2.2.1 Systemness

In CDS theory, “a system is a whole of some sort made up of interacting or interdependent elements or components integrally related among themselves” (Mobus & Kalton Reference Mobus and Kalton2015:73–74). This seems a rather trivial notion, but of course it constitutes a fundamental prerequisite for complexity to emerge. Systems are to be understood by knowing the units they are built of (and possibly relevant associated attributes) and the relations between them.

It is self-evident and axiomatic in linguistics that this holds for language as well, that languages are systems. The idea was introduced by Saussure (1916), the founding father of structuralism, who argued that language is a “system of signs” (and signs, in turn, are defined as pairings of signifié and signifiant, meaning and formal expression). Systemic relations mean that synchronically, at a given point in time and in a single language, units on several levels (phonemes, morphemes, lexemes) enter mutual relations of various kinds. Fundamentally, these relationships are either syntagmatic or paradigmatic. In syntagmatic (“chain”-type) relations items follow each other and jointly build a higher-order unit. For example, sounds build words (/m/+/æ/+/n/ > man), words build phrases (the+old+man is a prototypical noun phrase), and phrases build clauses ([The old man]_NP [is writing]_VP [a letter]_NP.). Paradigmatic (“choice”-type) relations hold virtually between units that can be exchanged for each other (can undergo “substitution” and stand in “contrast” to each other): they can fill the same slot, they function equivalently (and thus functionally defined constituent classes can be identified). For example, in the word man, the final consonant can be substituted by /t/, yielding another word, mat, and in the clause the object noun phrase slot after writing could also be filled by a single word (something) or a clause (what he cannot tell), showing the functional equivalence of these types of units. Evidently, any linguistic construction in any language builds on and exemplifies relationships of that kind. And the examples illustrate the important distinction and relation between function (what a unit does in context) and form (how it can be realized), similar to Saussure’s sign concept and leading to the notion of constructions (see Section 3).

2.2.2 Complexity

This property, embedded in the name of CDS, is characterized by a very large number of independent agents or objects with specific attributes who keep interacting in manifold ways and exchanging information (including attributes) locally, but without a steering authority. In these interactions, which may be both collaborative and competitive, they may congregate to build larger, more complex entities and hierarchies to serve specific contextual needs and functions (e.g. cells build organs, and organs are core parts of a body). It is noteworthy, however, that being complex in the CDS sense is not the same as merely being “complicated,” that is, consisting of many entities with relations that are difficult to understand.Footnote ⁶ The latter certainly constitutes a condition for the former, but complex systems are “quantitatively difficult to keep track of,” involving “properties of many different independent components” (Jensen Reference Jensen2022:3). They are aggregates, characterized by the entire set of properties discussed here, most importantly emergentism and auto-organization. In other words, it is characteristic that new entities are being built, that the emerging whole is more powerful than the sum of its parts,Footnote ⁷ and cannot simply be deduced from knowing its components. In principle, however, complexity itself is a complex notion, a view of a world perched between order and disorder that is best looked at through the lens of a specific discipline applying it (Krakauer Reference Krakauer2023).

The growth of complexity is of course closely related to the development of network relations and the emergence of new higher-order entities, possibly with new functions. Complex systems evolution therefore tends to increase diversity and functional specialization (Holland Reference Holland2014:44–45).

The relevance of this principle to languages is equally self-evident and uncontroversial. As stated earlier, in languages, units combine to build larger sequences and constituents, steered by functional needs, and the resulting complex entities are often not transparent and information-richer than the set of components. In this process, patterns and components of various kinds and magnitudes interact, both fundamentally in CDSs and in language as an application case. Consider Krakauer’s (Reference Krakauer2023) statement: He illustrates “complexity science” as steering a compromise, at a certain level of organization, between constraints of construction, constraints of matter, and fit defined as the level of granularity of observations. To me this sounds very much like a description of language evolution, where (as many examples in this Element show) we observe interactions between language levels (e.g. phonological changes triggering morphosyntactic rearrangements), between constituents and constructions (e.g. changes of one sound pattern influencing articulatory conditions for another, words copying meanings, or constructions expanding their properties and range of applications), or between languages (in multilingual contact settings). All of these processes increase complexity.

Complexity in languages has been a fashionable topic of linguistics in recent years (see Miestamo et al. Reference Miestamo, Sinnemäki and Karlsson2008; Sampson & Trudgill Reference Sampson, Gil and Trudgill2009; Kortmann & Szmrecsanyi Reference Kortmann and Szmrecsanyi2012; Culicover 2013; Newmeyer & Preston Reference Newmeyer and Laurel2014), though often this relates merely to different levels of being “complicated” as just defined. Tamaredo (Reference Tamaredo2020:ch. 2) offers a useful and competent survey of these discussions, including the tendency to reject the erstwhile axiomatic assumption of all languages being equally complex, definitions of the notion based on the difficulty of acquisition and use (for speaker and hearer), the role of contact (generally assumed to cause simplification; see Schreier Reference Schreier2005), and attempts at measuring complexity quantitatively. Lund and colleagues (Reference Lund, Fossali, Mazur and Ollagnier-Beldame2022) repeatedly thematize the relationship between a wider, general and a more narrow, technical, CDS-inspired understanding of complexity.Footnote ⁸

Complexity in language operates on two levels, the intralinguistic and extralinguistic ones.

Language-internal relations are conventionally understood as hierarchies of units that are mutually constitutive and that build higher-order functional units, as just shown. And this relation expands beyond sentence constituents to the level of utterances and discourse: clauses are constituents of utterances and sentences, which build discourse and texts. An additional, recursive level of complexity arises from the fact that some of these units can be mutually hierarchically constitutive: phrases are constituents of clauses, and clauses can be constituents of phrases. In line with CDS definitions of complexity, its understanding in linguistics correlates with the number of units involved and the nature of their hierarchies. Strictly language-internal complexity also arises from ease of processing and acquisition including features like predictability and regularity. For instance, Tamaredo (Reference Tamaredo2020) mentions “irregularities, new semantic/pragmatic distinctions into the grammar, or distinctions that are not transparently coded” (35) as characteristic features increasing complexity.

Language-externally, speakers can be viewed as agents in line with CDS conditions: there are large numbers of interactants, with different acquisition input histories and sociocultural backgrounds and consequently idiosyncratic idiolects, adopting varying identities and social roles and attributing different prestige assignments to their utterances. Clearly this offers room for all kinds of complexity to emerge. And the two levels also combine and interact in multiple ways: the usage of specific language forms co-varies with sociolinguistic parameters like class, gender, or style; speakers may select and enregister particular forms as increasingly indexical of social parameters; and the entire set of formal choices to encode a message and functional associations undergoes continuous variation and change – which leads us to the next CDS property.

2.2.3 Perpetual Dynamics

CDSs are always in flux, marked by continuously ongoing changes, adjustments, and modifications of system components in interaction and their relations over time, a perpetual process that has no starting point and never stops. In fact, in complexity science, it is not the agents, components, or “objects” that are viewed as central but the processes themselves; the world is “perceived as a nested hierarchy of processes” (Jensen Reference Jensen2022:13).Footnote ⁹ One interesting question is whether change proceeds fairly continuously or rather via “punctuated equilibria,” with stretches of stability (periods of stasis) typically interrupted and altered by bursts of intense change (in line with the S-curve pattern in Figure 1; cf. Kretzschmar Reference Kretzschmar2015:122–23).Footnote ¹⁰ “[T]he dynamics of complex systems commonly involve intermittent bursts of activity” (Jensen Reference Jensen2022:71).

There are some general principles and options of how systems may or tend to evolve (see Jensen Reference Jensen2022:ch. 3), but due to the large number of variables involved, the possibility of new players (e.g. through mutations) emerging on the scene, and possible special effects of non-linear but also linear dependencies, forecasting future system states is practically impossible. Many systems tend to oscillate between forces strengthening order or disorder, respectively, and thus to return to near-equilibrium states. Typically, there are “lever points” (Holland Reference Holland2014:25), where small directed actions cause substantially magnified effects (remaining within the same systemic arrangements) through aggregation. However, disruptive processes leading to different system states, often rather quickly, are also possible, for instance when some variable moves beyond a critical “tipping point” producing abrupt changes, or when exponential increase with feedback effects (a “more-begets-more” process) produces a distribution of a different magnitude. Also, neither effect sizes nor agent properties are stable, so, for instance, mutated properties may substantially affect the evolutionary directionality of a system or subsystem.

Again, this is equally self-evident and uncontroversial for languages and English(es). Language transmission from one generation to the next, also involving constant processes of change, is equally perpetually rolling. In fact, linguistic forms and habits are not even or mainly passed on across generations, but this transmission process happens all the time, irrespective of time scale: every single utterance always manifests and realizes the system, reiterates and reproduces linguistic usage; and every utterance is shaped by and at the same time influences and has the potential to modify future utterances produced by surrounding speakers (see Section 3).

Historical linguistics, the discipline studying the evolution of languages across different periods way back in time, tends to conceptualize select stages (such as Old English) as holistic, fairly stable units, but that is an idealization, too. In reality, transmission from caregivers to infant acquirers proceeded then as it does now, all the time and everywhere – there has been an unbroken chain of transmission, for example, from “Germanic” to “Old English” to “Middle English” to “Early Modern English” to “Modern English,” without any interruption. Parameters and precise conditions varied at different points in time and space, of course, so depending on extralinguistic history more or fewer contact effects will have filtered in, but fundamentally the wheel of evolution and change has always been in motion, for each individual and, compositionally, in society at large.

This process of transmission comprises both elements of continuity, perpetuation across time without substantial modification, with sounds, words and patterns passed on and shared largely identically across periods and varieties, and discontinuity (i.e. change, innovation, and the adoption of elements from contact languages (with which speakers got in touch as a consequence of external conditions)). To illustrate continuity, practically all varieties of English have a sound /p/, a lexeme hand, and a NP pattern with a Det-Adj-N constituent sequence (this interesting novel). In addition, there is always a potential for continuous change, including internal innovation and modification as well as a possible impact of contact effects, bringing in novel forms and habits from other languages. There are numerous publications in linguistics both on principles of language change (e.g. Aitchison Reference Aitchison2012; Chambers & Schilling Reference Chambers and Schilling2013; Bybee Reference Bybee2015) and on processes and effects of language contact (e.g. Thomason Reference Thomason2001; Winford Reference Winford2003; Matras Reference Matras2020; Hickey Reference Hickey2020; Mufwene & Escobar Reference Mufwene and Escobar2022; Adamou & Matras Reference Adamou and Matras2023), and of course there is a myriad of individual studies and documentations of either. Obviously, external, contact-induced modifications are particularly important for investigations of WEs, which always originated in contexts of migration and multilingual interactions, but the elementary internal processes of change are potentially effective there as well, of course.

As an illustration of these principles (with the story to be continued later as to the evolution of WEs), let me briefly recapitulate the development of modal verbs in English(es), from the beginnings to some present-day branches (see Lightfoot Reference Lightfoot1979; Denison 1993:292–339; Fischer & van der Wurff Reference Fischer, Van der Wurff, Hogg and Denison2006:146–52). As is well-known to language historians, in pre-Old English times some Proto-Indo-European perfect verb forms were semantically reinterpreted as (resultative) forms with present time meanings (the so-called “preterite-present verbs”). In Old English these verbs had full verb properties: they were main verbs, and some had direct objects, two properties which their German cognates have retained to the present day (e.g. ich kann das) but which English lost along the way (*I can that). Also, they had nonfinite forms, unlike today’s English modals. In the transition from Middle English to Early Modern English these verbs were reanalyzed as auxiliaries (modal auxiliaries, more precisely), increasingly gaining the distinctive so-called “NICE” properties (being used in special ways for negation, inversion in interrogatives, “code” in elliptical structures, and emphasis); this can be seen as an instance of self-organizing emergence (see 2.2.6). Furthermore, they no longer allow direct objects and cannot be used without full verb predicates (cf. I can do that); they come to lack nonfinite forms (*to can, *caning); and their semantics has changed in distinctive ways, getting restricted to epistemic or deontic meanings (here we see network relations between different verb classes as to their structural behavior in effect). In present-day English many modals have been shown to be losing ground, to the point of being partly disappearing (e.g. in the cases of may for permission, ought to, or shall); conversely, new modals (gonna, gotta, wanna, … ) keep emerging, and the use of semi-modals (be going to, have to, … ) increases (Krug Reference Krug2000; Leech et al. 2009:71–117). So, over more than 1,500 years usage conditions, associated attributes and relations between linguistic entities have continuously been changing, and the story continues.

