What makes a voice system? On the relationship between voice marking and case

International audience One of the major questions in Austronesian syntax concerns the relationship between voice marking, extraction, and case. Two common approaches have dominated previous literature. Either voice morphology marks extraction and case, or voice morphology feeds extraction and case. These positions are difficult to distinguish, because of the prominent one-to-one correspondence of voice, case, and extraction. In this paper, we bring new insights to this debate, with original data from Balinese and Dinka, a Nilotic language of South Sudan, which we show exhibits a familiar Austronesian-type voice system. We observe environments in these languages where the correspondence between voice and case and voice and extraction breaks down, in a manner that we argue provides evidence that voice marks extraction. Unlike in other extraction-marking languages, however, voice also affects case in Austronesian-type voice systems. We suggest that this is because extraction targets a case position. We account for the changes in case marking in the clause by suggesting that, when the external argument is not extracted, languages must employ alternative strategies to license it.