GIVENness and local contexts Ò GIVENness is checked compositionally, via operators in syntax Simon Charlow (Rutgers) Ò GIVENness is sensitive to the local context (assignment) SALT 29 @ UCLA May 18, 2019 Ò Generally: grammatical constraints like GIVENness must not unbind bound variables. Doing otherwise has paradoxical consequences. 1 2 S Upshot: indices matter less for deaccenting/ellipsis than thought. They S but S fix values for variables. But it’s the values that matter (Jacobson 2009). Ò This dissolves puzzles as old as the ellipsis literature, and helps Cally ⇤ she1 DIDN’TF leave ground a simpler theory of ellipsis based on perfect identity. 1 S Ò Sheds new light on impossible ACDs, focused bound pronouns. t1 VP said S she1 left 3 4 Damian blocked Steph, and then. GIVENness and anaphora (1) SETHF blocked Steph. (2) *STEPHF blocked Damian. underfocused (3) *SETHF blocked STEPHF. overfocused 5 6 Schwarzschild (1999) F-mark all and only material in B without a correspondent in A: Ò GIVENness: If B isn’t F-marked, it must be GIVEN a Damian blocked Steph SETHF blocked Steph õ Ò B is GIVEN iff it has an antecedent A B : ÇAÉ ÇBÉf õ = 2 a ÇDamian blocked StephÉ ÇSETHF blocked StephÉ 2 f Ò AvoidF: F-mark as little as possible (w/o violating GIVENness) block(damian, steph) block(x, steph) x : e ¨ a 2 { | } ÇBÉf is the focus set gotten by varying F-marked things in B (Rooth 1985, Kratzer 1991) 7 8 This sees indexical differences, even when blurred in a context. That is õ a problem. Suppose we’re in a context where g(1) g(2). = (4) I saw her1 and YOUF saw *HERF,2. We need a definition of GIVEN that’s explicit about assignments. Here’s Possible reply: forbid ‘Redundant’ assignments (Schlenker 2005)? one possibility (after Heim 1997: 206, cf. Schwarzschild 1999: 152): Or suppose we’re in a context where g 1 mary. g g ( ) A B g : ÇAÉ ÇBÉ = õ () 8 2 f (5) I saw Mary and YOUF saw *HERF,1. Possible reply: names are variables? But then we’re back to (4). Occasionally g is entertained (cf. Tomioka 2008, Griffiths 2018). Schwarzschild 1993, 1999; see also Jacobson 2004, 2009 9 9 10 Our improved evaluates A and B at the same “global” assignment. So õ õ unbinds variables, can’t distinguish free and bound occurrences. Pronoun meaning in context is what matters (Schwarzschild 1993, 1999): g g Even if g(1) mary, in the yellow, though eventually not the orange. A B at g ÇAÉ ÇBÉ = õ õ () 2 f (8) Newt likes her cat . *CALLY [1 t likes her cat ] too. This makes better predictions: 1 F 1 1 (6) I saw her1 and YOUF saw *HERF 2. g(1) g(2) é overfocused (9) Steph hopes I cite him1 . *SETHF [1 t1 hopes YOUF cite him1 ]. , = (7) I saw Mary and YOUF saw *HERF,1. g(1) mary é overfocused = Maybe unintuitive. But ok? No: in (10) GIVENness is satisfied, period! (10) Cally [1 t1 said she1 left ] but *she1 DIDN’TF leave. 11 12 Empirics aside, this is weird. Bound variables have values in their local contexts (though from the ‘outside’ the idea that they have values may There is an alternative. Rooth’s (1992a) ‘interprets focus’ in situ, seem strange, cf. Fine 2003). GIVENness shouldn’t ignore these values! ⇠ requiring its associate ↵ to be congruent with the value of a variable n: Ò And def shoudn’t make their values the global-contextual ones g g g ÇBÉ if g(n) ÇBÉ ÇB nÉ : 2 f ⇠ = ( undefined otherwise After all, presupposition satisfaction is checked in a local context: B and the ÇAÉ stored at n may be eval’d at different assignments. (11) If there’s an escalator in 18SEM, the escalator in 18SEM is hidden. (12) Each of these studentsi brought theiri laptop. If the congruence constraint was a kind of presupposition (as has often been proposed), it would be surprising if it was not also checked ‘in situ’. I’m adopting a semantic theory of alternatives for concretness, but these points apply equally to syntactic theories of alternatives (Katzir 2007, Fox & Katzir 2011). 13 14 g(1) mary g(1) mary = = S S S and S S but S g1,x g1,x Newt VP2 CALLYF ⇤ é λx ÇSÉ Cally ⇤ é λx ÇSÉ S 2 ⇠ likes her1 cat 1 S 1 S she1 DIDN’TF leave t1 VP t1 VP VP 2 said S2 ⇠ likes her1 cat she1 left 1,mary 1,cally 1,cally 1,mary Çlikes her1 catÉ Çlikes her1 catÉ # Çshe1 leftÉ Çshe1 DIDN’TF leaveÉ # 62 f 62 f 15 16 Akan data due to Augustina Owusu (p.c.): Updating Schwarzschild (1999) (13) Kofi re-pa Kwame ho. Ò GIVENness: If B isn’t F-marked, it must be GIVEN Kofi PROG-pass Kwame body a ‘Kofi is overtaking Kwame’ Ò B is GIVEN iff it is the sister of ⇠ a Deebi! KWAME na "-re-pa KOFI ho no. Ò AvoidF: F-mark as little as possible (w/o violating GIVENness) No! Kwame FOC 3SG-PROG-pass Kofi body DEF a ‘No, KWAME is overtaking KOFI.’ See Büring (2016) for discussion of a very similar system. No, though usually analyzed as a cross-categorial definite/familiarity marker (cf. Renans 2018 on Ga) can mark clauses that are GIVEN. 17 18 We stand in need of one more revision: The basic patterns are reproduced with indexical expressions: (14) Steph [1 t1 liked his1 shot ] and SETHF [2 t2 liked his2 shot ]. (15) (I’m the best.) No, IF am! We do not need to stress the second his. But while the latter orange node (16) (I ran a marathon.) Yes, you did. is GIVEN, the latter yellow node is not. (Very much like ‘rebinding’.) But there is a striking disanalogy in index-dependency: 1,steph 2,seth Çhis1 shotÉ Çhis2 shotÉ # 62 f (17) In ’92 the Potus was a Bush. #In ’04 [the POTUS]F was a Bush. GIVENness must be weakened, on pain of being unsatisfiable: GIVENness relates meanings via . We’ve seen ample evidence that the ⇠ meanings of pronouns (indexicals) saturate the assignment (context). Ò GIVENness: If B isn’t F-marked, it must be GIVEN, or dominated by a node that is GIVEN a Data like (17) suggest meaning doesn’t saturate the index: This is a new argument for something similar to “Maximize Background”. c,g c,g,(w,t) Ç↵É ...λ w t ... not Ç↵É ... = ( , ) = Truckenbrodt 1995, Wagner 2006, 2012, Büring 2008, 2016 19 20 Ellipsis requires identity. (18) I saw an elk from France. Did YOUF (see an elk from France)? (19) I saw her, but YOUF DIDN’TF (see her). A better theory of ellipsis Sloppy readings are easy to accommodate: (20) Mary [λi ti likes heri office], but SUEF DOESN’TF (λj tj like herj office). Sag characterized A and E here as ‘alphabetic variants’, a relation inspired by the λ-calculus notion of ↵-equivalence (though distinct). Keenan (1971), Sag (1976), Williams (1977) 21 22 Sloppy pronouns don’t need to be bound inside E (‘rebinding’): Two-part theory of ellipsis licensing (Rooth 1992b) (21) Johni’s mom likes himi. BILLF,j’s mom DOESN’TF (like himj). (22) Bagelsi [I like ti]. DONUTSF,j [I DON’TF (like tj)]. Ellipsis is licensed whenever the following two conditions are satisfied: (23) Every dog thinks I like it . Every CAT thinks I DON’T (like it ). i i F,j F j Ò Syntactic: A E Syntactic identity up to variable names ⇡ (24) If I see a cati I pet iti. If I see a DOGF,j I DON’TF (pet itj). Ò Semantic: Γ [A] ∆[E] A and E are (in) congruent structures õ Same range of interpretations available under deaccenting. Note that is the ex situ congruence relation. õ See Hirschbühler 1982, Evans 1988, Jacobson 1992, Rooth 1992b, Hardt 1993, Fiengo & May 1994, Tomioka 1999, Takahashi & Fox 2005, and many others. 23 24 John1 [t1 likes his1 mom] BILLF 2 [t2 does (like his2 mom)] õ , Hard to oversell how successful, illuminating this approach has been. likes(j, mom(j)) likes(x, mom(x)) x : e 2 { | } Ò Congruence is a feature of grammar not specific to ellipsis (Schwarzschild 1999, Büring 2016, cf. Tancredi 1992, Fox 1999). John1 [t1 mom likes him1] BILLF 2 [t2 mom does (like him2)] õ , The syntactic condition is unfortunate (e.g., Merchant 2001). There are likes(mom(j), j) likes(mom(x), x) x : e 2 { | } reasons to think the ellipsis-specific identity relation is exact. Binding in the elliptical clause guarantees that congruence is satisfied. In general, the interaction of binding and alternatives creates complications (Poesio 1996, Something akin to congruence is present even in dissenters from the overall Roothian Shan 2004, Romero & Novel 2013, Charlow 2018). This won’t affect any of my points. picture (e.g., Merchant 2001, Kehler 2000, building on Hobbs 1979). 25 26 Why not just coindex the sloppy pronoun and its correlate in A? No Meaningless Coindexing (NMC) (Heim 1997: 202) Mary1 [t1 likes her1 office ] SUEF 1 [t1 does (like her1 office) ] õ , If an LF contains an occurrence of a variable v that is bound by a node ↵, likes(m, office(m)) likes(x, office(x)) x : e 2 { | } then all occurrences of v in this LF must be bound by the same node ↵. Actually, this needs to be ruled out: Sag defined a context-sensitive sense of ‘alphabetic variance’ distinct from the λ-calculus notion, to similar effect.