Remarkably, there is a strong and explicit parallel in modelling the spread of innovations in society between CDS theory and linguistics: both have identified the “S-curve” as a typical representation of the diffusion dynamism, a model of the phase and state transitions that an innovation passes through. An S-curve (see Figure 1) is a sigmoid function, mathematically speaking, which has been shown to describe processes of spread in populations. The horizontal axis shows the progress of time, the vertical axis the intensity of the process in question, e.g. the spread of an innovation.

Figure 1 The S-curve as a diffusion model

(produced by Alan Taylor 2013, Wikimedia Creative Commons 3.0 license)

Figure 1 shows how new properties spread in a population over time, affecting an increasing proportion of a population. S-curves of diffusion show typical phases: a slow start, then a quick rise through the population, and finally a slow trailing off (typically leaving some residuals unaffected by the innovation, a cause of irregularity). This has been observed to be a common type of process in many domains. For CDS, Mobus & Kalton (Reference Mobus and Kalton2015) state: “The S-shape of the logistic function is generated when processes are characterized first by an exponential rise followed by an exponential deceleration to level off at a maximum” (215; see their graph 6.1, which applies the scheme to population biology). Interestingly, the growth of the number of connections in a network, sometimes called the “connectivity avalanche,” follows the same pattern (Seeley Reference Seeley, Bossomaier and Green2000:64), and this applies to the diffusion of innovations in a society in general (Rand Reference Rand2024: section 4).

In sociolinguistics, the S-curve is also the generally accepted model for how linguistic innovations spread in social space, in a characteristic “slow-quick-quick-slow” pattern (Kroch Reference Kroch1989; Labov Reference Labov1994:65–67). A classic application to the history of English is the work by Ellegård (Reference Ellegård1953), who plotted the spread of different innovative functions of the verb do between 1400 and 1700 and also found a rather straightforward S-curve. Change is thus understood as incremental, not discontinuous – but non-linear! A meta-study by Nevalainen (Reference Nevalainen2015) puts the model to the test and finds it largely confirmed and basically descriptively adequate for the vast majority of longitudinal changes investigated. Hence, both in biological CDSs and in language innovations diffuse in identical ways, in a S-curve shape. The fact that both are obviously related, being driven by the same evolutionary principle, can be seen as an important piece of support for the claim that language is also a CDS.

2.2.4 Network Relationships

In CDS theory, “systems are … networks of relations,” “the components are connected in various relations” (Mobus & Kalton Reference Mobus and Kalton2015:137), often indirectly, in a chain-like fashion.Footnote ¹¹ Multiple agents are involved and connected with each other as in a network, but there is no overarching steering authority; all changes happen as consequences of small-scale local interactions, with agents adhering to the same or similar rules. Interactions involve the exchange of information (and possibly as a consequence adjustments of an agent’s attributes). Typically, agents interact only with their immediate neighbors, and have no access to information on the system as a whole or on attributes of distant agents. Still, after many iterative runs of local information exchanges governed by certain rules of mutual interaction and influence, larger-scale systematic patterns may emerge: iterations give way to complexity (Mitchell Reference Mitchell2021, units 2.2–2.4 and 5). Examples of real-life networks vary widely in character, comprising, for example, neurons, airline routes, the internet, the power grid, banks, etc. (Mitchell Reference Mitchell2021, 9.1) Most real-life networks are “small-world” networks, with few links required to any other member via long-distance links and hubs (Mitchell Reference Mitchell2021, 9.3; Buchanan Reference Buchanan2003).

Networks depend on links between the individual agents, and typically in collective systems links are associated with shared properties which are unequal in weight and importance. For understanding the evolution of a network it is therefore important to identify the nodes and assess their relative strength and character. Typically, networks then can be broken down to sub-systems on several levels, based on subsets of commonalities, which participate in the interaction dynamics.

The growth of networks proceeds through several phase changes, often in S-curve-like massive connectivity increases, and it is thus closely related to the notions of emergence and self-organization, criticality (since in some states adding further nodes may tip the balance and cause the system to change states), and transitions between stability and chaos (Seeley Reference Seeley, Bossomaier and Green2000). Structures emerge through the incremental build-up of nodes in digraphs, and possibly circuits, and ultimately more complex configurations. Seeley (Reference Seeley, Bossomaier and Green2000) provides a complex survey of network dynamics and types, marked by different kinds of links and nodes and leading to self-organization and emergence.

Evidently, languages also operate in network relations between many different entities, marked by “complete interconnectedness” (de Bot et al. Reference de Bot, Wander and Verspoor2007:8). As stated before (in Section 2.2.2), this applies and can be viewed from two perspectives, extra- and intra-linguistically: networks between many individuals as speakers, and network interrelationships between linguistic forms and functions.

Language is always communication between speakers, and normally there are many of them, for example in a speech community. And they constitute a social network, obviously: The language-external perspective recognizes the fact that the development of linguistic forms depends upon social relationships and interactions between speakers as agents in a CDS of language. Thomason & Kaufman (Reference Thomason and Kaufman1988), for instance, explicitly stated that “the history of a language is a function of the history of its speakers” (4). And usage-based thinking, advocated here (see Section 3), sets language production by speakers in interaction, and that means network relations, at its center.

In a language-internal perspective, network-like interrelationships exist between linguistic items, i.e. forms and constituents, constructions, and language levels (see Hoffmann Reference Hoffmann2022:9 for an example of a network of ditransitive constructions). Both syntagmatic and paradigmatic relations are cases in point, involving many linguistic units of various kinds to either build something larger together or to fill an information slot; network relations hold between entities both in a chain relationship of compositionality and in a choice relationship of mutual exclusion. For example, in John built a hut the word hut stands in a network relation both with John, build and a (with which it builds a clause) and, by implication, with words like palace, condo or cupboard (which John could have built but, we are told indirectly, hasn’t).

Such network-like intralinguistic systemic relations may hold between individual entities but also between composites of entities and more abstract composites, levels, and attributes, for example across levels of language organization (like phonology, morphology, lexis, or grammar). As a nice example of this, let us have a brief look at the loss of endings in the history of English, and consequently the change of the character of English from a largely synthetic language (with many inflectional endings) to a largely analytic one (with very few suffixes left), a most fundamental transition. Indo-European, Germanic and Old English were strongly synthetic languages, rich in grammatical suffixes.Footnote ¹² The change started inconspicuously, with the “Germanic Main Stress Rule,” which fixed stress on first syllables of words in pre-Old English times. Stressed initial syllables cause a weakening of the phonation stream in later, especially final syllables, which, in turn, produces the weakening of full vowel qualities (to schwa, or, later, nothing) in final syllables. In Old English, this reductive tendency was amplified (presumably) by intensive contact with the Scandinavians’ Old Norse, supporting simplification, here the loss of endings. So gradually, most inflectional endings disappeared (and lack of marking of course also decreases the amount of order in the sub-system). For example, Old English sunu, with a full final vowel, became weakened Middle English sune and ultimately uninflected modern English son. However, the endings had had grammatical functions, expressing intrasentential relations (of identifying the subject or object in a clause, for instance), so their job needed to be done in some other way, compensated for by new grammatical means of expressing syntactic functions. Consequently, English developed a fixed Subject-Verb-Object word order and the strengthening of function words (a process of auto-organizing emergence, if we so wish). A former accusative ending became replaced by the direct object position immediately after the verb; the dative ending gave way to the preposition to plus the noun; the verbal subjunctive was weakened substantially, its function being taken over by modal verbs and modal adverbs, and so on. So the radical change of the character of English, unique among the Germanic languages, was motivated by a series of interacting causes and step-by-step influences, leading from stress to phonology to morphology and to syntax. Clearly these sequences and connections constitute a network, between individual linguistic forms and, more abstractly, systemic patterns (manifested by forms).

2.2.5 The Interplay of Order and Chaos

Chaotic systems, “whose behavior appears to follow a regular pattern but not entirely” (Mobus & Kalton Reference Mobus and Kalton2015:202), marked by reiterations of nonlinear equations and thus ultimately unpredictable feedback loops (251–52), constitute one major form of CDS. In chaotic systems many state variables are coupled, typically (though not obligatorily) non-linearly, sometimes adaptively (so that the conditions of connectedness change, with strategies and principles of evolution modified during the process due to internally selective factors). But their behavior is not completely random (though unpredictable) – after a large number of iterative re-runs they tend to approximate or reach “attractor” states (see Section 2.2.8), where the system returns to self-similar periodic behavior for a while.

Such systems are characterized by the co-existence of order and chaos (in different sub-systems and at different times): there are “pockets of order” inside “chaos” (Gleick 1987), “a complicated mix of ordered and disordered behavior” (Johnson Reference Johnson2009:15), with order arising “in both time and space” (21). This is a consequence of the fact that especially in the worlds of nature and culture noise and regularity compete (Krakauer Reference Krakauer2023). Broadly, relatively orderly sub-systems and relations can be understood as having comparatively few interactants with rather clear functional assignments, while chaotic sub-systems consist of large numbers of interactants with random, non-functional behavior. Transition from order and/or simplicity to chaos and/or complexity proceeds via phases of “turbulence” possibly producing “phase transitions”; CDSs tend to oscillate between both states. Nature exploits chaotic structures in such processes as “an important source of novelty and variety in living systems” (Green Reference Green, Bossomaier and Green2000:45).

While such pockets of order may (and tend to) occur unpredictably, they are not here to stay – “there is a natural tendency for something that is ordered to become disordered as time goes by” (Johnson Reference Johnson2009:25) – but not the other way around. This fact is ultimately caused by a very fundamental law of nature, the second law of thermodynamics, which states that in a closed system entropy (commonly taken to be a measure of disorderliness) always increases, and this process cannot be reversed. In other words, earlier orderly states will never return, but new ones are likely to arise somewhere. A central force generating increased order is feedback within system components, typically involving memory (which is “a form of feedback,” Johnson Reference Johnson2009:67) of earlier interaction effects and possibly boosting it via circular feedback loops (cf. my earlier remarks on S-curve diffusion and the avalanche of network interconnectivity).

Obviously, this applies to languages and Englishes as well: At any point in time in a given variety, there are sub-systems which are relatively orderly (i.e. consist of a relatively small number of units with clear functional assignments and clear mutual delimitation) and sub-systems which are relatively disorderly (marked by a large number of units, irregularity, fuzziness, and functional overlaps).Footnote ¹³ And this is constantly subject to change: ordered subsystems may be disrupted and break down, but conversely systematicity may emerge in formerly chaotic sub-systems.Footnote ¹⁴An example of the first type of change is provided in the next paragraph; in general, it is assumed that rule-governed sound changes, especially if different laws of change interact and compete, may lead to less coherent, more disorderly sets of forms. The mechanism that materializes the second tendency is an increase of conventionalized associations between meaning (function) and form (pattern) in usage, clearly a manifestation of an increase of network relations and order based on feedback and memory as posited in CDS theory.

An example from the history of English, extending back across millennia, is the dissolution of the English strong verb system. For Indo-European, the conventional, generally accepted reconstruction, disregarding variability, posits a perfectly regular and predictable system, with verbs having four functionally distributed principal forms which had different degrees of stress assignment and, correspondingly, vowel variants. By Old English, internal effects caused disequilibria, and the system moved toward turbulence: Various interfering conditioned sound changes, triggered by different following sound environments, produced a rather complex but still somewhat predictable set of strong verb classes and forms, with different modified stem vowels. In the transition to Middle English (ME) the reduction of four principal verb forms to three caused the breakdown of whatever order may have been left. Variability as to which forms survived produced the rather chaotic set of several hundred unpredictable and idiosyncratic irregular verb forms which we find in English today. Even more turbulence, irregularity and chaosFootnote ¹⁵ characterize regional and social dialects, where forms such as bring – brung, fight – fit, sit – sot, fetch – fotch, catch – cotch, take – tuck/tooken, and others appear (Schneider Reference Schneider1989:90–114). On the other hand, English(es) have also developed counter-tendencies toward regaining order. One is regularization (since the establishment of “weak verbs” with dental suffixes for past in Germanic), adopting the cognitive principles of regularity, compositionality, and isomorphism, a consistent association of form (dental suffix –ed) and meaning (‘past’). Some verbs became re-classified accordingly (e.g., Old English helpan–healp–hulpon–holpen became Modern English help-helped-helped), and again, even more so in dialects (where we find forms like knowed, growed, gived, etc.). Secondly, some new regularities of past tense formation have emerged. For example, Cheshire (Reference Cheshire, Stein and van Ostade1994) showed how the “ideaphone” /ʌ/, originally found in past tense forms like flung, wrung, or stung, became more widely associated with the notion of ‘past-ness’, contributing to the formation of further, especially dialectal past tense forms such as run, struck, wrung, done, sunk, drug, tuck, or brung (see Schneider Reference Schneider1989:90–114), and, for example, to the ongoing massive spread of snuck for sneaked in Canada documented by Chambers (Reference Chambers2006–07).

