2.1 Are Farsi sluices derived from clefts?

One possibility is that Farsi sluices are derived not through movement – of a yet unknown variety – but from a cleft structure. Sluicing-like constructions have long been known to exist in Japanese, Korean, and Mandarin Chinese, all languages lacking obligatory wh-movement. But there is a great deal of evidence suggesting that, at least for these languages, the source of the sluicing-like construction is not an ordinary constituent question but rather a clefted question. Merchant (Reference Merchant, Alexiadou, Fuhrhop, Law and Kleinhenz1998), following earlier work, makes this proposal for Japanese (similar approaches are taken in Nishiyama, Whitman & Yi Reference Nishiyama, Whitman, Yi, Akatsuka, Iwasaki and Strauss1996 for Korean, and Adams Reference Adams, Arunachalam and Scheffler2004 for Mandarin Chinese). He dubs the slucing-like construction found in Japanese pseudosluicing, an example of which is given in (22).

1. (22) Dareka-ga sono hon-o yon-da ga, watashi-wa [_CP [_TP pro dare da/de-aru] ka] wakaranai. someone-nom that book-acc read-past but I-top  who be-pres q know.not ‘Someone read that book, but I don't know who it is.’ (Merchant Reference Merchant, Alexiadou, Fuhrhop, Law and Kleinhenz1998: 91)

What looks here like the wh-remnant of a sluice is actually just a wh-phrase in the pivot of a cleft. Since the expletive subject and copula are both null and the cleft clause (the part that looks like a relative clause) is only optionally present, the construction in (22) looks like sluicing in English.

This analysis of pseudosluicing relies crucially on the fact that the cleft clause is optional. In English, too, either (23) or (24) is a suitable answer to the question Who lives in Paris?

1. (23) It's Aurélie who lives in Paris. full cleft

2. (24) It's Aurélie. truncated cleft

The exact relationship between the constructions in (23)–(24) has not been decisively settled. Some accounts relate the truncated cleft in (24) to the full cleft in (23) derivationally, while others posit no relation whatsoever (see Mikkelsen Reference Mikkelsen, Geist and Rothstein2007 for discussion, references, and an analysis of truncated clefts as specificational copular clauses). While the structural analysis of clefts is orthogonal to my purpose here, it is important to keep the truncated and full varieties apart conceptually. The two constructions differ in the restrictions they place on their pivots, restrictions that will be useful in figuring out whether what looks like sluicing in Farsi is a cleft.

Farsi has a productive clefting strategy. The question in (25) can be answered with either a full cleft (answer 1) or a truncated cleft (answer 2).

1. (25)

   • Q: che kesi dar zad? what someone door hit.3sg ‘Who knocked?’

   • A1: rostam-e ke dar zad. Rostam-is that door hit.3sg ‘It's Rostam who knocked.’ full cleft

   • A2: rostam-e. Rostam-is ‘It's Rostam.' truncated cleft

Constituent questions can be formed on the pivot of either type of cleft:

1. (26) ye kesi in ketāb-o xunde vali ne-midunam ki bud ke ketāb-o xund. one someone this book-obj read.3sg but neg-know.1sg who was that book-obj read.3sg ‘Someone read this book, but I don't know who it was that read the book.’

2. (27) ye kesi in ketāb-o xunde vali ne-midunam ki bud. one someone this book-obj read.3sg but neg-know.1sg who was ‘Someone read this book, but I don't know who it was.’

If we are trying to derive a sluicing-like structure from one of the clefts above, the truncated cleft in (27) seems like the more promising source. The cleft clause is already missing and Farsi, as a pro-drop language, does not have expletives (see Karimi Reference Karimi2005: 89–94 for discussion). The only difference, then, between (27) and a sluice is the presence of the copula. But while the copula is optional in Japanese, there is no general process of copula omission in Farsi. Leaving -e ‘is’ out in a predicational copular clause, as in (28), or a full cleft, as in (29), is ungrammatical. (For ungrammatical examples, I provide the closest grammatical English translation possible.)

1. (28) māshine sohrāb qermez*(-e). car Sohrab red-is Intended: ‘Sohrab's car is red.’

2. (29) rostam*(-e) ke dar zad. Rostam-is that door hit.3sg Intended: ‘It is Rostam who knocked.’

Two conceptual arguments militate against positing a process of copula deletion here. First, as an elliptical operation, it would be quite strange, applying to a constituent that is not a phrase. Second, copula deletion would only target truncated clefts, a restriction that is nothing more than a stipulation.

A number of empirical arguments can also be brought to bear on the issue (introduced originally in Merchant Reference Merchant, Alexiadou, Fuhrhop, Law and Kleinhenz1998, Reference Merchant2001: 115–127).Footnote ⁴ First, truncated clefts do not allow wh-adjuncts in pivot position, though they are fine as the remnant of a sluice. This is illustrated for English in (30).

1. (30) He fixed the car, but I don't know how/why/when/where (*it was). (Merchant Reference Merchant2001: 121)

An identical constraint is found in Farsi, as shown for four different wh-adjuncts in (31)–(34).

1. (31) navid ye jur-i javāher-o dozdide. ne-midunam chetor (* bud). Navid one way-ind jewels-obj stole.3sg neg-know.1sg how was ‘Navid somehow stole the jewels. I don't know how.’

2. (32) vis māshin-o be ye dalil-i taamir karde vali ne-midunam cherā (* bud). Vis car-obj to one reason-ind repair did.3sg but neg-know.1sg who was ‘Vis repaired the car for some reason, but I don't know why.’

3. (33) rostam māshin-o ye moqe-yi taamir karde vali ne-midunam kei (*bud). Rostam car-obj one time-ind repair did.3sg but neg-know.1sg when was ‘Rostam repaired the car sometime, but I don't know when.’

4. (34) royā javāher-o ye jā-i qāyem karde vali ne-midunam kojā (*bud). Roya jewels-obj one place-ind hiding did.3sg but neg-know.1sg where was ‘Roya hid the jewels somewhere, but I don't know where.’

If sluicing in Farsi is derived from a truncated cleft, then the contrast in grammaticality when the remnant is a wh-adjunct is unexpected.

