1 Introduction

The question of how language evolved is probably as old as the scientific study of language itself, and it continues to stir controversial debates that often relate to the very nature of language. Any account of the evolution of language has to answer the question of what it actually is that evolved: is it an innate capacity or a complex configuration of domain-general cognitive skills? Such questions are at the centre of language evolution research, an interdisciplinary field that has been enjoying growing popularity since the 1990s. In this Element, we offer a brief review of pertinent research on language evolution from the perspective of cognitive linguistics, and we discuss what a cognitive-linguistic perspective can add to the study of language evolution. In doing so, we follow up on previous work, including our own, that has argued that cognitive linguistics, and particularly Construction Grammar as arguably the most influential approach under the broad umbrella of ‘cognitive linguistics’, provides a suitable framework for studying the evolution of language (e.g., Reference ArbibArbib 2012; Reference HurfordHurford 2012; Reference Pleyer and WintersPleyer & Winters 2014; Reference SinhaSinha 2017; Reference Pleyer and HartmannPleyer & Hartmann 2020; Reference VerhagenVerhagen 2021). We will also discuss how findings from language evolution research can, in turn, inform cognitive-linguistic theorising.

Before doing so, we briefly need to define the scope of language evolution research. The term language evolution is notoriously ambiguous as it can either refer to the evolutionary emergence of language or to the continued development of fully fledged human language(s) (Reference Tamariz and KirbyTamariz & Kirby 2016). Reference HaspelmathHaspelmath (2016) has criticised such a conflation of ‘origins of language’ on the one hand and ‘language change’ on the other. But as Reference Mendívil-GiróMendívil-Giró (2019: 24) points out, such an apparent conflation follows naturally from an approach that conceives of languages as social and cultural objects (as does cognitive linguistics). According to such an approach, ‘there is every reason to suppose that the very first grammatical constructions emerged in the same way as those observed in more recent history’ (Reference BybeeBybee 2010: 202). Domain-general cognitive, social, and interactional processes are the basis of the emergence of linguistic structure both in language change and language evolution (also see Reference IbbotsonIbbotson 2020: 16). As such, we argue that the scope of language evolution research encompasses both the origins of language and the processes that lead from very early forms of language, often captured in terms of a hypothesised protolanguage (Reference Tallerman, Tallerman and GibsonTallerman 2012), to fully fledged human language. Consequently, we will cover both aspects in this Element.

The relationship between cognitive linguistics and language evolution is a bilateral one. On the one hand, language evolution research is highly relevant to our understanding of language when we take the ‘cognitive commitment’ (Reference LakoffLakoff 1990) of cognitive linguistics seriously. According to the cognitive commitment, analyses and theories in cognitive linguistics need to take into account what is known about cognition from other disciplines (cf. Reference Evans and GreenEvans & Green 2006). This means that cognitive linguistics also needs to be informed by what is known about the cognitive underpinnings of language in language evolution research. Such an approach also goes hand in hand with a ‘commitment to look for converging evidence’, that is, a commitment to actively seek out evidence from other disciplines on the role of explanatory cognitive factors used in cognitive linguistics (Reference Evans and GreenEvans & Green 2006). In his call for a ‘cognitive evolutionary linguistics’, Reference VerhagenVerhagen (2021: 11) re-conceptualises these commitments in terms of a ‘biological commitment’. It specifies more precisely what is already implicit in these commitments, namely that what we say about communicative behaviour ‘should also be compatible with what we know about communication in organisms in general and about cognition in general’.

On the other hand, cognitive linguistics also has a vital role in specifying the cognitive and interactional factors central to human language, whose evolution needs to be explained by any account of language evolution (Reference Győri, Wen and TaylorGyőri 2021). This is expressed in Reference Jackendoff, Yamakido, Larson and DéprezJackendoff’s (2010) dictum, ‘Your theory of language evolution depends on your theory of language’. For example, a usage-based view of language sees knowledge of language as represented in a learned network of constructions of differing degrees of schematicity and complexity, which is mutually shared to the degree that it enables interlocutors to co-create contextually scaffolded meaning in interaction. From an evolutionary perspective, the question then becomes how the cognitive abilities, as well as the social structures supporting them, evolved (Reference Pleyer and HartmannPleyer & Hartmann 2020).

In Section 2, we will take a closer comparative look at human language and other animal communication systems. Much previous research comparing human language and animal communication has claimed that there is a ‘qualitative’ difference between the two (e.g., Reference BickertonBickerton 1990; Reference Berwick and ChomskyBerwick & Chomsky 2016). However, more recent work has shown that many animal communication systems are much more complex than previously thought. Overall, this research indicates that there is much more continuity between human language and animal communication systems in terms of the different structuring mechanisms giving rise to communicative structures as well as regarding the importance of pragmatics, context, and social learning (e.g., Reference Engesser and TownsendEngesser & Townsend 2019; Reference Pleyer and HartmannPleyer & Hartmann 2020; Reference Hobaiter, Graham and ByrneHobaiter et al. 2022).

Section 3 discusses what we can learn from decades of research on language-trained animals (e.g., Reference Lyn, Shackelford and VonkLyn 2012). Here, we will focus again on cognitive similarities (e.g., the ability to learn symbolic associations and basic pragmatic and structural sensitivity) and differences. Similar to our discussion of animal communication, these differences are seen more as a matter of degree, not of kind (e.g., differences in the size of storage of associations, the schematicity, and abstractness of such stored associations, and differences in pragmatic and contextual influence on interpretation).

Section 4 will introduce the influential hypothesis that language builds on joint attention, cooperation, and shared intentionality (e.g., Reference TomaselloTomasello 2008, Reference Tomasello2014). Here, we will focus on the fact that many other animals have highly sophisticated social cognition (e.g., Reference Seyfarth and CheneySeyfarth & Cheney 2015; Reference Bettle and RosatiBettle & Rosati 2021). For instance, many non-human animals can understand others’ goals and attention and show some degree of perspective-taking and cooperation. However, there seem to be differences in that human communication is characterised by a much higher degree of cooperativeness, perspective-taking, and the negotiation and awareness of shared goals and the use of pragmatic inferences (e.g., Reference Tomasello, Hare and YamamotoTomasello 2017).

Section 5 gives an overview of how hypotheses about language evolution have been tested experimentally, mainly focusing on behavioural studies with human participants but also taking computational modelling work into account. We focus on two influential paradigms that have been used extensively in research on human symbolic evolution in general and on language evolution in particular: experimental semiotics is an approach in which participants are required to communicate without language, which often entails that they are required to ground novel communication systems in interaction. In iterated learning (IL), on the other hand, participants are trained on an artificial language that is then transmitted from participant to participant, thus simulating the historical chain of transmission characteristic not only of language but virtually all kinds of cultural artefacts. The two paradigms have been combined in many ways and given rise to complex experimental set-ups that combine learning, communication, and transmission, thus investigating a wide variety of system-internal as well as social and environmental factors that influence linguistic structure.

In Section 6, we turn to the real-world dynamics of language, focusing on two complementary domains: on the one hand, we discuss the implications of historical language change for understanding language evolution. On the other hand, we discuss what we can learn from developing sign languages. The so-called ‘emergent’ sign languages such as Nicaraguan Sign Language (NSL) and Al-Sayyid Bedouin Sign Language have often been cited as paragon examples of the evolution of a new language. However, some of the guiding assumptions in this line of research have recently been questioned, especially as they tend to exoticise the languages in question as well as their users, which in turn has implications for our understanding of these languages as ‘special’ cases of language evolution.

We conclude with a brief discussion of the relationship between cognitive linguistics and language evolution, arguing that the usage-based perspective of cognitive linguistics and the current mainstream view of language evolution as a process of cultural evolution are highly compatible.

2 Comparing the ‘Design Features’ of Language and Animal Communication

This section adopts a comparative perspective on human language. We will compare human language with animal cognition and communication and ask to what degree there are similarities and differences in the underlying cognitive and interactional processes and mechanisms.

Reflecting the general approach of this Element, the aims of this section are again twofold: on the one hand, we will ask what can be learned about human language by comparing it to other animal communication systems. Specifically, we will look at the implications of current research on animal communication for cognitive linguistics. That is, to what degree are cognitive-linguistic conceptualisations of language, and its cognitive and interactional underpinnings, compatible and convergent with research on animal communication and cognition? And to what degree can such research contribute to theory-building in cognitive linguistics? This reflects the cognitive commitment and the commitment to seeking converging evidence (see Section 1).

