What vs. who and which: Kind-denoting fillers and the complexity of whether-islands

We present results from three acceptability judgement experiments investigating the effect of discourse linking (d-linking) and animacy on whether-islands and interactions with resumption in Greek and English. Based on Anagnostopoulou’s referentiality hierarchy, we test the acceptability of four types of wh-phrases, what, what X, which X, which of X in a range of configurations (simple questions and questions involving extractions out of (non-island) that-clauses and whether-islands). We further test interactions between animacy and d-linking in English. Our results show that d-linking improves whether-islands in both Greek and English. However, d-linking does not alter the overall interactions: whether-islands remain mostly less acceptable than that-clauses. While acceptability increases overall as predicted by the referentiality hypothesis, we obtain two unexpected contrasts: (i) a contrast between which X and what phrases and (ii) an independent effect of animacy; who is better than what, on a par with which X phrases. These contrasts affect the acceptability of whether-islands but not that-clauses. We propose that what sets what phrases apart, is the contrast between kind-denoting (what) and ordinary individuals (which,who), which can be triggered by d-linking or animacy. This denotational hypothesis predicts that the distinction is only relevant for scopal islands like whether-islands. The denotational contrast affects the processing complexity of whether-islands. Kind-denoting wh-fillers have higher integration costs (in the sense of Gibson’s complexity model). The denotation of the filler interacts with its complexity (e.g. who vs. which X); together, they may improve the acceptability of whether-islands; however, they cannot cancel their overall complexity, as they cannot cancel the main scope island (i.e. the question intercepting the filler-gap dependency); thus, manipulations of the filler cannot restore the acceptability of whether-islands. Finally, a crucial overall conclusion is that alleviation of integration costs of the filler (through d-liking/animacy) has a stronger effect on improving whether-islands compared to cancellation of locality costs (through resumption).