A parallel argument can be made from the incompatibility of truncated clefts with pivots that correspond to the implicit argument of a preceding clause. In (35), the object of eat in the first clause is not overtly expressed; the truncated cleft in the second clause is accordingly ungrammatical. A sluice is, of course, possible (these are Chung, Ladusaw & McCloskey's (Reference Chung, Ladusaw and McCloskey1995)sprouting cases).

1. (35) They said they had already eaten, but I don't know what (*it was).

Farsi exhibits the same restriction, with one small caveat. For reasons that are not entirely clear to me, a simplex verb like xordan ‘eat’ must always take an object, as shown in (36), even if it is a noun with little semantic content distinct from the verb, such as qazā ‘food’.Footnote ⁵

1. (36)

   1. (a) giti qazā xord. Giti food ate.3sg ‘Giti ate.’

   2. (b) #giti xord. Giti ate.3sg

Complex predicates in Farsi (also called light verb constructions; see Farudi Reference Farudi2005 and references contained therein) do not have this restriction. The internal argument of a complex predicate like otu zadan ‘to iron’ (lit. ‘iron’+‘to hit’) can be implicit, as in (37).

1. (37) giti otu zad. Giti iron hit.3sg ‘Giti ironed.’

A complex predicate's implicit object argument cannot be questioned with a truncated cleft, as shown in (38), though a sluice formed on it is fine.

1. (38) giti dāre otu mizane vali ne-midunam chi(*-e). Giti have.3sg iron hit.3sg but neg-know.1sg what-is ‘Giti is ironing, but I don't know what.’

The third piece of evidence that sluicing in Farsi is not derived from a truncated cleft comes from case restrictions on the pivot. The closest thing that Farsi has to case is the enclitic rā, illustrated in A1 of (39), which occurs on specific inanimate and all animate object DPs (it is, in other words, a differential object marker). Note that, while the citation form of this morpheme is rā, in colloquial speech it can be realized as o or ro depending on the identity of the final segment of the word to which it attaches. Phrases bearing rā can never be pivots, as shown by the ungrammaticality of A2.

1. (39)

   • Q: mahin ki-o daavat kard? Mahin who-obj invitation did.3sg ‘Who did Mahin invite?’

   • A1: sohrāb*(-o) daavat kard. Sohrab-obj invitation did.3sg ‘She invited Sohrab.’

   • A2: sohrāb(*-o) bud. Sohrab-obj was ‘It was Sohrab.’

In contrast, the remnant of a sluice can optionally be rā-marked, as in (40).

1. (40) mahin ye nafar-i-ro daavat karde vali be sohrāb ne-mige ki(-ro). Mahin one person-ind-obj invitation did.3sg but to Sohrab neg-say.3sg who-obj ‘Mahin invited someone, but she won't tell Sohrab who.’

The proper analysis of rā is a contentious issue (see, for instance, Karimi Reference Karimi1990, Dabir-Moghaddam Reference Dabir-Moghaddam and Stein1992, Ghomeshi Reference Ghomeshi1997b, Dalrymple & Nikolaeva Reference Dalrymple and Nikolaeva2005, and the references they contain for a variety of different approaches). I expect that, with further investigation, the optionality that it displays in the above example (40) will find an explanation.Footnote ⁶ For present purposes, it is enough that the distribution of rā is different in truncated clefts and sluicing.

Finally, the pivot position of a truncated cleft is restricted to DPs. As shown in (41), putting a PP in this position results in ungrammaticality.

1. (41)

   • Q: giti bā ki dāsht sohbat mikard? Giti with who had.3sg speaking did.3sg ‘Who was Giti speaking with?’

   • A: *bā sirus bud. with Cyrus was Intended: ‘It was with Cyrus that she was speaking.’

In contrast, PPs routinely serve as remnants in sluicing, as shown in (42).

1. (42) giti bā kesi dāsht sohbat mikard vali na-goft bā ki. Giti with someone had.3sg speaking did.3sg but neg-said.3sg with who ‘Giti was talking with someone, but she didn't say who.’

These four pieces of evidence make deriving sluicing in Farsi from a truncated cleft a difficult, if not impossible, proposition. There is, however, another clefting analysis that avoids many of the shortcomings of the previous one. Sluices might be derived from full clefts to which verb phrase ellipsis has applied. This analysis is represented schematically in (43).Footnote ⁷

1. (43)

   

This structure for full clefts comes largely from É. Kiss (Reference É. Kiss1998: 256–261). But while she analyzes the copula as the overt realization of F(ocus), I have made the more conservative assumption that it is a V. A surface structure that looks like a sluice is derived by eliding the VP. This deletes the copula and the cleft clause, leaving only the wh-pivot.

Two facts suggest that this analysis, too, is incorrect. First, I argue elsewhere (Toosarvandani, to appear) that, while Farsi has a species of verb phrase ellipsis, it does not apply to all types of verbs, but only to the language's complex predicates. This type of ellipsis deletes the phrasal complement of the light verb. In (44), the phrase headed by otu ‘iron’, the nonverbal half of the complex predicate, which contains the internal argument piran-o ‘shirt’, is elided, leaving behind the light verb zad ‘hit’.

1. (44) sohrāb piran-o otu na-zad vali rostam [_vP[_NP piran-o otu] zad]. Sohrab shirt-obj iron neg-hit.3sg but Rostam shirt-obj iron hit.3sg ‘Sohrab didn't iron the shirt, but Rostam did.’

The type of ellipsis involved in the analysis in (43) deletes a VP headed by a simplex verb, namely the copula, a type which is otherwise unattested in the language.

Second, recall that truncated clefts do not allow the pivot to bear the object marker rā. Full clefts behave similarly, as illustrated in (45). Sluicing does, however, allow the remnant to be rā-marked, as we saw in (40).

1. (45) *mahin ye nafar-i-ro daavat karde vali be sohrāb ne-mige ki-ro bud ke daavat karde. Mahin one person-ind-obj invitation did.3sg but to Sohrab neg-say.3sg who-obj was that invitation did.3sg Intended: ‘Mahin invited someone, but she won't tell Sohrab who it was that she invited.’