On the other hand, the aim is to take an explicitly cognitive-linguistic perspective on animal communication and cognition as well as the implications of research on animal communication and cognition on theories of language evolution. That is, what can cognitive linguistics contribute to the study of animal communication and cognition? In some way, this also reflects another commitment of cognitive linguistics, the ‘generalisation commitment’ (Reference LakoffLakoff 1990; Reference Evans and GreenEvans & Green 2006). According to this commitment, cognitive linguistics seeks to explain language in terms of general cognitive principles. From a cognitive-linguistic view, we can therefore try to generalise beyond humans and ask whether particular cognitive principles operate in human language as well as animal communication. Here, the question then becomes whether explanatory cognitive principles and capacities proposed in cognitive linguistics can also be applied to analyse properties of animal communication. In turn, this also means that if a particular cognitive ability used to explain aspects of language in cognitive linguistics is not found to the same degree in other animals, this has direct implications for theories of language evolution. Specifically, cognitive linguistics might stress the importance of particular cognitive and interactional mechanisms, which then might be found to be expressed more strongly in humans or even to be absent in other animals. This then would represent an ‘evolutionary target’ of processes and properties that must have evolved in the human lineage to make linguistic interaction possible. In other words, this can give potential insights into ‘what is special about human language’.Footnote ¹ Such a cognitive-linguistic view can help elucidate which cognitive and social prerequisites had to be in place for human language to evolve. Based on this, a comparative perspective can shed light on how the social and communicative behaviour of non-human animals compares to these prerequisites.

So far, detailed engagements with animal communication have been relatively limited, although it is increasingly gaining more attention. For example, in previous work, we have investigated the implications of animal communication for language evolution from a usage-based perspective (Reference Pleyer and HartmannPleyer & Hartmann 2020), reviewing how recent findings in animal communication can make contact with usage-based and constructionist approaches. Further, we have argued that applying concepts from usage-based and constructionist approaches can offer new perspectives for analysing human language and animal communication in a shared framework. Reference Christiansen and ChaterChristiansen and Chater (2022) also compare human language and animal communication from a broadly usage-based perspective. They argue that in contrast to animal communication, meaning in linguistic interaction is co-created in interaction and instantiated online. It is characterised by improvisation, creativity, and the cultural transmission of previously successful bouts of the co-creating of meaning in interaction. Reference Amphaeris, Shannon and TenbrinkAmphaeris et al. (2021) argue that critical cognitive processes demonstrated by cognitive linguistics to be foundational for language, such as social cognition and symbolic cognition, can be found in animal cognition to differing degrees. In their view, this supports the theory that language evolved on the basis of animal cognition. Lastly, Reference Amphaeris, Shannon and TenbrinkAmphaeris et al. (2022) propose to capture the relationship between human language and animal communication in terms of prototype theory. They explicitly contrast this proposal to an all-or-nothing view of a feature of human language being present in animal communication or not, as well as to a view that sees animal communication and human language as lying on a continuous spectrum. Instead, they advocate a prototype-based view in which phenomena are categorised in terms of family resemblances, graded typicality, as well as overlaps (see also Reference Wacewicz, Zywiczynski, Hartmann, Pleyer and Benítez-BurracoWacewicz et al. 2020).

3 Signing Apes and Talking Birds: Language-Trained Animals

We have seen that animals exhibit complex communicative behaviour in the wild. However, certain features found in human language seem not to be present, such as complex compositional constructions and the use of an open-ended inventory of form–meaning pairings to collaboratively co-create meaning in interaction. However, it is possible that animals have more complex cognitive capabilities than those evident in their communicative behaviour in the wild. Given the complex cognition of some animals (e.g., Reference De WaalDe Waal 2016), especially other primates (e.g., Reference Seed and TomaselloSeed & Tomasello 2010), is it possible that they might be able to grasp (aspects of) human language? This has been one of the most popular as well as controversial questions in animal communication (e.g., Reference Seidenberg and PetittoSeidenberg & Petitto 1979; Reference Sebeok and Umiker-SebeokSebeok & Umiker-Sebeok 1980; Reference AndersonAnderson 2004). This is also reflected in the popularity of popular press books and media attention that this kind of research received in the past. In addition, it is also evident in the fact that many introductory linguistics textbooks (cf. Reference Wacewicz, Pleyer, Szczepańska, Ewa Poniewierska and ŻywiczyńskiWacewicz et al. 2023a) and introductions to language acquisition discuss this research extensively (cf. Reference Pleyer and HartmannPleyer & Hartmann 2020). Interest and active research in ‘ape language’ research have declined significantly over the years, especially following some highly critical evaluations of these experiments (e.g., Reference Terrace, Ann Petitto, Jay Sanders and BeverTerrace et al. 1979) as well as increasing ethical concerns (e.g., Reference HuHu 2014).

Nevertheless, the existing research is still of great interest for language evolution research as a way of probing deeper into animals’ communicative and cognitive capacities and comparing the human capacity for language with the abilities of other animals. Yet even after over seventy years of research, the controversy about the analysis and implications of ‘animal language studies’ is unabated. As Reference Tomasello, Hare and YamamotoTomasello (2017) summarises, ‘The “ape language” studies have come and gone, with wildly divergent claims about what they have shown’. However, at least in some respects, there is relative consensus, which we will discuss in the following. Overall, there are two key questions regarding language-training experiments with great apes: (a) are they able to learn symbols, and (b) are they able to learn aspects of the combinatorial structure of language, that is, syntax? Regarding the first question, many researchers argue that great apes do indeed show evidence of this, whereas the second question is much more contested.

The first systematic attempt at teaching an ape language was by Reference Hayes and HayesHayes and Hayes (1951). They tried to raise a female chimpanzee, Viki, as much as a human child as possible with the addition of explicit language teaching, including shaping Viki’s lips for sound production. After three years, according to Reference Hayes and HayesHayes and Hayes (1951), Viki was able to produce three words: ‘cup’, ‘papa’, and ‘mama’. However, although Viki produced specific sounds intentionally or at least in response to prompts, it requires interpretation to categorise them as English words. ‘Papa’ sounded like two consecutive lip smacks, and at least in existing videos, ‘cup’ was produced with Viki putting her hand in front of her mouth and touching her lips to make the sound.Footnote ⁴ In addition, they note that Viki sometimes confused her three words with each other.

One takeaway researchers took from this study was that chimpanzees are probably not able to produce spoken language due to their vocal tract anatomy. However, recent research has shown the high flexibility of mammalian vocal tracts, and simulations have demonstrated that a non-human primate vocal tract, in principle, can produce speech sounds or, in other words, is ‘speech-ready’ (Reference Fitch, de Boer, Mathur and GhazanfarFitch et al. 2016).Footnote ⁵ This means that their inability to produce speech is not due to vocal tract anatomy but instead due to the lack of human-like neural control systems for fine-grained vocal motor control.

Nevertheless, these considerations led several researchers to try to teach language to great apes using a different modality. These experiments were much more successful, but how much more successful is a matter of debate (see Reference Lyn, Shackelford and VonkLyn 2012 for a review). Reference PremackPremack (1971) tried to teach a chimpanzee, Sarah, a ‘visual language’. This ‘language’ was based on pieces of plastic with different colours, shapes, and sizes, which had different meanings and specific ordering rules. He reported the productive use and comprehension of more than ninety-eight signs after intensive training. Rumbaugh and colleagues (Reference RumbaughRumbaugh 1977) tried to teach another chimpanzee, Lana, an artificial visual language called ‘Yerkish’, whose vocabulary consisted of graphical symbols on a lexigram keyboard, reporting a productive and comprehension vocabulary of more than 123 symbols. Using a similar design but a more complex lexigram keyboard and a different teaching regime, including English language immersion, Reference Savage-Rumbaugh, Pate, Lawson, Smith and RosenbaumSavage-Rumbaugh et al. (1983) taught two chimpanzees, Austin and Sherman. They reported a productive use of more than sixty-eight symbols and comprehension of more than sixteen symbols.