A notable subtype of chaotic behavior which is also obviously shared by CDSs and language is bifurcation. This term describes the fact that a single category splits into two.Footnote ¹⁶ It represents one of possible forms of behavior in chaotic systems described in “catastrophe theory” developed by the mathematician René Thom as a branch of chaos theory (cf. Guastello & Liebovitch Reference Guastello, Liebovitch, Guastello, Koopmans and Pincus2009:12–15). Bifurcations are a simple (”first-order”) type of “catastrophy,” a functional discontinuity and sudden qualitative change at a specific, mathematically derived point, a transition point into chaos where a system may or may not change direction and may suddenly split into two co-existent states.

Obviously, this principle manifests itself in languages and Englishes as well, e.g. in historically documented instances of a single unit splitting into two, which is possible on several levels. An example from English phonology is the fact that Old English had a single high back short vowel /u/, which, however, in the course of time and in different sound environments split into today’s /ʊ/ (as in put, pull, butcher, … ) versus /ʌ/ (in but, pulp, butter, cut, … ). Splits have also occurred in lexis, with one word acquiring two different shapes and meanings – so flower/flour, urban/urbane, metal/mettle, shade/shadow, skirt/shirt, catch/chase, and other pairs go back to originally the same cognate.

2.2.6 Emergentism and Self-Organization

Emergentism and “auto-organization,” often termed “twin processes,” are core properties of CDSs (related to network building) and central components of evolution in general: smaller, less complicated entities interact, form new linkages, and jointly build gradually more complex functional units. This can be seen as a manifestation of the importance of a co-operative, complexity-building principle in life which can be taken to counter an emphasis on competition (e.g. the “survival of the fittest”) and fragmentation (as in structuralism) as basic organizing forces. Such new complex, multi-unit organizational entities adapt mutually and with their environment to fulfill specific functions, producing higher-order, more complex sub-systems and functional organization levels – so the system overall is self-organizing toward creative emergence.Footnote ¹⁷ A “spiraling cycle of increasing complexity,” forming structures and possibly leading to order (from chaos), has been observed. Mobus & Kalton (Reference Mobus and Kalton2015:459–61, 476) describe it as follows: “systems can self-organize (i.e. become more complex) … evolution is understood as a systematically produced trajectory of increasing complexity that need not be teleologically headed anywhere.” Typically, emergence proceeds via steps of gradual synchronization of properties, a kind of diffusion process. For example, the brain requires synchronization of neurons (cf. Minelli Reference 81Minelli, Guastello, Koopmans and Pincus2009), and the financial market builds on uniform herding behavior of traders (cf. Jensen Reference Jensen2022:28–33), “individual oscillatory motion becoming coordinated across many components” (59). As a rather transparent example, consider the growth and development of big cities. Typically, they start out as a small group of settlers staying at a place to exploit its advantages (strategic location, fertile soil, natural resources, etc.). Success begets more success: It keeps growing, attracting more people, requiring the construction of roads, more powerful housing units, and congregation places, the development of specialized professions, the need for an administration, a security force, an educational system, health services, and so on, with each of these sub-systems complexifying and auto-organizing effective rules of their own and the sub-systems interacting with each other.

Importantly, emergence operates on specific levels (Krakauer Reference Krakauer2023), within a certain range of operational parameters, “functionally closed protectorates” where the necessary and sufficient parameters to understand a system’s future states, encoding some degree of entailment, are available. For example, to understand an AI system playing the game of Go it is sufficient to consider the board, moves and strategies, and there is no need to look at software code or hardware components. Similarly, understanding the mind does not require exact knowledge of the physiology of the brain. Emergence operates internally, and in some cases may be seen as something like a “black box” from the outside. Krakauer’s discussion of CDS ontology implies that linguistic systems and sub-systems can be investigated and understood as emergent CDSs without needing to recur to bodily states, sound properties or the quantum-physical composition of speakers. Emergence results from interaction, not substance: “certain aspects of the emergent collective behaviour depend on so-called many-body effects, namely the cooperative collaborative effects generated by the interactions between the components and to a much lesser degree on the internal properties of the components” (Jensen Reference Jensen2022:22).

The diffusion of evolutionary novelty typically proceeds via “symmetry breaking” (Jensen Reference Jensen2022:section 2.5): Individual agents adopt properties against the probabilities defined by the collective of agents, and such a “mutant offspring” (33) may then attract others and spread, thus introducing a sub-system with different properties. “Emergence of structure associated with the breaking of symmetry is a fundamental phenomenon which is seen as underlying the emergence of structure in hugely different situations” (34). In theories of language change this appears to correspond to Croft’s (Reference Croft2000) distinction between “altered replication” (or innovation, when a linguistic entity is being modified in specific reproduction instances) and “propagation” (or diffusion, when a new linguistic form becomes preferred over others, i.e. spreads in the community).

In these processes, adaption, evolution and emergence go hand in hand, highlighting different aspects of largely the same or similar processes. Jensen (Reference Jensen2022) states: “Adaptive evolutionary dynamics can be viewed as both an example of emergence and a very effective motor for generating emergent collective structure” (39), referring to evolution through Darwinian natural selection as a classic example (60–64). Characteristic processes include the emergence of a characteristic scale (at which an innovation most readily manifests itself), robust collective parameters, symmetry breaking (as just defined), the growth of networks, and intermittent dynamics, e.g. via a punctuated equilibrium (39). Adaptation is seen as a co-evolutionary selective pressure which gives direction to random mutations and variation by increasing the formation of similar types (e.g. species). Compared to evolutionary biology, in social dynamics natural selection through adaptation operates not through genome changes but through the reproduction of (modified) properties in interdependent agents, who modify their strategies based on experience (Holland Reference Holland2014:6). This includes cultural evolution – the growth, decline and innovation of “the popularity of an idea and the invention of new concepts and ideas” (Jensen Reference Jensen2022:61). All of these aspects, obviously, apply to the evolution of a language variety as well, when an innovation originates and gets preferred in a community for some reason (perhaps because it is more transparent or simpler in its encoding of a meaning intended). The reproduction rate depends on interaction density with other agents and the fitness of mutations in context, which determine whether properties of co-evolving agents will multiply, mutate further, or die out.Footnote ¹⁸

Again, this principle is evidently applicable to English(es) and languages as well. Larsen-Freeman (2013) states: “As new forms emerge through adaptation and co-adaptation, they self-organize into coherent patterns” (104). Beckner and colleagues (Reference Beckner, Blythe, Bybee, Christiansen, Croft, Ellis, Holland, Jinyun, Larsen-Freeman and Schoenemann2009) put it pointedly: “Linguistic patterns are not preordained by God, genes, school curriculum, or other human policy. Instead, they are emergent” (18). Early work along these lines is a paper by Hopper (Reference Hopper1987), who sees grammar as emergent (discussing essentially what has come to be called “grammaticalization”).

A basic manifestation of this property in languages is simply the principle of compositionalityFootnote ¹⁹: small units jointly build larger constituents with a different function. This is illustrated by any higher-level construction (words built from sounds, phrases from words, clauses from phrases, etc.), e.g. in the build-up of complex lexemes (like washing machine, unhappiness, … ), idioms (e.g. spill the beans, kick the bucket), complex prepositions (in spite of; cf. Bybee Reference Bybee2010), and chunking and phraseologisms (i.e. constructions like To tell you the truth; What is more; After all; … ). And the principle then remains effective and further expands to more abstract compositional patterns. In Construction Grammar, for example, a progress from rather simple and concrete form-meaning pairings (words) toward the growth of increasingly abstract schematic constructions is posited (Hoffmann & Trousdale Reference Hoffmann and Trousdale2013). Examples of such schematic constructions include the indirect object constructions with the schema SVO_indirO_dir (e.g. he gave me a book) or the comparative correlative construction the Xer, the Xer (e.g. the older, the merrier) (Hoffmann Reference Hoffmann2019). In fact, Hoffmann (Reference Hoffmann2021) showed that developmentally more advanced World Englishes varieties show increased productivity in realizing schematic constructions, clearly a process of construction emergence. Grammaticalization, when formerly independent units come to co-occur regularly and ultimately fuse into single, complex units (with additional semantic and syntactic properties associated) also constitute one fairly clear manifestation of emergence in languages (cf. Beckner et al. Reference Beckner, Blythe, Bybee, Christiansen, Croft, Ellis, Holland, Jinyun, Larsen-Freeman and Schoenemann2009: 6–7; Kretzschmar Reference Kretzschmar2015: 109–12). Chernyshova et al (Reference Chernyshova, Piccoli, Ursi, Lund, Fossali, Mazur and Ollagnier-Beldame2022) argue that recurrent, increasingly conventionalized patterns and multimodal practices in conversational interactions can be viewed as emergent complex and collaborative routines.

Two wider, related concepts in linguistics are worth mentioning in particular as early and indirect manifestations of the observation of such self-organizing tendencies in languages. In a classic book on Language Edward Sapir (Reference Sapir1921:ch. 7) introduced the notion of “drift” in languages, an inbuilt trajectory of “unconscious” long-term change toward more consistent typological properties (e.g. the loss of endings in English, a process rolling consistently for more than a thousand years by now). Keller (Reference Keller1994) posited the operation of an “invisible hand” in language change, an unintended directionality of evolutionary processes of linguistic change toward more consistent systemic relationships as the product of collective human agency, without any steering authority.Footnote ²⁰

2.2.7 Nonlinearity and Fractals

Chaos and complexity theory originally derive from the investigation of the properties of nonlinear equations in mathematics. Ideally, relationships between components and agents can be captured by non-linear functions, typically with feedback loops (so that the output of an activity / run becomes the input to a later re-run of the same function). Consequently, changes can spread very quickly and exponentially, and can also produce qualitative leaps. In reality, though, real-life systems (like the fairly well-known predator-prey cycle) have been shown to approximate nonlinear systems, but applications have often applied these concepts somewhat metaphorically (since often too many factors to be controlled for may have an impact), leaving strictly mathematical modelling to idealized simulation systems.Footnote ²¹

A characteristic property of chaotic systems, resulting from nonlinear equations, is what is technically known as “sensitivity to initial conditions” and has come to be popularly widely known as “butterfly effect” (illustrated as the flapping of the wings of a butterfly causing a tornado some time later at a totally different place). Hence, a minimally slight difference of initial states leads to unpredictably large divergence of a system (or rather, two variants of a system) down the road: amplification in feedback loops may lead to a far-reaching qualitative leap.Footnote ²² This has important consequences for the time flow of such systems: as a matter of principle their evolution is explainable from hindsight but unpredictable in advance. Several phenomena commonly associated with CDSs ultimately derive from nonlinearity. Guastello and Liebovitch (Reference Guastello, Liebovitch, Guastello, Koopmans and Pincus2009) list “attractors, bifurcations, chaos, fractals, self-organization, and catastrophes” (36) as an inventory of new concepts which can be modeled as nonlinear functions offering entirely new ways of understanding change.

Consequently, as applied to a language as a CDS, the history and development of English (or any language) can be explained but not predicted, because small-scale distinctions (possibly via affecting other levels of organization) may have far-reaching, qualitative effects. A very strong example of such a process in the history of English, the loss of endings and its causes and steps, was showcased earlier in Section 2.2.4 on network relations: a small-scale phonotactic change (the fixing of stress) almost 2,000 years ago has ultimately led to the fundamental transformation of the typological character of English, from synthetic to analytic. Elsewhere (in Schneider Reference Schneider, Mauranen and Vetchinnikova2020b) I spelled out details of what can be seen as another example of a butterfly effect, the long-time consequences of West Germanic “replacive lengthening” in English versus German, where a minimally widened variant condition of the sound environment in Anglo-Frisian for words to be affected by a sound change at that period (some 1600 years ago) produced a consistent phonotactic difference in lexical shapes between cognates in English (where nasals got lost in words such as five and us) but not German (compare fünf, uns).

