Extraction from Relative Clauses: An Experimental Investigation into Variable Island Effects in English—or—This is a Dissertation That We Really Needed to Find Someone Who'd Write
This dissertation centers around the islandhood of relative clauses in English and aims to determine whether relative clauses in English ever permit extraction of a relative clause argument to a relative clause‑external position. It picks up a thread in English that started with studies on Mainland Scandinavian languages, which are well‑known for permitting extraction from relative clauses under certain conditions. The main contributions of this work are empirical and methodological in nature, but it also makes minor but contentful theoretical contributions.
On the empirical side, the dissertation presents findings from eight acceptability judgment experiments, which together present a challenge to the idea that relative clauses in English are always strong islands. Experiment 1, which was run in the early stages of this work, shows that the definiteness of the nominal phrase that contains a relative clause has no independent impact on that relative clause’s transparency to extraction (porosity, as it is called in this work). Experiments 2 and 3 represent perhaps the strongest challenge to the traditional idea that relative clauses are always strong islands. Those experiments probe three environments in which a nominal phrase can reside—the pivot of an existential assertion, the predicate of a nonverbal clause or sentence, and the direct object of transitive verb—and using a factorial definition of island effects (Sprouse et al. 2012), shows that in the former two environments, island effects are reduced nearly completely. The two following experiments, Experiment 4 and Experiment 5, investigate the islandhood of infinitival relatives formed on the relative clause subject. These relative clauses are found to be porous in any environment, indicating a lack of selective islandhood and a lack of islandhood in general, a finding which I believe to be novel but which is compatible with Bhatt’s (1999) analysis of subject infinitival relative clauses. Experiment 6 through Experiment 8 turn back to finite relative clauses, focusing specifically on relative clauses under two sets of transitive verbs. The first set is composed of eight “ordinary” transitive verbs and the second is composed of six transitive verbs which each have a felicitous use as an “evidential existential” verb—one that can be used to indirectly make an existential assertion (Rubovitz‑Mann 2000). The predicted results fail to obtain in those experiments, from which it is tentatively concluded that there are no transitive environments which facilitate extraction from a relative clause within the direct object of one of these verbs.
On the methodological side, the dissertation serves as a sort of educational tool (Chapter 3) and case study (Chapters 4–6) for three different experiment designs and methodologies. The first is an experiment design, due to Sprouse et al. (2012), that is referred to in this work as the length by structure design. I present a thorough overview of the design and how it can be extended to compare relative clause island effects in different syntactic and semantic environments. The design faces some challenges for relative clauses specifically, and an alternative design is presented which permits measurement of relative, but not absolute, island effects. The alternative design is referred to as the dependency by environment design. Finally, a computational quantitative modeling method is presented, referred to here as mixture modeling, which can be used to gain insight into the nature of ratings distributions in acceptability judgment experiments. The mixture modeling method is described in detail and is used to argue that the results of certain conditions in Experiment 3 are the result of a sizeable chunk of participants rating the condition as genuinely grammatical and another sizeable chunk rating the condition as genuinely ungrammatical, as opposed to all participants giving roughly similar ratings to each other.
On the theoretical side, the dissertation addresses several different families of hypotheses that aim to explain the extraction phenomena described in this work. The hypotheses are separated into two families: those which take acceptable extraction from relative clauses to be not a grammatical issue but an issue concerning the mapping between acceptability and grammaticality and those which take it to be a grammatical issue. The former family of hypotheses is rejected. Within the second family of hypotheses are two subfamilies of hypotheses: so‑called “reductionist” hypotheses, which generally aim to explain away island effects as non‑grammatical phenomena, and grammatical hypotheses, which take the stance that island effects are the result of grammatical constraints. The hypothesis advocated for in this work is the latter, and the experiment results presented here are argued to support that hypothesis.
All supplemental materials for this work (or links to them) can be found on my website: https://jakewvincent.github.io/dissertation.html.