Given that great apes show evidence of flexible control of their hands, several studies in the 1960s and 1970s tried to teach several great apes a natural language in the visual modality: American Sign Language (ASL). For chimpanzees Washoe (Reference Gardner and GardnerGardner & Gardner 1969) as well as Moja, Tatu, and Dar (Reference Fouts and Tukel MillsFouts & Mills 1998), the productive vocabulary reported in peer-reviewed studies ranged from 119 to 160 signs. Chimpanzee Nim Chimpsky was said to produce around 125 signs (Reference Terrace, Ann Petitto, Jay Sanders and BeverTerrace et al. 1979). Gorilla Koko was reported to produce 85–150 signs. In non-peer-reviewed publications, the orangutan Chantek was reported to produce around 150 signs (Reference Miles, Parker, Mitchell and MilesMiles 1999). For gorilla Koko, the claim has been made that she could produce over 1,000 signs, although these claims were not reported in peer-reviewed publications (Reference Lyn, Shackelford and VonkLyn 2012).

However, here, we have to make a significant caveat that, unfortunately, is often misrepresented in the literature. Very often, we find the statement in the literature that these chimpanzees were taught, or use ‘sign language’ or ‘American Sign Language’. However, ASL is a complex natural human language. But most of the teachers in these experiments were not native signers, and many, if not most of them, were not even fluent signers (Reference AndersonAnderson 2004: 276). This means that ‘in practice the apes were taught signs borrowed from ASL with English word order, not true ASL’ (Reference KaplanKaplan 2016). This is somewhat acknowledged by Patterson, who stated that Koko was taught a modified version of ASL called ‘gorilla sign language’ (Reference NewmanNewman 2013), but to what degree this system should be called a ‘sign language’ is not clear. However, the fact remains that, as Reference AndersonAnderson (2004: 280) puts it, the great apes in these experiments had ‘virtually no evidence for the grammatical mechanisms of true ASL’. One Deaf assistant in the Nim Chimpsky study also recollects that the other (hearing) trainers were overly generous in what they recognised as a sign even if they only partially resembled ASL signs (Reference NeisserNeisser 1983; cf. Reference KaplanKaplan 2016; though see Reference StokoeStokoe 1978 for a more charitable interpretation). Moreover, Reference KaplanKaplan (2016) points out that for the chimpanzees that were studied, many of the gestures interpreted by the researchers as ASL signs were actually naturally occurring gestures in the wild, such as COME/GIMME and HURRY. The latter is particularly significant as in at least one study (Reference Fouts, Fouts, Gardner, Gardner and Van CantfortFouts & Fouts 1989); HURRY comprised 65 per cent of the 206 produced signs interpreted as ASL signs.

These considerations are also of fundamental importance when considering the second question, ‘Can an ape create a sentence?’ (Reference Terrace, Ann Petitto, Jay Sanders and BeverTerrace et al. 1979), that is, if there are ordering principles in great ape ‘sign language’ production. At least some research indicates incipient tendencies towards broad semantic ordering principles (serial order) in some subjects. The best evidence for this comes from bonobo Kanzi, ‘the star pupil’ of ape language research. It is generally agreed upon that Kanzi showed the most sophisticated skills of any language-trained primate (Reference Tomasello, Hare and YamamotoTomasello 2017). He was not only reported to use more than 256 lexigram signs productively and understand ±179 symbols, but he was also tested on English comprehension, in which he was immersed. Over a six-month period, Reference Savage-Rumbaugh, Murphy and SevcikSavage-Rumbaugh et al. (1993) tested both Kanzi and Alia, a human child aged eighteen to twenty-four months during testing, on their comprehension of requests such as Can you put some toothpaste on your ball? and Carry the rock to the bedroom. They reported his overall accuracy across different types of requests with different construction properties at 71.5 per cent, compared with Alia’s 66.6 per cent of accurate responses. This led them to the conclusion that Kanzi’s spoken English comprehension is similar to that of a 2.5-year-old child (cf. Reference HurfordHurford 2012; Reference Lyn, Shackelford and VonkLyn 2012). However, as Reference TruswellTruswell (2017) points out, Kanzi’s performance is not the same across the board. In fact, Truswell shows that ‘Kanzi doesn’t get NP coordinations’ (Reference HurfordHurford 2012: 495). For noun phrase-coordination constructions such as Give the water and the doggie to Rose or Give the lighter and the shoe to Rose, his accuracy falls to 22.2 per cent. For example, in both cases, Kanzi only gave Rose one of the two items. In contrast, Alia’s performance was the same as her baseline (68.4 per cent). Reference TruswellTruswell (2017: 410) states this ‘suggests a species-specific, construction-specific deficit’. Most researchers, therefore, agree that none of the language-trained apes shows evidence of grammatical structuring or hierarchical structure in production or comprehension (e.g., Reference HurfordHurford 2012; Reference Lyn, Shackelford and VonkLyn 2012; Reference TruswellTruswell 2017). In Construction Grammar terms, there seems to be a difference in the schematicity and abstractness of stored constructions when comparing humans and language-trained animals (Reference Pleyer and HartmannPleyer & Hartmann 2020).

The question of animal grammatical abilities has also been tested in the paradigm of ‘artificial grammar learning’, especially with primates and birds (see, e.g., Reference ten Cateten Cate 2017; Reference ten Cate, Petkov and Hagoortten Cate & Petkov 2019 for reviews). These studies showed evidence of a basic continuity between animal cognition and human language learning capacities, especially in the domain of statistical learning, and some limited ability to detect regularities and dependencies in structured sequences of stimuli (Reference Petkov and ten CatePetkov & ten Cate 2020). However, the current evidence is seen to be ‘insufficient to arrive at firm conclusions concerning the limitations of animal grammatical abilities’ (Reference ten Cateten Cate 2017). One of the most impressive results so far comes from a study of two rhesus monkeys trained to indicate correct sequences of flashing coloured dots on a hexagonal spatial grid that followed a complex ‘mirror grammar’ (Reference Jiang, Long and CaoJiang et al. 2018). This grammar had the supra-regular pattern of AB|BA and ABC|CBA. Monkeys saw the first half of the grammar on a screen and were trained to complete the sequence following the mirror grammar. For example, a flashing dot would first appear in location 1, then location 2, and then location 5. If they then completed the sequence using the correct grammar, in this case, by first touching location 5, then 2, and 1, they were rewarded with water or juice. They were even able to transfer this grammar to longer sequences with a length of 4 (ABCD|DCBA) and 5 (ABCDE|EDCBA) and to novel geometrical layouts, such as a pyramid, a horizontal line, or two hexagons. These results indicate cognitive capacities in animals that approach the hierarchical cognitive complexity of human language. However, one crucial difference to humans is that these monkeys needed literally tens of thousands of trials to learn this grammar (Reference FitchFitch 2018; Reference Jiang, Long and CaoJiang et al. 2018). To put this into perspective, Reference Jiang, Long and CaoJiang et al. (2018) also taught this grammar to human five to six year olds. Not only were they vastly more accurate than the monkeys, they also learned it after five trials. Interestingly, Reference Jiang, Long and CaoJiang et al. (2018) suggest that this is due to preschoolers using a different strategy, namely chunking, which in usage-based and cognitive-linguistic approaches has been shown to be central to how language is learned (e.g., Reference TomaselloTomasello 2003; Reference BybeeBybee 2010; Reference IbbotsonIbbotson 2020).

Symbol comprehension and production have also been tested in other symbol-trained animals. For example, border collie Rico was shown to understand about 200 sound-item mappings (Reference Kaminski, Call and FischerKaminski et al. 2004), and could successfully retrieve toys from another room upon hearing the appropriate label. Border collie Chaser even retrieved 1,022 toys with different labels (Reference Pilley and ReidPilley & Reid 2011). Interestingly, both dogs showed evidence of ‘fast-mapping’ – learning a label after a single exposure – and learning through inferential reasoning by exclusion. This means that upon hearing a novel label, they retrieved the correct toy if it was the only one in the set they hadn’t learned a label for, and retained knowledge of that label afterwards. A grey parrot, Alex, learned and produced correct labels for ‘>50 objects, seven colors, five shapes, quantities to eight, three categories (color, shape, material) and used “no,” “come here,” “wanna go X,” and “want Y” (X,Y being appropriate location or item labels). He combined labels to identify, request, comment on, or refuse>150 items and to alter his environment’ (Reference Pepperberg, Shackelford and VonkPepperberg 2012: 297). There are also studies of dolphins indicating that they can learn the referential function of novel symbols and gestures, and even some evidence for comprehension of semantic ordering (Reference Pack, Herzing and JohnsonPack 2015).

