A Compositional Semantics for wh-ever Free Relatives Aron Hirsch (MIT) 1. Introduction. This paper proposes a new compositional semantics for wh-ever free relatives: (1) I will talk to [whoever comes to the party]. Building on Rawlins’ (2008, 2013) analysis of unconditionals, I argue that wh-ever FRs have a question denotation: they denote sets of propositions. They undergo covert movement from argument position to restrict a modal in a conditional-like structure, while their in situ copy is interpreted like an E-type pronoun. The proposal unifies wh-ever across the constructions where it occurs, and accounts for the disparate set of properties wh-ever FRs show. 2. Rawlins (2008, 2013). The point of departure is Rawlins’ analysis of unconditionals, (2). He takes the wh-ever adjunct in (2) to denote its Hamblin set, i.e. the set characterized in (3), and takes (2) itself to be a Lewis-Kratzer-Heim conditional, i.e. (4) containing a restricted modal. (2) Whoever comes to the party, it will be fun. (3) pst . x [p = w . x comes to the party in w] (4) Op [[whoever comes to the party ] the party is fun] (slightly simplified) Rawlins assumes that whoever comes to the party and  compose by Pointwise Functional Application: each proposition in the set characterized in (3) is taken pointwise as the restrictor argument of . The resultant meaning is a set of conditionals of the form in (5), and that set is universally quantified over by a higher operator, Op in (4). (5) If x comes to the party, the party will be fun. The sentence in (2) naturally conveys that the speaker is ignorant about who will, in fact, come to the party. Departing somewhat from Rawlins, I take this to be derivative of the effects of two presuppositions: (i) a presupposition that some element of the set characterized in (3) be true at the actual world (i.e. the existence presupposition of the question), and (ii) a presupposition encoded in  that its restrictor argument not be an epistemic certainty. Given (i), (ii) is satisfied in (2) only if the set characterized in (3) contains at least two propositions (e.g. p’ = that John comes to the party, p’’ = that Mary comes to the party), and there are epistemically accessible worlds at which p’ is true and p’’ false, and worlds at which p’’ is true and p’ false. 3. Towards FRs. The construction in (6) is similar to (2), except there is a pronoun in the main clause whose interpretation is linked to that of the wh-ever XP. (6) Whoever comes to the party, I will talk to them. The analysis given above for (2) extends to (6), provided that them receives an E-type interpretation as the (maximal) entity that comes to the party in w’, with w’ bound by . The interpretation of (6) is then the set of conditionals of the form in (7), universally quantified over. (7) If x comes to the party, I will talk to the (maximal) entity that comes to the party. 4. Proposal. (1) is different from (6) in that the wh-ever XP appears in argument position, but I argue that (1) and (6) have a similar LF. I take it that (1), like (6), contains a covert modal: (8) [ I talk to whoever comes to the party] The wh-ever XP, being of type , is not interpretable in situ complement to talk to, so undergoes covert movement to the position where the wh-ever XP appears overtly in (6), i.e. the wh-ever XP moves to restrict the modal: (9) Op [[whoever comes to the party ] I talk to whoever comes to the party] I assume the copy theory of movement, and develop a proposal that the in situ copy of the whever XP is converted into a definite description by a modified notion of trace conversion. When a generalized quantifier over individuals undergoes QR, Fox (2002) suggests that its in situ copy is converted into a definite description by (i) replacing the quantifier with a definite determiner, (10b), and (ii) introducing a variable into the definite description bound by the quantifier, (10c). I suggest that (i) and (ii) are separable, and that in FRs, only (i) occurs (cf. Schueler 2008).

a. [every boy x John talked to every boy] b. [every boy x John talked to the every boy] c. [every boy x John talked to the boy x] The internal composition of the wh-ever XP in (9) is as follows: wh-ever denotes (11) and come to the party (12); they compose by Intensional Functional Application, yielding (3) above. (11) [[wh-ever]] = f . p . x [p = w . f(w)(x)] (12) [[comes to the party]] w = x. x comes to the party in w When trace conversion applies to the in situ copy of the wh-ever XP in (9), wh-ever is deleted and replaced with the definite determiner, (13). (12) is taken as the argument of the definite determiner, yielding (14), convergent with the E-type interpretation of the pronoun in (6). (13) Op [[whoever comes to the party ] I talk to the whoever comes to the party] (14) x[x comes to the party in w] The world parameter in (14) is bound by , and the overall meaning, as in (6), is a set of conditionals of the form in (7), universally quantified over by Op. 5. Advantages. It has been observed that wh-ever FRs have properties in common with questions and definite descriptions (Jacobson 1995, Dayal 1995, Rullmann 1995), and appear to be modalized (Dayal 1997, von Fintel 2000). The parallel with questions is established e.g. by (15), which shows that wh-ever FRs, like questions, can take the hell, which is plausibly analyzed as operating on a set of propositions. Modalization is indicated by the fact that wh-ever FRs convey ignorance (16a) or indifference (16b). (15) a. Who the hell came? b. I will talk to [whoever the hell comes]. (Tredinnick 2005) (16) a. I will talk to whoever comes to the party, #namely John. b. I picked up whatever tool was handy. (It was the hammer, but I didn’t care.) The proposal accounts for the disparate set of properties: the wh-ever XP has a question denotation; its in situ copy is converted to a definite description; and the construction is conditional-like in that it contains a modal. The proposal unifies wh-ever in FRs with wh-ever in questions and concessives, which were unified with one another in Rawlins (2008, 2013): the wh-ever XP has a question denotation across constructions. With regard to ignorance and indifference, I adopt the core insight of von Fintel (2000) that the difference between them traces to a difference in modality: ignorance involves an epistemic modal presupposing that its restrictor argument not be an epistemic certainty; indifference involves a counterfactual modal. Von Fintel localized modality in a presupposition of wh-ever, which left (17) as a puzzle: presuppositions project out of a conditional antecedent, but indifference in (17) is interpreted within the antecedent. In the present proposal, modality is asserted and (17) has the LF in (18), where the counterfactual modal merges in the antecedent. (17) Unless Jack simply voted for whoever was at the top of the ballot, he must have spent at least 5 minutes in the voting booth. (18) [Unless [whoever was at top of the ballot ] John simply voted for the whoever … ballot] The proposal also captures von Fintel’s (19), involving an ignorance reading. Epistemic modals resist embedding, (20), and I take it that the modal responsible for the ignorance reading merges at a wide scope position, (21). Although conditional antecedents are generally scope islands, I suggest that island effects only arise when the higher copy binds a variable in the in situ copy, which is not the case in (21). Ignorance is derived by presupposition, and the assertion is the set of conditionals of the form in (22), universally quantified over — the right meaning for (19). (19) Unless there’s a lot of garlic in whatever Arlo is cooking, I will eat out. (20) ??Unless it must be raining, I will go out tonight. (von Fintel 2000) (21) [[whatever A is cooking ] [unless there’s a lot of garlic in the whatever A is cooking …]] (22) If Arlo is cooking x, unless there’s a lot of garlic in the thing Arlo is cooking, I will eat out. (10)