How PRO to Control PRO?

How PRO to control PRO? Distribution of Infinitives/Gerunds (1) V/A-Complement a. John promised Mary [PRO to leave early]. b. John persuaded Mary [PRO to leave early]. c. John wondered [when PRO to leave]. d. Mary considered [PRO leaving early]. Sentential Subject/Intraposition e. [PRO to eat/eating too much lettuce] is not healthy. f. Mary realized that [PRO speaking of herself/himself] would embarrass John. Extraposition g. Mary realized that it embarrassed John [PRO to speak about himself/*herself]. h. Mary realized that it helped John [PRO to speak about himself/herself]. Adjunct i. John resigned after [PRO delivering his speech]. j. Mary1 helped us2 without [PRO1 being rewarded/*PRO2 rewarding her]. Rationale/Purpose Clause k. John brought the book (in order) [PRO to read it on the flight]. l. John brought the book (*in order) [PRO to read on the flight]. Infinitival Relative m. Something [PRO to eat] would be nice now. N/P-complements n. We were told about the decision [PRO to evacuate the island]. o. I read a story about [PRO falling in love with a foreign girl]. 1 Non-Arguments for PRO (2) Understood Coreference (Haegeman 1994) a. Poirot1 abandoned the investigation [in order PRO1 to save money]. b. Poirot1 abandoned the investigation [without PRO1 giving an explanation]. Problem: Predication can achieve the same effect. There is no obvious entailment from understood coreference to syntactic representation. Compare: (3) c. d. John ate the meat raw. John1 underwent an operation (on him1). e. We entered the room1. The walls (of it1) were covered with blood. Together (Haegeman 1994): “Requires a local plural NP in an A-position”. a. b. I saw the boys/*boy together. [PRO to leave together] would be stupid. Problem: The local NP can be the controller, not necessarily PRO. This is possible in (b), with an implicit controller (stupid of X), though not in cases of real arbitrary control (...would insult Mary). (4) Binding: The infinitive behaves like a binding domain (GC), hence contains a subject. a. John1 wanted [PRO1 to talk about himself1/*him1]. Problems: i) The argument is invalid. If the infinitive is a binding domain, then the contrast in (a) follows. But not vice versa: The contrast in (a) could also follow from the fact that the infinitive lacks a subject and the binding domain is the whole matrix. ii) It needs to be shown, independently, that condition A requires a syntactic and not an implicit binder (true, but not trivial, and false for conditions B & C). 2 b. Mary1 thought that John talked *(to her1) about herself1. c. d. Mary1 thought that John said (to her1) [PRO1 to wash herself]. John never talked (*to her1) about Mary1/her1. (5) Theta Criterion: Every θ-role must be assigned to an argument. Problem: Implicit arguments satisfy this criterion, why must PRO be projected? (6) EPP: The subject must be projected. Problem: The reasoning is circuler. The claim that the EPP is exceptionless in itself assumes that null subjects can satisfy it. But if they can’t, then the EPP is violable. One cannot motivate the existence of PRO on the basis of a principle whose status crucially relies on the existence of PRO. Genuine Arguments for PRO (7) Predication (Haegeman 1994): Secondary redicates require a local syntactic subject. a. b. c. d. (8) [PRO to arrive exhausted at a party] is terrible. John ate *(the meat) raw. * Dinner was served angry at the guests. [cf. ... PRO wearing aprons]. [PRO to serve dinner angry at the guests] is not a good idea. Object control + embedded subject-oriented anaphor. Russian (Natalia Strahov, p.c.) a. John1 otpravil Mary2 k svocj1/*2 sestere. John sent Mary.ACC to SELF’s sister.DAT ‘John sent Mary to self’s1/*2 sister’ b. John1 otpravil Mary2 navestit svoju1/2 sestru. John sent Mary.ACC to-visit SELF’s sister.ACC ‘John sent Mary to visit self’s1/2 sister’ 3 c. John1 ubedil Mary2 navestit svoju*1/2 sestru. John persuaded Mary.ACC to-visit SELF’s sister.ACC ‘John persuaded Mary to visit self’s*1/2 sister’ (9) (10) Dative control + binding of embedded anaphor unbindable by a dative. German (Wurmbrand 1998) a. weil der Hans der Maria sich1/*2 auf dem Photo zeigte. since the John-NOM Mary-DAT SELF in the picture showed ‘since John showed Mary himself / *herself in the picture’ b. Sie hat dem Hans erlaubt [PRO1 sich1 den Fisch mit Streifen vorzustellen]. She has John-DAT allowed PRO SELF the fish with stripes to-imagine ‘She allowed John to imagine what the fish would look like with stripes’ Split antecedents. a. b. (11) [Koster & May 1982] * John1 suggested/showed Mary2 to each other1+2. John1 suggested to Mary2 [PRO1+2 to help