Since their discovery by Ross (1967), island constraints have attracted a great deal of research interest. Nonetheless, they are only poorly understood and a lot of questions remain to be answered. Are they derivational syntactic constraints, representational constraints at PF, or merely processing limitations? When and how can island violations be ameliorated by resumption? Why are some islands weak and others strong? Which role do properties of the extracted element play in incurring an island violation or not? -- In this talk, I will present a pattern of island-(in)sensitivity found in focus movements in Asante Twi (Kwa) and Limbum (Grassfields Bantu) that seems to indicate that islands may be selectively active for some categories (VP, PP) but inactive for others (NP, DP). While the island-insensitivity of at least some nominal arguments might be explained as a repair by resumption in Asante Twi (Korsah & Murphy 2019), such an explanation seems implausible in Limbum. What is more, novel data indicate that even in Asante Twi, there are some noun types for which repair by resumption is unavailable but which nevertheless are insensitive to island constraints. I conclude from this that islands can be sensitive to the category of the extracting element.