So what does this research tell us about the human ability to learn and use language, and its evolution? In other words, what can we learn from this about the potential ‘evolutionary baseline’ that human cognition evolved on top of? Basic symbol learning seems to be relatively widespread in different animals. What is more, animals also seem to be able to learn more abstract relations such as ‘same’ and ‘different’ (e.g., Reference HurfordHurford 2007), something that has even been demonstrated in bees (Reference GiurfaGiurfa 2021; cf. Reference Pleyer, Kuleshova, Qing Zhang, Goldwater, Anggoro, Hayes and OngPleyer et al. 2023). They also have been shown to understand relations between symbols, which according to Reference DeaconDeacon (1998) is the key feature of truly symbolic cognition. For example, Chaser was able to learn hypernymic common nouns with one-to-many and many-to-one relations (‘frisbee’, ‘ball’, and ‘toy’). Specifically, she knew each tested item by its proper-noun name, and also was able to categorise all these items under the category ‘toy’ and a subset of them (e.g., balls of different sizes and colours) under ‘balls’. Overall, the evidence suggests that animals, and especially apes, possess many of the cognitive skills requisite for language, such as statistical and sequential learning, categorisation, some semantic ordering principles, and basic symbol learning (Reference Tomasello, Hare and YamamotoTomasello 2017).

However, then the question becomes what differences explain humans’ capacity for language. For one, as we have seen, there seems to be a difference in humans’ ability for hierarchical processing. However, there is also a range of other differences that animal language studies demonstrate. One concerns the size of the repertoire these animals learn. The peer-reviewed reported range of productive vocabularies for language-trained apes was between 68 and more than 256 (Reference Lyn, Shackelford and VonkLyn 2012). However, this pales in comparison to the number of constructions that humans know. For example, Reference Brysbaert, Stevens, Mandera and KeuleersBrysbaert et al. (2016) estimate that ‘an average 20-year-old native speaker of American English knows 42,000 lemmas and 4,200 non-transparent multiword expressions’ with around 6,000 lemmas added from twenty to sixty years of age. From a constructionist perspective, which would also count semi-filled, unfilled, and more abstract constructions, this estimate would even be much higher (cf. Reference PleyerPleyer 2017). By twenty-four months, the number of words children know already ranges from 100 to 600 (Reference Fenson, Dale, Reznick and BatesFenson et al. 1994), clearly beginning to outnumber even the highest scores claimed for language-trained animals. This point is definitely reached when children enter school, a time by which they know about 14,000 words (Reference TemplinTemplin 1957). Even the non-peer-reviewed claims of Koko knowing 1,000 signs and Kanzi being able to understand 3,000 words, as well as Chaser’s documented 1,002 items, pale in comparison. Humans are therefore exceptional in their ability to learn a network of constructions. The capacity for ‘massive storage’ of constructions in memory seems to be an important evolutionary development in humans (Reference HurfordHurford 2012; Reference Pleyer and HartmannPleyer & Hartmann 2020). This is of particular interest from a usage-based perspective, in which aspects of human memory play a crucial role in explaining the organisation and acquisition of linguistic knowledge (e.g., Reference DivjakDivjak 2019; Reference SchmidSchmid 2016).

However, one other significant difference shown by these studies lies in the social domain. More specifically, many differences between the acquisition of symbols in great apes and humans might stem from the fact that humans have special prosocial motivations and sociocognitive and pragmatic abilities. For example, human linguistic interactions are characterised by their ‘Mitteilungsbedürfnis’ (Reference FitchFitch 2010), their desire and motivation to share perspectives and experiences and co-create meaning in interaction. So whereas humans use declarative gestures and declarative utterances from very early on, the vast majority of the utterances produced by language-trained apes are requests (Reference TomaselloTomasello 2008). In addition, the apes did not show much interest in the perspectives and contributions of their communicative partners. This is evidenced in their lack of turn-taking, a feature of communication central to human interaction (Reference LevinsonLevinson 2016). For example, more than half of Nim Chimpsky’s utterances interrupted his teacher, and he would also frequently sign at the same time as the teacher, indicating no awareness of interlocutor turns (Reference KaplanKaplan 2016). We will turn to this special role of cooperation and social cognition in language and language evolution next.

4 Cooperation and Communication: The Joint Attention Hypothesis

Language acquisition in humans is fundamentally social. It rests on infants’ and young children’s sociocognitive abilities as well as their interactional and social motivations. Humans communicate triadically. Much face-to-face communication involves the producer and the recipient attending jointly to a third entity (Reference TomaselloTomasello 1999). This capacity for joint attention is foundational for language acquisition. For example, from as early as twelve months onwards, infants use gaze following to learn about objects and events (Reference Flom and JohnsonFlom & Johnson 2011). By eighteen months, infants learn to associate a new word not with the object they are currently interested in but with the object the adult is looking at (Reference BaldwinBaldwin 1993; Reference Baldwin and MosesBaldwin & Moses 2001). Indeed, the role of joint attention is already evident before the emergence of language in children. For example, twelve-month-olds already point declaratively to share attention and interest, and to ‘share their perspective’. Importantly, when an interesting event occurs that infants point to, they are only satisfied if it leads to triadic joint attention, that is, if the experimenter exchanges looks between them and the event, not when the experimenter only attends to the event without acknowledgement (Reference Liszkowski, Carpenter, Henning, Striano and TomaselloLiszkowski et al. 2004). As we have seen, this declarative communicative behaviour seems fundamentally different from how language-trained great apes communicate.

However, children’s use of social cognition in language acquisition goes much further than that. They also start to take interlocutors’ intentions and knowledge states into account. For example, twenty-four-month-old children learn a new word for the adult’s intended action, not the failed one they actually performed (Reference Tomasello and MichelleTomasello & Barton 1994). At the same age, they also learn that a new word refers to something that is new to the adult but not to them (Reference Akhtar, Carpenter and TomaselloAkhtar et al. 1996). Again, these foundations are already evident in pre-linguistic infants. By fourteen months, infants understand declarative–cooperative pointing gestures (Reference Behne, Carpenter and TomaselloBehne et al. 2005), and exhibit knowledge of what ‘we’ have experienced together. For example, if being asked to hand the experimenter a toy the experimenter seems to be excited about, infants at this age hand them the toy that is new to the experimenter but not to the infant (Reference Tomasello and HaberlTomasello & Haberl 2003; Reference Moll, Carpenter and TomaselloMoll et al. 2007). At fourteen to eighteen months, they also show more complex evidence of using context and shared experience to interpret gestures. For example, Reference Liebal, Behne, Carpenter and TomaselloLiebal et al. (2009) had children play a ‘cleaning up’ game, in which an adult points at an object and an infant puts it into a box. They then had another adult enter the room and point at an object. Infants did not interpret this pointing gesture egocentrically. They did not put the object into the box but instead handed it to the adult. These kinds of experiments offer evidence for a sociocognitive foundation of language acquisition in humans. As Reference Tomasello, Carpenter, Call, Behne and MollTomasello et al. (2005: 690) put it:

Tomasello and colleagues describe this infrastructure as ‘shared intentionality’: motivations and abilities to engage with others in cooperative, collaborative activities with shared goals, plans, and intentions, and to share attention, experiences, and other psychological states with others (Reference Tomasello, Carpenter, Call, Behne and MollTomasello et al. 2005). It is this shared intentionality infrastructure that non-human primates lack, and which explains the difference between human language acquisition and the performance of language-trained apes and animal communication systems in the wild: ‘What they lack are the skills and motivations of shared intentionality – such things as joint attention, perspective-taking and cooperative motives – for adjusting their communicative acts for others pragmatically, or for learning symbols whose main function is pragmatic’ (Reference Tomasello, Hare and YamamotoTomasello 2017: 95).

