This thesis is a cross-linguistic investigation into the nature of null arguments in radical pro-drop languages where null arguments are claimed to be derivable via the process called argument ellipsis, which directly elides arguments. I examine whether null arguments in such languages involve internal structure by testing extraction possibilities out of them on the basis of the hypothesis that the possibility of extraction indicates the presence of internal structure in anaphora sites. Based on novel data from Chinese, Japanese, Korean, Mongolian, and Turkish, I show that elliptic arguments exhibit a hitherto unnoticed pattern of extraction out of ellipsis sites, a pattern that does not follow from either Hankamer and Sag’s (1976) surface anaphora such as VP-ellipsis or deep anaphora such as do it anaphora. Specifically, I show that null arguments in the languages in question uniformly disallow overt extraction out of them, while they uniformly allow covert extraction (more precisely, extraction that does not affect word order). Having established the overt/covert extraction asymmetry out of null arguments under consideration, I show that the extraction pattern in question has several consequences for the PF-deletion versus LF-copying debate in the literature on ellipsis. Specifically, taking the possibility of overt extraction out of anaphora sites as an indication of the presence of internal structure in overt syntax and the possibility of covert extraction as an indication of the presence of internal structure in covert syntax/LF, I argue that the relevant extraction asymmetry out of null arguments in question is best analyzed under the LF-copy analysis, which provides ellipsis domains with internal structure only in covert syntax/LF so that only covert movement is possible out of the relevant domains. I argue that PF-deletion is also an available strategy and propose that the dichotomy between PF-deletion and LF-copying is related to the phasal status of ellipsis domains, building on Bošković’s (2014) phase-based analysis of ellipsis, where only phases and phasal complements can undergo ellipsis. In particular, I claim that ellipsis of phases, including argument ellipsis, is implemented by LF-copying and ellipsis of phasal complements is implemented by PF-deletion, which is shown to be a by-product of the phase theory. Finally, I show that the LF-copy analysis of elliptic arguments has consequences for the proper analysis of a number of phenomena, including wh-in-situ, control, and the timing of null operator movement.