This contrast is essential to ruling out the verb phrase ellipsis analysis of Farsi sluicing, since none of the other diagnostics for truncated clefts applies to full clefts. Adjuncts can appear in the pivot of a full cleft (46), questions formed on the pivot can ask about the implicit argument of a preceding clause (47), and PPs are permitted in pivot position (48).

1. (46) rostam māshin-o taamir karde vali ne-midunam kei bud ke māshin-o taamir kard. Rostam car-obj repair did.3sg but neg-know.1sg when was that car-obj repair did.3sg ‘Rostam repaired the car, but I don't know when it was that he repaired the car.’

2. (47) giti dāre otu mizane vali ne-midunam chi-e ke dāre otu mizane. Giti have.3sg iron hit.3sg but neg-know.1sg what-is that have.3sg iron hit.3sg ‘Giti is ironing, but I don't know what it is that she is ironing.’

3. (48) giti bā kesi dāsht sohbat mikard vali na-goft bā ki bud ke dāsht sohbat mikard. Giti with someone had.3sg speaking did.3sg but neg-said.3sg with who was that had.3sg speaking did.3sg ‘Giti was talking with someone, but she didn't say with whom it was that she was talking.’

2.2 Are Farsi sluices derived by movement?

It seems, then, that Farsi sluicing cannot be assimilated to a cleft structure. There are numerous restrictions on the pivot of a cleft that simply do not hold of the remnant in a sluice. In this respect, Farsi sluicing patterns with its English analogue. There are a number of other parallels suggesting that sluicing should be analyzed in essentially the same way in both languages – as involving movement of the interrogative phrase to a left peripheral position followed by deletion of the rest of the clause.

A weak argument for syntactic movement of the remnant in sluicing, due to Merchant (Reference Merchant2001: 48–50), comes from its position with respect to the verb. While Farsi generally has SOV word order, CP arguments of the verb occur to the right (49). DP arguments – including, as in (50), CPs embedded under in ‘this’ – occur in the canonical preverbal position.

1. (49)

   1. (a) midunam [_CP ke sohrāb bastani-sh-o na-xorde]. know.1sg  that Sohrab ice.cream-his-obj neg-ate.3sg‘I know that Sohrab didn't eat his ice cream.’

   2. (b) *[_CP ke sohrāb bastani-sh-o na-xorde] midunam.  that Sohrab ice.cream-his-obj neg-ate.3sg know.1sg

2. (50) [_DP in-ro [_CP ke sohrāb bastani-sh-o na-xorde]] midunam. this-obj  that Sohrab ice.cream-his-obj neg-ate.3sg know.1sg ‘I know that Sohrab didn't eat his ice cream.’

The remnant in a sluice, too, can only occur to the right of the verb, as shown in (51).

1. (51)

   1. (a) sohrāb ye chiz-i xorde vali ne-midunam chi. Sohrab one thing-ind ate.3sg but neg-know.1sg what ‘Sohrab ate something, but I don't know what.’

   2. (b) *sohrāb ye chiz-i xorde vali chi ne-midunam. Sohrab one thing-ind ate.3sg but what neg-know.1sg

The parallel distribution of remnants and CP arguments of the verb follows directly from a movement-plus-deletion account. Since the CP from which the sluice in (51a) would be derived is positioned to the right of the verb, the remnant, too, would end up to the right. An alternative analysis, like that of van Riemsdijk (Reference Riemsdijk1978: 231–254), under which sluicing does not contain any deleted structure and the remnant chi ‘what’ is just a DP, predicts incorrectly that the remnant should occur where all other DP arguments occur, to the left of the verb.Footnote ⁸

The strongest evidence for movement comes from situations where the remnant in a sluice behaves just like its nonelliptical counterpart. Merchant (Reference Merchant2001: 89–107) discusses this class of facts under the rubric of form–identity generalizations. If, for instance, the interrogative phrase of a question bears a certain case, say accusative, then the remnant in the corresponding sluice should also bear accusative case. Even in English, a language lacking most inflectional morphology, this generalization holds. In the subject question of (52), accusative whom is not allowed regardless of whether or not the rest of the clause is pronounced.

1. (52) Somebody from Kankakee is going to be invited to the party by Ralph, but they don't know who/*whom (is going to be invited to the party by Ralph). (Ross Reference Ross1969: 254)

Farsi is also impoverished in its case morphology. The only candidate for case marking is the object marker rā, introduced in section 2.1, which appears on all animate and specific DPs in object position. It is thus obligatory on ki ‘who’ in (53). But in the corresponding sluice in (54), repeated from (40) above, the presence of rā on the remnant is merely optional.

1. (53) ki*(-o) mahin 〈ki-o〉 daavat karde? who-obj Mahin  invitation did.3sg ‘Who did Mahin invite?’

2. (54) mahin ye nafar-i-ro daavat karde vali be sohrāb ne-mige ki(-o). Mahin one person-ind-obj invitation did.3sg but to Sohrab neg-say.3sg who-obj ‘Mahin invited someone, but she won't tell Sohrab who.’

While this optionality is clearly unexpected under the movement-plus-deletion analysis of sluicing, it does not constitute an argument against it.Footnote ⁹ Granted, the movement-plus-deletion account will have to be augmented to account for the distribution of rā under sluicing – specifically, why rā can be absent on a wh-remnant that, in a nonelliptical clause, would require it – but, as far as I can see, such an effort must be made no matter what analysis one pursues. If instead the animate DP remnant in the sluice in (54) is base generated as the complement of the verb, then the normal case licensing mechanism will have to be prevented from always assigning rā.

The second form-identity generalization, involving preposition stranding, is more successful as a diagnostic for movement. If the remnant in sluicing arrives at its position by movement then it should obey the usual constraints on movement. If prepositions must normally be piedpiped, then when the correlate in a sluice is a PP, the remnant should be a PP as well. For languages that allow preposition stranding, we expect the reverse: it should be possible for a DP remnant to have a PP correlate. For the most part, this seems to be right (though see Almeida & Yoshida Reference Almeida and Yoshida2007 for a counterexample). In English, a preposition can be stranded in a regular question (55), as well as in a sluice (56).