An additional core property of chaotic systems, resulting from nonlinearity and iteration, is the fact that they are typically fractal, i.e. self-similar independent of scale: Patterns and relationships between components are self-similar irrespective of the magnitude of the observational level, whether microscopic or pertaining to the universe. Fractals do not follow a normal, Gaussian distribution (with some small, many average and some big objects) but break down to smaller and smaller similar patterns (trees to branches to twigs, for instance). A classic example, widely reproduced, is the Mandelbrot set (pictured in Figure 2, with colored surroundings), named after the concept’s inventor: This is a two-dimensional set produced by a mathematical formula where the visualization is always the same, no matter how deeply one zooms in in recursive magnifications. In nature, many phenomena have been shown to follow the same behavior (coastlines, snowflakes, leaves of ferns, river systems, blood vessels), though there the self-similarity operates only on a limited number of scales – they are “fractal-like” (Mitchell Reference Mitchell2021: unit 3.2; cf. Jensen Reference Jensen2022: section 2.1 on systems with characteristic scales), or statistically but not geometrically self-similar (Guastello & Liebovitch Reference Guastello, Liebovitch, Guastello, Koopmans and Pincus2009:17).

Figure 2 The Mandelbrot set

(CC BY-SA 3.0, https://commons.wikimedia.org/w/index.php?curid=16063)

Fractals are characterized by dimensions which are, unlike the standard dimensions in space, not integers, i.e. they are not one-dimensional lines or two-dimensional planes but have a dimension between 1 and 2, and this dimensionality is a measurement of a fractal’s density compression. In nature, fractals are an extremely efficient way of squeezing in an enormous amount of material (say, veins, or brain parts) into a small amount of space, with densely packed curves packed into higher-level self-similar curve lines (Mitchell Reference Mitchell2021: units 3.3 and 3.4). Scaling theory implies that there are optimal levels of compression (Krakauer Reference Krakauer2023), where “effective coarse-graining” ontologically offers a reasonable compromise between noise and order, losing some but not too many parameters, thus optimizing principles in local environments. A related, helpful concept is that of being “nested,” every (sub-)system being part of another system (de Bot et al. Reference de Bot, Wander and Verspoor2007:8).

Fractals in language are not as directly visible as other CDS properties, but they appear to be identifiable and are possibly caused by communicative efficiency needs, a balance between saving cognitive and articulatory energy and sufficiently rich information encoding. A case in point is Zipf’s law, a “power law” stating that the frequency of occurrence of a word in a corpus is inversely proportional to its rank in a frequency listing (Larsen-Freeman & Cameron Reference Larsen-Freeman and Cameron2008:109–111; Kretzschmar Reference Kretzschmar2015:83–86). Hence, according to a “principle of least effort” this strikes a balance between hearer (who wants maximally efficient decoding, thus preferring familiar forms) and speaker (who needs a choice between simple and efficient encoding or the encoding of special, less predictable information) needs.

It is noteworthy that not all properties of CDSs are scale-free; in fact, exponential functions may produce and emergent systems may be marked by characteristic scales (Jensen Reference Jensen2022:23–26, 50–54).

2.2.8 Attractors

After a large number of recurrent iterative runs (time units) dynamic systems tend to approximate specific types of “attractors,” subsets of their phase space – relational shapes which are strongly self-similar but never fully identical (mathematically a consequence of nonlinear equations). Chaos theory distinguishes fixed-point, periodic, and chaotic (or “strange”), fractal attractors – self-similar space states which can be perceived as relatively orderly, so this property interacts with the interplay between order and chaos referred to earlier, and it is intrinsically connected with the notions of dynamics, change, and emergence: A “system diverges from its initial state and after a transient period settles into some attractor state,” which may be “a simple equilibrium” or “a strange attractor” (Green Reference Green, Bossomaier and Green2000:26).

While the notion of attractors is rather precisely defined in chaos theory, its metaphorical application in other fields is often handled more loosely, indirectly building upon the meaning “something that attracts” in a wider sense. Cooper (Reference Cooper1999) proposes the notion as relevant to language acquisition and change, with applications (involving the Northern Cities Shift in US dialects) strongly leaning toward mathematical, computational and information-theoretic procedures and viewing attractor patterns as competing with each other in a sociodynamic perspective. Larsen-Freeman and Cameron (Reference Larsen-Freeman and Cameron2008) define attractors as “a region of a system’s state space into which the system tends to move” (50), and apply it culturally to concepts as diverse as the gaits of a horse-and-rider system (trotting, galloping, etc.), British A-level examinations (resisting change as generally accepted behavior), explanatory notions (such as the knowledge that the earth rotates around the sun) or artefacts (e.g. a violin; 50–55). In a second-language acquisition context they also view “preferred paths within individual performances,” counter-acting intra-individual variability, as “attractors” (153), and identify specific discourse patterns or fossilized states as attractors (180–185, 214; cf. de Bot et al. Reference de Bot, Wander and Verspoor2007:10, 15). This seems open to discussion: When the concept is applied to any kind of pattern or convention it is in danger of losing its conciseness and explanatory power.

Thus, the character and role of the attractor concept in languages calls for further consideration. Very basic conceptual relations between systemic choices may be regarded as fixed schemes which serve as attractors – for example, contrast as an elementary type of paradigmatic relation (for example, voice in English is either active or passive; and in the case of mediopassives form is pulled to the active but meaning to the passive frame). Individual linguistic items and structures that speakers aim at and produce in the process of information encoding may be seen as attractors. More generally, this may be taken to be closely related to the wider concept of “constructions” as in Construction Grammar (Hoffmann & Trousdale Reference Hoffmann and Trousdale2013; Hoffmann Reference Hoffmann2022; Laporte Reference Laporte2021; Ungerer & Hartmann Reference Ungerer and Hartmann2023): the conceptual schemas in construction grammar (illustrated in Section 2.2.6, for example) clearly “attract” further lexical material to be used in the very same way (cf. Hoffmann Reference Hoffmann, Buschfeld, Hoffmann, Huber and Kautzsch2014, Reference Hoffmann2021). Attractors in language can thus be also understood, fully in line with CDS thinking, as emergent patterns, “a probabilistic profile … out of the state space of possible outcomes” (Kretzschmar Reference Kretzschmar2015:137) in a dynamic evolutionary process. And they can drive language change, for example when a system “transitions from the basin of attraction of one attractor to the basin of another” (Cooper Reference Cooper1999:137).

4.3.1 Systemness

The presence of systemic relations is obvious and uncontroversial in all kinds of varieties of English. As stated in Section 2.2.1, syntagmatic, co-occurrence relationships build larger units, and paradigmatic relations determine meaningful choices between functionally equivalent potential fillers of any slot. Here is a simple example from WEs, a prototypical structure found in Colloquial Singaporean English:

1. (1) I dunno lah. (from an informal conversation; source: Schneider Reference Schneider2020a: 169)

Even this simple three-word pattern allows us to identify a number of systemic language-internal relations between clause constituents of several kinds. The words represent a basic syntagmatic pattern (S-Aux-Neg-V-A_{[sentence adverbial / discourse marker]}). The “underlying” items do + not + know are syntagmatically fused to yield dunno. The (typically Singaporean) discourse marker lah takes a syntagmatic role as sentence adverbial modifying the matrix clause proposition, with specific semantics (‘assertion’); and in itself it represents an actively selected element of a subsystem, potentially contrasting with further Singlish discourse markers such as lor (which would mean ‘compromise’), meh (indicating ‘uncertainty’), and others.

Systemic relations operate mainly within a given variety; across varieties they can be compared, and of course they can influence each other (in a network-like relationship) and they vary in terms of the complexity levels involved. An example of a subsystem that is ostensibly related to and only moderately more complex than the metropolitan one is the set of demonstratives and distinctions expressed by them. Mainstream English encodes a binary distinction of near vs. far relative to the speaker (this/these vs. that/those; see Section 4.3.5). Some dialects (e.g. on the Shetland and Orkney Islands) are moderately more complex by adding a third term, yon, meaning ‘distant from both speaker and hearer’ (Siemund Reference Siemund2013:96). A case of a quite different and much more complicated systemic set-up is the pronoun system of Bislama, an English-derived creole spoken in Vanuatu. Influenced by Indigenous languages (and thus also showing network complexity), Bislama distinguishes not only singular and plural but in addition has dual (for exactly two referents) and trial (for three) forms, and for all of them except the singular the variety distinguishes inclusive from exclusive forms (i.e. whether the addressee is included in the referent set or not). On the other hand, there is no distinction by gender or subject/object/possessive form. So, for example, hem is ‘he’, ‘she’, ‘him’, ‘her’, ‘his’, and so on; yumitrifala, the inclusive trial, means ‘me, you, and somebody else’, mifala (first person plural exclusive) means ‘we (more than three, not including you)‘, and so on (see Siemund Reference Siemund2013:97).

Systemic and network relations are obviously also involved in functional re-assignments of pronoun forms found rather widely (i.e. subject forms used as objects and vice versa, subject or object forms used as possessives, and various modifications of gender reference [eWAVE f1–2,5–6, 18–27, 29–31]). Very many varieties have developed distinct second person plural pronouns (thus resolving the ambiguity of standard English you) such as y’all, youse, you’uns, you guys, unu, and so on, thus also changing systemic relations in the set-up and meaning of pronouns. The same applies to other modifications of pronominal uses and relations in several WEs (e.g. the emergence of a distinction of inclusive versus exclusive non-singular first person pronouns [just shown for Bislama; eWAVE f36]), or various types of pro-drop (eWAVE f42–44). Over many centuries the relativization system of English has undergone fundamental systemic changes triggered partly by other reorganization processes in English. Examples include the intrusion of French-/Latin-derived wh-relativizers into a formerly invariant position (Old English Þe, related to modern that), the reorganization of formal choices based on animacy (who vs. which), various constraints on omitting a relative pronoun (”gapping,” acceptable in standard English in object and prepositional complement position [the man I saw, the man I talked to] but only in some dialects as a subject [the man Ø talked to me helped me]), the loss of case marking over time (cf. whom vs. object who and whose vs. of which), and more (Schneider Reference Schneider2020c). All of these processes interact and are realized and structured somewhat differently in WEs (see the documentation in eWAVE, f185–199), involving rearrangements based on underlying systemic relations, affecting the orderly character of partial systems, involving network relations, and often self-organization. For example, the genitive relativizer of which, as opposed to whose, is used substantially more frequently in Indian English (IndE) than in other varieties (Schneider Reference Schneider2020c:98). Developments increasing order (addressed in Section 4.3.5), such as regularization or the establishment of clear distinctions in demonstrative relations, also relate to and result from systemic pressures and tendencies.

4.3.2 Complexity

Language-externally, it seems reasonable to see speakers as the agents of a CDS, and their language production as associated attributes. Larsen-Freeman (Reference Larsen-Freeman, Jenkins, Baker and Dewey2018a) mentions “individual human beings, their contact zones, and globalized networks” as relevant parameters (54). In line with CDSs, in WEs, as in all languages, there is an enormously large number of speakers who keep interacting, a prerequisite for complexity. Speakers are individuals and act as such, and the principle of local agency without a steering authority that is central to CDSs clearly applies: people speak with their family, friends, neighbors and other interaction partners, mostly (at least in informal contexts) without having communicative conventions on a larger social, say national, level in mind, but by doing so they inadvertently contribute to the conventionalization of the forms they use, and to language evolution. In addition, a level of social complexity results from the fact that they are embedded in, shaped by and actants of group relationships marked by sociolinguistic parameters, and also agents in sociopolitical processes, thus representing a higher-level order of conglomerates. Relevant individuals in WE-producing settings were, for example, Indigenous low-status people, members of established local elites, settler merchants, missionaries, and so on. Consequently, in addition to the myriad of individual input experiences and factors this also introduces some level of order, so the interactants’ linguistic performance is not only to be seen as potentially chaotic idiolects but also as partly determined by levels of social organization, cooperation, and possibly low-level predictability. For example, specifically with respect to WEs, in the Dynamic Model of the evolution of Postcolonial Englishes (Schneider Reference Schneider2007), political factors determine identities of and changing relationships between settlers and Indigenous groups, and the sociolinguistic interactions between them are decisive for the process of “structural nativization,” the emergence of locally characteristic language patterns, and the evolution of NEs – clearly processes with many agents, hierarchies, and levels of complexity involved. For example, local leaders and their sons (hardly ever daughters, I am afraid, in paternalistic social structures) were lured to associate with colonists in Britain’s characteristic “indirect rule” leadership pattern and thus offered access to English schools and better acquisition of the language, while typically Indigenous workers and lower-class people had no such opportunity and acquired English not at all or in natural interaction contexts and in “grassroots” format (Schneider Reference Schneider2016a). Social organization thus was one of the major input factors that were instrumental in linguistic evolution and complexity emergence.

