In 1965 Isidore Dyen published a lexicostatistical classification of some 245 Austronesian languages ? the most ambitious undertaking of its kind ever attempted. The purpose of this classification was to specify, in accordance with an implicit family tree model of linguistic differen tiation, the hierarchy of pairings between all languages and language groups in the Austronesian family. By definition the structure of the resulting dendrogram was taken to mirror the history of language splits from the break-up of the most remotely reconstructible ancestor (Proto Austronesian) to the attested languages. The problem confronted was thus the traditional problem of subgrouping in comparative linguistics. The method employed, however, was one only recently developed by Swadesh (1950, 1955), and this study inevitably became an important test not only of the correctness of received views on Austronesian sub grouping, but of the reliability of lexicostatistics itself. Dyen's results were strikingly at variance with the traditional view that Austronesian speakers originated on the Asian mainland and gradually expanded through island southeast Asia into the farthest reaches of the Pacific. To account for the reported distribution of lexico statistical percentages ? where more than 30 prima facie primary branches of the Austronesian family were represented in Melanesia ? Dyen concluded instead that the Austronesian expansion had begun in the area of New Guinea and the Bismarck archipelago. Some Austrone sian speakers remained in this area over the succeeding millennia, he maintained, while one descendant group (Proto-Malayo-Polynesian) set out on great voyages of discovery both to east and to west. Almost immediately (Grace 1966) it was recognized that Dyen's conclusions not only clashed with what had long been supposed on the basis of physical and cultural evidence, but also clashed with linguistic evidence of a more traditional and better-established kind. The result has been that over the past decade and a half researchers in the field
Stored only when both papers are in the corpus — 78 citations exist upstream.