1. (55) Who was Peter talking with 〈who〉?

2. (56) Peter was talking with someone, but I don't know (with) who. (Merchant Reference Merchant2001: 92)

Farsi is not a preposition-stranding language. If a wh-phrase is scrambled for information structure reasons to clause-initial position, the preposition must be piedpiped along with it (57a); stranding the preposition is severely ungrammatical (57b). The sluice in (58) is also grammatical only when the preposition of the remnant is piedpiped.Footnote ¹⁰

1. (57)

   1. (a) bā ki ali 〈bā ki〉 harf mizad? with who Ali  speech hit.3sg ‘Who was Ali talking with?’

   2. (b) *ki ali bā 〈ki〉 harf mizad? who Ali with  speech hit.3sg

2. (58) ali bā kesi harf mizad, ammā ne-midunam *(bā) ki. Ali with someone speech hit.3sg but neg-know.1sg with who ‘Ali was speaking with someone, but I don't know who.’ (Merchant Reference Merchant2001: 96)

There is one restriction on movement that has not been presented here: island constraints. This is because sluicing in English is famously able to void all sorts of island violations. In the Appendix, I show that sluicing in Farsi also does not obey island constraints. For reasons of space, I am not able to contribute here to the resolution of why, if sluicing is derived by movement, it is able to ignore island constraints (see Merchant Reference Merchant2001, Reference Merchant and Johnson2008 for extensive discussion).

If the remnant in Farsi sluicing gets to its position outside of the deleted phrase by movement, what kind of movement is it? In the next section, I argue that the syntactic operation responsible for fronting wh-phrases in sluicing contexts is associated with focus. This analytical connection will lead to an examination of the interpretative effects of this type of movement.

3.1 Farsi as a wh-in situ language

Farsi is usually said to be wh-in situ, and if one looks only at simple monotransitive sentences, this appears to be true. A declarative sentence with SOV word order like (59) can be questioned as in (60)–(61). Subject interrogative phrases occur in their normal sentence-initial position (60). Object interrogative phrases occur to the left of the verb but to the right of the subject (61).

1. (59) sohrāb moz-o xord. Sohrab banana-obj eat.3sg ‘Sohrab ate the banana.’

2. (60) ki moz-o xord? who banana-obj eat.3sg ‘Who ate the banana?’ subject question

3. (61) sohrāb chi-o xord? Sohrab what-obj eat.3sg ‘What did Sohrab eat?’ object question

Indirect objects have a more complicated distribution. Noninterrogative indirect object PPs can occur either to the left or the right of the verb (62).Footnote ¹¹ But, as shown in (63), the corresponding interrogative phrases only occur to the left of the verb.

1. (62)

   1. (a) hasan ketāb-o dād (be) ali. Hasan book-obj gave.3sg to Ali ‘Hasan gave the book to Ali.’

   2. (b) hasan ketāb-o be ali dād. Hasan book-obj to Ali gave.3sg ‘Hasan gave the book to Ali.’ (Kahnemuyipour Reference Kahnemuyipour2001: 47)

2. (63)

   1. (a) hasan ketāb-o be ki dād? Hasan book-obj to who gave.3sg ‘Who did Hasan give the book to?’

   2. (b) *hasan ketāb-o dād be ki? Hasan book-obj gave.3sg to who

Locative PPs exhibit a similar pattern. They can either precede or follow the verb (64), while their interrogative counterpart kojā ‘where’ is only found preverbally (65).

1. (64)

   1. (a) ali ketāb-o gozāsht ru miz. Ali book-obj put.3sg on table ‘Ali put the book on the table.’

   2. (b) ali ketāb-o ru miz gozāsht. Ali book-obj on table put.3sg ‘Ali put the book on the table.’ (Kahnemuyipour Reference Kahnemuyipour2001: 48)

2. (65)

   1. (a) ali ketāb-o kojā gozāsht? Ali book-obj where put.3sg ‘Where did Ali put the book?’

   2. (b) *ali ketāb-o gozāsht kojā? Ali book-obj put.3sg where

To account for these facts, Kahnemuyipour (Reference Kahnemuyipour2001) proposes that all interrogative phrases raise and adjoin to vP. While this movement is sometimes string vacuous, it ensures that all wh-phrases end up to the left of the verb. Under this account, Farsi is, strictly speaking, not wh-in situ, since interrogative phrases do not surface in the same position where they are merged; they undergo short-distance movement to Spec-vP.Footnote ¹²

This movement, however, is not enough to derive sluicing. In all the examples just given, the interrogative phrase, while not in its base position, is still lower in the structure than the subject. Assuming that subjects raise to Spec-TP,Footnote ¹³ the structure of a nonsubject constituent question under Kahnemuyipour's analysis can be given schematically as (66).

1. (66)

   

The structure in (66) cannot serve as the input to sluicing since there is no constituent that contains everything in the clause except the wh-phrase. Specifically, since the subject is in a structurally superior position, if sluicing targets the sister of the wh-phrase, then we predict – falsely – that the subject will always be stranded. For our purposes, then, Farsi is effectively a wh-in situ language.Footnote ¹⁴

This is not to say that interrogative phrases are fixed in place. They undergo the same information-structure-driven movement processes that noninterrogative phrases do. It is one such process – focus fronting – that I will argue is responsible for moving the remnant to a position where it can be stranded in sluicing.

3.2 The syntax of focus fronting

Major sentence constituents in Farsi are subject to scrambling for information structure reasons. In one type of scrambling, which I call focus fronting, a phrase fronts to a clause-initial position where it receives a pitch accent (indicated with capitalization), as shown in (67).

1. (67) giti midune ke pesTE sohrāb 〈peste〉 xaride.Giti know.3sg that pistachio Sohrab bought.3sg‘Giti knows that Sohrab bought pistachios.’

