Language evolution research in the year 2020 A survey of new directions

This introductory paper reviews recent advances in language evolution research and summarizes the contributions of the special issue “New Directions in Language Evolution Research” in the broader context of these developments. Specifically, we discuss the increasing role of multimodality and iconicity, the more integrative view of lan-guagedynamicsthathasarguablybroadenedthescopeof languageevolutionresearch, and recent methodological innovations that allow for a more fine-grained study of e.g. typological distributions or behavioral patterns that can give clues to some of the key questions discussed in the field.


Introduction
Research on language evolution is undoubtedly among the fastest-growing topics in linguistics.This is not a coincidence: While scholars have always been interested in the origin and evolution of language, it is only now that many questions can be addressed empirically.Researchers can now draw on a wealth of data and a multitude of methodological approaches that have been developed in the different disciplines that try to find answers to what has been called "the hardest problem in science" (Christiansen & Kirby, 2003).Importantly, any theory of how language may have emerged requires a solid understanding of how language and other communication systems work.As such, the questions in language evolution research are not just manifold but also interface with key open questions in historical and theoretical linguistics: Where does human language come from and what makes it unique compared to animal communication systems?How do cognition, communication and transmission shape grammar?Which factors can explain linguistic diversity?How and why do languages change?To what extent is the structure of language(s) shaped by language-internal and external environmental factors?
The aim of this introductory article is to give a brief overview of current new directions in language evolution research that are taken to address these questions.As Dediu & de Boer (2016) have noted, the field of language evolution research is currently coming of age, and has developed a rich toolkit of methods both for comparative research, which investigates the commonalities and differences between human language and animal communication systems, and for studying the cumulative cultural evolution of sign systems in experimental settings, including both computational and behavioral approaches (see, e.g., Tallerman & Gibson, 2012;Fitch, 2017).In addition, large-scale typological studies have gained importance in recent research on language evolution (e.g., Evans, 2013;Blasi et al., 2016;Majid et al., 2018;Jackson et al., 2019).Here, we survey important recent developments in the field, and discuss how the papers collected in this special issue of Language Dynamics and Change can be seen as representative of these new directions.Note that the special issue is split into two volumes, which is why some of the papers summarized here will appear in the next issue of LDC.

2
Quo vadis language evolution research?An overview of new directions The view that there has been a "gap" in language evolution research between the infamous ban of discussions relating to the emergence of language by the Societé de Linguistique de Paris in 1861 and an alleged resurgence of the field in 1990s has long been identified as a myth (see, e.g., Fitch, 2010: 16).However, it is true that the field has seen a surge since the late 20th century, and it has Language Dynamics and Change 10 (2020) 3-26 evolved considerably over the past two decades.In this section, we briefly outline a number of major trends that are also taken up in the contributions to the special issue: the increasing importance of multimodality and iconicity in accounts of the evolution of language; taking into consideration contextual and environmental factors more thoroughly; and the growth of interdisciplinary approaches in both theory and method.
Regarding the relationship between theory and method, the question of causality has been discussed intensively.Various interacting factors have led to a shift in perspective here: First, the notorious problem of spurious correlations has been pointed out in various papers (e.g., S. Roberts & Winters, 2012, 2013).Second, the "replication crisis" (e.g.Open Science Collaboration, 2015) has affected virtually all domains of empirically-based scientific inquiry, including linguistics, and third, language evolution research often relies on very indirect evidence.This makes it attractive not only for theoretical "just-so stories" but also for both far-reaching and far-fetched empirical hypotheses, which increases the problem of spurious correlations and makes the sanity checks that have been called for in the wake of the replication crisis all the more necessary.
