I began the preface to The languages of Australia (LoA, published 1980) by stating that it was, in several ways, premature. By this I meant that more descriptions of languages would be forthcoming during the 1980s and 1990s (as, indeed, they have been), which would provide a surer basis for generalisation. I now realise that LoA was most importantly of all, conceptually premature.

I had learnt the principles of historical linguistics from my teachers Warren Cowgill and Calvert Watkins, and from reading Meillet, Benveniste and others. And I had assumed that the methodology which applies so well for the languages of Europe and North America and Oceania would also be appropriate for the linguistic situation in Australia. It is not, but it took me a long time to realise this.

I sometimes wondered whether my lack of success in applying the established methodology of historical linguistics to the Australian linguistic situation was a feature of that situation, or a reflection on my abilities. Then, in the 1990s, I did intensive field work on Jarawara, spoken in southern Amazonia, and undertook a comparative study of the six languages of the Arawá family, to which it belongs. I found that here the established methodology worked perfectly (it was like a dream, after my struggles with the Australian situation). I was able to establish correspondence sets, compare their distributions, and then to reconstruct the phoneme system, more than four hundred lexemes, quite a bit of morphology, and some of the syntax for proto-Arawá.