Following Karimi (Reference Karimi2005: 131–158). I assume that the object DP in this example, peste ‘pistachios’, raises to the specifier of a dedicated focus projection, Spec-FP. This focus projection is located above T but below C:

1. (68) giti midune …

   

Focus-fronted elements thus end up sandwiched between the subject and the complementizer ke.

Evidence that ke is, in fact, a complementizer comes from two facts. First, ke always occurs to the left of all other elements in the clause. This is what we expect if, as the overt realization of C, it heads the clause.Footnote ¹⁵ Second, in accordance with they way Rosenbaum (Reference Rosenbaum1965: 41) originally defined the category of complementizer, ke is found only in subordinate clauses – e.g. sentential complements (67) or relative clauses (69) – but not in matrix clauses (70).Footnote ¹⁶

1. (69) rostam az māshin-i [ke sohrāb xarid] xosh-esh miyād.Rostam from car-rel that Sohrab bought.3sg happy-his comes.3sg‘Rostam likes the car that Sohrab bought.’

2. (70) *ke sohrāb māshin-esh-o furuxt.that Sohrab car-his-obj sold.3sgIntended: ‘Sohrab sold his car.’

3.3 The semantics of focus fronting

With a syntax for focus fronting in hand, we can now turn to its semantics. A proper exposition of how all focus-fronted elements are interpreted would require more space than is available, so I confine my discussion here to how interrogative phrases are interpreted in Spec-FP, since it is interrogative phrases that are relevant to sluicing.Footnote ¹⁷

Consider the questions in (71)–(73). The interrogative phrases in these examples have raised to a position above the subject where they receive a pitch accent, a position that I have identified as Spec-FP. Intuitively, these fronted interrogative phrases are interpreted as standing in a contrastive relationship with another phrase in the preceding clause.

1. (71) midunam ke sohrāb ye ketāb xarid vali ne-midunam CHE ketāb-i-ro sohrāb 〈che ketāb-i-ro〉 xarid.know.1sg that Sohrab one book bought.3sg but neg-know.1sg what book-ind-obj Sohrab  bought.3sg‘I know that Sohrab bought a book, but I don't know what book he bought.’

2. (72)

   1. (A) ne-midunam sohrāb che romān-i-ro dust dāre.neg-know.1sg Sohrab what novel-ind-obj friend have.3sg‘I don't know what novel Sohrab likes.’

   2. (B) na, man az shomā porside budam che FILM-i-ro sohrāb 〈che film-i-ro〉 dust dāre.no I from you asked was.1sg what movie-ind-obj Sohrab  friend have.3sg‘No, I had asked you what movie he likes.’

3. (73) midunam sohrāb vis-o kojā mixād shām bebare vali yād-am nist KEI mixād vis-o 〈kei〉 beresune xune.know.1sg Sohrab Vis-obj where want.3sg dinner take.3sg but memory-my is.neg when want.3sg Vis-obj make.arrive.3sg home‘I know where Sohrab wants to take Vis to dinner, but I don't remember when Sohrab wants to bring Vis home.’

In (71), the determiner of che ketābi-ro ‘what book’ contrasts with the determiner of ye ketāb ‘a book’. In (72), the restriction of che filmi-ro ‘what book’ contrasts with the restriction of che romāni-ro ‘what novel’. In (73), the entire interrogative phrase kei ‘when’ contrasts with kojā ‘where’.

If the focus-fronted interrogative phrase, or some part of it, must be contrastive, then we expect that focus fronting will be infelicitous in out-of-the-blue linguistic contexts where there is nothing for the interrogative phrase to contrast with. This seems to be correct. When the focus-fronted question in (74a) is uttered without any preceding discourse, it is infelicitous. The same question with neutral word order is fine (74b).

1. (74)

   1. (a) #CHI sohrāb 〈chi〉 āvord?what Sohrab  brought.3sg‘What did Sohrab bring?’

   2. (b) sohrāb chi āvord?Sohrab what brought.3sg

The obligatory contrastive focus on fronted interrogative phrases can be modeled formally using Rooth's (Reference Rooth1985, Reference Rooth1992) alternative semantics. As a warning to the reader, the machinery of Rooth's theory may seem a bit excessive at this point for the task at hand, but the technical implementation of focus fronting developed below is an essential prerequisite to the discussion in section 4.2.

In Rooth's semantics for focus, all natural language expressions have two semantic values: an ordinary semantic value provided by the interpretation function 〚.〛° and a focus semantic value given by the focus interpretation function 〚.〛^f. When an expression does not contain a focus, its focus semantic value is simply the set containing its ordinary semantic value. Thus, the focus semantic value of Mary likes Sue is 〚Mary likes Sue〛^f={like(sue)(mary)}, or the set containing the proposition that Mary likes Sue. When a focus is present, the focus semantic value is derived by making a substitution in the place marked by focus. For MARY likes Sue, the focus semantic value is 〚MARY likes Sue〛^f={p|∃x[p=like(sue)(x)∧x∊D _e]}, or the set of propositions of the form x likes Sue, where x is in the domain of entities.

The focus semantic value of an expression is always present alongside the ordinary semantic value. By itself, though, it does not enter into the truth conditions of the sentence. Focus semantic values are used by a focus interpretation operator, ~, which for Rooth is the only semantic object able to make reference to focus values. The ~ operator is adjoined freely at LF, taking a focus in its scope (we can also call the scope of a ~ operator its domain). The operator makes reference to focus semantic values through a presupposition relating its two arguments: the phrase φ to which it is adjoined and a free variable, either a set &Ggr; or an individual γ. The presupposition that the focus interpretation operator introduces is given in (75).

1. (75)

   1. (a) Set case. φ~&Ggr; presupposes that &Ggr; is a subset of the focus semantic value for φ and contains both the ordinary semantic value of φ and an element distinct from the ordinary semantic value of φ.