each other1+2]. Case concord with quirky subjects. Icelandic (Sigurðsson 1991, ex.8) a. Strákana vantað i alla í skólann. the boys.ACC lacked all.ACC.PL.MS in the school ‘The boys were all absent from school’ b. Strákarnir vonast til [að PRO vanta ekki alla í skólann]. the boys.NOM hope for to PRO.ACC lack not all.ACC in the school ‘The boys hope not to be all absent from school’ c. Strákana leiddist öllum í skóla. the boys.DAT bored all.DAT.PL.MS in the school ‘The boys were all bored in school’ 4 b. Strákarnir vonast til [að PRO lei ðast ekki öllum í skóla]. the boys.NOM hope for to PRO.DAT bored not all.DAT in school ‘The boys hope not to be all bored in school’ Distribution of PRO (12) *PRO in case positions. a. b. c. d. e. (13) * John invited PRO. * John talked about PRO. * John said that PRO can leave. * John believed PRO to be intelligent. * [For PRO to leave abruptly] would be impolite. *PRO in governed but caseless positions. a. b. c. d. e. * There was invited PRO. * A picture PRO. (cf. a picture *(of) Mary) * John’s belief PRO to be innocent. * It is likely PRO to be angry. * John was proud PRO (cf. ...*(of) Mary) Conclusion: PRO is ungoverned. [Chomsky 1981] (14) The PRO-Theorem [Chomsky 1981] Condition A: An anaphor must be bound in its local domain. Condition B: A pronoun must be free in its local domain. The local domain includes a governor and a subject. PRO is [+anaphor,+pronoun]. ----------------------------------------------------------------------∴ PRO has no local domain PRO (in OC) has a subject ----------------------------------∴ PRO is ungoverned 5 (15) Problems for the PRO-Theorem i) Under Chomsky’s (1986), the local domains of anaphors and pronouns can be different. In (a), the object DP is the minimal CFC with a governor in which the pronoun can be free, but in (b) the entire clause is the minimal CFC with a governor in which the anaphor can be bound. We therefore expect PRO to be possible there, simultaneously satisfying conditions A and B. But it is not, see (c). a. They1 like [DP their1 pictures]. b. They1 like [DP each other’s1 pictures]. c. * They1 like [DP PRO1 pictures]. ii) The exclusion of [-Agr] Infl from the class of governors is completely ad-hoc, voiding “government” as a structural relation. iii) If PRO is caseless and the Visibility Condition of Chomsky (1981) obtains, then PRO violates the θ-criterion. On the other hand – why should the Visibility Condition obtain, anyway? iv) If PRO is caseless, and NP-movement is case-driven, then we fail to account for John wanted [PRO to be elected]. However, NP-movement may be driven by the PRO-theorem as well, or – more likely – by the EPP. v) As we have seen in (11), PRO is case-marked (hence governed) in some languages. Referential properties of PRO (16) “Weather”-it & extraposition-it vs. there-expletive. a. b. c. Sometimes it1 rains without [PRO1 snowing]. It bothered John [PRO without actually offending him] that he wasn’t elected. There occured three more accidents without [there/*PRO being any medical help available on the premises]. 6 (17) PROarb is [+human]. a. b. (18) [Chomsky 1981] [For coffee prices to fall in August] is very unusual. [PRO to fall in August] is very unusual. Semantic plurality in partial control. a. John told Mary that he wanted to meet (*each other) at 6. b. John told Mary that he wanted to work together / *become partners. (19) De se / de re. a. b. The unfortunate expects that he will get a medal. The unfortunate expects [PRO to get a medal]. Implicit Control Dative Control (and locality effects) (20) a. Mary1 said it was difficult [PRO1 to get herself a visa]. b. * Mary1 said it was difficult for Bill2 [PRO1 to get herself a visa]. (21) a. * Mary knew that it had been decided (by John) [PRO to behave herself]. b. Mary knew it had been prohibited [PRO to reveal herself in public]. (22) Omissibility of Controllers a. b. (23) John recommended *(to us) to leave. Gil himlic (lanu) la’azov. Hebrew Apparent Arbitrary Control (Williams 1980, Bouchard 1984, Huang 1989) a. b. c. John said (*to Mary) [PROarb to behave oneself]. It is fun [PROarb to eat ice-cream]. Genx [it is fun for x [for x to eat ice-cream]]. 