Proto-Australian

TL;DR

• Australia has some 250 languages at contact, and for fifty years linguists argued about whether they could be traced back to one ancestor. The best-known sceptic, R. M. W. Dixon, said they could not.
• Mark Harvey and Robert Mailhammer have now done it. Their 2024 book rebuilds Proto-Australian with the ordinary comparative method: a consonant system with five places for nasals, a first person *ŋaj, five noun classes, an aspect-based verb and a core vocabulary that includes *maɻ “hand”, *tir “tooth” and *pukari “dream”.
• The family has 39 branches. Pama-Nyungan, which covers nine-tenths of the continent, is one of them. The other 38 are crowded into the north, and that is where the language was born.

The question that we asked was did all of those languages come from a common source. — Robert Mailhammer, 2018

Can Australia’s languages have one ancestor? #

Draw the languages of Australia on a map and the continent splits in two. Across nine-tenths of it, from Cape York to Perth and down to Victoria, lie the Pama-Nyungan languages: Warlpiri, Arrernte, Pitjantjatjara, Gamilaraay, Yolŋu Matha and most of the rest. They are as clearly one family as Germanic or Slavic. Their ancestor, Proto-Pama-Nyungan, was probably spoken around the Gulf of Carpentaria something under 6,000 years ago (see The Cradle of Pama-Nyungan).

The other tenth is a different world. Across the Top End and the Kimberley, in a band a few hundred kilometres deep, sit some twenty small families and a scatter of isolates: Gunwinyguan, Iwaidjan, Worrorran, Nyulnyulan, Mirndi, Tangkic, and single languages such as Gaagudju, Laragia and Wageman. Many of them build words the opposite way to Pama-Nyungan. They stack prefixes onto the verb, so a single word says who did what to whom. Pama-Nyungan languages mostly use suffixes.

Is the north a collection of separate families, or the old branches of the tree that Pama-Nyungan grew from? Dixon, the field’s great synthesiser, went back and forth. In 1980 he was confident that all but two or three of Australia’s languages belonged to a single family. By 2002 he had abandoned the family tree altogether. In his new view, Australian languages had diffused into one another for so long that the comparative method could not see through them, and Pama-Nyungan itself was no family. Most of his colleagues disagreed about Pama-Nyungan, but no one had shown the deeper tree either. In 2017 Claire Bowern, reviewing Australia’s isolates, judged that the time had come for a new assessment of Proto-Australian.

That same year Harvey and Mailhammer published the first instalment of one.

How do you rebuild a language nobody wrote down? #

You compare its daughters. The method is the one that rebuilt Proto-Indo-European: find words of similar meaning across the daughter languages, and look for sound differences that recur regularly. If Latin p keeps answering to English f (pater, father; piscis, fish; pes, foot), the words are inherited, and you can work back to the sound that produced both.

Australia makes this hard in two ways. Its languages share a sound system so uniform that chance resemblances are everywhere. And they borrow words heavily, especially through name taboos, which retire any word that sounds like the name of someone who has died. Dixon’s scepticism rested on both.

Harvey and Mailhammer’s 2017 paper in Diachronica answered with grammar rather than vocabulary. They took the noun-class markers of the northern languages, the little prefixes that sort nouns into kinds. In 36 of the 67 well-recorded northern languages the prefixes matched in form and meaning, and Yanyuwa, a Pama-Nyungan language, kept a full set. They then set out two tests that let a small set of matches carry weight:

• Where the matches fall. Matches concentrated in one region weigh against chance. Matches scattered across regions that never touched weigh against borrowing.
• How irregular they are. Languages do not borrow whole paradigms complete with their quirks. When the same odd gaps and irregularities turn up in related paradigms of languages far apart, they were inherited.

Their conclusion was blunt: “we show that inheritance is favoured over chance and contact.” The matches, they added, are “concentrated in only 10% of the continent”, which is not what chance produces. They claimed only “a prima facie case” for a proto-language “ancestral to 36 NPN languages and the PN family”, and left the rest of the grammar for later. The 2024 book, Proto-Australian: Reconstruction of a Common Ancestor Language, carries the method through the whole grammar. It has chapters on sounds, nouns, verbs and syntax, a statistical test of cognacy, and appendices of cognate sets and paradigms for each branch.

