Origin and development of language

Robin Dunbar (Dunbar 1996) proposed the management of an exponentially growing network of social relationships as the source of evolutionary ‘pressure’ for a larger brain and as the nursery of language, but it has no explanation of its origin. For additional environmental context and a richer view of the important social variables, I recommend Chris Knight’s great opus Blood relations (Knight 1995) on the origins of humanity from early hominids. [A reasonable summary in this review]

The requisites for the invention and development of language are not that difficult and do not require some magical power of manipulating symbols with recursive rules (pace Noam). The reference here is Harris (1991:357-405).

  1. Ostensive denotation and onomatopoeia (including sound symbolism) make an obvious starting place. It is well established that many species associate distinct ‘cries’ with specific environmental variables, and these associations are collectively controlled. Hominids and early hominins may be presumed to have had at least an inventory of simple nouns and
  2. at least a few utterances exhorting action on the order of “come!”, “watch out!”, “flee!”. Even in pre-syntactic speech, these latter are first-order words which are only said if one or more of the zero-order words are also said (or intended and understood), e.g. someone flees from a hawk. From this the expansion of the inventory of both zero- and first-order words would follow individual and social needs, e.g. someone eats berries. With this, simple first-order syntax emerges, predication.
  3. Dunbar and Knight describe circumstances calling for second-order words which assert relationships, attitudes, commentary, and the like “about” the first-order predications, constituting second-order predication. (Note that no recursive rules are required for this.)
  4. Harris proposes that we are probably only at the early stages in the development of language, and identifies informational and other properties of language that might be augmented, properties of discourse and sublanguages, of metadiscourse and argumentation, and the language-like systems that result from modifying, limiting, or eliminating some of these properties, or adding explicit functions for indexing (in logic and mathematics), temporal coordination (in music and poetry), and so on.

The requisites for the origin of language amount to no more than the same processes of collective control that are well documented for language change. Reconstruction of the prehistory of language families is limited to about 6000 years (Callaghan 2014). The ultimate basis of this indeterminacy is in the nature of collective control, as for example the socially imposed segmentation of the continua of speech sounds and speech gestures into phonemes, with variability of reference values and perceived outputs both in individual and collective control. Reconstruction of an unwritten ancestor language is a projection from records of daughter languages, a projection that is more or less well informed by study of current language variability and change. Regularities in the differences between cognate words are interpreted as sound changes by which the languages diverged through time, and a common ancestral proto-language is reconstructed. The methodology has been validated many times, e.g. by reconstructing a Latin word from vocabulary found only in the descendant Romance languages and then digging up an inscription with that word in it. Reconstructed proto-languages may then be compared and cognate words found. But as yet earlier proto-languages are reconstructed from those comparisons, and those compared in turn, the indeterminacies mount up the farther back the projections go.

Also, as Harris remarked there is no a priori reason to suppose that language was originally invented only once and only in one place. Much as e.g. “person, persona” derives from a pre-Indo-European language, Etruscan, by way of Latin, we may have dim echoes of Denisovian or Neanderthal speech lurking in our everyday vocabulary, or even that of Homo Erectus. Language may have been invented on branches of the human tree that died out, yet leaving traces in the speech of other groups in contact and perhaps intermarriage.

We see some of these factors in children as they learn language. Children encounter their language(s) as an environmental fact like any other, but inherent in the nature of language is that we constantly recreate it as we use it and mutually calibrate our usage with one another, and children especially must invent it for themselves each time in order to participate.

I posted the following to the net a year or so ago as [Bruce Nevin 2018-12-03_14:36:33 UTC]. The podcast linked in the first paragraph below is fascinating, and is important, maybe essential, for fully following the sequel.

Interviews with researchers in this podcast suggest that language has a more fundamental and essential role in cognition than I had realized, stitching together what are disparate ‘islands’ of perception in animals, small children, and (a startling story!) in adults who lack language.

I did some more research into the story in the podcast about the role of 8 to 10 year olds in creating a new language de novo.

