[8] Using computational phylogenetics, Bouckaert, Bowern & Atkinson (2018) posit a mid-Holocene expansion of Pama–Nyungan from the Gulf Plains of northeastern Australia.

Pama–Nyungan languages generally share several broad phonotactic constraints: single-consonant onsets, a lack of fricatives, and a prohibition against liquids (laterals and rhotics) beginning words.

An exception is Kala Lagaw Ya, which acquired both fricatives and a voicing contrast in them and in its plosives from contact with Papuan languages.

In his 1980 attempt to reconstruct Proto-Australian, R. M. W. Dixon reported that he was unable to find anything that reliably set Pama–Nyungan apart as a valid genetic group.

Part of Dixon's objections to the Pama–Nyungan family classification is the lack of obvious binary branching points which are implicitly or explicitly entailed by his model.

However, the papers in Bowern & Koch (2004) demonstrate about ten traditional groups, including Pama–Nyungan, and its sub-branches such as Arandic, using the comparative method.

In his last published paper from the same collection, Ken Hale describes Dixon's scepticism as an erroneous phylogenetic assessment which is "so bizarrely faulted, and such an insult to the eminently successful practitioners of Comparative Method Linguistics in Australia, that it positively demands a decisive riposte.

"[12] In the same work Hale provides unique pronominal and grammatical evidence (with suppletion) as well as more than fifty basic-vocabulary cognates (showing regular sound correspondences) between the Proto-Northern-and-Middle Pamic (pNMP) family of the Cape York Peninsula on the Australian northeast coast and Proto-Ngayarta of the Australian west coast, some 3,000 km apart (as well as from many other languages), to support the Pama–Nyungan grouping, whose age he compares to that of Proto-Indo-European.

Third, and related to this, we do not have to assume in this model that there has been intensive diffusion of many linguistic elements that in other parts of the world are resistant to borrowing (such as shared irregularities).Additional methods of computational phylogenetics employed by Bowern and Atkinson[10] uncovered that there were more binary-branching characteristics than initially thought.