The key question, then, becomes: Which hypothesized causal relationships are robust and what kind of evidence can support those causal links?Many of the variables discussed in language evolution research are causally linked and interact with each other in different ways.Therefore, in order to avoid the problem of spurious correlations, a new approach, which S. Roberts (2018a) terms the 'maximum robustness approach' , is gaining momentum.Instead of focusing on simple relationships, e.g., environmental variable X correlating with linguistic trait Y, the idea is to systematically construct more complex and coherent causal graphs (Pearl, 2009) that are based on all the available evidence as well as hypothesized links between multiple interrelated variables.CHIELD, a novel database for hypotheses in language evolution, allows for exploring such graphs and identifying complementary or conflicting relationships as well as empirical gaps that lack evaluation, which can help design empirical research addressing those issues and uncovering actual causal mechanisms (S.Roberts, 2018a, S. Roberts et al., 2020).

Multimodality and iconicity
The modality where language originated has been hotly debated: According to the "gesture-first" view, "language evolved initially from manual gestures with vocal elements gradually added" (Corballis, 2012: 383; see also, e.g., Corballis, 2003, Tomasello, 2008, Arbib, 2012, among others;see McNeill, 2012, for an overview and criticism).The "speech-first" view defended by e.g.Dunbar (1996) and MacNeilage (2008), by contrast, argues that language must always have existed in the vocal-auditory modality given its present-day dominance (see, e.g., Zlatev et al., 2017: 453).These different theories, which go back even further than the works cited here (see, e.g., Fitch, 2010 for a historical overview), are now increasingly replaced by more nuanced and fine-grained views that argue for a complex interplay between auditory and visual channels in the early multimodal beginnings (e.g.Kendon, 2017;Perlman, 2017).For example, regarding the visual-spatial modality, the idea of "pantomime-first" has been put forward as "a theoretical proposal that is arguably distinct from both gesture-first and multimodal-first" (Zlatev et al., 2017: 473; also see Żywiczyński, Wacewicz & Sibierska, 2018).
In line with the strongly empirical outlook of modern language evolution research, a number of studies have set out to test the predictions of those different theories empirically (e.g., Zlatev et al., 2017;Macuch Silva et al., 2020).Across modalities, the key role of iconicity has received increased attention (e.g., Perniss, Thompson & Vigliocco, 2010;Perniss & Vigliocco, 2014;Dingemanse et al., 2015;Pleyer et al., 2017) with the similarities and differences between modalities in the affordances for iconicity being highlighted (Perlman et al., 2018) and methodological innovations developed to assist systematic study (Motamedi et al., 2019a).
Importantly, these theories go in tandem with a more general development in the language sciences noted by e.g.Wacewicz & Żywiczyński (2017), namely that face-to-face interaction represents "the core ecological niche for language" (Torreira, Bögels & Levinson, 2015: 1).Language is therefore more and more regarded as a multimodal phenomenon.In particular, the role of gesture has received increasing interest (see, e.g., Müller et al. [eds.], 2013).The affordances of the gestural modality for grounding communication systems have been investigated quite extensively.On the one hand, experimental studies have shown how participants develop new gestural communication systems from scratch (Galantucci, Garrod, & G. Roberts, 2012;Motamedi et al., 2019b).On the other hand, the emergence of new sign languages such as Nicaraguan Sign Language (NSL) and Al-Sayyid Bedouin Sign Language has supplemented theories of language emergence with natural data.
Carla Hudson Kam and Oksana Tkachman add to this line of research by investigating the role of interpretability.In particular, they focus on the use of spatial modulations to encode concrete spatial relations on the one hand and more abstract grammatical relations on the other.Starting from the finding that the way entities are placed in signed space seems to facilitate the way grammatical relations, but not actual spatial ones, are understood in NSL, they conduct a series of experiments with hearing non-signers-adults and children-testing Language Dynamics and Change 10 (2020) 3-26 the hypothesis that the locative use of space is less directly interpretable as it requires more interpersonal coordination and thus takes more time to conventionalize.In a first experiment, participants were presented with a short narration mentioning two characters, with a pronoun referring back to one of them.Depending on the condition, the narrator used no gesture, the same gesture for the character and the pronoun referring to this character, or different gestures for the character and the pronoun.Hudson Kam and Tkachman's results show that both children and adults interpret pronouns differently based on co-speech gestures.In a second experiment, participants also watched short narratives involving two characters, with the narrator producing co-referential localizing gestures indicating the spatial location of the two characters (left vs. right).In a picture selection task, participants were then asked to indicate which picture best described the story they had heard.They were shown two versions of the same scenario that differed in the relative spatial location of the characters.While the adults showed some limited consistency in their interpretation, the children did not.They conclude that adults and older children, even if not familiar with a signed language, take in spatial information conveyed via co-speech gesture while younger children do not, possibly because the interpretation of spatial information is quite demanding in that it requires inferences about spatial configurations as well as about perspectivation.
