2 Assemblages and the Evolution of Languages

[T]here is no language in itself, nor are there any linguistic universals, only a throng of dialects, patois, slangs, and specialized languages. There is no ideal speaker-listener, any more than there is a homogeneous linguistic community. Language is, in Weinreich’s words, ‘an essentially heterogeneous reality.’ There is no mother tongue, only a power takeover by a dominant language within a political multiplicity. Language stabilizes around a parish, a bishopric, a capital. It forms a bulb. It evolves by subterranean stems and flows, along river valleys or train tracks; it spreads like a patch of oil.

Deleuze and Guattari, A Thousand Plateaus1

Approaching the study of languages from the point of view of assemblage theory is a difficult task because linguistic entities operate at several levels of scale at once. First, words and sentences are component parts of many social assemblages, such as interpersonal networks and institutional organisations, interacting not only with the material components of those assemblages but also with non-linguistic expressive components. Second, some linguistic entities (religious discourses, written constitutions) have the capacity to code all the components of a given assemblage. While in the first case linguistic entities are variables of a social assemblage, in the second case they become a parameter of it. Finally, a given language may itself be studied as an assemblage, exhibiting the characteristic part-to-whole relation: sounds or letters interact to form words, with irreducible semantic properties of their own, and words interact to form larger wholes, sentences, with their own semantic and syntactic properties. When studying real languages as assemblages it is important to emphasise that there is no such thing as ‘language in general’. That is, we must replace the reified generality ‘language’ with a population of individual dialects coexisting and interacting with individual standard languages. Each member of the population can then be examined for its composition. The expressive components of a given language include not only the meanings of its words and sentences, but also other non-semantic sources of expressivity: tone, stress, rhythm, rhyme. The material components are either acoustic matter – pulses of air produced in the larynx and shaped by tongue and palate, teeth and lips – or physical inscriptions, like carvings on stone, ink on paper, or the ones and zeroes that code language into electricity flows in the Internet.

Although for analytical purposes it is convenient to distinguish these three different levels – a language as a variable of a social assemblage; a language as a parameter of a social assemblage; and a given language as an assemblage – it is clear that in any concrete case all three levels operate simultaneously and influence one another. In this chapter the three levels will be described separately, starting with language as a variable, or as an expressive component, of concrete social assemblages like communities and organisations. Any assemblage component is characterised both by its properties and by its capacities. While the phonological, semantic, and syntactic properties of a given language may remain constant at different levels of scale, its capacities do not. The distinction between properties and capacities is roughly that between what an entity is and what it can do. In a famous text, appropriately named How to Do Things With Words, the British philosopher John Austin introduced the idea that sentences (or more exactly, the statements expressed by sentences) have the capacity to perform acts that generate social commitments by their very utterance: promises, bets, apologies, threats, warnings, commands, marriage vows, death sentences, war declarations, and a large variety of other ‘speech acts’.2

Speech acts are often performed by statements that have certain semantic and syntactical properties: performative statements like ‘I promise that X’ or ‘I command you to do Y’. But these properties are not sufficient to endow them with their (illocutionary) capacities. Take for example a judge’s utterance of a death sentence: ‘I declare you guilty and condemn you to die.’ This statement has a capacity to affect only if the person who utters it is someone with legitimate authority who is part of the right organisation: a judicial, and not a legislative or executive, governmental organisation. Moreover, the statement must be addressed to someone with the capacity to be affected. Animals other than humans, for example, cannot be declared guilty in a court of law, and in many judicial organisations a mentally ill person cannot be condemned to die. Commands, too, presuppose the existence of an authority structure, like the chain of command of a military organisation. For the speech act to have binding capacities, those who give orders as well as those who receive them must have the right positions in that hierarchy: one must be a subordinate and the other his or her legitimate superior. Other speech acts presuppose not the existence of institutional organisations but of communities. In a tightly knit community, for example, word of mouth about dishonoured commitments travels fast and these violations can be punished by ridicule or ostracism. Hence, a community is the assemblage within which promises, bets, apologies, and assertions have a capacity to affect, and it is mostly its members who have the capacity to be affected. An outsider who is just passing by a small town may make statements with the right performative properties, but he or she cannot be held accountable for breaking a promise or failing to heed a warning. In other words, the outsider’s speech acts do not create a commitment that is enforceable.

