities required for routine conversation are extraordinarily impressive. The speech stream typically flows at somewhere between 135 and 180 words per minute (Everett 2017, chapter 8). In managing that stream, the speaker has to decide what he or she wants to say; compose that thought into a specific ordered and structured set of lexical items; and turn that representation of what to say into a sequence of motor commands that drive articulation. This cognitive feat is executed under time pressure, as participating in a conversation requires one to match its normal flow seamlessly, and the utterance the speaker launches must be both responsive to and constrained by the previous contributions to the conversation. Inserting a comment requires the speaker to be tuned, at least to some extent, to the attention, interests, and knowledge of the audience.¹ In the evolution-of-language literature, there has been a recent focus on pragmatics, making much of the fact