Whereas Hudson Kam & Tkachman are concerned with integral features of sign languages that do not arise in spoken language, the study by Katie Mudd, Connie de Vos and Bart de Boer uses data from an emerging sign language for modelling language persistence.Building on a mathematical model of sign language persistence (Aoki & Feldman, 1991), they inform their agent-based model with real-world data.In particular, they investigate the impact of extralinguistic factors such as the number of deaf individuals in the population, the number of hearing carriers of the deaf allele, population size, and mating patterns.In their model, the sign language does not persist in a small population, while the proportion of deafness in the population only leads to non-persistence if it reaches an extremely low threshold.However, they show that for both results, there are counterexamples in existing emerging sign languages, and they discuss additional factors that may be necessary to explain the persistence of sign languages.

2.2
An integrative approach to language dynamics Another important trend in language evolution research is that its scope has extended quite significantly.This can partly be seen as a by-product of a different but related development, namely the increasingly widespread view of language as a complex adaptive system, i.e. a system whose global properties arise from multiple independent interactions on the local level (e.g., Steels, 2000;Beckner et al., 2009;Kirby, 2012), which entails a more unified approach to the ontogentic, phylogenetic, and "glossogenetic" (Hurford, 1990) dimensions of language dynamics.This is why the term "language evolution" is sometimes used without terminological consistency, as Haspelmath (2020) notes: On the one hand, it can refer to the biological evolution of the biological capacity for language (called "linguisticality" by Haspelmath), on the other hand, it can refer to the cultural evolution or diachronic change of particular language systems (also see Mendívil-Giró 2019, Hartmann in press).While Haspelmath (2020) argues that "[t]he comparison of languages does not lead to immediate insights about human linguisticality", various scholars have defended the view that investigating present-day distributional patterns can yield valuable clues to the emergence of language.For instance, Heine & Kuteva (2007, 2012) argue that "typological generalizations on language structures" (Heine & Kuteva, 2007: 55) can be used to reconstruct the "genesis of grammar" with the help of grammaticalization theory.In a similar vein, Bybee (2010: 202) emphasizes the relevance of investigating historical language change for understanding language evolution: "there is every reason to suppose that the very first grammatical constructions emerged in the same way as those observed in more recent history".However, note that both Bybee and Heine & Kuteva hedge their hypotheses carefully, emphasizing that their accounts can only give clues to the origins of grammatical units, not language per se.However, as e.g.Antonio Benítez-Burraco points out in his paper, investigating both typological and grammaticalization patterns can point to environmental and social driving forces in cultural evolution, which in turn can indeed be relevant for understanding the origins of the human capacity for language.While Haspelmath's (2020) distinction between "evolution of linguisticality" and "evolution of languages" reminds us that both biological and cultural factors are at play in language development, it is also important to keep in mind that we cannot clearly tease apart the emergence of the language capacity from the cultural evolution of language(s).This point is made in a particularly forceful way by coevolutionary models, which assume that the evolution of language, culture, and biological features interact (see, e.g., Evans, 2013;Kirby, 2017).In the present special issue, Benítez-Burraco makes a strong case for this hypothesis.Arguing against the view that "the evolution of language" on the one hand and "the evolution of languages" on the other can and should be teased apart, he points to recent evidence according to which (i) human cognition might not be as uniform as is usually assumed and might have changed over our recent history, and (ii) the structure of languages is partly shaped by social, cultural, and environmental factors.He therefore argues for an eco-Language Dynamics and Change 10 (2020) 3-26 evo-devo account of language evolution, emphasizing the role of human selfdomestication, i.e. the emergence of traits typically associated with domestication in biological evolution, but without external influence.