Equally, of course, there are all kinds of intra-linguistic network structures – mutual relations and similarities between linguistic forms and constituents of all kinds. As an example, let us look at the evolving functions of the form one in time and space. It started out in Old English as a plain numeral, and in the course of time it adopted a range of new functions including pronoun (buy one), determiner (one person I know), adjective (her one concern), and a noun phrase head substitution form sometimes called “propword,” with specific semantic constraints of rejecting an attribute and replacing it by another (an interesting book and a boring one; Rissanen Reference Rissanen, Rissanen, Kytö and Heikkonen1997; Schneider & Buschfeld Reference Buschfeld2020). Clearly, these functions relate to each other (and to other constituents in the same slots) in systematic and increasingly complex ways; and this trend continues into WEs. In Cyprus or Namibia, for example, one functions as a presentational indefinite article (My sister is one art teacher; Schneider & Buschfeld Reference Buschfeld2020:136). The most interesting and creative innovation, however, indicative of dynamism and network complexity, can be found in Colloquial Singaporean English, where one functions as something like a clause-final relativizer mirroring and modifying an underlying, motivating structure from Chinese (e.g. those wear black one ‘the ones who wear black’; Bao Reference Bao2015:104; see eWAVE f195). Gwee (Reference Gwee2018), a book written entirely in “Singlish,” abounds in such examples, including, apparently, some with wider functions: for example, “they last time got say ‘chap-barang’ one” (60); “this fella is ang moh pai one” (116); “words … that can be used in so many ways one?” (143). The last example is particularly interesting from a network perspective since it obviously combines two distinct relativization strategies (which thus interact to reinforce each other rather than being mutually exclusive), English that and Singlish one.

An increase of complexity is also involved in the formation of multi-morpheme words (in word formation), chunking and phraseologisms described in Section 4.3.6, or in the growth of more variable schematic constructions shown by Hoffmann (Reference Hoffmann2021; see Section 4.3.6).

4.3.3 Perpetual Dynamics

The relevance of this point is obvious, I think: The ongoing transmission of English from Old English times to the present day, mentioned earlier, comprises all elements of the “English Language Complex” (Mesthrie & Bhatt Reference Mesthrie and Bhatt2008:1–10). English has continuously been passed on (and partly been transformed in this process) to regional and colonial dialects, and in that sense, as mentioned earlier, “Singlish,” IndE, Nigerian Pidgin English and all other NEs and contact varieties are equally continuations and dynamic re-instantiations of what once was Old or Middle English, daughter varieties of or complex interacting branches of this overall CDS of “English(es)”!

In addition to the examples of continuous diffusion (see Section 2.2.3), there is also always discontinuity (i.e. change, innovation, and the adoption of elements from contact languages). For example, Singlish has subject omission in Can!, IndE has the lexeme dhoti, and Nigerian Pidgin English has a second-person plural pronoun una – all “picked up” at some point along the way through time and space for some reason, and not shared by the others.

The perpetual dynamism of WEs as CDSs involves most immediately quantitative changes: in certain varieties some forms or patterns may become more frequent, in others they may fall into disuse. Kretzschmar (Reference Kretzschmar2015) rightly posits that “‘change’ in the complex system of speech will consist of an alteration in frequency of any particular feature, instead of the selection of one form over another” (113); and I would add that selection, one form winning out and an older one vanishing, may become a later option, thus with ultimately qualitative effects. In general, qualitative changes (new forms and schemes being introduced, like the ones mentioned in the previous paragraph) are also possible; and, as just stated, an interaction between both (when frequency differences gradually become so drastic that alternative choices in different varieties result) is another option. It is noteworthy that this applies to the relationship between the major metropolitan varieties, BrE and AmE, as well: While it is often posited that both are distinguished by clearly distinct choices (petrol – gas, pavement – sidewalk, etc.; nonrhotic vs. rhotic pronunciation of words like car, card; /a:/ vs. /æ/ in dance, etc.), Algeo (Reference Algeo2006) showed that the vast majority of differences are quantitative in nature, with the same forms being used on both sides of the Atlantic but frequencies and preferences varying drastically in many instances (cf. Schneider Reference Schneider, Durgasingh and Meer2025b).

Here is a small-scale but suggestive example of such an interaction between frequency fluctuation and the possibility of qualitative preferences shifting, regarding lexicosemantic change in IndE. I investigated interactions between word meanings and frequencies of words or word meanings in different varieties (see Schneider Reference Schneider, Mauranen and Vetchinnikova2020b), and I found emerging strong preferences in some varieties. For example, let us look into the meanings expressed by the verb learn. I posit there are three core meanings: ‘receive information’ (I learned you had an accident), ‘store information’ (I learned some Malay words), and ‘acquire an ability’ (Suzy learned how to swim). In IndE we can observe change based on data from the 1970s (the “Kolhapur Corpus”) and from around 2000 (the ICE corpus component). As Figure 3a shows, in those roughly two decades the core meaning ‘store information’ gained ground quite strongly (from 206 out of 258, i.e. 79.8% of all tokens to 325/381, i.e. 85.3%), while both other meanings remained roughly constant in frequency. It is noteworthy that in all ESLs in that study the proportion of this core meaning is higher than in BrE (though not as prominently as in India), which suggests that in the transmission into NEs a “focusing” process seems to be going on, favoring the further growth of dominant variants at the expense of weaker ones, which may be falling into disuse over time. Similar developments and shifting preferences were observed with frequency relationships between near-synonyms. For example, assume and suppose denote roughly the same mental process, but in India, much more than elsewhere, suppose has come to be the preferred lexeme: In the Kolhapur corpus suppose occurs only moderately more frequently than assume (124 versus 87 tokens, i.e. 58.8% of the sum of both), but some twenty years later it predominates strongly (508 over 59 tokens, =89.6%), with assume having become substantially dispreferred (see Figure 3b).

Figure 3aLong description

The x-axis labels are: learn, receive info, learn, store info, and learn, acquire ability. The y-axis ranges from 0 to 350 in increments of 50. The bars for learn (receive info) show 28 for Kolhapur 1978 and 24 for ICE 2000. The bars for learn (store info) show 206 for Kolhapur 1978 and 325 for ICE 2000. The bars for learn (acquire ability) show 24 for Kolhapur 1978 and 32 for ICE 2000.

After Schneider (Reference Schneider, Mauranen and Vetchinnikova2020b:36, 34).

Figure 3bLong description

The x-axis labels the two words whose frequency is compared: assume and suppose. The y-axis represents frequency, ranging from 0 to 600 in increments of 100. The data are as follows: assume, for India 1978 Kolhapur, the frequency is 87 and for India 2000 ICE, the frequency drops to 59; suppose: for India 1978 Kolhapur, the frequency is 124 and for India 2000 ICE, the frequency significantly increases to 508.

Figure 3a and 3b Shifting frequencies of preferred meanings and lexical choices in India, 1978 vs. ~2000.

Bernaisch’s (Reference Bernaisch2015) study of Sri Lankan English documents a wide range of lexicogrammatic properties and innovations that are mainly characterized by significant frequency preferences, and consequently the author discusses norm issues, the question of how much difference in frequency is required to claim endonormativity (376), or the emergence of structural profiles indicative of a semiautonomous status (ch. 6).

In Section 2.2.3 I illustrated the ongoing dynamism in the evolution of English by outlining shifts and changes in the sub-system of modal verbs. This story can easily be continued by following it up into WEs, where the permanent quantitative fluctuation and ongoing processual readjustments between individual forms can equally be found. In regional Inner-Circle (L1) dialects, alternative options and properties to the developments in Standard English have been retained or have emerged. For example, in Southern AmE double modals can be found (e.g. I might could do that), in a pattern which violates the modern English rule of allowing only one modal in a clause. However, this may be seen as continuing the Old English property of being able to combine preterite presents in the same predication, with special pragmatic functionality (e.g. implying tentativeness). In WEs, frequency shifts of modal usage are common; for example, shall occurs very frequently in Kenya but is rare in Hong Kong (Collins & Yao Reference Collins, Yao, Hundt and Gut2012). Hansen (Reference Hansen2018) provides a comprehensive investigation of variation and change in modal verbs in three second-language NEs (from Hong Kong, India, and Singapore, based on ICE corpus data and bringing in a sociolinguistic perspective, most importantly via an apparent-time methodology), focusing on the domain of obligation and necessity. She shows, for example, that in Hong Kong have to is used most frequently, with quite some difference, by younger speakers aged 17–25, and that the root (obligation) meaning of must decreases sharply through all age cohorts (272–73; similarly in India, 281). All varieties, most strongly India, use must commonly to express epistemic modality (307). As a strong causal factor she identifies “profound influence of the substrate languages” (309). In general, she finds consistent frequency differences of uses of specific modals (e.g. the semi-modal have to) across varieties (310–11). The ESL varieties are shown to follow native varieties in some core trends, notably the decrease of core modals (especially deontic must) and an increase of some semi-modals, especially have to (less so have got to and need to). In a similar vein (and looking into the same varieties), Laporte (Reference Laporte2021) shows how different varieties come to prefer different constructions involving the core verb make, and she points out frequency-based emerging syntactic, lexical and semantic idiosyncracies in the NEs investigated (298).

So, it’s a perpetuum mobile, a continuous reshuffling process with old forms retaining old properties only in some varieties but gaining new functions in others, with the frequency of individual forms waxing and waning from one variety to another, with relations constantly changing in a process that never ends.

4.3.4 Network Relations

In Section 2.2.4, I argued that the notion of network in languages can be regarded as applying in an extralinguistic and an intralinguistic perspective.

In the language-external perspective, speakers are agents with specified network relations between them, and interactions between them manifest network behavior. In the so-called “second wave” of variationist sociolinguistics the concept of social networks became crucially integrated into the discipline’s thinking in work by Jim and Lesley Milroy in Belfast, which explicitly showed that speech behavior and language forms used depended on speakers’ networks, their communication patterns and interaction habits (Milroy Reference Milroy1980). Similarly, interaction patterns between locals and colonists as described in the Dynamic Model of Postcolonial Englishes (Schneider Reference Schneider2007; see Section 4.3.2) constitute networks of various degrees of density, and the unilateral implication between factors such as politics, identities, interaction patterns and, consequently, constructions establishes a complex network dependency leading from external conditions to internal structural properties.

Language-internally, many interrelationships and mutual or unilateral causal dependencies between specific constructions can also be identified. Consider my earlier example introduced in Section 2.2.4, the longitudinal change of character of English from synthetic to analytic, which involved a large-scale loss of endings throughout its history. It can be readily shown that today in some WEs (e.g. China English) this trend continues further: the -s suffix on verbs (for the third person singular) or nouns (for plural) is often omitted (Schneider Reference Schneider, Curzan and Adams2011). This is obviously due to substrate effects of analytic Sinitic languages: Mandarin Chinese lacks endings for verb concord or noun pluralization, and this linguistic habit remains to some extent when Chinese people acquire and speak English. We are confronted with a crosslinguistic network relationship here: a morphosyntactic property of Chinese relates to and interferes with Chinese speakers’ language production in English.

An impressive example of such a crosslinguistic, typologically motivated network relation is offered by the work of Brunner (Reference Brunner2017). The author convincingly documents a strong typological effect of local Indigenous languages on the emerging structural choices in noun phrases in the Englishes of Singapore and Kenya. Notably, such a connection would not be sensed by speaker intuition but surfaces only with the help of a rigorous quantitative methodology, although it is shown to be robust and highly statistically significant. Asian substrate languages in Singapore (Baba Malay, Hokkien, Mandarin, and Cantonese) are predominantly head-final, while the East African Bantu, Nilotic, and Cushitic languages in Kenya are characterized by a head-initial character and postnominal dependents. As the author shows (Brunner Reference Brunner2017:170–205), this preference gets transferred in a network-like relation to the emerging regional Englishes: in Singaporean English premodifiers within the noun phrase occur more prominently and tend to be longer and internally more complex (Example 2a), while in Kenyan English postmodifiers are found more frequently, with the same formal properties (Example 2b).