What did Proto-Australian sound like? #

Much like an Australian language today, which is itself a finding. Here is the consonant system they reconstruct:

Read across the nasal row and you have the Australian signature. It has a nasal at every place where it has a stop, five in all, including the ŋ of English sing, which in Australia can begin a word. Speakers made the tip-of-tongue sounds two ways, flat and curled back, so t and ʈ, n and ɳ, l and ɭ were different sounds. The stop made with the blade of the tongue was a single sound that daughters split two ways, so the authors write it two ways, t̪ and c. There are no fricatives at all: no s, f or h. And there is no voicing contrast, so p and b are the same sound.

The vowels are the surprise: five, i e a o u, where most Australian languages have three. The other difference from Pama-Nyungan is the shape of words. Proto-Australian allowed words of one syllable ending in a consonant, such as *maɻ “hand”, *tir “tooth” and *maj “food”. Proto-Pama-Nyungan did not. Every word needed two syllables, so it added a vowel: Barry Alpher’s Proto-Pama-Nyungan has *maɻa “hand” and *maji “food”. That small rule says something large. The forms of the north are the old ones, and Pama-Nyungan, for all its size, is the innovator.

How did it say “I”? #

As *ŋaj. The free pronoun “I” has descendants almost everywhere: Warlpiri ngaju, Gamilaraay ngaya, Burarra ngay-pa, Proto-Worrorran *ŋaju, Proto-Pama-Nyungan *ŋaju. The oblique form, used for “me” and “my”, was *ŋar. In the ASJP word lists, about four Australian languages in five still begin “I” with ŋ. Worldwide the figure is about one in thirteen. Whether that ŋ reaches beyond Australia is the subject of The First Person of Sahul.

The free pronouns were for emphasis. The everyday work was done by a second set, short pronouns that leaned on the first word of the clause. Harvey and Mailhammer call this the clitic complex. It was obligatory in every finite clause and sat first or second in the sentence, the way the auxiliary does in Warlpiri today. Into it went who was involved, how many there were and what kind of thing they were:

The two columns show that Proto-Australian was ergative. It marked the doer of a transitive act differently from the subject of an intransitive one, so “I” in “I sleep” is not the “I” of “I spear the wallaby”. The plural forms were irregular, and they had to be doubled by a general plural marker *ra=.

This is the book’s boldest structural claim. Over time the clitic complex slid onto the verb and fused there, and the result is the prefixes of the north. The Gunwinyguan or Mirndi verb that packs subject and object into its front is the Proto-Australian clitic complex glued in place. In much of Pama-Nyungan it never fused. Warlpiri still carries it as a separate little word in second place, and many Pama-Nyungan languages attach a similar string of clitics to the first word of the clause. The two great types of Australian language are two fates of one system: glued to the verb in the north, left loose or worn away in the south. The linguist Brett Baker, working on the northern verbs, takes the proposal “that the NPN prefixing systems developed from an earlier clitic system of the kind found in Warlpiri” as his starting point. He also names its sharpest problem: some fused northern forms that combine subject and object look older than the theory allows (Baker 2025, pp. 378–381).

What kinds of things did it sort? #

Five. Proto-Australian had noun classes, as many northern languages still do, and the markers were these (the same five shapes in the 2017 paper and the 2024 book):

Three animate classes, two inanimate. The vegetation class is the one to watch. Harvey and Mailhammer insist that it covered all plants and plant products, not just vegetable food. Its ma- survives in 27 northern languages, and in 25 of them it still marks plants, vegetable food and things made from plants. It covered “plant products (e.g. spears etc.)”, in the 2017 paper’s own example. In the Kimberley’s Worrorran languages it also names kinds of country, the way English speakers say “pandanus country”. A speaker sorting the world this way put yams, spears, digging sticks and bark under one heading.

