The aim of this paper is twofold. We first provide a reassessment of the voice systems of Kanakanavu and Saaroa, two Austronesian languages spoken in southern Taiwan), and in particular show that their voice forms that were previously identified as patient focus (patient voice or undergoer voice: patient), locative focus (locative voice or undergoer voice: locative), and beneficiary/instrument focus are actually nominalized forms. Our findings allow us to reconsider the position of Kanakanavu and Saaroa among the Formosan languages. We take as a point of departure Ross’s (2009) subgrouping hypothesis whereby Proto-Austronesian (PAn ) includes four primary offshoots: Tsou, Rukai, Puyuma, and all other Austronesian languages. This later subgroup, dubbed Nuclear Austronesian, is identified on the basis of the “nominalization-to-verb” innovation, whereby the PAn affixes *-en, *<in>, *-an, *Sa-/*Si-, which were only used in forming nominalizations, were expanded to encode verbal usage in Proto-Nuclear Austronesian. Under this hypothesis, Kanakanavu and Saaroa are both viewed as being Nuclear Austronesian languages. We try to map our findings along with Ross’s (2009) reconstruction, and in so doing we are led to place Kanakanavu and Saaroa higher up in Ross’s (2009) subgrouping tree and to propose a new hypothesis for the higher phylogeny of the Austronesian languages.