1. 1. (a) a four month kind of attachment (ICE-SIN; Brunner Reference Brunner2017:180)

   2. (b) any barrier hindering you from that kind of marriage (ICE-EA; Brunner Reference Brunner2017:191)

Thus, network relations between structural associations based on Indigenous Asian and African languages, respectively, extend to emerging linguistic habits in the evolution of Singaporean and Kenyan English, respectively, and in a most subtle manner affect speakers’ associations and habits in these varieties.

Another example of emerging, expanding, and changing language-internal network relations is provided by the innovative spread of “intrusive as” especially in IndE (and also, less well established, in other South Asian Englishes), as documented by Lange (Reference Lange2016) and Koch and colleagues (Reference Koch, Lange and Leuckert2016; see also Schneider Reference Schneider, Mauranen and Vetchinnikova2020b). As these authors show (see Example 3, from Lange Reference Lange2016), in complex transitive verb complementation structures (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985), where verbs are followed by an object and an object complement which classifies the object (as in I consider her my advisor), a connecting particle as has become established in IndE and South Asian Englishes with verbs that do not allow this construction in metropolitan varieties.

1. (3) The main temple is called as Rang-Mahal.

   a teacher named as Mr. Keating

It has been shown that this construction occurs most commonly in IndE, less frequently in other South Asian Englishes (notably Sri Lanka), and also elsewhere, especially in learners’ utterances; and in IndE it is associated most effectively with certain verbs that commonly occur with it (mainly term as, call as, name as). This is a complex, perhaps even strange construction, since the word class of as in this pattern is by no means clear, even counterintuitive: It can be viewed as a preposition or a conjunction, but in fact in these constructions it expresses, surprisingly, “a copular relation” (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 1200), a “predicative” function (Aarts Reference Aarts1992), since I regard him as my friend corresponds to a relation prototypically expressed by a copula (cf. I think he is my friend; cf. Schneider Reference Schneider, Fries, Müller and Schneider1997b). The point is that this construction has a source model in standard English (I consider/regard him as my friend), but only with a small set of verbs, being barred from this structure with many other, semantically identical or similar ones (*I think/believe him as my friend). Now, whatever the reason (forces of analogy, an increase of transparency, contact-induced origins, etc.), it is clear that new associations have been built (between this construction type and verbs suitable for it), at varying degrees of strength in India, South Asia, and elsewhere, and network connections between words, structural meanings, and construction types are being built and expanding, producing slightly more complex and ever-evolving systemic relations.

A simpler example of a network relationship plainly between formal constituents is the inventory of reflexives found across varieties of English. The forms of Standard English are somewhat erratic – for instance, myself is derived from the possessive pronoun my but themselves from the personal pronoun object form them. In WEs, all kinds of combinations, both of personal and possessive stems in different persons and of -self/-selves independent of number, and also just simple forms, occur (e.g. meself, hisself, ourself, theirself, him, etc.); for documentation, see Siemund (Reference Siemund2013:26–33; eWAVE f11–16). All these constituents and relations are thus connected with each other as network elements, showcasing a lot of creative potential.

Obviously, some of the other structures discussed in this Element involve network relations as well, for example the Malaysian/Singaporean English kena passive, which affects the paradigmatic relationship with passive formations with be or get (see Section 4.3.5), or the different emerging constructions around the core verb make documented by Laporte (Reference Laporte2021). Other cases in point, which also relate to systematicity and complexity, are pronoun uses in widened functions, such as she/her or he/him for inanimate referents or generalized subject or object pronoun forms (eWAVE f1–2, 5–6) or the well-known alternation (and prescriptivist discussions about it) between me and I (John and me/I) in coordinate subjects (eWAVE f7).

4.3.5 The Interplay of Order and Chaos

Broadly, “ordered” relations can be understood as sets of a small number of choices with clear functional assignments, while chaotic relations involve a large number of entities with unpredictable properties and modes of behavior. An example from the evolution of English which remains effective in WEs, of course, was briefly characterized in Section 2.2.5 – the breakdown of a once orderly system of functionally predictable verb forms as opposed to today’s co-existence of many irregular verb forms.

A rather simple example with effects in WEs is the relation between the simple past verb form (I wrote) and the present perfect (I have written). In standard BrE (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985) the former marks an event as completed, having occurred at some point in the past, whereas the perfect implies some sort of ongoing current relevance for the present moment of time (with the activity in question being ongoing, recent with immediate effects on the present situation, or having occurred within a time frame that spans up to now). This systematic relation (order as a firm form-meaning implication) became established during the Early Modern English period, very roughly between the sixteenth and the eighteenth centuries, so in postcolonial varieties which branched off during that period, including AmE, it is known to be less firmly established as a grammatical habit (Werner Reference Werner2014; Werner et al. Reference Werner, Seoane and Suárez-Gómez2016; cf. Siemund Reference Siemund2013:115–16, eWAVE f99–100). Predictable implications between structure and meaning, as in standard BrE, can be understood as orderly relations, while variable, fuzzy relations, where it is not clear what exactly a perfect or past structure entails, represent a step toward a chaotic set-up.

A similar relationship obtains with another grammatical feature of English which historically is fairly young, the distinctive function and meaning of the progressive (She is running.) While in metropolitan English the progressive clearly marks an activity as ongoing, incomplete, and viewed from inside (as opposed to the simple form which implies an outside perspective on a completed or stative fact), it is well known that in NEs the progressive is used more widely, also for stative predications (I am knowing that. She is owning a car; for documentation, see Siemund Reference Siemund2013:138–40, eWAVE f88). The effect is the same as described in the previous paragraph: an orderly relationship, with a specific form signaling a distinctive meaning, is dissolved, so the overall system, the syntax-semantics interface, becomes a little more unpredictable and chaotic.

As a further example of the oscillation between orderly and chaotic properties (involving other CDS principles as well), let us have a short look at the discourse particles of Colloquial Singapore English, which have been widely investigated and documented (e.g. Lim Reference Lim2004:5.3). Singlish is known to have an abundance of such forms, which denote varying attitudes to the proposition expressed (which then is implied to be evident, annoying, surprising, etc.), beginning with the ubiquitous and generally well-known lah particle (Can, lah!). These items have been borrowed over decades from various regional donor varieties like Baba Malay, Hokkien, or Cantonese, and the set of such forms has been continuously expanded and integrated into the same grammatical slot of discourse markers (Lim Reference Lim2007): speakers of Singlish know that items such as meh, lor, hor, and others, often in clause-marginal positions, are “subjective comments” on the statement made, expressing a personal stance toward it. This associative, by now stable grammatical knowledge conventionalized among Singlish speakers constitutes an element of order in this variety: Speakers know how to interpret these forms in context, they understand that by using them their interlocutors subtly express a stance, position themselves relative to their utterance. And this is a young, recently emerged pocket of order: It is known that these particles have emerged over the last few decades, together with the growth of Singlish as a distinct marker of a local cross-cultural identity. This orderly relation serves as an attractor, since it integrates other forms and expressions, and in this process affects and streamlines their usage condition. For example, Wee (Reference Wee2003) showed that the single form know, derived and shortened from the English parenthetical you know, has been pulled and integrated into this discourse marker slot, and it now functions increasingly like other monosyllabic discourse markers. And the set of such forms is productive and open to expansion, which adds a slightly “chaotic,” potentially de-stabilizing option to the partial system. A relatively recent addition to the Singlish set of discourse markers is the form sia, for instance (Hieramoto et al. Reference Mie, Daniel Wong Gonzales, Leimgruber, Jun Jie, Jessica, Ngefac, Wolf and Hoffmann2022).

A specific manifestation of chaotic systems, referred to in Section 2.2.5, are the so-called “catastrophies” described by Thom, and among these in particular the phenomenon of bifurcation – the case that a single entity splits into two (i.e. either one form develops two functions or one function comes to be expressed by two different formal means).

An example of the former process is the emergence of locally distinctive, innovative functions of the forms only and itself in IndE as focus markers (Bernaisch & Lange Reference Lange2012; Lange Reference Lange2012). The form only in IndE has developed a novel function called “presentational,” placed immediately after the constituent (mostly noun phrase) that it focuses on (Sailaja Reference Sailaja2009:55; see Figure 4). Since this innovation has not dispensed the older, original use it may thus have one of two different functions and meanings, either the distinctively Indian presentational one (as in Example 4a and Figure 4) or the mainstream “contrastive” one (as in Example 4b) – it “does double duty as contrastive and as presentational focus marker “ (Lange Reference Lange2012:185). Similarly, the form itself has evolved into an intensifying focus marker following its referent, which, unlike in other varieties of English, need not be a noun phrase but can, for example, be an adverbial, as in (Example 4c). As Bernaisch & Lange (Reference Lange2012) show, this presentational, focus-marking function of itself has also spread to other South Asian Englishes (see also Lange Reference Lange2012:193–95), and it is possibly motivated by functionally similar focus particles in South Asian languages, thus building crosslinguistic network relations.

Figure 4 Example of presentational use of only in IndE (from Mehrotra Reference Mehrotra1998:115). Reprinted with permission.

Figure 4Long description

On the left, a bald man with a striped shirt points his finger, looking somewhat bewildered. On the right, a couple appears to be speaking to him. The man on the right has a distinctive hairstyle, a prominent nose, and is wearing a patterned shirt. The woman next to him has dark hair pulled back and is wearing a sari with polka dots. A speech bubble above the couple reads, in bold, capitalized text, WE LIVE IN BANDRA ONLY!!

1. 1. (a) These women only said this. (Sailaja Reference Sailaja2009:55)

   2. (b) I saw only him [not anybody else].

   3. (c) you should start going to the gym from now itself (Lange Reference Lange2012:186).

Another example of a case of bifurcation, starting from a functional category, can be selected from the grammar of Singaporean English (similarly in Malaysian English). In standard English the passive is built by a form of be plus a past participle (The mail was delivered); alternatively (more informally and less commonly) be can be replaced by get as an auxiliary, implying a stronger component of suffering from the activity expressed (John got beaten). Singlish has borrowed an additional passive auxiliary from Malay, kena, which carries additional semantic constraints: The activity expressed by a kena-passive must be something unpleasant that the subject undergoes involuntarily. An example from a Singlish conversation between young males is he kena play out (‘he was cheated’; Schneider 2020:170, 172). The Singlish book mentioned earlier (Gwee Reference Gwee2018) also offers a number of examples of this construction, such as “Poor guy kena whacked a lot” (38); “you kena influenced by ang moh culture” (112). The semantic space of passivity is thus subdivided in a novel way in that variety, introducing a new level of complexity (and order/lack of order) when compared to metropolitan English.

Stenroos (Reference Stenroos2008) documents another instance of “order [emerging] out of chaos” in a historical phase of BrE dialect. The loss of grammatical gender between Old and Middle English is generally assumed to have resulted in “confusion” (445). The author subscribes to the variationist axiom “that linguistic variation is, on the whole, orderly rather than chaotic” (446), assuming that orderly, systematic patterns are “in principle describable, even though they may be very complex” (446) – fully in line with CDS principles as applied and understood in this Element. Based on thirteenth-century texts from the Southwest Midlands, she shows that in that specific domain the assignment of new pronominal gender reference followed semantic distinctions based on animacy, individuation and natural categorizations. Thus, again we witness a semantically motivated reorganization from an orderly stage via a moderately turbulent transition period to the establishment of newly orderly notional distinctions.

Actually, a look into the history of English offers another beautiful example of the oscillation between orderly and chaotic sub-systems; many grammars of Old and Middle English touch upon this phenomenon, but McColl Millar (Reference McColl Millar2000) traces and documents it in exemplary detail (and also through standard and nonstandard sub-varieties). Old English had a complex but fundamentally orderly system of articles/demonstratives, with a very large number of different forms, depending upon whether the form was “simple” or “compound” (i.e. unmarked or the demonstrative act enforced), and then varying by three genders, four cases, and to some extent different noun classes and regions. By the end of the Old English period this set of forms and their functional assignments broke down completely (perhaps due to contact with Scandinavian languages), so for a while this development seems to have resulted in pure chaos in the identification of a noun phrase referent. But, remarkably, toward Middle English a new, beautifully orderly system of an article (the) and only four demonstratives emerged, distinguishing only two parameters (proximity, i.e. near vs. far to the speaker, and number, i.e. singular vs. plural), expressed by this – that/these – those, respectively. Interestingly enough, these individual forms all can be traced back to forms that had existed in the Old English system (with totally different functions), so fragments from an earlier system survived the in-between chaos and were re-assigned new functions in an emerging new pattern of order. In a wider perspective, see the large number of features in eWAVE involving interchanges or functional reassignments of articles, determiners, and demonstratives, all, of course, somehow affecting systemic and network relations and thus also the amount and character of order in a given variety (eWAVE f59–71).

