The performance of the Automated Similarity Judgment Program (ASJP) method of

language classification has been tested quantitatively across the world's language

families, as well as through more detailed, qualitative inspections of ASJP trees,

comparing them with classifications of individual families by experts. Different

quantitative performance evaluations all point to a relatively poor overall performance in

the case of Austronesian. In order to investigate why Austronesian appears to be so

recalcitrant, we identify the individual Austronesian language groups that are responsible

for the discrepancies between ASJP and expert classifications-the 'black sheep' of the

family-using a simple technique called jackknifing. It turns out that many of the

languages which induce a poor fit between the expert and ASJP classifications belong to

subgroups of Austronesian that are problematic in various ways. Thus, inaccuracies in the

experts' classification of Austronesian must, at least partly, be responsible for the added

amount of error in the ASJP classification when it comes to Austronesian.