   2. (b) Individual case. φ~γ presupposes that γ is an element of the focus semantic value for φ distinct from the ordinary semantic value of φ. (Rooth Reference Rooth1992: 93)

Setting aside momentarily the question of precisely how the free variable gets its value, from the presupposition in (75) we already know something about what this value must be. The free variable's value will be either: (1) a subset of the focus semantic value of φ that contains not only φ but something else as well; or (2) a member of the focus semantic value of φ that is distinct from φ itself. The presupposition is stated disjunctively in order to unify the interpretation of different kinds of focus structures, including contrastive focus, the focus that shows up in question-answer pairs, and the focus that is associated with adverbs like only. While Rooth (Reference Rooth1992: 90f.) suggests a way of getting rid of this disjunction, I will leave the definition as is for reasons of concreteness. Only the (b) disjunct (the individual case) comes into play in the course of this paper.

Focus fronting is defined by the adjunction of a focus interpretation operator ~ to the element in Spec-FP. To show how the focus structure of a fronted wh-phrase is derived, I give a partial LF structure for (76), repeated from (71) above, in (77).

1. (76) midunam ke sohrāb ye ketāb xarid vali ne-midunam CHE ketāb-i-ro sohrāb 〈che ketāb-i-ro〉 xarid.know.1sg that Sohrab one book bought.3sg but neg-know.1sg what book-ind-obj Sohrab  bought.3sg‘I know that Sohrab bought a book, but I don't know what book he bought.’

1. (77)

   

A focus interpretation operator is adjoined to DP₃ in Spec-FP, the interrogative phrase CHE ketābi-ro ‘what book’, which contrasts with DP₂, the indefinite ye ketāb ‘a book’. By the presupposition in (75b), DP₂ must therefore be a member of the focus semantic value of DP₃. Since it is the interrogative determiner of DP₃ that bears a pitch accent, the focus semantic value of the entire phrase is obtained by a making a substitution in the position of the determiner. Thus, 〚CHE ketābi-ro〛^f={P|∃[P=(book)]}, or the set of expressions of the same type as an interrogative phrase whose restriction is book. P is a variable of the type of interrogative phrases, and is a variable of the type of interrogative determiners.

I have left the types for these variables unspecified since giving appropriate denotations for the indefinite and interrogative determiners is a significant challenge. In Rooth's theory of focus, in order for the focused interrogative determiner in (76) to contrast with the indefinite determiner in the antecedent clause, the two must be of the same type. Romero (Reference Romero1998: 29–36) gives denotations for which and how many in the domain of determiners, 〈〈e, st〉, 〈〈e, st〉, 〈st〉〉〉, such that their alternatives include one another as well as an existential option. This is sufficient to account for the example in (76), but Romero's account must be expanded in order to account for the contrastive relationships that the other wh-phrases enter into (e.g. who, when, where).

Kratzer and Shimoyama (Reference Kratzer, Shimoyama and Otsu2002) and Kratzer (Reference Kratzer, Carlson and Pelletier2005) provide another option within a Hamblin semantics, in which indefinites and interrogative phrases both denote sets of individuals. Other semantic objects denote sets of traditional denotations. Function application occurs pointwise: a functor taking an indefinite or interrogative phrase as its argument applies to each of the individuals in the set denoted by these expressions. At the sentence level, this schema produces a set of propositions to which operators of the desired force – question, existential, etc. – can apply.

Whatever semantics for interrogative phrases and indefinites one chooses, the data presented here require that they be alternatives to one another. In the context of (76), this means that 〚ye ketāb〛°=a(book) must be in 〚CHE ketābi-ro〛^f.

3.4

In this section, I have argued for the existence in Farsi of a process of focus fronting. When an interrogative phrase raises to Spec-FP, it must stand in a contrastive relationship with another phrase of the same type. This understanding of the syntax and semantics of focus fronting will be of use in the next section, where I argue that focus fronting is responsible for moving the remnant out of the deleted constituent in Farsi sluicing.

4.1 The proposal

All the pieces we need to derive sluicing in Farsi are now in place. I propose that it proceeds as follows: first, an interrogative phrase undergoes focus fronting to Spec-FP; then, the sister of F, TP, which contains the rest of the clause including the subject, is deleted (at PF). As illustrated in (78), this produces the correct surface string. The proposal is shown schematically in (79).

1. (78) rāmin ye chiz-i xaride. hads bezan [_FP chi [_TP rāmin 〈chi〉 xaride]].Ramin one thing-ind bought.3sg guess hit.2sg  what Ramin  bought.3sg‘Ramin bought something. Guess what.’

1. (79)

   

The primary syntactic evidence that the remnant in Farsi sluicing is in Spec-FP comes from its position with respect to the complementizer. Recall from section 3.2 that the focus projection is located above TP but below CP. If sluicing involves deletion of TP, then we expect that the complementizer ke should be able to appear in a sluice. This expectation is borne out, as shown in (80)–(81).Footnote ¹⁸

1. (80) mahin mixād ye chiz-i bexare vali yād-esh ne-miyād ke chi.Mahin want.3sg one thing-ind buy.3sg but memory-her neg-come.3sg that what‘Mahin wants to buy something, but she doesn't remember what.’

2. (81) bābā-m injā nist. xod-et miduni ke cherā.dad-my here neg.is self-your know.2sg that why‘My dad isn't here. You yourself know why.’ (Zire Tiq (Iranian television serial), 22 January 2007)

In both examples, ke occurs in its normal position to the left of the remnant.

The presence of the complementizer in sluicing is perhaps a bit surprising. Merchant (Reference Merchant2001: 61–82) shows that a wide variety of languages do not allow elements in C – e.g. complementizers, verbs, clitics, agreement morphology – to occur in a sluice. In some dialects of Dutch, for example, complementizers, which can otherwise cooccur with an interrogative phrase in Spec-CP (82), are excluded in a sluice (83).

1. (82) Ik weet niet, wie (of) (dat) hij gezien heeft.I know not who if that he seen has‘I don't know who he has seen.’

2. (83) Hij heef iemand gezien, maar ik weet niet wie (*of) (*dat).he has someone seen but I know not who if that‘He saw someone, but I don't know who.’ (Merchant Reference Merchant2001: 74f.)

Merchant captures this observation in the sluicing-comp generalization, which he states as follows: ‘In sluicing, no non-operator material may appear in COMP’ (62).