Regularization may be seen as a product of systemic pressure toward orderly relations – e.g. of noun plural formation (childrens, sheeps; eWAVE f48) or of verb past tense forms (like gived, knowed; Section 2.2.5).

4.3.6 Emergentism and Self-Organization

An example from English dialects illustrates the emergence of systematically organized subsystems through time: Anderwald’s (Reference Anderwald2001) work on an emerging polarity-based systemic opposition in past tense copula forms in dialects of BrE (based on the 10 million-word corpus of spoken colloquial English which is part of the British National Corpus [BNC]). Her starting point is the observation that the past tense forms of the English copula (was – were) show an irregular, unusual distribution, with the choice of a verb form based on the grammatical person of the subject, different from the behavior of main verbs (I was – you(sg.) were – she was – we/you(pl.)/they were). This is a case of irregularity without a function (a residual trace of an earlier language stage), and as such, thus, prone to be removed.

A simple “solution,” found in many dialects, is the choice of a single form throughout all grammatical persons – most likely the majority form was, which yields I/you/she/we/they was – a pattern which is common in some dialects (e.g. African American Vernacular English or Southern AmE). However, the relationship gets more complex when polarity, commonly expressed by phonologically weak morphological clitics (-n’t), comes in. Anderwald’s analysis shows that positive was but negative weren’t are spreading in all grammatical persons in BrE dialects (the same tendency has been documented for the dialect of the island of Ocracoke, NC; Wolfram & Schilling-Estes Reference Wolfram and Schilling-Estes1997:83–84). Since this is an unexpected, counterintuitive result it is worth asking why this is the case. The new distribution marks positive versus negative polarity (the important piece of information of “yes or no?”) not only by the (phonetically weak) clitic –n’t but strengthens it by distinct stem choices added: was always signals positive information, were marks negative propositions. In other words, this ongoing reorganization of a formal subsystem in BrE dialects marks priority given to the distinct expression of polarity over the traditional, synchronically unmotivated English person-number system. Clearly this can be interpreted as an evident self-organizing tendency in BrE dialects, based on the strength and impact of cognitive factors and principles, and also as a process introducing a small pocket of order (formal consistency based on polarity) into what in standard English is a slightly erratic (chaotic?) paradigm.

Emergence and auto-organization manifest themselves also in the construction of new, complex entities, either in word formation (Biermeier Reference Biermeier2008) or by chunking, the formation of multi-word items or “prefabs.” In Outer Circle WEs, on the lexical level new complex entities have been produced (i.e. local, partly patterned word formation products). We find new compounds (e.g. IndE chai wallah, rikshaw wallah, and many more wallahs; Malaysian English botak head; New Zealand English sharemilker; etc.), localized phraseologisms (e.g. Kenyan English look at someone with bad eyes, break a leg ‘make pregnant without being married’), and localized construction types (e.g. Singaporean English Can or not? – Can, lah!). The Philippine English forms to back-carry and to gift-give are particularly interesting, since these are verbal compounds, a word formation type that is not productive in standard English (such forms are rare and only produced by backformation). Thus, the auto-organizing process has not just generated new individual formatives but actually expanded formation rules and constraints, added a new layer to network types. Growing levels of organizational complexity manifest themselves in various ways – for instance, in Cameroonian Pidgin in what Hoffmann (Reference Hoffmann2022:16) calls the “’N belong N’ construction,” which is semantically transparent and productive: gras bilong hed ‘grass belong head’ = ‘hair’; wara bilong skin ‘water belong skin’ = ‘sweat’, and so on. Some varieties, notably Caribbean creoles, create possessives by combining a prefix fi (derived from for) and personal pronouns (e.g. fi-me ‘my’; eWAVE f17), or also by postnominal or even prenominal phrases (like for my sister husband) with for, with bilong, or with plain juxtaposition (eWAVE f74–77). Others build complex interrogatives by using additional formatives (who-all; eWAVE f39) or reduplication (who-who; eWAVE, f40). A low, possibly early manifestation of emerging chunking are observable co-occurrence tendencies, as for example in IndE, where of which tends to follow as a result more often than expected (Schneider Reference Schneider2020c:98). Vetchinnikova (Reference Vetchinnikova, Filppula, Klemola, Mauranen and Vetchinnikova2017) looks into chunking, lexical fixing in specific construction types and the formation of “lexical bundles,” and argues that this is a process characteristic of emergence in CDSs. All these, and many more, are products of a language variety’s self-organizing, emergent capacity.

Self-organization can also be taken to operate on more abstract levels, in the growth of new construction habits (function-form associations) in a speech community. Hoffmann (Reference Hoffmann, Buschfeld, Hoffmann, Huber and Kautzsch2014) hypothesizes that the emergence of mainly substantive constructions characterizes early phases of a new variety’s evolution in the Dynamic Model, as opposed to “meso-constructional” schematic constructions which occur at later stages. In his 2021 Element he shows this to be true: Broadly, the more advanced varieties are in their developmental trajectory, the more creative and liberally lexically filled their schematic constructions are. Hoffmann (Reference Hoffmann2021:21) calls this the “Dynamic Model Productivity hypothesis: Varieties in later phases of the Dynamic Model show (1) more productivity of the slots of (semi-) schematic construction than varieties in earlier developmental phases due to (2) less reliance on prototypical and frequent fillers.”

To document this advanced type of productivity (equivalent to auto-emergence), Hoffmann (Reference Hoffmann2022) employs “LNRE” (‘Large Number of Rare Events’) statistical models, and he shows that the hypothesis is basically valid, showing that self-organization shows the capacity of building abstract relations associated with variant forms in advanced network relations. Figure 5 shows a clear example. Based on several thousands of tokens from the well-known, monumental GloWbE corpus,Footnote ²⁸ it shows the range of lexical variability in the “as ADJ as a N” construction, exemplified by as quiet as a cemetery, as gentle as a lamb, or as fit as a fiddle. The lines of the graph represent type-token ratios in varieties assigned to different phases of the Dynamic Model (plus GB, the erstwhile donor variety). All varieties show lexical variability of Adj-N pairs, increasingly so with higher token numbers. Most interestingly, however, it is perfectly clear that there is a strong correlation between productivity and Dynamic Model phases: “Phase V varieties show the highest productivity rate (even higher than the BrE reference variety), while Phase II varieties have the lowest productivity. Phase IV and Phase III can be found in between these two extremes (with Phase IV, however, significantly superseding Phase III varieties)” (Hoffmann Reference Hoffmann2022:45). Clearly, this can be seen as emergent grammar patterns and effects of self-organizing forces in the course of time.Footnote ²⁹

Figure 5 Productivity of the ADJ-N slots across Dynamic Model phases

(Hoffmann Reference Hoffmann2021:45; CC-BY4.0 license)

Figure 5Long description

The x-axis represents tokens, and the frequency ranges from 0 to 25000 with increments of 5000. The y-axis represents types, and the frequency ranges from 0 to 15000 with increments of 2000. There are five distinct lines on top of each other, each representing a different category (varieties in the Second, Third, Fourth and Fifth phase of the Dynamic Model, and Great Britain), as indicated by the legend in the bottom right corner. Each line starts at zero types and zero tokens at the bottom left end and increases to 25000 tokens to the right, but rises to varying type heights. The line for phase Two varieties (very light grey dashed line) is lowest by far, reaching only about 5000 types at the right end of 25000 tokens. All other lines reach between 12000 and 15000 types but are also different in their elevation. The line for phase Three (light grey solid line with longer dashes) is lowest of these, reaching about 11000 types at 25000 tokens. The line for phase Four (medium grey solid line with longer dashes) is slightly higher, reaching approximately 12500 types at 25000 tokens. The line for Great Britain (black solid line with small dots) is next highest, reaching about 14500 types at 25000 tokens. The line for phase Five varieties (dark grey solid line with short dashes) is the highest of all and reaches about 15000 types at 25000 tokens.

Another example illustrating the same process, a simpler and less theoretically motivated one, is the expansion of ditransitives in IndE, shown by Mukherjee and Hoffmann (Reference Mukherjee and Hoffmann2006). Ditransitive constructions, with an indirect and a direct object (I gave you a book) are a common complementation pattern in English. As the authors documented, in IndE a much wider range of verbs allows this complementation pattern, thus building emergent schematic patterns and new networks between constructions and verbs (see Examples 5a–c).

1. 1. (a) the park would have provided the city the much-needed greenery

   2. (b) we presented each donor a travel bag

   3. (c) the FCI is not supplying us the foodgrain (all from Mukherjee & Hoffmann Reference Mukherjee and Hoffmann2006)

Of course, various other constructions and examples discussed in other sections also illustrate the effects of auto-emergence in association with other CDS principles. For example, the distinctively Singlish construction with postponed relativizer one discussed in Section 4.3.2 (produced by “filtering” processes through Chinese impact; Bao Reference Bao2015) can be taken to represent self-organization within this variety in association with network relations with other structures and with other substrate languages. Similarly, and interestingly, in some varieties the form finish has emerged as a marker of completeness of an activity (thus standing in paradigmatic network relation with the morphological past). This has been analyzed most thoroughly for children’s English in Singapore (Buschfeld Reference Buschfeld2020, as in Examples 6a and 6b), but it is also documented for other high-contact varieties (e.g. Examples 6c and 6d).

1. 1. (a) He comb his hair finished.

   2. (b) She tie finish her shoe laces. (both Buschfeld Reference Buschfeld2020:202)

   3. (c) I cook finish. (Tok Pisin)

   4. (d) You finish eat? ‘Have you eaten?’ (Indian South African English; both quoted from Siemund Reference Siemund2013:119)

This usage of finish showcases a wide range of CDS properties: It has auto-emerged via self-organization (thus also showing ongoing change processes), operates systemically within past time reference options, increases the complexity of time reference options, and stands in network relations of various kinds, including the meaning of the word in English, input options and models from substrate languages (in Singapore Chinese V+wán-(le); Buschfeld Reference Buschfeld2020:201), and the grammatical options for time reference.

In fact, the domain of tense-modality-aspect expressions seems particularly liable to the self-organized, network-related emergence of newly grammaticalized uses in some varieties. This is illustrated by perfective done and remote-time been as auxiliaries in Earlier African American English (Schneider Reference Schneider1989:114–27) and some West African and Caribbean pidgins and creoles, or by innovative future time markers derived from English forms with other functions, as in gwine (Gullah), busy (White South African English), preverbal go (Nigerian Pidgin), or (derived from the adverbial by-and-by) bae /bai in Bislama and Tok Pisin (see Siemund Reference Siemund2013:121–22, 141). Compare also the wide range of innovative habitual marking strategies documented in WEs (eWAVE 90–93) or other distinctive patterns involving progressives or perfectives (e.g. eWAVE f89, 94–95, 97–98.

Another formative strategy found in several varieties that can be seen as auto-organizing is reduplication (or even multiple repetition of a single word) to express intensification or duration (see Examples 7a–c).

1. 1. (a) I walk-walk-walk ‘I was walking’ (Singapore)

   2. (b) Ay ben wed wed wed wed ‘I waited for ages’ (Kriol; both quoted from Siemund Reference Siemund2013:144)

   3. (c) fool-fool ‘very foolish’, small-small ‘very small’ (Jamaican Patwa), proper proper ‘very properly’, talk-talk ‘discuss intensely’, same same (widespread in Asia) (Schneider Reference Schneider2020a:113, 157, 215)

4.3.7 Nonlinearity and Fractals

As stated earlier, CDSs are best modeled mathematically by nonlinear equations, with exponents raised to higher powers than 1. It seems difficult to pin down this precise understanding of the notion in language, however, unless the commonly observed ups and downs of frequency changes of forms through time are metaphorically accepted as nonlinear. Given the central role of usage in the evolution of speech, ideally what we would need, and what would need to be modeled, is a complete record of all utterances made in a universe under investigation, and their quantitative relationship to speakers and contexts – but this is a thought experiment at best. Trivially, however, it is not difficult to posit and believe that nonlinearity, the absence of a proportional relation of changes between input and output, applies: clearly, the multitude of individuals involved, linguistic choices available, and possible situational settings is not connected by direct, linear functional relations, and is likely to produce nonlinear interactions, supported and possibly produced by the feedback loops in usage.