Hence, the capacity of speech acts to affect members of an organisation or a community depends on the existence of enforcement mechanisms. After a promise has been made, a bet agreed upon, or an apology accepted, the status of a neighbour in a community is changed by that event in the sense that he or she is now regarded by other members as having acquired a specific social obligation: to keep the promise, to pay the bet, to behave in a forgiving way. And similarly for speech acts performed in organisations: a guilty verdict changes the status of someone from innocent to guilty with all the consequences this has; and once a command has been given in a military organisation the status of the soldier who received the order is changed by it, creating an obligation to carry out the command or be punished for insubordination. Deleuze and Guattari refer to the events in which the capacities of language are exercised, the events that create social obligations, as incorporeal transformations. They constitute an important expressive component that interacts with the assemblage’s material components: human bodies and the ‘bodies’ of buildings and machines. As they write in relation to a judge’s sentence:

In effect, what takes place beforehand (the crime of which someone is accused), and what happens afterward (the carrying out of the penalty), are actions-passions affecting bodies (the body of the property, the body of the victim, the body of the convict, the body of the prison); but the transformation of the accused into a convict is a pure instantaneous act or incorporeal attribute that is the expressed of the judge’s sentence. Peace and war are states or interminglings of very different kinds of bodies, but the declaration of a general mobilisation expresses an instantaneous and incorporeal transformation of bodies.3

Although in their treatment of language Deleuze and Guattari borrow the concept of a speech act from Austin, they also add some elements of their own. Specifically, they invent the concept of an order-word to connect speech act theory to the theory of assemblages. On the one hand, order-words refer to the capacities of statements to create commitments: ‘Order-words do not concern commands only, but every act that is linked to statements by a social obligation. Every statement displays this link, directly or indirectly. Questions, promises, are order-words.’4 On the other hand, order-words are not related to the communicative functions of language, functions that imply the existence of a subject or person with intentions to communicate, but to the impersonal transmission of statements (and their illocutionary capacities). When a command is given in a military hierarchy, for example, the intentions of the commander issuing the order may be important, but after it is issued it is the transmission of the order all along the chain of command that matters. And when the dishonouring of a commitment is witnessed in a community, the intentions of the person witnessing a broken promise or a false assertion are only the start of a process of transmission via word of mouth, an impersonal process that gives the community as a whole its capacity to store the reputations of its members. As Deleuze and Guattari write:

If language always seems to presuppose itself, if we cannot assign it a non-linguistic point of departure, it is because language does not operate between something seen (or felt) and something said, but always goes from saying to saying. We believe that narrative consists not in communicating what one has seen but in transmitting what one has heard, what someone else said to you. Hearsay . . . Language is not content to go from a first party to a second party, from one who has seen to one who has not, but necessarily goes from a second to a third party, neither of whom has seen. It is in this sense that language is the transmission of the word as order-word, not the communication of a sign as information.5

To conclude this brief account of languages as variables (or components) of social assemblages it will be useful to clarify some of the terminology used by the authors. In particular, rather than simply distinguishing the material and expressive components of a community or organisation, Deleuze and Guattari sometimes use different terms for the assemblages themselves. Communities and organisations consist of more than just persons and their organic bodies. They are composed of heterogeneous material bodies, including tools and machines, food and water, buildings and furniture. The word that the authors use to refer to the assembling of heterogeneities as such is ‘machinic’. But these material bodies are linked by the enforceable social commitments created by the enunciation of speech acts. Thus, a community or organisation is, on the one hand, ‘a machinic assemblage of bodies, an intermingling of bodies reacting to one another; [and] on the other hand, it is a collective assemblage of enunciation, of acts and statements, of incorporeal transformations attributed to bodies’.6 In what follows I will stick to the simpler usage, speaking of assemblages and their material and expressive components, rather than making a distinction between types of assemblage.

