John Joseph on Limiting the Arbitrary

04 Jun 2017 | all notes

I’ve long had a thing for the more historiographical aspects of science and the history of linguistics in particular – possibly to do with my entering the language sciences via historical linguistics, a subfield where the mythologisation of early researchers and their foundational contributions to the discipline is fairly strong. While one sometimes finds references to the (pop) history of science embedded in the generally more modern-oriented research of the cognitive sciences, these often strike me as rather generic examples of the ‘look how silly pre-enlightenment scientists used to be’ type, rather than being treated as relevant cautious tales regarding our beliefs and commitments to present-day scientific paradigms (an exuberant confidence in modern science that I liken to being blinded by the enlightenment).

Either way, it was out of that same interest that John Joseph’s Limiting the Arbitrary: Linguistic Naturalism and its Opposites in Plato’s Cratylus and Modern Theories of Language (2000) had been on my reading list for a long time, and I was really happy to (temporarily) get my hands on a copy earlier this year. The book does not have one overarching story structure but is more of a retracing of the different ideas that linguistic scholars had about the ‘arbitrariness’ vs. ‘naturalness’ of linguistic signs over the centuries. Possibly the strongest take-home message for myself was that the history of this question and its present-day incarnations (e.g. iconicity) is long and complex and not in any way done justice by most linguists’ standard reference to Saussure (1916), since the post-humously jumbled together Cours de linguistique générale might represent one of the most confused and unhelpful standpoints on the matter. (This is just my own soundbite summary of the brief passage on Saussure in the present book and does not mean that Joseph is in any way carelessly dismissive of him – unlikely given that he is also a Saussure biographer (2012)).

So while my reading of Limiting the arbitrary started off as nothing but a private interest detour in the history of science (rooted in a potentially mythology-based fascination with historiography) I was really taken back when more than half way through the book Joseph comes up with this striking argument for the relevance (and importance) of knowing the history of (one’s) science, and particularly its terminology:

The ancient debate over whether language is arbitrary or natural is neither dead nor irrelevant to current linguistic theory, as the conventional wisdom holds. Rather, it has been absorbed into some of the most solidly established tenets of present-day theory, with other effects being discernible in the structure and use of languages themselves.

It is only by acknowledging our engagement in this age-old debate that we can genuinely begin to pass beyond it. Until then we, like so many of our predecessors, are apt to make little real progress in understanding the most basic questions about language — only the rhetorical veneer of progress, as we trip into the old intellectual ruts while our attention is distracted by the gleam of the latest models and metaphors.

It is not enough to simply declare the debate irrelevant, or to claim that nature is itself a conventional construct, or to say that language operates partly through natural and partly through conventional means and leave it at that. Until we have wrestled with this debate in all its historical manifestations and metamorphoses, any position we take is liable to recapitulate some earlier version of the debate. To get beyond it will require a major conceptual leap, and as any broad-jumper knows, preparing for a leap requires taking a few steps backwards.

(p.139-140, emphasis mine)

The remark about the gleam of the latest models in combination with the following, timeless question to be ‘kept in mind’ during the reading of the later chapters:

Can any version of naturalness itself be ‘natural’, or must it be known to us always and only as a historical product?


reminded me of something I’ve previously wondered about: in the present age of ‘big data’ any statistical analysis that reveals even slight preferences for one type of linguistic form over another is presented as evidence against the arbitrariness of signs, but just how small an effect size are we actually willing to accept as still giving us some meaningful insight into how language works? (This is leaving aside the fact that experimental results regarding the effects of iconicity are related to individual processing (Emmorey 2014) rather than to linguistic signs per se (Joseph 2015), a difference that many are bound to overlook given how sidelined semiotics is in most current linguistics curricula.)