Studies investigating historical language change have shown that methods from evolutionary biology can be fruitfully applied to the reconstruction of language history (see, e.g., Gray, Drummond & Greenhill, 2009, Syrjänen et al., 2016, Bouckaert, Bowern & Atkinson, 2018).Additionally, theoretical approaches like Croft's (2000) evolutionary approach to language change or Blevins' ( 2004) and Ritt's (2004) evolutionary approaches to phonology have shown the potential of mutual cross-fertilization between linguistics and evolutionary theory.Language itself has come to be viewed more holistically.On the one hand, as pointed out in Section 2.1, language is increasingly seen as a fundamentally multimodal phenomenon.On the other hand, despite its undisputed special status, language is no longer seen as the "odd one out" among our cognitive and cultural traits.Instead, an integrative view sees language as bearing important commonalities with other phenomena, which is why phenomena as different as the construction of paper airplanes (Caldwell & Millen, 2008), the evolution of a collaborative art project (Müller & Winters, 2018), diachronic changes in ancient coin designs (Pavlek, Winters & Morin, 2019), and the evolution of language(s) can be investigated in a shared theoretical framework.The 2016 founding of the Cultural Evolution Society bears witness to the appeal of such an integrative approach to language, cognition, and culture.
These ideas are of course not entirely new.Among others, Humboldt (1836) has already argued for a close connection between language, cognition, and culture, even though many of his ideas are by now outdated, to say the least (e.g., the assumption that the complexity of a language reflects a nation's degree of civilization).What is (relatively) new, however, are many of the methods that can now be used to study these relationships in more detail.This arguably also entails a more thoroughly interdisciplinary outlook: While language evolution research has been a highly multi-disciplinary field from the outset, it could be argued that this interdisciplinarity has often consisted in combining insights from studies that themselves largely remained within disciplinary boundaries.Many more recent studies, however, cross these boundaries by taking a strongly interdisciplinary approach already in the design of their studies.For example, Dediu & Levinson (2013, 2018) argue that Neanderthals had vocal language on the basis of archaeological as well as genetic evidence, and Tylén et al. (2020) combine the analysis of archaeological findings with behavioral experiments to investigate the evolution of symbolic communication.
The interpretation of archaeological findings for studying language evolution entails some obvious challenges: For instance, it is a matter of debate what kinds of archaeological records constitute evidence for symbolic or even linguistic behavior (see, e.g., d 'Errico & Vanhaeren, 2009;Botha, 2012).This is why Cory M. Stade proposes to use Theory of Mind (ToM) as a proxy for language ability, rather than relying on symbolism.She argues that ToM might be better identifiable in the archaeological record as certain social behaviors have been shown experimentally to require theory of mind.She mentions a number of advantages, e.g. the strong correlation that has been found between ToM and language, which has led to the widely-shared assumption that both have co-evolved (see Woensdregt, Cummins & Smith, forthc.for a computational model).Stade's study is particularly relevant in light of recent discussions about ToM in general (see e.g.François & Rossetti, 2020;Sulik & Lupyan, 2020) and may give rise to methodological discussions about how ToM can be defined and operationalized more thoroughly.
Representing the explorations of interdisciplinary connections, Gareth Roberts and Betsy Sneller make a case for increased collaboration between sociolinguists and language evolution researchers by comparing the five foundational problems in language change (Weinreich, Labov & Herzog, 1968) and the four foundational questions in ethology (Tinbergen, 1963).The paper discusses resemblances between sociolinguistic and cultural evolutionary approaches that both deal with population-level problems like the spread and propagation of features as well as opportunities for integrating them in a mutually informative way.Roberts and Sneller further show how experimental paradigms that are usually used to study language evolution can be used to test sociolinguistic theories of language change, and how the study of natural language change can in turn inform experiments in cultural language evolution.An interesting point that they note is that the questions driving language evolution research today were already embraced by Weinreich, Labov & Herzog in their formulation of a general science of language change.