The story in brief: Prior to 1977, deaf children were with their separated families, and each developed idiosyncratic ‘home signs’ to communicate basic needs. Then the government mandated that all Nicaraguan children must learn to speak Spanish, regardless of their native tongue. In 1977, a special education center for congenitally deaf children was established in Managua, Nicaragua. They extended this policy to these children with no ‘tongue’, who had never learn any language, being deaf. From 50 chidren in 1977, the school population had grown to 100 in 1979, the year of the Sandinista revolution. In 1980, the Democratic Socialist government opened a vocational school for deaf adolescents, and the population in the two school had increased to over 400. Like Alexander Graham Bell, the teachers were sold on ‘oralism’ vs. ‘signing’ (two battling camps in the world of pedagogy for the deaf), but they had only very limited success teaching lip reading, and their students just didn’t get what Spanish was about.

However, they were not prohibited from signing among themselves, especially outside the classroom, and over time their disparate ‘home sign’ ways converged into a common way of signing.

“By combining gestures and elements of their home-sign systems, a pidgin-like form and a creole-like language rapidly emerged. They were creating their own language. This “first-stage” pidgin has been called Lenguaje de Signos Nicaragüense (LSN) and is still used by many who attended the school at this time.” (Wikipedia)

(You will presently see the significance in this story that those early attenders still use this primitive pidgin 40 years later, despite exposure to later developments.)

The staff couldn’t understand what the children were saying, but they didn’t ask for help until 1986, when they called on a sign language linguist from MIT, Judy Kegl.

By that time, there had been nine ‘generations’ of incoming students since the 50 in 1977, six ‘generations’ since the 400 in 1980. The annual arrival of newly-exposed children is crucial to this story, especially those under age 10.

Other linguists after Dr. Kegl have got involved. References that I’ve listed at the end of this post describe how they have observed the refinement and sophistication of this pidgin sign language progressively with each new generation of children. In brief:

  • Signing gestures at first were close to the miming done by congenitally deaf adults who lack language (another part of the podcast). These relatively large and expressive/depictive gestures became more compact, stylized, and conventionalized to signs with each generation.
  • Children under about age 10 innovate as they learn the sign language of the community that they move into.
  • The change that is the focus of the referenced studies is the emergence of spatial orientation as a modifier of signs. This is a well-known feature of ASL and other sign languages that have been created by people with prior knowledge of language. Signing before the chest has straightforward (so to speak) meaning, and the same sign to the right or to the left denotes a modification of the basic sign.
  • Orientation is used to indicate co-reference of two signs to the same individual or to the same temporal context. Hereby hangs the central entry point of my particular interest in this.
  • This side channel for information is overloaded, in that there is no overt way to distinguish e.g. same person from same time, or same person from same utensil that the person was using for eating.
  • This is because this metalanguage device–the means for signing about signs themselves ( this present sign has the same referent as that prior sign)–is not part of the signing system itself. The metalanguage is separate from the language. And it is because it lacks ‘words’ (signs), but rather is only a modifier of signs, its capacity as a metalanguage is impoverished.
  • Worse, this side channel for information is also overloaded for another reason. It is used for purely expressive purposes, e.g. turning rhythmically from side to side in the progress of a narrative.

In natural language, the metalanguage is part of the language. We can use words to refer to words and to assert co-reference and the like. For example, I could say any of the following, in progressively more explicit form:

I read the paper by Kocab et al.

I read the paper which is by Kocab et al.

I read a paper; it is by Kocab et al.

I read a paper; a paper (previous word same as a prior word) is by Kocab et al.

The last paraphrase is unnatural because previous same as prior is asserted explicitly in words, rather than the immediately previous occurrence of paper being reduced to the pronoun it (which carries the metalanguage information ‘same as something said nearby’), or being reduced to the - ich part of which (which carries the metalanguage information ‘same as as a closely preceding word’), plus (by a longer route) the definite article in the paper . Even the which can be reduced to zero because to do so is conventional and the metalanguage assertion of same reference is understood because the remaining juxtaposition of paper and by does not otherwise admit of any predication there.