7 Agent Control (24) a. * John was promised (by Mary) [PRO to leave]. b. It was decided [PRO to leave]. c. The game was played [PRO wearing no shoes]. (25) Rationale Clauses a. The boat was sunk [PRO to collect the insurance]. b. * The boat sank [for the owner/PRO to collect the insurance]. c. * The ship was sunk to become a hero. d. Grass is green to promote photosynthesis. Lebeaux (1984) (26) Main idea a. All types of PRO are anaphoric, including so-called PROarb. b. (27) The differences between locally bound lexical anaphors and PRO in NOC are not due to their different constitution (both are anaphors), but due to the different mechanisms of anaphor binding. Long Distance (LD) controlled PRO is not a pronoun a. His1 / *PRO1 having shaved already shows that Mary arrived more than 5 minutes after John1 did. The antecedent of PRO must command it (i.e., the first S dominating the antecedent dominates PRO). (28) Narrowing down “anaphorhood” Properties of anaphors are of two different types. The dependency property An anaphor is referentially dependent on an antecedent. 8 The binding properties a. b. c. d. [5]-[6] The antecedent must be local (inside GC / CFC). The antecedent must c-command the anaphor. The antecedent can’t be split. The anaphor gets only sloppy reading under VP-ellipsis. (29) Lebeaux: Only the dependency property is inherently linked with anaphors. The binding properties follow from a particular mechanism of binding (predication). Concretely, local binding exhibits the binding properties, non-local binding doesn’t. (30) Local binding Lebeaux sets up the definitions [6]-[7] such that the domain of local binding for a lexical anaphor is the minimal NP/S containing it, whereas for PRO (a nonlexical anaphor) it’s the minimal NP/S containing the nonfinite clause. (31) Local binding a. b. John saw himself in the mirror. John tried [PRO to leave]. Binding domain = matrix S Binding domain = matrix S Non-local binding c. d. (32) John knew that [there were [pictures of himself] inside]. John told Mary that [[PRO leaving early] would be a mistake]. Nonlocally bound reflexives may take non-commanding or split antecedents, and allow strict readings under VP-ellipsis [10]; same for non-locally bound PRO [11]. Note: Some OC contexts allow split antecedents, cf. (10b). Lebeaux: Since all occurences of PRO are anaphoric (dependent), but some show binding properties (OC) and others don’t (NOC), the binding properties can’t be inherent to PRO. 9 PROarb is locally bound by an operator / PROarb is an A-bar anaphor (33) a. b. [PROx making a large profit as a slum landlord] requires [PROx exploiting the tenants]. [PROx to know him] is [PROx to love him]. The interpretation of the two PROarb subjects is linked, even if pragmatics prefers otherwise: c. * [PROx becoming a movie star involves [PROy recognizing you]. Lexical subjects need not be linked: d. (34) His getting up on time will require her setting the alarm clock. Since neither PRO binds the other PRO, they must be both variables bound by the same operator – with universal/generic force. PROarb is similar to the indefinite pronoun one. a. ∀x / Genx ((PROx making a large profit) requires (PROx exploiting the tenants)). (35) Important note: The universal operator is associated with necesseity predicates – require, entail, involve. When the predicate doesn’t imply necessity, the linked reading disappears. a. (36) [PROx going to the movies] beats [PROy staying home and eating popcorn]. Linked readings obtain in clausal contexts, where PRO exists, not in nominal contexts. a. b. PRO winning games requires PRO losing games. (winner of some games = loser of some games) The winning of games requires the losing of games. (If someones wins, someone (else) must lose). 10 If PROarb is an anaphor, what is its binding domain? (37) The binding domain of PRO is the minimal NP/S containing the nonfinite clause. a. b. * [BD [PROx winning the trust of the populace] requires [PROy having to serve in the army]]. [BD1 [PROx winning the trust of the populace] requires that [BD2 [PROy having to serve in the army] be abolished]]. Unifying LD and arbitrary control (38) In general, NOC comprises of either long-distance or arbitrary control. If no controller is available in the binding domain of PRO, then either LD or arbitrary control is allowed. Since the choice between them is not grammatical, Lebeaux suggests to unify both cases. (39) a. Closure: If PRO is unbound in its binding category, adjoin a null operator b. Op to the binding category and coindex it with PRO. Operator interpretation: Op may be coindexed with a grammatical antecedent or a discourse referent. Otherwise, Op is generic. See examples [35]-[38]. (40) Problem: (39a) predicts that a local potential antecedent rules out NOC. But we’ve seen examples like (a), where that is possible. a. b. Mary realized that it would help John [PRO to speak about himself/herself/oneself]. Mary realized that [PRO speaking about herself/himself/oneself] would embarrass John. Solution: Maybe John in (a) doesn’t c-command the extraposed clause. Notice that the operator in (b) would c-command John, violating [43]. 