Following the first three chapters on Plato’s Cratylus dialogue, the second half of the book traces the history of arguments on the ‘naturalness’ (and the many related and derived incarnations of that concept) of linguistic conventions in a quasi-chronological fashion. Below is, more for myself (now bereft of a full copy) than anything else, not so much a summary but a selection of the most insightful quotes and excerpts from the book. Page references at the end of blockquotes refer to Joseph (2000) unless specified otherwise.

Chapter 1: Nature and convention

Hermogenes: For no word is ever connected to anything by nature, but rather through the custom and habit of those who started calling it that and those who picked up the habit. (384c?-d7)

In this light [Hermogenes] raises what has always been the most powerful argument for conventionalism. If norms of language, law, or any other area of human activity are rooted in nature, the they should be universal across human communities, since nature is the same everywhere. But in fact norms vary widely from community to community.


The movement from nomos to nomothetēs, from law to lawgiver, is a movement from present (synchrony) to past (diachrony), a historicization of the question of linguistic correctness.


Chapter 3: Imitation and essence

Cratylus’ view seems absurdly extreme: if someone calls him by a name that is (wholly or partly) other than his own, that person is not performing any linguistic act, any more than a pot that makes a noise when struck. But this absurdity is the ultimate consequence of a naturalistic view of language, and Plato was the first, so far as we know, to recognize it. Any speech which fails to embody true reason must be dismissed by the linguistic naturalist as not speech at all. And if speech is true by definition, the whole issue of truth and falsehood in speech dissolves into nothingness.


The two contradictory positions Socrates refers to are: (1) words are correct if they imitate things in sounds, and (2) words cease absolutely to be correct if any sound changed. […] Note that (1) is a semantic and functional claim about language, while (2) is a purely formal one. Socrates points out in effect that the functional aspect of language inherently contradicts any absolute claim of a purely formal nature — an insight as timely today as it was 2500 years ago.


Cratylus’ persistence reflects the stubbornness of physis-based views of language in the history of linguistic thought. No nomos-based argument, however strong logically and however buttressed with the authority of ‘science’, has made a dent in the physis-based beliefs about language held by the population at large, which fears arbitrariness as a retreat to a primitive, pre-cultural state.


Socrates’ reply here has helped lead more than a few readers of the Cratylus to believe that Plato is finally endorsing the physis position. We have already seen plenty of evidence to indicate otherwise. Note what precisely Socrates has agreed to: not that language does operate by likeness to things rather than by chance signs, but that likeness (whether or not it is actually an option) is preferrable to chance. He has accepted a view of what language should be, not what it is


Communication through language appears to take place in spite of words evolving away from their ‘correct’ form. Confronted with this troublesome detail, Cratylus tries to eliminate it by tossing it disdainfully into the wastebasket of ethos, “habit” or “usage”.6

6 In much the same way, Saussurean structuralists would use the wastebasket of parole, and Chomskyan generativists the wastebasket of ‘performance’ and later ‘periphery’ (on which see Chap. 6).


It is clear that the search for linguistic correctness must move beyond physis as appearance and nomos as convention. It must also shift from the many who use words to the one who created them, which means a move from the present to the past — a historicization. The question ‘What is linguistic correctness?’ has become: What did the lawgiver have in mind when making language, and how did he embody it in the words he created?


One of the frustrations of the Cratylus is that Socrates does not return at the end to the first two questions that were posed to him. But his answer to them is implicit in later questions and answers, and may be formulated as follows:


Do words belong to things by nature or convention? Broadly speaking, both; technically neither. Physis and nomos are complementary, not contradictory forces. Neither alone is sufficient to account for the operation of actual languages. In the ideal language every word would make the essence of its meaning clear through its form, and would do so ‘naturally’ through direct imitation (memēsis) of that meaning. We could say then that ideally physis should be the operation principle of language — provided we take physis to indicate not just observable physical nature but also (and above all) the transcendent forms that underlie it. Again, that is the ideal, but in reality the masses are satisfied (as indeed they should be) to follow any nomos that has been established, regardless of how well or poorly it meets the criterion of correctness just outlined.