So far order-words have been treated as one of the variables characterising social assemblages, variables capable of taking as many values as there are possible speech acts. But how are we to conceive of other, larger linguistic entities that seem to code every component of a community or organisation? In the case of a judge’s sentence, for example, the court (as part of a larger whole composed of other judicial organisations) is not only characterised by discrete events in which statements exercise their capacities to affect but also by a larger body of law that codifies past legal knowledge and precedent. It is by referring to that legal code that a particular prison sentence acquires its legitimacy. Similarly, in a military organisation (and many civilian organisations) there exists a written constitution codifying the functions, rights, and obligations of all incumbents, whatever their role or function in the organisation. This constitution forms the background against which particular commands can be given with legitimate authority. Although these written codes also operate, on a day-to-day basis, through the transmission of order-words, they are not so much variables of these assemblages as parameters quantifying their overall environment. And similarly for communities: a religious discourse, codifying places into sacred and profane, codifying food into permissible and taboo, codifying days of the year into ordinary and special, affects every expressive and material component of the social assemblage. The distinction between variables and parameters comes from mathematical models of physical processes. In these models, temperature can appear as a variable, as in the internal temperature of an animal’s body, for example, as well as a parameter quantifying the temperature of the animal’s environment. In the previous chapter, we borrowed this distinction to parametrise the concept of assemblage, building into it ‘control knobs’ with variable settings. We can use the same manoeuvre here to characterise the way in which legal or religious categories are systematically assigned to all components of a social assemblage. Parametrising the concept allows us to capture the inherent variability of the identity of all historically constituted entities: we can use different settings to characterise one and the same community or organisation at different points in their history. But it also makes it easier to think about the way in which this variation is distributed statistically in a plurality of communities or organisations inhabiting the same city or country. We can simply picture in our minds a conceptual population, consisting of a large number of assemblages with different settings of their parameters.

Being able to think about entire assemblage populations, and about the statistical form in which their variation is distributed, is crucial to the application of assemblage theory to the evolution of languages and dialects. Although the above discussion emphasises the transmissibility of order-words among the members of organisations or communities, it is clear that order-words are also transmitted across generations, from parents or teachers to children. In this respect words and sentences are similar to genes, that is, they are replicators. Biologists have known for a while that genes do not have a monopoly on evolution. Learned patterns of behaviour, that is, behavioural patterns that are not genetically hard-wired, can be transmitted from one generation of animals to another by imitation. The best-studied example of this evolutionary process is bird songs, like the complex songs of nightingales or blackbirds. These replicating behavioural patterns are referred to as ‘memes’.7 In human populations memes are exemplified by fashion (dress patterns, dancing patterns) but a more important type of replicating entity, one that replicates not by imitation but by enforced social obligation, is the sounds, words, and grammatical patterns that make up a language: although babies may at first aim at imitating the sounds coming out of their parents’ mouths, they soon learn that speaking their mother tongue is not optional but obligatory, and that there is a norm (the dialect spoken in their community) to which they must conform.

In all cases of evolution variation is indispensable. If genes replicated exactly, if there were no copying errors or other mutations, organic entities would not have the capacity to evolve because the selection pressures that promote the replication of some variants at the expense of others, leading to the slow accumulation of adaptive traits, would have no raw materials on which to operate. And this is also true of any other kind of replicating entity. In the case of dialects the amount of variation, and the extent to which it is centripetal or centrifugal, is in many cases determined by the intensity of enforcement. The capacity of communities to store reputations and detect violations of local norms, coupled to the use of ridicule and ostracism as informal forms of punishment, gives them the capacity to enforce and preserve their local linguistic identity.8 A young member of a small town who goes to a big city to study and returns with a different accent will not be treated with respect, as bringing a more sophisticated variant to enlighten the locals, but as an outsider who must be ridiculed. But enforcement can also be performed by an institutional organisation, like the Academies of Language that, starting in the late sixteenth century, were chartered to create standard versions of the dominant dialects of regional capitals (Florence, Paris, Madrid). The homogeneity of these artificial languages was enforced through the publication of official dictionaries, formal grammars, and books of correct pronunciation.9