The book also revises the 2017 paper. In 2017 these were agreement prefixes, carried by adjectives and demonstratives but not by nouns. Now they are optional clitics that grew out of generic nouns, words like “man”, “woman”, “meat” and “tree” put before a specific noun to say what kind of thing it was. Many Australian languages still do this, saying “meat kangaroo” or “tree spear”, and in the north the generic shrank into a prefix.

The nouns took case suffixes much as Pama-Nyungan nouns do: ergative *-ji, dative *-ku and genitive *-kana.

What did its verbs do? #

They tracked how an event unfolded more than when it happened. The verb root took a short suffix that marked aspect: a completed past (*-m, *-ɲ, *-j, *-ŋ), an ongoing state or present (*-n, *-ɭ, *-ŋ), and a bare root for commands. Verbs fell into conjugations by the consonant they took. A small set of these inflecting verbs did the work of many, joined by uninflected coverbs marked *-j. That is still the pattern of the Kimberley and the Victoria River, where a few dozen inflecting verbs carry hundreds of coverbs. Over time the system drifted toward tense, and the daughters remade their futures independently, so no single future can be rebuilt.

What words survive? #

The cognate appendix runs to many pages. About fifty of its reconstructions circulate in public sources; here is a selection, with Barry Alpher’s Proto-Pama-Nyungan form (written here in the same phonetic letters) and a northern descendant where one is on record.

Most of this is the body and the hearth, as basic vocabulary should be. One entry is not. Proto-Australian had a word for “dream”, *pukari. Across much of the continent its descendants still mean just that: Yolŋu bukaway in north-east Arnhem Land, Karajarri pukarri south of Broome, Yandruwandha pukuru in the Lake Eyre basin (Bowern and Atkinson 2012 data). In the West Kimberley the same word names the Dreaming. In Yawuru, bukarri is “dream” and bukarri-garra, literally “many dreams”, is the Dreamtime (Hosokawa 1991, pp. 29, 57). Up the Fitzroy in Nyikina, bugarrarra is “the Dreamtime”, also glossed “the Beginning” (Stokes 1982). Stokes and McGregor reconstruct the Nyulnyulan ancestor as *bukarri, “dream, dreamtime” (2003, p. 62). The Dreaming sense belongs to this one corner of the continent; elsewhere the word means a dream and nothing more, and the Dreaming goes by other names, such as Yolŋu wangarr. Bowern’s team codes the Kimberley word as a loan into Nyulnyulan from its Pama-Nyungan neighbours (Hunter-Gatherer Language Database). Inherited or borrowed, the word the Kimberley uses for the time when the world was laid down and the law was made is Proto-Australian’s word for dreams.

Who are its children? #

Thirty-nine branches. One is Pama-Nyungan. The other 38 are northern: families such as Gunwinyguan, Iwaidjan, Worrorran, Nyulnyulan, Mirndi, Garrwan and Tangkic, and no fewer than 24 isolates, single languages with no close relatives. Each was once as distinct a line as Pama-Nyungan. The others simply did not spread.

Download the data (CSV)

Three languages are left out. The languages of Tasmania were cut off when Bass Strait flooded, and nothing in their scanty records connects them to the mainland. Tiwi, on Bathurst and Melville Islands off Darwin, is an isolate that does not belong. Neither does Meriam Mir, of the eastern Torres Strait, which is a Papuan language of the Trans-Fly.

Then there is the question of where the family was born. The rule of thumb in historical linguistics is that a family’s homeland is where its branches are most varied, because that is where it has had the longest to split. Harvey and Mailhammer put it plainly: “The area with the most isolates has the greatest diversity. Within Australia, that area is the Top End.” Therefore, “the Top End is the most likely homeland for Proto-Australian” (Harvey and Mailhammer 2025). Nick Evans made the same observation about the northern families two decades earlier. The far north holds “perhaps 90% of Australia’s linguo-genetic diversity in an eighth of its land area” (Evans 2003, p. 1).

What does the book leave open? #

How old the language is, and how it spread. Harvey and Mailhammer’s best guess is that Proto-Australian “was most likely spoken about 6,000 years ago in the Top End”. The figure comes from Tiwi. The Tiwi Islands were cut off by the sea about 6,000 years ago, and on the book’s classification Tiwi is not a Proto-Australian language, so the family seems to have spread after the cut. That dates the spread, not the language, which could have been spoken for a long time before its daughters left home. The comparative method has no reliable clock to say how long.