Farsi constitutes a prima facie counterexample to this generalization, though it is not alone in this respect. Merchant offers his own counterexample from Hungarian, which, as illustrated in (84), allows the complementizer hogy to appear optionally in a sluice.

1. (84) A gyerekek találkoztak valakivel de nem emlékszem, (hogy) kivel.the children met someone.with but not I.remember that who.with‘The kids met someone, but I don't remember who.’ (Merchant Reference Merchant2001: 82)

The unexpected behavior of complementizers in Hungarian and Farsi may derive from a shared property of the two languages. While Hungarian is not a wh-in situ language like Farsi, wh-movement is not to Spec-CP as in English. Rather, interrogative phrases obligatorily raise to a focus projection located below the complementizer (Horvath Reference Horvath1986: 44–51, É. Kiss Reference Katalin1987: 56–61).

The conclusion that emerges is that the sluicing-COMP generalization holds only when the remnant of the sluice is in Spec-CP. For languages that do not involve the C domain in sluicing, the generalization simply does not hold. The analogous constraint for Hungarian and Farsi would ban the overt reflex of F from occurring in a sluice. Unfortunately, since there is no overt realization of F in Farsi, we cannot test this hypothesis.Footnote ¹⁹

4.2 The focus structure of sluicing

If the remnant in Farsi is situated in Spec-FP, we expect it to exhibit the same phonological and semantic properties as a focus-fronted interrogative phrase in nonelliptical contexts. This seems generally to be correct. Just like the nonelliptical examples of focus fronting in (71)–(73), the remnants in (85)–(87) all bear a pitch accent.Footnote ²⁰

1. (85) man midunam ke sohrāb ye ketāb xaride va rāmin midune CHE ketāb-i.I know.1sg that Sohrab one book bought.3sg and Ramin know.3sg what book-ind‘I know that Sohrab bought a book, and Ramin knows what book.’

2. (86) sohrāb be man goft che ketāb-i-ro dust dāre vali na-goft che FILM-i-ro.Sohrab to me said.3sg what book-ind-obj friend have.3sg but neg-said.3sg what movie-ind-obj‘Sohrab told me what book he likes, but he didn't say what movie.’

3. (87) faqat midunam kojā sohrāb dustdoxtar-esh-o did; ne-midunam KEI.only know.1sg where Sohrab girlfriend-his-obj saw.3sg neg-know.1sg when‘I only know where Sohrab saw his girlfriend; I don't know when.’

In (85), the interrogative determiner of the remnant is in a contrastive relationship with the determiner of its correlate. In (86), the restriction of the remnant contrasts with the restriction of its correlate. In (87), the entire remnant contrasts with its correlate.Footnote ²¹ Leaving off the pitch accent on the remnants in these examples results in ungrammaticality, as shown in (88)–(90).

1. (88) *man midunam ke sohrāb ye ketāb xaride va rāmin midune che ketāb-i.I know.1sg that Sohrab one book bought.3sg and Ramin know.3sg what book-ind

2. (89) *sohrāb be man goft che ketāb-i-ro dust dāre vali na-goft che film-i-ro.Sohrab to me said.3sg what book-ind-obj friend have.3sg but neg-said.3sg what movie-ind-obj

3. (90) *faqat midunam kojā sohrāb dustdoxtar-esh-o did; ne-midunam kei.only know.1sg where Sohrab girlfriend-his-obj saw.3sg neg-know.1sg when

There is a class of sluices, however, that seem not to bear out this prediction. Consider the examples in (91)–(92). The remnants in these sluices do not bear a pitch accent, which is the usual phonological realization of focus.

1. (91) midunim che ketabā-i-ro sohrāb xaride va rāMIN ham midune che ketāb-i.know.1pl what book.pl-ind-obj Sohrab bought.3sg and Ramin also know.3sg what book-ind‘We know what books Sohrab bought, and Ramin also knows what books.’

2. (92) mā midunim sohrāb chandtā ketāb xaride vali rāMIN hanuz ne-midune chandtā.we know.1pl Sohrab how.many book bought.3sg but Ramin yet neg-know.3sg how.many‘We know how many books Sohrab bought, but Ramin doesn't yet know how many.’

Nonetheless, the remnants in these examples are perceptually distinct from surrounding material. Impressionistically, they are louder, indicated here with italics. I argue that the remnants in (91)–(92), while lacking pitch accents, do indeed contain foci, though not of the ordinary kind. They are instances of what is known as second occurrence focus (Partee Reference Partee, Moore and Wyner1991, Reference Partee, Bosch and Sandt1999; Rooth Reference Rooth1992, Reference Rooth, Galloway and Spence1996; Hajičcová, Partee & Sgall Reference Hajičová, Partee and Sgall1998; and much subsequent work).

In certain contexts, foci do not receive a canonical phonological realization with a pitch accent. Consider the English example in (93).

1. (93) Our grad students only quote the FAculty_F. No, the UNdergrads_F only quote the faculty _F. (modified from Büring Reference Büring2006: 7)

The adverb only is focus sensitive, associating with a focused element somewhere in its scope. In the first sentence of (93), only is associated with the faculty, which bears a pitch accent as expected. (The sentence expresses universal quantification over the people who the grad students quote.) I have marked the fact that it is a focus with a subscripted F. In the second sentence of (93), only occurs another time, again associating with the faculty. This is the second occurrence focus, which is not realized with a pitch accent like a canonical focus, but rather with increased energy (it is louder) and increased duration (Rooth Reference Rooth, Galloway and Spence1996, Bartels Reference Bartels, Kamp and Partee2004, Beaver et al. Reference Beaver, Clark, Flemming, Jaeger and Wolters2007). As above, I indicate this type of phonological realization with italics.

While the formal source of second occurrence focus is still obscure, the environment that licenses it is well understood. Building on a proposal by Rooth (Reference Rooth, Galloway and Spence1996), Büring (Reference Büring2006), argues that whether or not a focus will be realized as a second occurrence focus is determined by the principle in (94).