Let’s illustrate this with a few examples. Before the collapse of the Roman Empire, all the regions the Romans had conquered, and on which they had imposed Latin as the official language, were relatively linguistically homogeneous. The area under Roman control was so vast that there must have been local variations, but the very presence of Roman troops and officials (as well as of a variety of government organisations) kept the variation centripetal: to be able to address a government official, for example, the locals had to use the Latin of Rome, and this gave them a standard to use as a norm. But the moment the empire ceased to exist, and with it the organisations that enforced its rule, the variation turned centrifugal, as local communities replaced imperial organisations as enforcement mechanisms. Vulgar Latin, the Latin spoken by the conquered masses, began to change and gave birth to a large variety of Romance languages: several forms of Italo-Romance, of Franco-Romance, and of Hispano-Romance.10 At first, this explosive divergence went unnoticed: people in the Latinised areas of the former empire thought they were all speaking Latin, and there were hardly any names for the emerging variants. Awareness of the existence of new versions of oral Latin involved an organisational intervention: the reforms that the court of Charlemagne introduced in AD 809. A professional grammarian named Alcuin was hired to report on the state of language in the kingdom, and when he returned he informed the king that something new existed outside the walls of the castle, a new language he called ‘rustica romana’.11

By the year 1000, many romance dialects coexisted in Europe in what is called a ‘dialect continuum’.12 The dialect of medieval Paris, for example, was connected to the dialect of Florence by a continuum of French, Franco-Provencal, and Gallo-Italian dialects. Although relatively sharp transitions (isoglosses) did exist in this continuum, the divergent set of Franco-Romance, Hispano-Romance, and Italo-Romance dialects was a highly deterritorialised entity, particularly when compared to the form of Latin spoken in governmental and ecclesiastical organisations. This other, much more prestigious, form of Latin was taken from classical books and was given an official spoken form during the Carolingian reforms. Considered as an assemblage, it was highly territorialised and coded: its internal homogeneity could be preserved by reference to Roman texts; its borders policed by aristocratic and religious communities; and its uses (reading the Bible aloud during Mass, writing laws and edicts) formally codified. To distinguish the different states in which a given language can exist, Deleuze and Guattari introduced the concepts of major and minor languages. As they write:

Must a distinction be made between two kinds of languages, ‘high’ and ‘low’, major and minor? The first would be defined precisely by the power of constants, the second by the power of variation. We do not simply want to make an opposition between the unity of a major language and the multiplicity of dialects. Rather, each dialect has a zone of transition and variation . . . [It] is rare to find clear boundaries on dialect maps; instead there are transitional and limitrophe zones, zones of indiscernibility . . . The very notion of dialect is quite questionable. Moreover, it is relative because one needs to know in relation to what major language it exercises its function: for example, the Québecois language must be evaluated in relation to standard French but also in relation to major English, from which it borrows all kinds of phonetic and syntactical elements, in order to set them in variation . . . In short the notion of dialect does not elucidate that of a minor language, but the other way around; it is the minor language that defines dialects through its own possibilities of variation.13

The relativity of the notions of major and minor language can be captured by a parametrised concept in which the values of the parameters can vary historically. Thus, while all the vernacular forms of Romance continued to be minor relative to major classical Latin in the first half of the second millennium, some dialects became major relative to others. The commercial revolution that took place in Europe from the eleventh to the fourteenth centuries, and the diversification of governmental functions in the proliferating city-states, multiplied the uses of writing: licences, certificates, petitions and denunciations, wills and post-mortem inventories, commercial and financial contracts began to be written with increased frequency.14 Because this rising demand was not matched by an equivalent supply of classical Latin scribes, the governments of regional capitals commissioned the creation of writing systems for their own dialects.15 The appearance of a written form for the vernaculars had several consequences: it reduced the intensity of variation in those dialects, decelerating their evolution relative to those without writing; it increased awareness of their unique identity among their users, by making comparisons between written documents possible; and it augmented their level of prestige relative to classical Latin. In short, writing had territorialising effects that gave some members of the continuum discrete boundaries and constant content, transforming them into major languages relative to the remaining minor ones, even if they were still minor relative to major Latin.