In more recent work, Tomasello and colleagues further elaborate their system for capturing the sociocognitive differences between humans and other apes and the development of shared intentionality in human children. Specifically, they distinguish between joint intentionality on the one hand and collective intentionality on the other. Joint intentionality is a second-personal, interactive mode in which infants and young children base their communication especially on their interlocutor’s attentional, intentional, and knowledge states. This is children’s mode of engagement before age three. Starting around age three, children develop an understanding of ‘collective intentionality’. They begin to understand that conventions are based on collective agreements that guide and normatively coordinate social behaviour. This understanding of collective conventions is evident in children’s complex sociocognitive behaviours, such as emerging concerns for social evaluation, impression management, and the enforcement of social norms (Reference Engelmann, Herrmann and TomaselloEngelmann et al. 2012; Reference TomaselloTomasello 2014). It also scaffolds children’s understanding of other linguistic behaviours based on the creation of interactional conventions, such as conceptual pacts (Reference Matthews, Lieven and TomaselloMatthews et al. 2010), pretend play (Reference PleyerPleyer 2020), and politeness and impoliteness norms (Reference Pleyer, Pleyer, Roberts, Cuskley and McCrohonPleyer & Pleyer 2016, Reference Pleyer, Pleyer, Ravignani, Asano and Valente2022).

Animals, especially non-human primates, also exhibit complex social cognition (Reference Seyfarth and CheneySeyfarth & Cheney 2015). For example, chimpanzees understand seeing, knowledge, and ignorance. They also know that others make inferences and understand others’ goals, perceptions, and intentions (see, e.g., Reference TomaselloCall & Tomasello 2008; Reference Bettle and RosatiBettle & Rosati 2021 for reviews). Current research even indicates that the controversial question ‘Does the chimpanzee have a theory of mind?’ (Reference Premack and WoodruffPremack & Woodruff 1978), that is, whether they can attribute mental states to others, can be answered with a yes.

Research with eighteen-month-old human infants has shown that they can generalise their own perceptual experience to the experience of others. In a study by Reference Meltzoff and BrooksMeltzoff and Brooks (2008), two groups of infants had experience with two different kinds of blindfolds. In one group, it was an opaque blindfold. But in the other group, it was a trick blindfold that looked opaque from the outside but was actually see-through. Infants in the opaque blindfold condition expected a person not to see an object when wearing it, whereas children in the see-through condition expected a person to be able to see the object. Follow-up research showed that children also used this experience to calculate what a person knew. In an experiment by Reference Senju, Southgate, Snape, Leonard and CsibraSenju et al. (2011), eighteen-month-olds saw a typical ‘false belief’ test in which an object was first placed into a box by a teddy bear but then was taken out of the box again. Crucially, an adult shown watching the scene put on a blindfold after the object was put into the box. Eye-tracking showed that infants having experience with the opaque blindfold expected the adult to reach for the toy where they had seen the teddy bear put it. But infants having experience with the trick blindfold had no such expectation, as generalising from their own experience indicated that the adult had seen the toy being removed. Interestingly, Reference Kano, Krupenye, Hirata, Tomonaga and CallKano et al. (2019) replicated this study with chimpanzees, bonobos, and orangutans, who had experienced either an opaque barrier or a trick, see-through barrier. Eye-tracking data showed that most of the apes expected the adult to reach for the item in the box only in the opaque barrier condition but not in the trick barrier condition.

This question of theory of mind is of particular relevance because theory of mind and other related aspects of social cognition are intricately connected to specifically human communication making use of ostension, inference, and the recognition and expression of intentions. These capacities, then, are also argued to be crucial for the evolution of language (e.g., Reference TomaselloTomasello 2008; Reference Heintz and Scott-PhillipsHeintz & Scott-Phillips 2023). However, mastery of a more complex theory of mind seems to rely to some degree on children acquiring language and their enculturation in linguistic interactions (e.g., Reference Astington and JodieAstington & Baird 2005), suggesting a more complex co-evolutionary picture (Reference Rubio-FernandezRubio-Fernandez 2023).

Overall, great apes show sophisticated sociocognitive skills, and the recognition and expression of intentions seem to be part of the evolutionary platform for the evolution of human language (Reference MooreMoore 2017). However, there still seem to be crucial differences between great ape social cognition and motivations and those of humans. Specifically, non-human primates show evidence of complex social cognition mostly in competitive contexts. For example, their understanding (and production) of declarative pointing seems to be extremely limited (though see Reference Leavens, Racine, Hopkins, Botha and KnightLeavens et al. 2009). As mentioned, fourteen-month-olds understand a pointing gesture as informative for the question of which of two locations hides a toy. Chimpanzees, on the other hand, ‘follow the point to the bucket and say, in effect, “A bucket. So what? Now where’s the food?” They do not understand that the pointing is intended to be ‘relevant’ to the searching as a shared activity’ (Reference Tomasello and CarpenterTomasello & Carpenter 2007: 122).

Interestingly, in a similar context, when an experimenter makes a prohibitive gesture towards a bucket and then leaves the room, chimpanzees are able to infer a competitive motive and become interested in this bucket, and not the other (Reference Herrmann and TomaselloHerrmann & Tomasello 2006). They also become interested in a location if they see a chimpanzee try to reach towards it, as opposed to when a human experimenter points at it (Reference Hare and TomaselloHare & Tomasello 2004). Reference Bettle and RosatiBettle and Rosati (2021), along similar lines as Tomasello and colleagues, propose that other primates possess complex social cognition for competition. They argue that humans share this sociocognitive platform with other primates, but that humans have evolved additional social cognition for cooperation, which includes joint attention, sustained attention to others, attributing cooperative and shared intentions, and complex cooperative perspective-taking. This sociocognitive infrastructure for cooperation has been argued to represent a foundational, ‘species-unique contribution to the language acquisition process’ (Reference IbbotsonIbbotson 2020: 116). Another important aspect of this sociocognitive infrastructure for cooperation is that it includes a propensity for cultural learning and cumulative cultural evolution (Reference TomaselloTomasello 1999), of which language evolution and change are prime examples (Reference PleyerPleyer 2023). Cumulative cultural evolution, based on the cooperative infrastructure for cooperation, therefore seems to be a foundational process in how language evolved.

However, it has to be noted here that a focus on shared intentionality and its relation to cultural evolution is ‘not the only game in town’. The question of which (socio)cognitive abilities are the foundation of cultural evolution is very much an active field of research (see, e.g., Reference Heyes, Moore, Tehrani, Kendal and KendalHeyes & Moore 2023). The same holds for the question of how cultural evolution in turn transforms our cognitive processes, and leads to the emergence of new cognitive mechanisms which in turn influence cultural evolution in a dynamic feedback loop. For example, one such feedback loop that has been investigated concerns the way that language and social cognition co-develop and co-evolve (Reference Rubio-FernandezRubio-Fernandez 2023).

One way to investigate how cultural evolution can lead to the emergence of structure, and the factors influencing this process, is experimentally. Indeed, the cultural evolution of language has increasingly been tested in laboratory settings. We will discuss this research in the next section.

5 Language Evolution in the Lab

While studies comparing humans and non-human animals as reviewed in Sections 2–4 can give valuable clues to the evolutionary scaffolding of language, it does not answer the question of how linguistic structure comes about over the course of cultural evolution. Although the diachronic development of existing languages (see Section 6) can prove informative here, it is hardly sufficient as we are dealing with fully fledged human languages that have developed over generations. In the absence of any records of the earliest stages of language, laboratory experiments are used to approximate situations in which, in one way or another, communicative systems are created from scratch. In this section, we give an overview over two of the most influential paradigms that investigate the evolution of human symbolic communication systems in laboratory settings: experimental semiotics on the one hand, and artificial language learning studies on the other. These two approaches overlap to some extent and have been combined in many ways, especially in recent research. Section 5.1 introduces experimental semiotics; Section 5.2 discusses Iterated Learning (IL) as arguably the most influential artificial language learning paradigm; Section 5.3 reviews more recent developments in experimental approaches investigating the different factors that shape the evolution of communication systems.

6 Real-World Language Dynamics: What Language Emergence and Change Reveal about Evolution

As mentioned in Section 1, to what extent the dynamics observed in present-day languages can be subsumed under the umbrella term ‘language evolution’ is an open question. But especially if we take a usage-based perspective on language evolution (see Section 7), the cultural evolution of language is at least as important as the evolutionary developments that led to the emergence of the cognitive mechanisms underlying the human capacity for language, especially given the assumption that both are closely intertwined, and that at least some of the cognitive principles that underlie language use and language dynamics in the present must have played a role in the biological evolution of the human capacity for language. As such, analysing variation and change in present-day languages can prove insightful for understanding processes of cultural evolution. In the subsequent sections, we will mostly focus on research that has been conducted in the framework of Construction Grammar, which in turn has been singled out as a highly promising approach for studying language evolution by multiple authors (e.g., Reference ArbibArbib 2012; Reference HurfordHurford 2012; Reference Johansson, Roberts, Cuskley and McCrohonJohansson 2016; Reference Hartmann and PleyerHartmann & Pleyer 2021). But this focus on Construction Grammar does not mean that other cognitive-linguistic approaches could not offer equally relevant insights. To mention just one recent example, Reference SchmidSchmid’s (2020) entrenchment-and-conventionalisation model, which synthesises elements from multiple influential cognitive-linguistic approaches, can be considered a promising overarching framework for studies on language evolution as well.