Chapter 4: Natural grammar and conventional words

Varro’s compromise: nature versus will

No one has ever deconstructed the nature-convention debate as powerfully as Plato, yet it survived essentially intact, resurfacing in its Sophistic terms in Stoic thought (3rd-1st centuries BC). Looking into our prime source for Stoic language theory, the De lingua latina (47-45 BC) of Marcus Terentius Varro (116-27 BC), we find the association of the natural with grammar and the arbitrary with lexicon well in place.


Whereas in the 11th century Peter Helias had believed that here were as many grammars as there were languages, the 13th-century speculative grammarian Michel de Marbais professed that “whoever knows grammar in one language knows it in another, with regard to everything that is essential to grammar”, all the differences among languages being merely accidental (Jolivet 1969: 1458-9). This view, which would have important resonances in the 17th and again in the late 20th centuries, undoes what has always been the most powerful argument against the connectedness of language — the existence of different languages — by saying in effect that languages only appear to be different, because of superficial trivialities that divert attention away from their underlying identity.


In the following passage from Gottfried Wilhelm Leibniz’s (1646-1716) Nouveaux essais sur l’entendement humain (written 1703, published 1765), where Philalethes (the ‘lover of truth’) presents the rationalist Leibniz’s characterization of the views of contemporary empiricists, notably John Locke (1632-1704):


THEOPHILUS. I know that the Scholastics and everyone else are given to saying that the significations of words are arbitrary (ex instituto), and it is true they are not settled by natural necessity; but they are settled by reasons — sometimes natural ones in which chance plays some part, sometimes moral which involve choice.

(Leibniz 1981 [1765]: 278)

[…] for if ideas themselves are not arbitrary, neither can the language which expresses them be wholly so. If the mind is the mirror of nature, and language the mirror of mind, then language too should mirror nature.


By analogy with natural and accidental signs, human beings have also created ‘instituted’ or ‘artificial’ signs, which are ‘voluntary’ in two senses first because the connection between symbol and meaning has been chosen, and secondly because the individual can freely choose to use the particular symbol or leave it unused.


Herder’s Abhandlung über den Ursprung der Sprache won the Berlin Academy Prize in 1770 and thus was very widely read, unlike Rousseau’s Essai. Early on Herder positions himself in opposition to both Condillac and Rousseau by asserting that “the former turned animals into men and the latter men into animals” (Herder 1772: 31). He rejects any attempt to draw an analogy from the ‘naturalness’ of animal language to human language (ibid., p.37). Human language was invented, when man was “placed in the state of reflection which is peculiar to him”, and when the mind in this state of reflection was “for the first time given full freedom of action” (p.52). Neither this reflection nor this freedom is shared by animals, whereas for man they are “essential to his species; and so is language” (ibid.). Man does not at all ‘speak by nature’ in the way that animals do because his essential nature, language was “invented as naturall and to man as necessarily as man was man” (p.56). No aspect of its invention was arbitrary — Herder considers the very idea of the arbitrariness of language to be nonsensical (pp. 92-93).


The role of the Varronian tradition in the establishment of scientific linguistics would crystallize with Wilhelm von Humboldt (1767–1835), whose work made a deep impact on the 19th-century study of language, and who is chiefly remembered today for his posthumously published introduction to his study of Kawi, the sacred language of Java (Humboldt 1836). Kawi showed heavy lexical borrowing from Sanskrit, but had a grammatical-syntactic system that was purely Malayan. To the basic question — is Kawi a Malayan o a Sanskritic (Indo-European) language, or a combination of the two? — Humboldt answered that the grammatical system of a language is what determines this regards. Hence Kawi is a Malayan language with Sanskritic borrowings, not the other way round, nor is it a mixed Sanskritic-Malayan language.