2.3
Methodological innovations In his recent detailed review, Fitch (2017) already lays out the wide varieties of methods and interdisciplinary approaches that are now available to researchers from various fields to study the components of the human language faculty, drawing on comparative biology, neuroscience, paleontology and genetics.Similarly, there has been an explosion in experimental studies addressing the cultural evolution of linguistic structure in the lab, ever since pioneering studies showed that both the grounding of communication systems from scratch in horizontal interaction (Galantucci, 2005), and the verti-Language Dynamics and Change 10 (2020) 3-26 cal cumulative cultural evolution of language over generations (Kirby, Cornish & Smith, 2008) can be simulated with human participants in controlled settings.These approaches are now referred to as 'experimental semiotics' (Galantucci, Garrod & G. Roberts, 2012) and 'iterated learning' (Kirby, Griffiths & Smith, 2014).They initially served as empirical validations of early computational models that suggested linguistic structures could emerge through interaction and transmission (e.g., Steels, 1995;Kirby & Hurford, 2002).However, since then, studies using artificial communication systems have matured into a range of paradigms that have been used to study the emergence and evolution of linguistic structure and the pressures shaping it (Tamariz, 2017).General observations include different pressures for simplicity/learnability and expressivity/informativeness that can result from horizontal interaction or vertical transmission (Kirby et al., 2015;Motamedi et al., 2019b;Carr et al., forthc.).Other recent studies have scrutinized in what way communication systems are sensitive to properties of the signaling space (Little, Eryilmaz & de Boer, 2017;G. Roberts, Lewandowski & Galantucci, 2015), properties of the meaning space (Carr et al., 2017), the referential context (Winters, Kirby & Smith, 2015, 2018;Tinits et al., 2017), factors related to the mode of communication (Nölle et al., 2018), or information-theoretic principles (Kanwal et al., 2017, Stevens & Roberts, 2019).
This includes, for instance, how languages order semantic constituents (e.g., Goldin-Meadow et al., 2008;Schouwstra & de Swart, 2014;Özçalışkan, Lucero & Goldin-Meadow, 2016;Meir et al., 2017) or modifiers within the noun phrase (Culbertson & Adger, 2014;Martin et al., 2019) as well as how the social environment, e.g., group size (Raviv, Meyer, Lev-Ari, 2019) or social biases (Sneller & Roberts, 2018;G. Roberts & Fedzechkina, 2018), affect the evolution of linguistic systems.In this way, artificial language learning, communication and transmission studies have become powerful tools to study the mechanisms underlying linguistic universals and diversity more generally.This point is echoed by G. Roberts and Sneller, who argue that artificial language experiments could have a significant impact on sociolinguistics and studies of language change.
Besides cross-fertilization with other linguistic subfields, multidisciplinary approaches to language evolution are now commonplace.This is in part due to the realization that language does not evolve in a void but is rather "the product of a multiscale communicative niche construction process at the intersection of biology, environment, and culture" as Coupé et al. (2019: 1) formulate it.There is now increasing evidence that a large number of features that vary widely across languages, such as specific speech sounds (e.g., Everett, Blasi & Roberts, 2015;Moisik & Dediu, 2017) or lexical or semantic categories (e.g., Language Dynamics and Change 10 (2020) 3-26 Brown & Lindsey 2004;Regier, Carstensen & Kemp, 2016) might not just reflect arbitrary cultural 'choices' , but are rather motivated by and sensitive to the specific local physical, sociocultural or technological environment (see Lupyan & Dale, 2016 for an overview).While this kind of 'linguistic adaptation' was initially suggested due to observed correlations, such as the inverse relationship between morphological complexity and population size (Lupyan & Dale, 2010), such findings are now corroborated by phylogenetic evidence suggesting that the historical diversification of language families is better explained by environmental factors such as a climate rather than neutral drift (Bentz et al., 2018;Hua et al., 2019).An ambitious paper by Blasi et al. (2019) illustrates just how rigorous and transdisciplinary more recent approaches to linguistic adaptation have become: Combined results from biomechanical modeling of bite configurations with phylogenetic analyses of language change and ethnographic data suggests that changes in diet during the neolithic revolution have affected the distribution of labiodental sounds such as [f] and [v].These phonemes are significantly easier to produce with an overbite that resulted from the transition into agriculture leading to a softer diet.