7 A Usage-Based Perspective on the Evolution of Language

In this Element, we have discussed a number of key topics central to integrating cognitive linguistics and language evolution research. Through this, we have illustrated how both fields can enter into a productive dialogue. On the one hand, this concerns how cognitive linguistics can be informed by – and correlates with – research on language evolution, especially when it comes to research on (a) the cognitive and social foundations of language and interaction (Sections 2, 3, and 4), and (b) the evolutionary dynamics of the emergence of structure in language (Sections 5 and 6). On the other hand, it concerns how cognitive linguistics can inform the study of language evolution with regard to the key topics discussed here. Minimally, we have shown that these topics, as well as others, need to be integrated if we want to arrive at a cognitive-linguistic account of language evolution. The massively interdisciplinary nature of language evolution research of course makes this a highly complex undertaking. So while we have given a comprehensive overview of central topics from a perspective integrating cognitive linguistics and language evolution research, we have in essence still only scratched the surface of relevant topics where both fields can cross-fertilise each other.

Here, in conclusion, we want to adopt a broader perspective and show two principal ways in which cognitive linguistics and language evolution research can be integrated: one concerns the utility of cognitive linguistics in spelling out the foundations of language evolution. The other concerns the broader contribution of cognitive linguistics, usage-based approaches, and evolutionary linguistics to understanding language as a complex adaptive system.

The first principal way of integrating cognitive linguistics and language evolution research entails a kind of ‘shopping list’ approach where cognitive linguistics helps in outlining the cognitive and interactional mechanisms and processes that have to be present in a ‘language-, interaction-, and construction-ready brain’ (Reference ArbibArbib 2012; Reference PleyerPleyer 2023) in order to make the evolution of language possible. This includes, for example, the human ability for symbolic cognition, a capacity for massive storage of a network of constructions in memory, statistical learning and pattern-finding, processes of entrenchment of constructions in memory related to frequency and usage effects, capacities for abstraction and schematisation, and sociocognitive capacities such as perspective-taking, ostensive–inferential communication, shared and collective intentionality, the ability to converge on shared conventions, the dynamic, interactive co-creation of meaning in interaction, as well as others. On the other hand, this relates to specifying the processes that lead to the emergence of structure over repeated interactions in communities of practice and over the course of cultural transmission. It is here that cognitive linguistics and usage-based approaches can make significant contributions to language evolution research, as they show how language and structure emerge from interaction and usage. As discussed before, it is especially work on historical language change that can yield insights into processes that have not only led to the emergence of structure in living languages but can also explain the emergence of the first (protolinguistic) constructions both in interactional face-to-face encounters and their increasing conventionalisation and transmission within communities.

Cognitive-linguistic, usage-based approaches on the one hand, and evolutionary linguistics on the other, are highly compatible in their overall conceptualisation of language. Specifically, usage-based approaches (e.g., Reference Beckner, Blythe and BybeeBeckner et al. 2009) and evolutionary linguistics (e.g., Reference SteelsSteels 2011; Reference Kirby, Tallerman and GibsonKirby 2012) see language as a complex adaptive system. In a complex adaptive system, the global characteristics of the system emerge out of the complex local interactions of different factors in different dimensions and on different timescales. Four timescales are of particular importance in language evolution (cf. Reference Kirby, Tallerman and GibsonKirby 2012; Reference Hartmann and PleyerHartmann & Pleyer 2021; Reference PleyerPleyer 2023; Reference EnfieldEnfield 2014, Reference Enfield2022; Reference SinhaSinha 2015):Footnote ⁸

• The enchronic timescale of language use in context and social interaction

• The ontogenetic timescale of individual language development across the lifespan

• The diachronic (or glossogenetic) timescale of social/cultural historical change

• The phylogenetic timescale of bio-cultural language evolution.

For all timescales, cognitive linguistics and usage-based approaches can make important theoretical contributions. For example, regarding the ontogenetic timescale, usage-based accounts of language acquisition have amassed a wealth of relevant research on the cognitive and social-interactive processes and mechanisms children use to ‘construct a language’ and acquire a network of constructions (e.g., Reference TomaselloTomasello 2003; Reference IbbotsonIbbotson 2020; see also Section 4).

Concerning the ‘to-and-fro of social interaction’ (Reference EnfieldEnfield 2022) that makes up the enchronic timescale, one interesting proposal of how structure emerges in interaction is that of ‘ad hoc constructionalization’ (Reference Brône and ZimaBrône & Zima 2014). In ad hoc constructionalisation, local patterns and temporary constructions emerge and can be used as a shared linguistic resource for the duration of the encounter. Reference Du BoisDu Bois (2014), Reference PleyerPleyer (2017, Reference Pleyer2023), and Reference VerhagenVerhagen (2021) argue that the emergence of such patterns in interaction can serve as the starting point for their increasing recurrence in subsequent interactions, leading to their increasing conventionalisation in a community of practice. This, in turn, as briefly outlined in the model sketch that we will discuss later in this section, then can serve as the foundation for the cumulative cultural evolution of language. As this shows, and as we have seen throughout this Element, a usage-based approach to the diachronic timescale can uncover many of the evolutionary dynamics in language change, as well as shed light on the cognitive and social factors that influence it. One particularly promising model that combines the enchronic, ontogenetic, and diachronic timescales is Reference SchmidSchmid’s (2020) entrenchment-and-conventionalisation model. This approach models the interaction of the cognitive (entrenchment) and the social (conventionalisation) dimension of language, which are connected in a feedback loop through the driving force of usage. On the cognitive and individual dimension, repeated encounters with a particular structure strengthen its storage and representation in memory. On the community dimension, frequently occurring structures diffuse and become established and socially shared within a community of practice. The interaction and mutual reinforcement of these processes leads to language change. As these processes take domain-general cognitive processes as well as social interaction and usage as its starting points, this approach also has the potential to help explain the emergence of the first entrenched and conventionalised protolinguistic structures. In other words, it can be used as a model for the emergence of language through interaction, entrenchment on the cognitive side, and conventionalisation on the community side (Reference PleyerPleyer 2023). Such an approach takes seriously the fact that language is both a cognitive and a social phenomenon (Reference DąbrowskaDąbrowska 2020), and sees their interaction in usage as a key driving force in language evolution. In doing so, it also takes the interaction between individual and population into account more thoroughly, which has been another focus of recent cognitive-linguistic research (e.g., Reference Petré and Van de VeldePetré & Van de Velde 2018; Reference Petré and AnthonissenPetré & Anthonissen 2020). As Reference VerhagenVerhagen (forthcoming) points out, population thinking plays a crucial role in evolution but has arguably often been neglected in linguistic theorising. In particular, he argues that even some recent approaches do not adequately take into account the key insight of population thinking, namely ‘that the emergence of conventionality is a community level causal process distinct from the emergence of cognitive units and routines for speaking and understanding’ (Reference VerhagenVerhagen forthcoming). However, some approaches, including the entrenchment-and-conventionalisation model, but also Reference Baxter and CroftBaxter and Croft’s (2016) account of language change across the lifespan or Reference DąbrowskaDąbrowska’s (2020) adaptation of Keller’s invisible-hand approach, have started to make explicit proposals as to how the relationship between individual and population can be modelled. Also, first attempts have been made to distinguish the levels of individual and population in empirical research. In their analysis of the grammaticalisation of going to, for example, Reference Petré and Van de VeldePetré & Van de Velde (2018) try to tease apart essentially non-social mechanisms of innovation from inherently social mechanisms of propagation.

