We provide novel data from Asante Twi (AT, Kwa) which show that its pattern of argument encoding under argument extraction is more complex than previously described (Saah 1994, Korsah 2017, Korsah & Murphy 2019): (a) One and the same type of extraction may result in both a gap or a resumptive pronoun (RP) (/pace /claims for AT that extraction of NP-arguments always leaves behind an RP); (b) the choice between RP/gap is determined by the referentiality of the extracted XP (novel observation). Thus, the choice between RP/gap is not only determined by extraction type (e.g. movement vs. topicalization) or the properties of the extraction site (inside an island or not), but also by the properties of the extractee. Moreover, AT exhibits a preference to use an RP over a gap if possible - a pattern that is in conflict with economy constraints such as AvoidPronoun. We propose an analysis of the referentiality-driven RP/gap-choice  that is based on a structural difference between ref./non-ref. XPs and partial copy deletion. If you have any questions or comments, please don't hesitate to contact either of the authors or just leave a comment in the project. Unfortunately, we won't be able to provide realtime interaction, but we will check in on the comments every day and try to answer questions and respond to comments.