Such endeavors are to some extent enabled by a recent surge in large-scale databases enabling what Bickel (2015) calls 'distributional typology' , a datadriven comparative approach that addresses the what, where and why questions regarding language structure with statistical techniques.With Glottolog (Hammarström, Forkel & Haspelmath, 2019), linguists have access to an accurate classification for approximately 6500-8000 linguistic varieties including information on (sub-)families and dialects (see Hammarström, 2016, for an introduction and overview) and there are now many specialized databases for linguistic features (e.g., WALS, Dryer & Haspelmath, 2013), phoneme inventories (PHOIBLE, Moran & McCloy, 2019), basic vocabulary word lists (ASJP, Wichmann, Holman & Brown, 2018) as well as databases that allow linking linguistic and cultural or environmental data (D-PLACE, Kirby et al., 2016).These and other data sources (see http://languagegoldmine.com/ for a comprehensive list) are now increasingly used to investigate general patterns found across languages, such as systematic sound-symbolic meaning associations in the lexicon (Blasi et al., 2016) or universals and variation in colexification of emotion words across thousands of languages (Jackson et al., 2019).Similarly, an evergrowing number of corpora enables detailed synchronic and diachronic analyses of single languages and allows characterizing individual-and populationlevel dynamics that are involved in historical language change (e.g.Petré & Van de Velde. 2018).Recent work using English corpora has, for instance, addressed to what extent sensory vocabulary reflects communicative efficiency across semantic domains (Winter, Perlman & Majid, 2018).
Language Dynamics and Change 10 (2020) 3-26 The paper by Andres Karjus, Richard A. Blythe, Simon Kirby, and Kenny Smith represents another example of this line of research.They discuss how large-scale corpora can help understand language dynamics and change.Previous work in the field of "culturomics" (e.g., Michel et al. 2011) has been heavily criticized because "these approaches often present sweeping accounts of the historical development of words and concepts with little or no attention paid to important linguistic clues and principles of good science."(Brezina, 2018: 224) Unlike those previous works, the "topical-cultural advection model" that Karjus et al. propose aims at teasing apart different explanatory factors in order to separate random from systematic effects.Two case studies show that the model is able to capture topical effects in frequency changes of linguistic elements over time.
With the availability of rich data sources, it is tempting to simply search for cross-sectional patterns.For instance, economists have recently begun linking linguistic features such as future-tense marking or pronoun-drop to economic behavior (e.g., Chen, 2013;Kim, Kim & Zhou, 2017;Feldmann, 2019).However, one argument for the 'maximum-robustness' approach discussed in section 2 above is that such large-scale correlational studies are not well-suited to make causal claims, since relationships between linguistic and languageexternal variables are often complex and confounded (e.g., due to historical or geographic relatedness of cultural traits).It has therefore been argued that experiments or case studies are a better way for verifying such relationships (S.Roberts, Winters & Chen, 2015;S. Roberts, 2018b).This is exemplified by the relationship between spatial language and the environment: For a long time, field linguists have speculated that the way in which speech-communities reason and talk about spatial relations could be dependent on the local topographic environment (e.g., Wassmann & Dasen, 1998).However, large-scale surveys did not find simple relationships between spatial linguistic strategies and environmental factors such as climate or subsistence style (Majid et al., 2004).Only more recently, detailed individual case studies and comparisons between environments and speech-communities have found that social and environmental variables such as second language, subsistence style or local topography do indeed explain some of the variance in spatial linguistic strategies (e.g., Bohnemeyer et al., 2015;Palmer et al., 2017;Shapero, 2017).However, it is hard to disentangle the exact causal relationships between such variables.Magnus Pharao Hansen and Carolyn O'Meara therefore argue for an integrated approach to the adaptation of spatial language, through a combination of ethnolinguistic fieldwork and historical studies of language change.They provide a detailed framework for selecting linguistic varieties with regard to their grammatical differences, ambient climate and landscape features, and available historical data that can be used to reconstruct whether an adaptation to environmental factors has taken place in the development of these varieties.Observations from real linguistic data could then be tested with computational models (see, e.g., Spranger, 2016) or human subject experiments testing the impact of the environment on linguistic conventions (e.g., Nölle et al., 2020a).