The evolutionary process that gave birth to the ancestors of today’s English dialects shows a similar pattern. The raw materials for this process were brought to England, starting in the fifth century, by several migratory waves of Germanic-speaking Jutes, Angles, and Saxons. In the settlements created by these migrants, a dialect continuum was created similar to the one that was forming on the continent. One component of this continuum, the dialect which today is referred to as ‘West Saxon’, was given a writing system and its prestige was greatly increased with the appearance of Beowulf, a literary masterpiece. Thus, West Saxon became a major language in relation to the other minor variants. But then the Normans, who spoke a variant of Franco-Romance, staged a brutal invasion in which the English governing elite was physically exterminated, and with it any organisational means to enforce the rising standard. Defeat at the Battle of Hastings in 1066 not only sealed the fate of that elite but also became a transformative event in the history of their language. In the words of historian John Nist:

As a result of the Norman conquest, the Old English nobility practically ceased to exist. Within ten years after the Battle of Hastings the twelve earls of England were all Norman. Norman clergy . . . took over the highest offices of the Church: archbishop, bishop, and abbot. Since the prestige of a language is determined by the authority and influence of those who speak it, the French of the Norman masters became the tongue of status in England for almost two hundred years. The Norman King of England and his French nobility remained utterly indifferent to the English language until about 1200. During that time the royal court patronized French literature, not English. When the court and its aristocratic supporters did finally pay attention to the native language of the land they dominated, that language was no longer the basically Teutonic and highly inflected Old English but the hybrid-becoming, Romance-importing, and inflection-dropping Middle English.16

The forceful removal of all territorialising and coding pressures on the remaining population of minor languages increased their variation and accelerated their evolution. Of all the different changes that the dialectal population underwent perhaps the most important was the loss of inflections. An inflection is a syllable that, as part of a word, carries with it grammatical information. Highly inflected languages, whether Germanic or Romance, use the last syllable to express gender and number in nouns, and person and tense in verbs. The English peasants who lived during the Norman occupation had inherited the habit of placing stress on the very first syllable. This habit can still be detected in contemporary English words from different origin. Thus, words of Germanic origin consistently stress the root syllable (as in ‘love’, ‘lover’, ‘loveliness’) while those borrowed from Romance do not (‘family’, ‘familiar’, ‘familiarity’). Without enforcement beyond that performed by local communities, and given this habitual positioning of stress, the last syllables became literally eroded away. And without inflections the emerging dialects were forced to use a relatively fixed word order to express grammatical function, a major structural transformation. As Nist puts it: ‘Unhindered by rules of proscription and prescription, the English peasants demanded stress in the root syllable and remodeled the language with tongue and palate.’17

What kind of linguistic models do we need to accommodate these historical metamorphoses? Or to put this differently, what does language have to be, what kind of properties must it have, in order for it to possess these transformative capacities? Some linguistic models cannot account for evolutionary processes, or rather, subordinate the evolution of language to that of the human species. Noam Chomsky’s model, for example, postulates the existence of a universal grammar, a constant core common to all languages, evolved genetically and residing in our brains. This innate universal grammar includes linguistic categories (sentence, noun, verb) as well as re-writing rules that can transform one string of words belonging to those categories into another string.18 But in order to model an evolutionary process in which the replicators are linguistic, not genetic, we must get language out of our heads, and more importantly, the relations between words in a sentence must be conceived as relations of exteriority. As we argued in the previous chapter, a relation of interiority is one in which the terms constitute each other by the very fact that they are related; or, to put it differently, one in which the very identity of the terms is constituted by their relation, so that the terms have no autonomous existence. Hegelian totalities have that character, as do grammatical relations in Chomsky’s linguistics: nouns and verbs are constituted by their very relation in a sentence, as part of the totality of a universal grammar.

A model of language in which words are related to each other in exteriority has been created by the linguist Zellig Harris. In this model, a word carries with it, in addition to its meaning or semantic content, non-linguistic information about its frequency of co-occurrence with other words. The term ‘information’ refers to physical patterns, like the patterns of ones and zeroes processed by computers, so it is a measure of order, or of the degree to which a pattern departs from randomness. The (non-linguistic) information carried by words reflects the fact that some words tend to occur next to other words more frequently as a matter of actual usage. For example, after a speaker has uttered a definite article like ‘the’, the listener can reasonably expect that the next word will probably be a noun or a nominalised phrase. Harris calls these relations ‘likelihood constraints’. At any given point of time, these constraints may simply reflect the word combinations that happen to be used in a community (and hence be optional) but these customary patterns can become standardised or conventionalised and turn into obligatory constraints: a user who starts a sentence with ‘the’ may now be required to supply a noun or nominalised phrase as the next word.19 Saying that a likelihood constraint becomes obligatory is not, of course, to imply that the words themselves enforce the social obligation, but rather that the words are used in a collective assemblage, like a community or organisation, possessing its own enforcement capabilities.