It may be that these signers cannot recite “This is the house that Jack built” because there are too many distinct same-reference links in the chain. The song “I am my own grandpa” may be inaccessible to them unless they have writing as a bridge to the metalanguage capacities of spoken language.

Now the particular personal interest that this story sparked in me. In 1994, Terry Langendoen’s review of Zellig Harris’s Language and information: A mathematical approach was published in the journal of the Linguistic Society of America (Langendoen 1994). At the end, breaking abruptly from the general tenor and flow of his review, he wrote:

… the concluding paragraph of section 10.7, ‘Non-linguistic systems; music’, says something so outrageous that I am compelled to quote it in its entirety (318):

'Finally, it seems that the sign language of the deaf does not have an explicit operator-argument partial ordering, nor an internal metalanguage, but rests upon a direct juxtaposition of the relevant referents. This applies to autonomous sign languages, developed by the deaf without instruction from people who know spoken language.

Lest there be any doubt about the implications of this paragraph, by ‘internal metalanguage’ Harris means the sentences which constitute the grammar of the language (359).

Harris was clearly referring to an endogenous sign language such as that in Nicaragua. The discussion so far has been about the absence of a metalanguage that is part of the language itself. The further part about an explicit operator-argument partial ordering suggests to me that in this sign language they couldn’t have a “he said-she said” conversation, or to remark that they once thought thus and so but because of such and such have concluded that so and so did it. The sign language would have to have acquired the capacity to have one verb assert something about another (“an explicit operator-argument partial ordering”), probably, but maybe not provably, from their instruction in Spanish language literacy.

That said, the signer can introduce metadiscourse and perhaps metalanguage assertions by the spatial orientation of the sign gestures relative to the sagittal axis of the body and relative to one another in the signing space in front of the body, as is done also in ASL. See Senghas & Coppola 2001), among the following resources in which you an begin to follow up the Nicaraguan case:

Senghas, A., & Coppola, M. (2001). Children Creating Language: How Nicaraguan Sign Language Acquired a Spatial Grammar. Psychological Science , 12(4), 323–328. doi: 10.1111/1467-9280.00359

There’s a paywall at the site that the DOI links to; it’s posted here:


That was 2001. More recent:

Senghas, Ann. 2011. The Emergence of Two Functions for Spatial Devices in Nicaraguan Sign Language Hum Dev. 2011 Jan; 53(5): 287–302. doi: [10.1159/000321455]

Kocab, Annemarie, Pyers, Jennie, & Senghas, Ann. (2015). Referential shift in Nicaraguan Sign Language: a transition from lexical to spatial devices. Front. Psychol. 2014; 5: 1540. Published online Jan 9. doi: [10.3389/fpsyg.2014.01540]

A general discussion, like the above linked Wikipedia article:


Callaghan, Catherine. 2014. Proto Utian grammar and dictionary, with notes on Yokuts. Series:Trends in Linguistics. Berlin: De Gruyter Mouton.

Dunbar, Robin. 1996. Grooming, gossip, and the evolution of language. London: Faber and Faber.

Harris, Zellig S. 1991. A theory of language and information: A mathematical approach. Oxford: Oxford U. Press/Clarendon.

Knight, Chris. 1995. Blood relations: Menstruation and the origins of Culture . London & New Haven: Yale University Press.

Langendoen, D. T. 1994. Review of A Theory of Language and Information: A Mathematical Approach by Zellig Harris. Language , Vol. 70, No. 3 (Sep., 1994), pp. 585-588.