Russell Richie, Matthew L. Hall, Pyeong Whan Cho, and Marie Coppola provide another example that shows why it is important to address hypothesized causal relations with a mixed-methods approach.Building on previous work from experimental semiotics as well as natural observation and computational modeling, they investigate the role of social network dynamics in language emergence.Using a dyadic gestural communication task, they investigate whether conventionalization differs between two conditions, one with a richly-connected and one with a sparsely-connected communicative network.In the richly-connected communicative network, each member of a group of four participants gets to interact with every other member, while in the sparse condition, one member is involved in all dyads, thus serving as a "hub".They show that conventionalization is greater and grows more quickly in the richly-connected network when measured per unit time, i.e. per round in the experiment; however, the effect disappears when measured per interaction (the richly-connected network condition contains more interactions than the sparsely-connected network condition).A possible explanation is that there are different mechanisms of conventionalization at work: In the sparse network, the person that serves as a "hub" can set a standard that other participants tend to converge on; while in the rich network, participants have to negotiate whose utterances they will converge on, which is why participants in richlyconnected networks do not conventionalize referring expressions faster, per interaction, than those in sparsely-connected networks.In contrast, in a rich network with more interactions, conventionalization happens faster than in the sparse network.
Another promising avenue is the use of novel technology.For a long time, experimenters had to navigate a trade-off between experimental control and ecological validity.However, with the availability of new methodologies like immersive virtual reality (VR), it is now becoming clear that it is possible to design tasks that are both tightly controlled and naturalistic (Peeters, 2019).VR holds a lot of potential for investigating cases like spatial language, e.g., by comparing spatial linguistic strategies across simulated large-scale environments that resemble those that have been speculated to affect natural languages (see Nölle et al., 2020b).As discussed in section 2, it is this interdisciplinary approach that can help address causality and model the actual mechanism underlying cross-linguistic variation more rigorously and incrementally.Lastly, artificial intelligence researchers have recently begun modeling the emergence of compositional structure using deep reinforcement learning algorithms with a similar approach to early evolutionary linguistic models (Kirby & Hurford, 2002).In such models, compositionality emerges in multi-agent learning environments with different signal and meaning spaces to enable coordination (Lazaridou, Peysakhovich & Baroni, 2017;Mordatch & Abbeel, 2018;Choi, Lazaridou & de Freitas 2018).While this trend in AI has begun independently, recent work is beginning to converge with findings in language evolution research, for instance by introducing generational transmission as a cultural evolutionary mechanism (e.g., Cogswell et al., 2019;Ren et al., 2019).

Conclusion
The papers in this special issue show that the field of language evolution research has evolved substantially in recent years.Various new approaches, both theoretical and methodological, have been developed that shed new light on the key questions of language evolution research.But the new directions we have sketched here have not only influenced the way we address these questions.Instead, we have also seen a shift in focus regarding the research questions that are asked: Traditionally, the question of when, how, and why the capacity for linguistic communication evolved has been at the center of language evolution research.As we have seen, these questions are still important: For instance, archaeological and genetic evidence keeps providing new insights on the antiquity of language.More and more, however, the focus has shifted to uncovering the co-evolutionary dynamics of language, cognition, and culture.This has also entailed a reconceptualization of the scope of language evolution research, which now involves the study of language dynamics and change more generally from a cultural evolutionary perspective.The contributions to this special issue reflect this tendency.They also show the strongly interdisciplinary outlook of modern language evolution research in discussing how insights from previously neglected fields can contribute to our understanding of how language evolved and continues to evolve.