The spread itself is a puzzle. A language that “displaced all the other languages spoken before” did so, they admit, with “no clear evidence for population movement or economic and technological change in the past 10,000 years”. Their own verdict: “we have a good question, but no good answers, which means our models need revision.” The book itself has yet to be reviewed in a journal, and the field has not had its say.

Proto-Australian’s “I”, *ŋaj, also turns up across the water in New Guinea, on the coast that faced the land bridge before the sea drowned it. What that might say about where and when the language was spoken is the subject of Proto-Australian and the Drowned North.

Sources #

1. Harvey, Mark, and Robert Mailhammer. “Reconstructing remote relationships: Proto-Australian noun class prefixation.” Diachronica 34, no. 4 (2017): 470–515.

2. Harvey, Mark, and Robert Mailhammer. Proto-Australian: Reconstruction of a Common Ancestor Language. Studies in Language Change 24. Berlin: De Gruyter Mouton, 2024.

3. Harvey, Mark, and Robert Mailhammer. “New evidence confirms our Indigenous languages have a common source, but how they spread remains a mystery.” The Conversation, 29 January 2025.

4. “Indigenous language comes from a single root tongue.” ABC News, 28 March 2018.

5. Harvey, Mark. “Reconstruction of pronominals among the non-Pama-Nyungan languages and its implications for genetic subgrouping in northern Australia.” In The Non-Pama-Nyungan Languages of Northern Australia, edited by Nicholas Evans, 475–513. Pacific Linguistics 552. Canberra: ANU, 2003.

6. Baker, Brett. “The path from cliticised to prefixed person marking in proto-Australian.” In Projecting Voices: Studies in Language and Linguistics in Honour of Jane Simpson. Canberra: ANU Press, 2025.

7. Dixon, R. M. W. The Languages of Australia. Cambridge: Cambridge University Press, 1980.

8. Dixon, R. M. W. Australian Languages: Their Nature and Development. Cambridge: Cambridge University Press, 2002.

9. Bowern, Claire. “Language Isolates of Australia.” In Language Isolates, edited by Lyle Campbell, 424–447. London: Routledge, 2017.

10. Bouckaert, Remco R., Claire Bowern, and Quentin D. Atkinson. “The origin and expansion of Pama-Nyungan languages across Australia.” Nature Ecology & Evolution 2 (2018): 741–749.

11. Alpher, Barry. “Pama-Nyungan: phonological reconstruction and status as a phylogenetic group.” In Australian Languages: Classification and the Comparative Method, edited by Claire Bowern and Harold Koch, 93–126. Amsterdam: John Benjamins, 2004.

12. Evans, Nicholas. “Introduction: Comparative non-Pama-Nyungan and Australian historical linguistics.” In The Non-Pama-Nyungan Languages of Northern Australia, 3–25. Pacific Linguistics 552, 2003.

13. Hosokawa, Komei. The Yawuru Language of West Kimberley: A Meaning-Based Description. PhD thesis, Australian National University, 1991.

14. Stokes, Bronwyn. A Description of Nyigina, a Language of the West Kimberley. PhD thesis, Australian National University, 1982.

15. McGregor, William B., and Alan Rumsey. Worrorran Revisited. Pacific Linguistics 600. Canberra: ANU, 2009.

16. Green, Rebecca. A Sketch Grammar of Burarra. BA honours thesis, Australian National University, 1987.

17. Stokes, Bronwyn, and William B. McGregor. “Classification and Subclassification of the Nyulnyulan Languages.” In The Non-Pama-Nyungan Languages of Northern Australia, edited by Nicholas Evans, 29–74. Pacific Linguistics 552, 2003.

18. Bowern, Claire, and Quentin Atkinson. “Computational Phylogenetics and the Internal Structure of Pama-Nyungan.” Language 88 (2012): 817–845. Data: lexibank/bowernpny.

19. Bowern, Claire, et al. Hunter-Gatherer Language Database, feature “Dreaming”.