This looks like staying true to the Varronian line in more than one way: for like Varro, Humboldt ties the lexical-functional divide directly to the operations of the mind:


It was not for convenience, nor through imitation of the animals, that man chose speech to formulate and communicate his thought, but rather because speech is natural in him, both in terms of its organic production and its expressive value. […] It is thus dreaming to imagine an initial state in which man did not speak, followed by another state where he attained the use of speech. Man is naturally speaking, as he is naturally thinking […]. (Renan 1848: 90-92)


On De Saussure:

Thus the social and the unconscious — connected in ways that are never made explicit — take over many functions that the natural previously filled, in particular the imperviousness of language to individual will. ‘Natural’ may disappear as a word, perhaps even as a concept insofar as lexical signs are concerned, but it is never structurally eliminated from early structuralist thought. So the nature-other dyad survives the 19th century, this time as a division between the arbitrariness of the sign on the one hand, and its status as a ‘social fact’, an autonomous and ahistorical entity, on the other.


Every word that a man speaks is the result of a real, though doubtless unconscious, act of his free will. We are apt to speak of gradual changes in languages, as in institutions or anything else, as if they were the result of a physical law, acting upon beings who had no choice in the matter. Yet every change of the kind is simply the aggregate of various acts of the will on the part of all concerned. Every change in speech, every introduction of a new sound or a new word, was really the result of an act of the will of some one or other. The choice may have been unconscious; circumstances may have been such as practically to give him but one choice; still he did choose; he spoke in one way, when there was no physical hinderance to his speaking in another way, when there was no physical compulsion to speak at all […].

The study of men’s languages is strictly an historical study, a study of facts over which the will of man has a direct control.

(Freeman 1909, [1879]: p.237-238)

And yet if the most recent version of Chomsky’s theory is right, this Universal Grammar restricted to syntax and immediately connected to mind is not only minimal in size and scope, but its operation is not obviously linguistic (in Chomsky’s very restricted sense of ‘linguistic’), consisting as it does of devices like feature-checking which could find obvious analogues in general intelligence much more readily than language-specific features — i.e., morphology — might. Moreover, it means that nearly all of syntax is learned, which once upon a time was the behaviourist position so thoroughly demolished by Chomsky that no one any longer remembers the need to avoid moving toward it.


If one were in fact to conclude that such languages are wasteful of mental energy, would this not suggest that those who speak them are somewhat mentally degenerate? All right, perhaps this is going too far; but it is reasonable to say that the ground is laid for linguists who look at language death and survival from a Darwinian perspective to connect the world-wide spread of English to its ‘naturalness’ relative to the innate language faculty in the brain.


Chapter 5: Natural dialect and artificial language

The following quote anticipates the question of whether iconicity is really a property of the linguistic signs themselves or rather properties of how they are processed by individuals, an argument made more elaborately in Joseph (2015):

This seems to be an important reason for not praising one language and criticizing another, since all come from the same source and origin, the caprice of men; and all have been formed by the same reasoning faculty toward the same end, so that we might signify to one another the conceptions and understandings of the mind. It is true that over time, some languages, on account of having been more carefully regulated, have become richer than others, but this must be attributed not to any intrinsic quality of the languages, but only to the craft and industry of men. (Du Bellay 1914 [1549]: 41; my translation: JEJ)


To assert the mythical nature of standard languages is by no means to deny or downplay their force. Quite the contrary, myths are the key driving force of human cultures, which they effectively constitute, and are fully capable of overriding any physical or natural impulse or need a group of people may have.


Cultural belief in standard languages as a primary reality, rather than as some kind of abstraction derived from what people do, is extremely strong. In other words, people do not conceive of Standard English in the same way they conceive of, say, politics. Politics is a rather loose abstraction for whatever politicians do. People do not comment on or teach what is or is not politics, as they teach what is or is not Standard English. As a concept, Standard English is more akin to ‘the law’, which is not generally thought of as what lawyers do, but as a fixed institution with a traditional existence that determines what is right and wrong. Of course, most lawyers do not see the law this way, any more than most linguists see standard languages this way, but this fact hardly makes a dent in the general cultural conception of either.