Statistical information about frequency of co-occurrence, added to the normal semantic content of a word, provides a non-genetic evolutionary mechanism for the emergence of word categories. When a given word is used very frequently after another word, listeners come to expect with very high confidence that the second word will occur after hearing the first word. This implies that uttering the second word may become redundant since the speaker can count on the listener to provide it. And once it becomes customarily eliminated by the speaker, entire words can be reduced to suffixes or prefixes attached to another word, or even omitted altogether. Moreover, an entire sentence may be compacted into a single word, as speakers eliminate all its redundant companions, confident that the listener can reconstruct the full semantic content of the original sentence. Harris refers to the relations between words that allow this as ‘reduction constraints’, and he shows that application of reduction constraints over many generations can give rise to new classes of words, like adjectives, adverbs, conjunctions, and prepositions.20 A model of linguistic evolution cannot take for granted that early words had the combinatorial capacity that words have today, that is, the capacity to generate an infinite number of sentences from a finite dictionary. So, we can hypothesise that the earliest words were monolithic symbolic artifacts incapable of combining with each other but able merely to co-occur next to each other. At first, these monolithic artifacts had the same probability of co-occurring, that is, their relative frequency was random. But then this original symmetry was progressively broken by successive departures from equiprobability, changing the co-occurrence patterns and slowly endowing the artifacts with combinatorial capabilities.

Once classes of words had established themselves, a final constraint could emerge, one establishing first customary, then obligatory relations between words with the capacity to modify (operators) and words with the capacity to be modified (arguments). This final constraint, the operator-argument constraint, is needed to model the action of adjectives on nouns, of adverbs on verbs, and of the predicate of a sentence on its subject. The operator-argument constraint is also informational in the non-linguistic sense: the more unfamiliar the argument supplied for a given operator, the more informative it is, and vice versa – once an argument becomes too familiar it becomes redundant and can become a target for a reduction constraint. Thus, Zellig Harris’s model not only treats words as material entities entering into relations of exteriority with one another, but it is an explicitly evolutionary model: language evolves in a non-genetic way through the obligatory social transmission of combinatorial constraints, as words and sentences compete for ‘informational niches’.21 It can be objected that Harris treats sentences as purely communicative devices without considering that using a descriptive sentence in the context of a community always carries some illocutionary force: the descriptive sentence is typically a report or an assertion, the truth of which I am committed to defend as far as my neighbours are concerned. But to fault him for this would be to confuse two different levels: that of linguistic entities operating as variables of a collective assemblage, and that of a given language considered as an assemblage.22 In the latter case, as long as we emphasise that the transmission of linguistic entities across generations is compulsory, that is, that the replicators are norms not memes, we are dealing with the transmission of order-words.

Therefore, the two approaches can be used in conjunction. What the work of Harris contributes is a clarification of how assemblage theory should be applied to the history and evolution of languages.23 But in addition to adding a historical dimension to Deleuze and Guattari’s account, we must correct the social ontology in which the original application of assemblage theory to linguistic matters was couched. In the previous chapter we saw that the authors tend to reduce social complexity to three levels: the individual, the group, and the social field. Yet many of the collective assemblages that figure in linguistic history do not belong to any of these three categories. Take, for example, the Royal Academies that produced the highly territorialised and coded assemblages referred to as ‘standard languages’. The Tuscan Academy of Language, founded in 1582, and its French and Spanish counterparts (founded in 1637 and 1713, respectively) were complex organisations staffed with linguists officially dedicated to homogenising and formalising the dominant dialect of a particular city (Florence, Paris, Madrid). The authoritative dictionaries, grammars, and rules of correct pronunciation that they published operated as true order-words as they propagated through the rising middle classes, anxious to speak like their aristocratic counterparts. The further spread of standard languages involved, in addition, larger assemblages comprising many organisations: extensive networks of elementary schools and compulsory primary education in the standard. Thus, it seems clear that Deleuze and Guattari’s assemblage approach to language could greatly benefit from a more detailed social ontology, and that an extension of their work to history would also require taking into account a greater variety of forms of social agency. On the other hand, the ease with which their assemblage theory can be implemented at larger spatial scales and longer temporal scales is a testimony to its value and resiliency.

Notes