11 (41) Prediction: Interrogative complements should fall under OC if a matrix controller is avaialble. Paradox: Arbitrary control is allowed, LD control isn’t. a. Mary said that John wondered [how PRO to introduce himself/oneself/*herself]. Landau (2000: Chapter 1): The basics (42) What classifies a construction as OC? a. Classical answer: Whether or not PRO is obligatory. b. Alternative answer: Whether or not PRO is obligatorily controlled. ⇒ OC may obtain even if PRO alternates with a lexical subject (e.g., prefer). (43) Two types of OC: Exhaustive vs. Partial control (We thought that...) a. The chair1 preferred [PRO1+ to gather at 6]. b. Bil1 regretted [PRO1+ meeting without a concrete agenda]. c. Mary1 wondered [whether PRO1+ to apply together for the grant]. (cf... * to apply as partners). (We thought that…) d. * John1 managed [PRO1+ to gather at 6]. e. * The chair1 began [PRO1+ meeting without a concrete agenda]. f. * Mary1 is able [PRO1+ to apply together for the grant]. (44) EC-infinitives a) b) c) implicative aspectual modal (dare, manage, forget, force...) (begin, stop, continue...) (able, capable, need, must...) 12 PC-infinitives a) b) c) d) (45) desiderative interrogative factive propositional (want, prefer, decide, demand...) (wonder, ask, guess, inquire...) (hate, regret, dislike, shocked...) (believe, imagine, say, declare...) Question: What distinguishes EC- from PC-infinitives? Answer: The absence or presence of tense (Landau 2000). EC-infinitives are untensed, PC-infinitives are tensed: a. * Yesterday, John managed to solve the problem tomorrow. imp. b. * Yesterday, John began to solve the problem tomorrow. asp. c. * Yesterday, John was able to solve the problem tomorrow. mod. d. Yesterday, John hoped to solve the problem tomorrow. des. e. Yesterday, John wondered how to solve the problem tomorrow. int. f. Today, John regretted having solved the problem last week. fac. g. Oggi Gianni crede di avere offeso Maria la notte scorsa. prop. today John believes to have offended Mary the night last ‘Today, John believes to have offended Mary last night’ [Italian] (46) Outline of the account: OC is Agree between a matrix controller and the embedded [Agr] or PRO. The C head of of PC infinitives is associated with [+T], and derivatively, [+Agr]. This [+Agr] is the target of partial control. Because it bears no specification of Semantic Plurality, it allows the controller and PRO to differ on that feature. In EC, it is PRO which is directly controlled (C bearing no [T] or [Agr]), hence it must match the semantic plurality of the controller. Super-Equi (47) a. b. c. d. Mary knew that it disturbed John [PRO to perjure himself / *herself]. Mary knew that it damaged John [PRO to perjure himself / herself]. Mary knew that [PRO to perjure himself / herself] would disturb John. Mary knew that [PRO to perjure himself / herself] would damage John. 13 (48) a. b. In a structure [... X ... [ it Aux Pred Y [S PRO to VP]], where Y and S are arguments of Pred: i) If Pred is psychological, Y must control PRO. ii) If Pred is non-psychological, either X or Y may control PRO. In a structure [... X ... [S [S PRO to VP] Pred... Y]], either X or Y may control PRO. In short, LD-control is susceptible to intervention only in extraposition, and only when the intervener bears an experiencer θ-role. (49) Outline of the account: In extraposition, at the relevant level where control is established, an experiencer controller will c-command the infinitive but a theme/goal controller will not. Hence, OC is linked to experiencers. In intraposition, no clausemate argument c-commands the infinitive, hence NOC. Note: In the book I assumed that “the relevant level” is already established at DS, now I believe that the experiencer c-commands the infinitive only at LF. (50) A substantive correlation: The goal/theme object may fail to to c-command the infinitive only if the latter is adjoined to VP. But adjuncts are islands. So we expect a correlation between NOC and islandhood. (51) a. b. c. d. It would kill the workers1 [PRO1 to build this dam]. What2 would it kill the workers1 [PRO1 to build t2 ?] It would kill the forest [PROarb to build this dam]. * What2 would it kill the forest [PROarb to build t2 ?] (52) a. Hillary thinks it damaged Bill1 [PRO1 to talk about himself on the Dave Letterman show]. That’s the talk show2 that Hillary thinks it damaged Bill1 [PRO1 to talk about himself on t2 ] . Hillary1 thinks it damaged Bill [PRO1 to talk about herself on the Dave Letterman show]. * That’s the talk show2 that Hillary1 thinks it damaged Bill [PRO1 to talk b. c. d. about herself on t2 ] . 14