Overall, this means that cognitive-linguistic and usage-based approaches can contribute to possible scenarios of language emergence by specifying critical processes and dynamics on the enchronic, ontogenetic, and diachronic timescale. To illustrate this potential, we briefly outline a model of how a cognitive-linguistic, usage-based view of language evolution can capture fundamental processes that have the potential to kick-start language evolution: one foundation for this is the aforementioned ‘language-, interaction-, and construction-ready brain’. That is, we take as our starting point basic cognitive and interactional capacities enabling the co-creation of meaning in interactions, and enabling interactants to converge on shared symbolic practices that they could add to open-endedly in repeated interactions. Hominins who first started to develop basic communicative solutions to recurring communication problems would retain those that proved successful. Successful solutions to previous communicative challenges would shape how these challenges were solved in the future (Reference Christiansen and ChaterChristiansen & Chater 2022). They would also make it more likely that these communicative strategies were used with different and more and more communicative partners, leading to their social diffusion across a community. Repeated use of these communicative solutions would then lead to these solutions becoming entrenched (on the cognitive side) and conventionalised (on the community side). Over time, usage-based factors would then lead to a repertoire of cognitively entrenched and socially conventionalised (proto)constructions. That is, the first protolinguistic constructions would emerge on the interactional timescale. Through repeated use, they would then become part of a shared community-wide system, becoming entrenched in individuals on the ontogenetic timescale, and be transmitted through the community, and change through the transmission process, on the diachronic timescale. As research on experimental semiotics, IL, and the evolutionary dynamics of language has shown, over time such systems become increasingly more structured and systematic. They also become more abstract and schematic in nature. This process, then, offers a mechanism that can lead to the gradual transition towards the language pole on the protolanguage–language continuum (Reference Hartmann and PleyerHartmann & Pleyer 2021; Reference PleyerPleyer 2023).

Here, cognitive-linguistic models have much to offer for fleshing out such a scenario.

Regarding the phylogenetic timescale, one further interesting contribution of usage-based approaches is a focus on memory processes, and how entrenchment influences linguistic structure. Reference DivjakDivjak (2019), for example, outlines three types of entrenchment that influence the mental storage of constructions:

• repeatedly activated structures are processed more rapidly and in a more automated fashion (‘What You Do Often, You Do Faster and with Fewer Errors’);

• frequently co-occurring structures achieve unit status, and can be retrieved and accessed simultaneously (‘Units that Occur Together, Refer Together’);

• co-occurring structures fuse together and become chunks (‘Units that Occur Together, Blur Together’).

This means that structural aspects of language are influenced by the nature of entrenchment in memory. To what degree could these processes also help explain the emergence of structure in protolinguistic hominin communication? This is a question to be further investigated. However, just as an illustration, it can already be brought into dialogue with a specific proposal by Reference Planer and SterelnyPlaner and Sterelny (2021) on the emergence of the first composite signs. Planer and Sterelny argue that Homo erectus populations (who lived around 1.9 mya to 100 kyaFootnote ⁹) had a basic mixed-modality protolanguage. In this protolanguage, they argue, repair signals would become increasingly frequent. Communicative repair is a pervasive feature in human interaction (e.g., Reference Dingemanse, Roberts and BaranovaDingemanse et al. 2015b) and can also be found in the communication of other animals, such as great ape gestural communication (Reference Heesen, Fröhlich, Sievers, Woensdregt and DingemanseHeesen et al. 2022). As the degree of collaboration and cooperation increased in erectines, for example in the domain of tool-making, this would lead to more repair sequences. In addition, Planer and Sterelny argue that with increased social cognition and ‘smartness’, interlocutors would anticipate the necessities for repair and integrated repair signals into their initial message. For example, a pointing gesture might at first have been a repair mechanism (‘No, this stone’), before becoming part of a composite sign (‘Pick up stone’ + ‘Pointing at stone/This stone’). Although Planer and Sterelny do not make explicit reference to processes of entrenchment, the process that is described here is captured quite well by the results of entrenchment explicated by Reference DivjakDivjak (2019): repeatedly co-occurring items are stored as units, enabling their simultaneous access and retrieval, and repeatedly co-occurring structures become fused and chunked together, becoming a ‘referential unit’.

A cognitive-linguistic approach also has the advantage that it can help bridge the gap between questions of language origins and language development. As mentioned at the beginning of this Element, it has sometimes been criticised that the term language evolution conflates both aspects. And indeed, it is largely research on the cultural evolution of language that immediately seems highly compatible with the usage-based approach pursued in cognitive linguistics. However, we have also argued that research on the communication systems of non-human animals can partly be understood using concepts from cognitive linguistics. This seems straightforward as cognitive linguistics is interested in the interconnection between language, cognition, and the social and cultural environments in which language use takes place. In addition, it makes sense not to draw a categorical distinction between linguistic and non-linguistic signs but to conceive of language as one semiotic resource among others. This follows out of the gradualist approach of cognitive linguistics, evident in its embracement of prototype theory (e.g., Reference LakoffLakoff 1987) and in concepts like the lexicon-syntax continuum that plays a crucial role in Construction Grammar (Reference Ungerer and HartmannUngerer & Hartmann 2023). This gradualist approach, then, also offers a heuristic framework for modelling the emergence of linguistic signs out of pre-linguistic precursors.

In our discussion of the ‘shopping list’ approach, we have listed some of the cognitive capacities that needed to evolve for a ‘language-, interaction-, and construction-ready brain’, but these of course also need to be complemented with references to physiological and anatomical changes supporting the evolution of language (cf. Reference Győri, Wen and TaylorGyőri 2021). There have been major biological changes during hominin evolution, which served as the platform to make the evolutionary emergence of language via processes of cumulative cultural evolution possible. One of the most obvious ones is the evolution of human brain size, which began significantly increasing around 2 mya. In absolute terms, with an average weight of about 1,400 g, the human brain is roughly three times bigger than the brains of other great apes. In relative terms, the human brain is also significantly bigger than expected for a primate our size (Reference Verendeev and SherwoodVerendeev & Sherwood 2018). The human brain also differs in terms of its neuroanatomical architecture and brain organisation, with an expanded prefrontal cortex (Reference DeaconDeacon 1998) as well as other parts of the cerebral cortex and significant rewiring of fibre tract connections involved in capacities such as tool-making, social cognition, and language (Reference Sherwood, Shackelford and Weekes-ShackelfordSherwood 2019; Reference Ponce de León, Bienvenu and MaromPonce De León et al. 2021; Reference LangdonLangdon 2022: 336–343). Other changes relevant to language include, for example, the evolution of neuroanatomical structures supporting increased fine-grained sensorimotor coordination, motor control, and sequential learning abilities, which are foundational for both spoken and signed language. There were also changes specifically related to the evolution of speech, such as changes in vocal anatomy (see, e.g., Reference FitchFitch 2010: 297–337 for a review), as well as increased vocal and breath control (e.g., Reference Maclarnon, Tallerman and GibsonMacLarnon 2012; Reference Fuchs and Rochet-CapellanFuchs & Rochet-Capellan 2021) and vocal learning abilities (e.g., Reference ZhangZhang 2017). Some of these would have served as pre-adaptations or enabling conditions that were exapted in the evolution of protolanguage. Others would represent adaptations once protolinguistic communication got off the ground (Reference FitchFitch 2010; Reference Maclarnon, Tallerman and GibsonMacLarnon 2012), kick-starting a co-evolutionary feedback loop between (proto)language and the capacities supporting it (cf. Reference JohanssonJohansson 2021). All these changes likely had a gradual and long evolutionary trajectory in the 1.5 million years since Homo erectus (Reference Levinson and HollerLevinson & Holler 2014; Reference Dediu and LevinsonDediu & Levinson 2018) with important foundations of speech present at least since the last common ancestor between Neanderthals and modern humans 500 kya (Reference Dediu and LevinsonDediu & Levinson 2013). Of course, these changes were not restricted to speech. For example, they also include changes influencing the multimodal nature of human communication, such as anatomical and physiological changes related to the biomechanic foundations and muscle systems involved in co-speech gesture (Reference Pouw and FuchsPouw & Fuchs 2022). Many changes likely had complex and wide-ranging effects, such as the evolution of the globular braincase characteristic of Homo sapiens. This development likely was related to significant changes in developmental programmes, neural organisation, and the genetic regulation underlying them (Reference Boeckx and Benítez-BurracoBoeckx & Benítez-Burraco 2014; Reference Meneganzin, Pievani and ManziMeneganzin et al. 2022).