Thanks for this. The trend of thought in the early part of this strongly suggests to me that the development of language from Harris’s “simple sound inventories” of non-linguistic animals to the complexity of language that includes mathematical notation is one of autocatalytic creativity. Autocatalyitc creativity is a process that has been hypothesized (and I agree) to be a process at the heart of the origin and increasingly complex evolution of life (S. Kauffman (1995) At Home in the Universe, Oxford UP) , and in the punctate evolution of the forms of community development from roving bands into vast conurbations with global interconnections, not to mention the several “Industrial Revolutions” including the ongoing technological one…

A fundamental characteristic of language is a certain kind of sequence control referred to as dependency. To understand an adjective such as red there must also be understood a noun of which it is predicated (red pencil). Some words have no such dependency (pencil, kelp, vacuum); some require one of these zero-order words (red, fall), or two (eat, hit), or three (give … to, steal … from). Some words require one of these first-order words (false, true, easy); or two (cause, imply); or a combination of one of each (believe, surprise). The dependent words are adjacent only in very simple sentences. For reasons that I’ll lightly touch on below, other words may intervene (in lawful ways that permit reconstruction of the word dependencies). In the following overview note and paper, these are called “non-adjacent dependencies”.

Cognitive elements of language have existed for 40 Million years

Stuart K. Watson, Judith M. Burkart, Steven J. Schapiro, Susan P. Lambeth, Jutta L. Mueller and Simon W. Townsend. Non-adjacent dependency processing in monkeys, apes and humans. Science Advances, 21. Oktober 2020. DOI: 10.1126/sciadv.abb0725

The authors adduce evidence in primates and other species for interrupted sequence control of the sort that is necessary for language, and conclude that this cognitive prerequisite for language originated millennia earlier than did language itself.

When this is understood as a capacity to hold an interrupted sequence in memory and then resume it after an interruption it may seem less remarkable than when it is described as “non-adjacent dependency”.

However, in language, the interruptions are not arbitrary intrusions outside of language, they are intervening words with their own dependencies, and those dependencies are not arbitrary either. On the one hand, they may be shared (in That you kept the red pencil surprised me both keep and red are dependent upon pencil); on the other hand, they typically are reduced forms of words and word dependencies that occurred earlier in the same discourse, which are thereby reaffirmed in a new context. This word repetition is what distinguishes coherent discourse from a random sequence of sentences.

The researchers are correct, that ability to control a sequence (a dependency) that is interrupted is prerequisite to these properties of language.

I hope this isn’t too far off topic, but I have been wondering about the language of corvids (crow, ravens, etc), particularly since a recent article in, i think, Science, showed that there are more parallels between a bird’s brain and our own than has been usually thought.

The initial experience, however, happened in early summer of 2019, when my wife and I followed a nearly empty (of people) boardwalk loop in the rain forest near Tofino (west coast Vancouver Island). We walk slowly, and when we were about halfway round the loop, crows (ravens, or whatever they were) began to chatter noisily apparently in dialogue with another group some distance away. As we continued on our way, the other group got slowly closer, and it soon became clear that they were following another group of walkers who were much faster than us, and who eventually passed us on the trail.

What struck me, having something of a background in psycholinguistics, was that the chatter was far from random, to the extent that apart from the gross spectral form, I would have sworn that the groups, or single birds within a group, were humans discussing with each other in a foreign language the people they were following. I mean the short and long-term rhythms and the prosody (pitch and loudness variations) were just like human dialogue.

I asked two years ago whether anyone had studied the vocal interactions of corvids, and whether anything was known of their language(s), but got no response. It seems to me that the advances in pattern recognition (sometimes called Artificial Intelligence) could make some start into the problem, if it hasn’t been done already.

If I am right about the apparent parallel between human and corvid languages, either there is something about natural constraints to the way language develops, or language really does go back a long, long, way in evolutionary terms. The absence of similar forms in our closest relatives argues against the latter. I’m wondering whether an AI analysis of the sounds of human, corvid, and porpoise dialogue would find the kinds of similarities that I thought I heard on that boardwalk.

I’ll look again when I get a chance. Fascinating prospect. In many cultures there are accounts of animal talk. This verges with hunting lore and woodscraft on the one hand and with the lore of signs and omens on the other, but whatever the interpretive story indigenous folk are close observers.

A quick search turns these up:

The last has some linked citations.