As stated earlier (in Section 2.2.7), nonlinear and fractal relations are therefore more difficult to identify and pin down in language evolution. Vetchinnikova (Reference Vetchinnikova, Filppula, Klemola, Mauranen and Vetchinnikova2017) assumes that the fractal nature of language manifests itself in “inherent similarity in shape” (279) across the social, structural and temporal dimensions (e.g. similarity between short-term and long-term changes, or similarities across idiolects [284]). Specifically, she views similarities in the patterns of chunking (multi-word unit patterning, e.g. the pattern “it is ADJ that” [289–97]) between the cognitive and the communal planes of language representation (which she postulates, observing similarities between fixing in an individual’s language and the emergence of communal phraseological conventions [302]) as an indication of fractality. More broadly, the core processes of entrenchment and conventionalization postulated by Schmid (Reference Schmid2020) can probably readily be viewed as fractal in nature – self-similar and emerging across manifestations at different levels of magnitude, from utterance type to conventionalized social settings to individuals, groups of speakers, and communities (and, on all levels of scale, associated utterances).

For regional and social variants of items in dialectology Kretzschmar (Reference Kretzschmar2009, Reference Kretzschmar2015) most decidedly argued for and documented the fractal, scale-free nature of frequency distributions of tokens, stating: “The scale-free property of complex systems predicts that there will always be sub-populations at different scales” (Kretzschmar Reference Kretzschmar2015:123). To put it more simply, irrespectively of how closely we zoom in, whether we consider a speech community, a large part or a small group of speakers in it, or an idiolect, the frequency relationship between alternative choices made will always be the same. He showed that the same frequency relationship, which he calls “A-curve” and is related to what he calls the “80/20 rule” in the business world, holds with respect to the variants of an item in his Linguistic Atlas data: Token frequency plots in a set of data often yield a small number of types with high token frequencies (the prototypical representatives of a category) as against a large number of rare alternatives (the “long tail”). Figure 6Footnote ³⁰ shows this type of distribution for lexical variants of the concept ‘thunderstorm,’ based on frequency charts of items recorded in the Linguistic Atlas of the Middle and South Atlantic States: One item predominates strongly (thunderstorm, found 759 times), others trail it with moderate frequencies (thundershower, 248; storm, 136; thundercloud, 110), but very many other types (such as squalls, rainstorms, blinger, hail storms, etc.) occur but are offered extremely rarely, often only once. As Kretzschmar (Reference Kretzschmar2009, Reference Kretzschmar2015) shows, the same type of distribution holds in a scale-free manner (i.e. independent of the magnitude and breadth of items represented, across subsets of subjects, and so on), and it is also found in other natural-language type-token relations such as frequencies of word collocates or of relativizer choices or construction types in text corpora (Kretzschmar Reference Kretzschmar2015:94, 102–103). There is always one or a small number of high-frequency predominant, prototypical form(s) in addition to a small set of intermediately frequent and many rare alternative choices. These distributions, he believes, are self-organizing patterns derived from neural network processes, a “sort of emergent order that we expect in language patterns from the operation of complex systems” (Kretzschmar Reference Kretzschmar2015:147). Kretzschmar (Reference Kretzschmar, Kytö and Siebers2022) argues that the entire set of dialects (not to be reduced to the small set with conceptualized names) from all-encompassing “American English” down to any speaker group in between, independent of size and thus “scale,” shows the same type of distribution. And, importantly, this fractal quantitative relationship holds the cradle of possible future change. He argues that any of the “great many variants in the long tail of the nonlinear frequency profile” (29) can become an “embryonic variant” which may spread in the community and ultimately result in systemic change.

Figure 6 A-curve distribution of variants of ‘thunderstorm’ in LAMSAS (Kretzschmar Reference Kretzschmar2009:98).

Reproduced with permission of the Licensor through PLSclear.

Figure 6Long description

The x-axis ranges from 0 to 109 in varied increments. The y-axis ranges from 0 to 800 with an increment of 100. The curve falls drastically at the left and has a long, almost horizontal tail to the right. It passes through the following data points: (1, 800), (10, 30), (19,10), (28, 5), (37, 1), (46, 1), (55, 1), (64, 1), (73, 1), (82, 1), (91, 1), (100, 1), (109, 1). All the values are approximate.

This property of fractal, scale-free frequency distributions in language behavior is robust and well documented, and it is in line with another basic frequency distribution principle in language, Zipf’s law (see Section 2.2.7). However, there is a limit to this observation: this type of distribution operates at certain dimensions, at several levels of magnitude (e.g. community, group, speaker) but not globally (unlike a fully fractal distribution like, for instance, the Mandelbrot set, where one can zoom in without limitations, with the pattern remaining the same). Hence, in line with what was observed in Section 2.2.7, it seems meaningful to understand many frequency distributions in language as “fractal-like.” Much of this has to do with the question of the level of granularity a researcher is aiming at – are we going for the “big picture” (focusing on frequent, mainstream phenomena), or are we interested in the observation and interpretation of small-scale variants and details? In this sense, a fractal perspective on WEs may also be implied in observing distributions across different levels of regional scale in varieties, given that, for example, there are several properties shared specifically across Asian Englishes (e.g. lack of articles), smaller-scale regional constellations (e.g., Malaysian and Singaporean English share the discourse marker lah and the kena-passive), particular variety types based on contact histories (Kortmann et al. Reference Kortmann, Lunkenheimer and Ehret2020; Kortmann Reference Kortmann, Aarts, McMahon and Hinrichs2021), or simply properties of individual varieties (Schneider Reference Schneider, Durgasingh and Meer2025b). Certainly, such similarities across several but not all magnitudes can be understood as quasi-fractal.Footnote ³¹

4.3.8 Attractors

In second language acquisition, Larsen-Freeman (Reference Larsen-Freeman2018b) views learners’ knowledge states as “attractors,” modified by phase transitions and nonlinearity: “when the language resources of the learner change from one relatively stable attractor state to another, the point of transition is marked with increased behavioral variability, the result of nonlinearity. Having passed through a phase transition, the resources self-organize or restructure, where the new organization may be novel, qualitatively different from earlier organizations” (83; cf. Verspoor et al. Reference Verspoor, Lowie and van Dijk2008). In this sense “attractors” in the evolution of varieties can be understood as “preferred states” (Larsen-Freeman Reference Larsen-Freeman, Jenkins, Baker and Dewey2018a:52): usage habits or constructions that are easier to process, supported by frequency in discourse, or building mental associations and network relations to other, established, similar structures, backed by evident parallelisms.Footnote ³²

This “attracting” force, favoring one form of expression over another, can operate on several levels of complexity. In lexis, simple and transparent forms and those with high frequency will be preferred over rarer, internally complex or opaque lexemes. In Figure 7, spontaneously formed compounds like working time and run time are coined and preferred over a more idiomatic (but less well known) alternative operating hours because they are readily transparent, compositional for any reader. In Figure 8, a head noun (preceding of properties) is required, but we find what normally would be understood as a verb participle form (lost of properties) instead, presumably because the verb form is much more frequent,Footnote ³³ thus more deeply entrenched and with an attracting force in a learner’s usage-based body of English forms remembered. On a slightly more complex level, we find the same principle operating with some complementation patterns, for example “redundant” preposition choices as attractors. As is well known, in very many WEs the verb discuss takes not a direct object (as in metropolitan varieties: discuss something) but a preposition phrase with about (discuss about something; see Leuckert et al. Reference Leuckert, Lange, Bernaisch and Yurchenko2023: section 5.3). This is clearly cross-motivated by the attractive force of the corresponding noun complementation (a discussion about something) and by the complementation of semantically very similar high-frequency verbs (talk about something), which thus invite cross-constructional mental associations. At the highest level of network complexity, we find schematic constructions which operate on the same formal and semantic associations as “attractors,” taking variant lexical fillers, such as in the “way” construction: donations find their way into the country; I … sleep my way to the top; we’ll work our way through the Workbook (Hoffmann Reference Hoffmann2021:33).

Figure 7 Simplified lexical choices in China

Figure 8 Word class transfer (lost < loss) on Langkawi (Malaysia)

An interesting example from my own recent work is a study of the complementation of the multi-word verb look forward to (Schneider Reference Schneider, Smith, Collins and Korhonen2025a). Based on what originally was a coincidental, serendipitous observation which I then decided to trace systematically through several corpora from different countries and time periods, I found that the complementation of look forward to by a plain infinitive form rather than the verbal -ing participle called for in metropolitan varieties (I look forward to meet you rather than to meeting you) seems to have originated in IndE in the 1970s and to have spread and gained in frequency from there through decades: In IndE newspaper texts from the early 2000s as many as 40 percent of all verbal complementation patterns show this structure, with proportions only slightly lower in other sources, in other South Asian varieties, and, decreasing in frequency, in Southeast Asian and African ESL varieties. I argue that this is to be explained by the effect of second language acquisition processes, which favor simple, frequent, and transparent structures and may produce some degree of fossilization. In particular, the form to, which is homonymic with two completely different readings (a preposition, as in to the holidays, and an infinitive marker, as in to go), is primarily associated by learners with the infinitive function, which occurs about twice as often (in COCA) and is simpler (calling for a plain verb complement rather than a noun phrase or, in this case, a bimorphemic verbal -ing form), and it is also backed by (=associated with) other, formally and semantically similar structures such as I hope to come. Hence, the infinitive marking function of to in this construction apparently serves as an attractor to invite the plain infinitive complementation of look forward to, at the expense of the more complex “preposition + reduced Ving clause” reading.

A rather abstract and deeply rooted but consistently effective attractor seems to be a distinction originally discovered and documented by Mesthrie (Reference Mesthrie2006).Footnote ³⁴ Some NEs tend to delete many constituents that in mainstream grammar are considered mandatory. Take Example 8a, a typical utterance in Singlish: No subject, no lexical main verb, no further complements or constituents. In contrast, other varieties are characterized by what Mesthrie calls “anti-deletion”: a tendency to restore features which are otherwise commonly deleted, such as the infinitive marker in Example 8b, to retain forms which are often deleted elsewhere (e.g. the complementizer that or a copula), and to insert additional grammatical morphemes (see Example 8c). Among other things, this corresponds to the “resumptive pronouns” identified by Siemund (Reference Siemund2013:266–67; see Example 8d).

1. 1. (a) Can or not? – Can! (Singlish)

   2. (b) I let him to speak Zulu. (Black South African English; Mesthrie Reference Mesthrie2006:122)

   3. (c) As it is the case elsewhere, … (Mesthrie Reference Mesthrie2006:127)

   4. (d) The guests whom I invited them have arrived (Nigerian English; Siemund Reference Siemund2013:266).

The “attractors” in his case are deeply rooted grammatical habits and associations, to be expressed as something like “Be maximally concise: avoid superfluous forms” in the case of the deleter varieties (mainly in Asia) as opposed to “Be maximally explicit: insert all possible forms to make the message clear” in anti-deletion varieties (primarily in Africa).

Again, the oscillation between “attractors” and the emergence of properties can be identified in association with other CDS principles, and thus with other examples discussed elsewhere in this Element. A case in point, for example, is the interplay between preferences for pre- or post-head positions of noun phrase modifiers in Asian versus African varieties discussed in Section 4.3.4. (Brunner Reference Brunner2017). The regionally predominant head-modifier sequences (e.g. Adj-N vs. N-Adj) as mentally entrenched preferred associations can be seen as constituting attractors, different ones in different world regions in this case. Similarly, the semantic force of the activity type readings associated with past tense and perfect forms, respectively, shaped in part by contextual conditions of migration and cross-varietal influences, can be regarded as attractors, with specific meanings intended calling for certain formal realizations, and, alternatively, forms chosen inviting particular temporal interpretations. The relative force of attraction in these cases varies by context and variety, and as we saw in Section 4.3.5, seems to have been diminishing in some NEs when compared to BrE.