For any scenario of language evolution and the evolution of human cognition, it is a central question which evolutionary forces drove the biological changes and changes in social organisation supporting these capacities. Many different forces and adaptive scenarios have been proposed as driving factors, with different proposals for changes in the evolutionary niche of hominins which acted as selection pressures for the evolution of the capacities underlying language. Many of them are framed in terms of adaptive solutions to ecological, technological, and social challenges faced by hominins in the course of evolution.

For example, some approaches have pointed to ecological challenges that required novel subsistence strategies. These include, for example, coordinated ‘power scavenging’, in which the predator responsible for the kill was chased away in a coordinated group effort (Reference BickertonBickerton 2009), or complex coordinated hunting (Reference TomaselloTomasello 2014), which necessitated more complex communication. Others in turn have highlighted challenges related to the transmission of technological information, with complex communication as a prerequisite for (teaching the skills involved in) the creation of complex tools (e.g., Reference Morgan, Uomini and RendellMorgan et al. 2015; Reference Lombao, Guardiola and MosqueraLombao et al. 2017; though see Reference ShiltonShilton 2019).

Regarding sociality as a driver of human evolution, an influential line of reasoning has connected the evolution of language and cognition to selection pressures associated with the increasing demands of navigating complex primate social groups, especially managing the complexities of bigger group sizes (e.g., Reference Dunbar and ShultzDunbar & Shultz 2007) as well as the growing importance of culturally acquired knowledge and skills (e.g., Reference SterelnySterelny 2012; Reference SinhaSinha 2015). Proposals have been made for competitive scenarios – as in the ‘Machiaviellian Intelligence Hypothesis’, which posits that bigger brains and increased social intelligence evolved in order to manipulate and use others in contexts of social competition (Reference Byrne and WhitenByrne & Whiten 1988) or scenarios focusing on the role of persuasion in communication in such contexts (e.g., Reference Ferretti and AdornettiFerretti & Adornetti 2021). Others have stressed the importance of cooperation in human evolution (e.g., Reference TomaselloTomasello 2008; see also Section 4), with conflict management and coordination of cooperation as primary drivers of the evolution of human behaviour (e.g., Reference LeeLee 2018; Reference Newson and RichersonNewson & Richerson 2021). Language, in these models, has evolved for the management of interactions in social groups. For example, Reference DunbarDunbar (1996) has proposed that language arose out of a form of vocal social grooming required to maintain bigger group sizes. In a similar vein, Reference Benítez-Burraco and ProgovacBenítez-Burraco and Progovac (2020) have argued that humans have undergone a process of self-domestication in which language evolved for social coordination and the management of aggression. Relatedly, Reference Wacewicz and ŻywiczyńskiWacewicz and Żywiczyński (2018) have argued that the evolution of a cooperative, community-wide ‘platform of trust’ was a fundamental prerequisite for the development of language-like systems in hominins.

Many of these approaches see social complexity, cooperative subsistence activities, and the transmission of changing cultural information as responses to ecological challenges that led to the evolution of specifically hominin ways of living and communicating (Reference SterelnySterelny 2012; Reference Newson and RichersonNewson & Richerson 2021; Reference Planer and SterelnyPlaner & Sterelny 2021). A number of approaches have specifically highlighted the role of climate variation as a driver in the evolution of culture. On this view, the heightened instability of hominin habitats acted as a selective pressure for the evolution of a ‘cultural niche’ scaffolding the development and acquisition of complex cultural behaviours and skills (cf. Reference PottsPotts 2013). Given the importance of acquiring cultural behaviours and knowledge to adaptively respond to a variable and insecure environment, this placed a special demand on children being able to acquire this knowledge in ontogeny and the reallocation of resources to create a developmental niche representing a suitable learning environment.

These ideas are linked to the framework of ‘niche construction’, which stresses that organisms can modify their environments in a way that has consequences for their own evolutionary trajectories (and those of other species). This includes not only changes in the physical environment (as, for example, in the case of beavers building dams) but also the social and ‘epistemic’ environment. Humans, on this view, have created a particular niche enabling high-bandwidth transmission of cultural information and the development of complex skills as well as cumulative cultural evolution (e.g., Reference SterelnySterelny 2012). This resulted in the evolution of an extended period of childhood dependency and longer lifespans, coupled with highly plastic brain development influenced by culture, as well as a change towards ‘cooperative breeding’, in which children were cared for by multiple caregivers. The coordination of such ‘allocare’ and the importance of acquiring knowledge during childhood necessitated more complex communication, and generally higher cooperativeness, which then ‘unlocked’ the cooperative information sharing characteristic of language (e.g., Reference HrdyHrdy 2009; Reference Burkart, Guerreiro Martins, Miss and ZürcherBurkart et al. 2018; Reference Isler and van SchaikIsler & van Schaik 2012). These changes would likely also bring with them adaptations for teaching and heightened sensitivity to ostensive–inferential referential communication (Reference Csibra and GergelyCsibra & Gergeley 2009; Reference Heintz and Scott-PhillipsHeintz & Scott-Phillips 2023). One aspect of this niche was that it also served as a ‘language-ready niche of development’, which favoured the development of language in hominins with a ‘language-ready brain’ (Reference Odling-Smee, Laland, Botha and KnightOdling-Smee & Laland 2009; Reference SinhaSinha 2015). Cumulative cultural evolution and the niches supporting it would also co-evolve, with more complex cultural behaviours influencing the social structures in which they developed and vice versa, in a feedback loop (Reference SterelnySterelny 2012; Reference SinhaSinha 2015; see also Section 4). This means that (proto)language itself likely acted as a further selection pressure for the evolution of the cognitive, anatomical, and social structures supporting language.

The proliferation of adaptive scenarios has been criticised in the past for being too unconstrained, giving rise to a high volume of potential ‘just-so stories’, which sound plausible but do not have enough actual evidence and theoretical motivation to back them up (e.g., Reference BickertonBickerton 2009; Reference FitchFitch, 2010; Reference JohanssonJohansson 2021). However, recent scenarios have increasingly made progress in integrating a wealth of evidence and theoretical considerations from different disciplines to strengthen their proposals, and as this Element has shown, cognitive linguistics can make significant contributions to such enterprises.

A promising project that must be mentioned here is the Causal Hypotheses in Evolutionary Linguistics Database (CHIELD; pronounced like ‘shield’) by Reference Roberts, Killin and DebRoberts et al. (2020). They have collected a large set of causal hypotheses proposed in theoretical models of language evolution that have been tested in the literature, especially, but not exclusively, in experimental studies (see Section 5). What makes this resource so valuable is that it allows for making connections between different studies, and exploring various hypotheses and the degree to which they have been substantiated in the literature. Importantly, the causal graph set-up of CHIELD forces researchers to make explicit the assumed causal links between different factors (Reference RobertsRoberts 2018).

One thing that should be kept in mind when discussing scenarios of language evolution is that, as noted by Reference Parravicini and PievaniParravicini and Pievani (2019), language considered as a trait is ‘a complex mosaic of sub-traits with different phylogenetic stories’ (see also Reference Boeckx and Benítez-BurracoBoeckx & Benítez-Burraco 2014). This means that the traits supporting language likely evolved following different trajectories and have different evolutionary histories. Similarly, the evolution of human language and cognition likely also was a process of mosaic evolution instead of a single ‘cognitive revolution’ (e.g., Reference Berwick and ChomskyBerwick & Chomsky 2016) representing a sudden ‘crossing of the Rubicon’ towards behavioural modernity (cf. Reference Meneganzin and CurrieMeneganzin & Currie 2022). Neither is it simply captured by a steady and gradual cumulative change towards behavioural modernity (Reference McBrearty and BrooksMcBrearty & Brooks 2000). Instead, the mosaic evolution of behavioural modernity, including language, likely represents a long, protracted, historically contingent process with fits and starts, and interchanging periods of stasis and innovation. It was highly reliant on cultural, contextual, and demographic conditions, which likely took a long time to stabilise and lead to the social environments required to support high-fidelity cumulative cultural evolution of cognitively modern behaviours, including language (Reference Meneganzin and CurrieMeneganzin & Currie 2022; Reference Scerri and WillScerri & Will 2023).

As this Element has shown, cognitive linguistics as a framework is ideally suited to help in understanding both the mosaic nature of the structures supporting language as well as their mosaic evolution.