answer to elaine: why the P stranding penalty? It shows that extracting the complement of a noun is not the same as extracting the complement of a preposition. One (syntactic) way to go would be to say that the ’subject island’ is ‘selective’ and allows PP but not NP to be ‘moved’. This is unlikely given the acceptability of NP extraction out of infinitival or gerund subjects: In his bedroom, which [to describe __ as small ] would be a gross understatement, he has an audio studio setup. (Chaves, 2013, 303) [...] phenomena which [to understand __] would take an amount of information processing beyond at least our current limit. (Chaves & Dery, 2019, 481) [...] the Joker is a fascinating character who [spending time with __] is a treat. (Culicover & Winkler, in prep) And some prep stranding cases have been reported as more acceptable than others: a. What were [pictures of __] seen around the globe? (Kluender, 1998, 268) b. Which problem will [a solution to __] never be found ? (Chaves, 2013, 301) So we suspect processing is at stake here. Chaves & Dery, 2019, 481 suggest that “the grammatical function of the fronted phrase PP is clearer from the onset than if NP were fronted, given the presence of the preposition: there are fewer potential gap sites that are consistent with the extracted constituent, aiding processing and improving acceptability” . Another hypothesis is that the difficulty in extracting from a preposition-stranded subject is due to the fact that most cases of P-stranding occur post-verbally in English: NP extractions from nominal subjects are very rare, and hence difficult to process based on syntactic surprisal (Hale, 2001; Levy, 2008; Chaves, 2013). Hi David: Thanks for the comment. Just to clarify: our project is about what makes certain structures “islands” for extractions. Our understanding of what an “island” structure is is one that is unacceptable (similar in ratings to other unacceptable / ungrammatical materials). In particular, we have been looking at extractions out of subject positions in English (and French in other work). In this particular project, we have found that PP extractions from subject positions are no worse than PP extractions from object positions in relative clauses, and both are much better than corresponding ungrammatical versions. So in our examples on the poster, B and F are a lot better than D and H, but there is no interaction between position (subject , object) and grammaticality (PP-extract, ungram). Consequently, we don’t think that the subject position is an “island" for extraction in RCs. (I could attach the model results here, but I think it’s pretty obvious from the graph that the analyses that I am saying here are correct: a main effect for extraction: PP-extractions better than ungrammatical baselines; and no interaction between position (subject, object) and extraction (PP-extraction, ungrammatical). (In other work, we show that the subject position is an “island" for extraction in wh-questions.) B / Subject PP extraction The politician wanted to explore the project, to which [the opposition _ ] might occupy the Senate for a month. F / Object PP extraction The politician wanted to explore the project, to which the Senate might discuss [the opposition _ ] for a month. D/ Subject-ungram The politician wanted to explore the project, which the opposition might occupy the Senate for a month. H/ Object-ungram The politician wanted to explore the project, which the Senate might discuss the opposition for a month. In your question, you are asking about a different baseline: the grammatical versions in A and E: A / Subject- RC The politician wanted to explore the opposition to the project, which _ might occupy the Senate for a month. E / Object- RC The politician wanted to explore the opposition to the project, which the Senate might discuss _ for a month. It sounds like you are wondering if there is an interaction between position (subject, object) and extraction type (PP from-subject, NP subject). It turns out that this interaction is not significant (see model results below). But even if it were significant, we don’t think that this kind of interaction would be addressing what counts as an “island” in English, since both the PP extraction from subject (B) and the PP extraction from object (F) are not close to the ungrammatical baseline. So this question wouldn’t exactly be about “islands” anymore, but about factors affecting English sentence acceptability, which is a reasonable thing to be wondering about, but wasn’t the focus of this study. In any case, the interaction isn’t close to reliable. Thanks again for your question. Scaled residuals: Min 1Q Median 3Q Max -1.97990 -0.64012 0.01786 0.65603 1.96850 Random effects: Groups Name Variance Std.Dev. Corr WorkerId (Intercept) 0.02509 0.1584 RC-PP 0.02899 0.1703 0.82 sub_vs_obj1 0.06592 0.2568 0.48 0.90 RC-PP:sub_vs_obj1 0.40469 0.6362 -0.66 -0.11 0.33 Item (Intercept) 0.09124 0.3021 RC-PP 0.05844 0.2417 -0.77 sub_vs_obj1 0.10160 0.3188 0.08 -0.69 RC-PP:sub_vs_obj1 0.33009 0.5745 0.28 0.10 -0.49 Residual 0.38453 0.6201 Number of obs: 271, groups: WorkerId, 34; Item, 16 Fixed effects: Estimate Std. Error df t value Pr(>|t|) (Intercept) -0.18733 0.08887 16.79700 -2.108 0.0504 . RC-PP 0.25283 0.10134 20.23300 2.495 0.0213 * sub_vs_obj1 0.06480 0.11877 16.61600 0.546 0.5926 RC-PP:sub_vs_obj1 0.19172 0.23592 17.69900 0.813 0.4272