In this paper, we explore different ways to account for the peculiarities of depictive secondary predication in English, which we think can be characterized as longdistance modification. Other than with resultative secondary predication, the depictive and its target, typically the subject or direct object of a verbal phrase, do not form a contiguous constituent. Instead, the depictive attaches to the verbal domain that also embeds the target phrase. This sibling configuration, together with the constrained flexibility in choosing a target, obviously poses a challenge to the syntaxsemantics interface and we therefore compare three general LTAG approaches to deal with this. We eventually favor a rather semantic approach that also allows for a more principled view in terms of Van Valin’s ACTOR-UNDERGOER distinction. Our analysis predicts that only the verbal arguments which are the respective lowest and highest entries in the ACTORUNDERGOER hierarchy can act as targets for a depictive. To our knowledge, this is the first work that investigates the treatment of secondary predication within the TAG framework.