They key point is that for any language standard to be established regarding word order, there must first be variation in the order, and the existence of variation means that the word order is ultimately arbitrary, even if we cannot resist seeking ‘natural’ explanations and accounts for the different orders and their frequencies.


Our processes of speaking and writing, that is, do not proceed word by word, but in larger ‘pre-packaged’ chunks. This observation has potentially important implications for how we imagine language being ‘stored’ in the brain. It has long been imagined in the form of a grammar and a lexicon being in our heads. This is wildly metaphorical — grammars and lexica are books, after all, and it is astonishing how thoroughly the metaphorical nature of their projection into our heads has been forgotten by some people. But the basic idea is that in one part of our brains is an inventory of words, understood as sound-meaning correspondences, and in another part are rules for putting the words together.


For linguists, there is no getting around the fact that their objects of study must be constructed, no matter how much or how little they try to idealize them. Ultimately, Chomsky’s I-language and E-language, the standard language and the literary language, are all constructs of the same type. Whether or not we ascribe a Platonic existence to them — as Chomsky does with I-language, as most other people do with E-language, and as most people except linguists do with the standard language), all we can know about them is what we can infer from actual instances of language production. Even Chomsky’s ‘native-speaker intuitions’ are actual instances of language productions, statements made about sentences which are themselves productions. This is, in effect, the conclusion of Plato’s Cratylus: however much we want to believe there is a real language beyond mere parole or performance, we cannot know it directly.


Chapter 6: Invisible hierarchies

Jakobson soon replaces the term ‘goal’ with ‘result’:

The relationship between a diachronic law and a synchronic law (cf. CLG 131) can be defined as the relationship between a means and a result obtained. We use the term result rather than goal not in order to deny that diachronic laws are tendencies, teleological in nature, but because in many cases te outcome does not coincide with the original problem; just as in other areas of human activity, especially collective ones, goals are not always attained. (1962 [1929]: 106; my translation: JEJ)

Jakobson further specified that the investigation of an historical change should be “limited to a linguistic system characterized by one and the same function, i.e., to linguistic entities which are functionally equivalent” (1962 [1928a]: 1).


There is no question that the exchange of letters with Trubetzkoy in 1930 marked the major turning point in Jakobson’s thought, from the kind of radically arbitrary structuralism called for by the CLG (which however entailed a cognitive dissonance with the CLG’s own treatment of diachrony, society, and relative motivation) toward a new kind of structuralism grounded in the nature of systems themselves — an idea whose metaphysical quality is at once its greatest attraction and shortcoming. It attracts those linguists who want to believe that they are unlocking the secret architecture of the human mind, which perhaps mirrors that of the universe, and repels those who regard unobservables as antithetical to science. Jakobson set out early to physicalize this nature of systems, locating it in the nature of the human mind and brain, as well as of the articulatory, auditory, and other perceptual faculties — but always keeping the metaphysical aspect on tap for its powerful rhetorical effect, as shown for example by the title of his most famous and important article on this subject, “Quest for the Essence of Language” (1966). Had he entitled it “Research into the Physical Basis of Language”, as he might well have done, the mystical attraction would have been lost, and with it perhaps Jakobson’s own fascination with the subject.


On one level, core-periphery is a philosophy of history masquerading as, or at least conflated with, a scientific model. What distinguishes the three components, UG, core grammar, and periphery, is how much of each is pre-determined by nature and how much is left open to historical circumstance, once again understanding ‘history’ in the sense of ‘things that happen as a consequence of (wilful) human activity’. UG is ahistorical, pure nature. Perophery contains the direct effects of history, modulated indirectly by some human parameters. Core grammar is nature having left some of her facets open to historical determination; history appears embedded here, but it is entirely under nature’s control. History is what introduces confusion into originally perfect structure: it is assumed that order comes from UG, from nature, and whatever is disorderly belongs to the periphery, the historical domain. This is in effect a version of the oldest dichotomy of language recorded in history, that between physis and nomos, nature and convention.