1. (94) Domain theory of primacy

   Among two foci in a sentence, the primary focus is the focus whose domain contains the domain of the other. (Büring Reference Büring2006: 8)

In other words, for a sentence that contains two foci, the primary focus, realized with a pitch accent, is the one whose domain is larger and contains the domain of the other focus, which is consequently realized as a second occurrence focus. The relevant notion of ‘domain’ here is the same as the scope of one of Rooth's ~ operators (see section 3.3).

Büring's account correctly derives the focus structure of the second sentence of (93), which is repeated in (95) with bracketing to mark focus domains. Only is identified with a focus operator, ~₄, that takes the verb phrase in its scope and is associated with the focus on the faculty (indicated through coindexation).

1. (95) [No, the UNdergrads_F3 only [quote the faculty _F4]~₄]~₃.

But there is a larger focus domain, that of ~₃, which takes scope over the entire sentence and is associated with new information. The specific conception of newness that Büring adopts is that embodied in Schwarzschild's (Reference Schwarzschild1999) definition of givenness:

1. (96) givenness

   An utterance U counts as given iff it has a salient antecedent A and

   1. (i) if U is of type e, then A and U corefer;

   2. (ii) otherwise: modulo ∃-type shifting, A entails the existential F-closure of U. (Schwarzschild Reference Schwarzschild1999: 151)

The root level operator, ~₃ in (95), associates with all nongiven material in the sentence, namely the undergrads. Any foci not associated with ~₃ are accordingly given. It follows from this that second occurrence foci are always given. In (95), the faculty is only associated with ~₄, whose domain is contained within the domain of ~₃. If the faculty were made nongiven by associating it with the root level operator, the principle in (94) would require that it be realized as a primary focus. The foci on the faculty and the undergrads would share a single domain, that of ~₃.

Crucially, Büring assumes that a single focus can associate with more than one ~ operator. This happens when the focus of a smaller domain is nongiven information, as in the first sentence of (93), repeated in (97).

1. (97) [Our grad students only quote [the FAculty_F1,2]~₂]~₁.

The faculty here is new information, which must be associated with the root level focus operator, ~₃, as well as with the operator identified with only, ~₂.

In the rest of this section, I show that the recalcitrant Farsi example in (92), repeated in (98) below, has a focus structure isomorphic to that of (95). In section 3.3, I proposed that Spec-FP constitutes its own focus domain with a ~ operator adjoined to the phrase that fills it. The remnant of the sluice, chandtā ‘how many’, which by hypothesis is situated in Spec-FP, is thus associated with ~₂. But the subject of the matrix clause, rāmin ‘Ramin’, also bears a focus that is associated with the root level focus operator ~₁.

1. (98) mā midunim sohrāb chandtā ketāb xaride vali [rāMIN_F1 hanuz ne-midune [chandtā _F2] ~₂]~₁.we know.1pl Sohrab how.many book bought.3sg but Ramin yet neg-know.3sg how.many‘We know how many books Sohrab bought but Ramin doesn't yet know how many.’

The configuration in (98) is precisely the one that licenses second occurrence focus. According to the definition in (94), the focus on rāmin ‘Ramin’ is realized as a primary focus with a pitch accent since its domain contains the domain of chandtā ‘how many’, which gets a noncanonical realization without a pitch accent. Sentences like (98), instead of presenting a problem for deriving sluicing in Farsi by focus fronting, thus constitute a strong argument for it. Büring's account of second occurrence focus only works for these examples if the remnant constitutes its own focus domain by being situated in Spec-FP.

Before moving on, I should mention that the literature on second occurrence focus has generally concentrated on foci associated with focus-sensitive adverbs like only (as in the original example in (93)). Büring's theory, which defines the licensing environment for second occurrence focus in terms of Roothian ~ operators, predicts that the phenomenon should not be restricted in this way. Any time a focus domain is contained within another, larger domain, a noncanonical realization of the smaller domain's focus should be possible, regardless of whether it is associated with an adverb or not. Contrastive foci, for instance, should be able to be realized as second occurrence foci given the right conditions. The adjectives in (99a) bear pitch accents since they contrast with each other. Rooth (Reference Rooth1992: 79–82) analyzes such cases as involving a focus interpretation operator adjoined to the DPs containing the adjectives. Since both green and red are nongiven in this example, they receive a canonical realization with a pitch accent. In the continuation in (99b), a similar contrastive focus structure is set up, but, while blue is new, red is given from the preceding sentence. It is not associated with the root ~ operator, thereby satisfying the condition in (94) for being a second occurrence focus.

1. (99)

   1. (a) OK, [so I'll press [the GREEN_F1,3 button] ~₁ when [the RED_F2,4 button] ~₂ starts blinking] ~_3,4.

   2. (b) No, [you press [the BLUE_F5,7 button] ~₅ when [the red_F6 button] ~₆ starts blinking] ~₇. (Büring Reference Büring2006: 17)

Büring's intution, which I share, is that red is indeed realized as a second occurrence focus, without a pitch accent but with greater prominence. If true, this suggests that second occurrence focus in English is not restricted to occuring only with focus-sensitive adverbs. I have made a parallel argument for focus fronting in Farsi. The focus on the element in Spec-FP, which can be realized as a second occurrence focus given the right conditions, is also not associated with a focus-sensitive adverb.

4.3

The preceding section has been an effort to understand how the interrogative remnant in Farsi sluicing escapes deletion. This happens, I have argued, by an operation of focus fronting. The question remains why this movement happens at all since, in contrast to English wh-movement, Farsi focus fronting is optional. Said another way, what is the reason for the contrast in (100)?

1. (100) rāmin ye chiz-i xaride.Ramin one thing-ind bought.3sg‘Ramin bought something.’

   1. (a) hads bezan [_TP rāmin chi xaride].guess hit.2sg  Ramin what bought.3sg‘Guess what Ramin bought.’

   2. (b) *hads bezan [_TP rāmin chi xaride].guess hit.2sg Ramin what bought.3sgIntended: ‘Guess what.’

The wh-phrase chi ‘what’ does not have to front in the nonelliptical clause in (100a). Leaving it in situ is ungrammatical, however, if TP is deleted as in (100b). In the next section, I propose that a formal property of sluicing itself forces the interrogative phrase to front.