On ‘Natural’ Phonology, Morphology and Syntax:

Adapting the terminology created by Charles Sanders Peirce, Jakobson proposed that language structure (conceived on a purely formal level) is iconic of meaning. […] But in all such cases exceptions are bound to appear. […] One is obliged to ‘peripheralize’ such cases somehow; for the most part this has been done by classifying them as historically unstable. […] As in the French 17th century, what is arbitrary in language is associated with what is capricious, and it is understood that, although a linguistic community could in principle adopt any rule, no matter how capricious and ‘crazy’, over the long haul nature will separate the wheat from the chaff, and eliminate any element of the language that is not grounded in the principles of iconicity and markeness. In the most extreme formulation I know of, that of Shapiro (1983), the entire structure of language is configured as a network of markedness patterns, and exceptions are treated as ‘markedness reversals’, apparently as needed, without recourse to any principle that might limit them.


The challenge for OT is to escape this inherent circularity, and at least some of those who comprehend the problem are trying to do this through restricting the positing of constraints and grounding recognized constraints in psycholinguistic, if not neurolinguistic, ‘reality’. THe problem now, as it was in Plato’s time, is knowing what that reality is independently of language. And like Saussure, the adherents of OT want to be able to deal with the observable fact that human languages (let alone human speech) show such an extraordinary range of structural configurations as to suggest that the make-up of a language is fundamentally arbitrary, while at the same time to analyze it as a system so rigorous and autonomous as to demand to be approached from the point of view of limiting the arbitrary.



The alert reader, and especially the alert disgruntled reader, will have noticed that the second half of this book has in common with Plato’s Cratylus the fact that it asks more questions than it answers. The criticism will be levelled that it throws into question various versions of (what I take to be) linguistic naturalism without offering a clear alternative to put in it splace — unless that alternative is to do the history of linguistics rather than linguistics as such. That is not in fact the alternative I am suggesting, my belief being that the ‘history’ and ‘doing’ linguistics are, or ought to be, indistinguishable from one another. It would not therefore be possible to do one in place of the other.


But it is inconceivable to me that our conceptions of the natural are not at least to some degree the products of convention, the principal evidence being the same as that cited in the Cratylus: the fact that those conceptions differ among different peoples and different individuals, and that they change over time. And that is as true with regard to the physical universe as to human behaviour: again, there is no scientific truth that is not falsifiable, so that any ‘natural law’, no matter how seemingly grounded in the reality of the universe, will eventually be shown up as at least partly conventional, so long as science continues to pursue progress in the form of revised, and apparently deeper, understandings.



Emmorey, Karen. 2014. “Iconicity as structure mapping.” Philosophical Transactions of the Royal Society B: Biological Sciences 369 (1651): 20130301. doi:10.1098/rstb.2013.0301.

Freeman, Edward Augustus. 1909. “Race and language.” In Essays: English and American, edited by Charles William Eliot. Vol. 28. The Harvard Classics.

Joseph, John E. 2000. Limiting the arbitrary: linguistic naturalism and its opposites in Plato’s Cratylus and modern theories of language. Vol. 96. Studies in the History of the Language Sciences. Amsterdam: John Benjamins Publishing Company. doi:10.1075/sihols.96.

———. 2012. Saussure. Oxford: Oxford University Press.

———. 2015. “Iconicity in Saussure’s linguistic work, and why it does not contradict the arbitrariness of the sign.” Historiographia Linguistica 42 (1): 85–105. doi:10.1075/hl.42.1.05jos.

Renan, Ernest. 1848. De l’origine du langage. Paris: Michel Lévy.

Saussure, Ferdinand de. 1916. Cours de linguistique générale. New Haven: Yale University Press.