Jack bought something, but I don’t know [CP whati [
IP Jack bought ti]].
Jack talked to someone, but I don’t know [CP whoi [
IP Jack talked to ti]].
Preposition Stranding Generalization (PSG)
A language L will allow preposition stranding under sluicing iff L allows preposition stranding under regular wh-movement.
The PSG is demonstrated in (1b), where the wh-sluice ‘who’ leaves a stranded preposition under sluicing, which corresponds to the fact that English is a P-stranding language. Merchant further demonstrates the descriptive power of the PSG by verifying its applicability to more than twenty languages. Examples drawn from other languages continue to confirm its validity (e.g., Almeida and Yoshida 2007, Stjepanović 2008, Rodrigues, Nevins, and Vicente 2009, Van Craenenbroeck 2010).
In this squib, I investigate Emirati Arabic (EA) in detail and argue that it provides cases in which the PSG can be falsified. In EA, while P-stranding is banned in wh-questions, sluicing is possible even when the underlying structure would contain a stranded preposition. For example:
John ʃərab gahwa [wɪjja ħəd], bəs maa ʕrf [mənu
John ʃərab gahwa [PP wɪjja ti]].
John drank coffee with someone but not 1.know who John drank coffee with
‘John drank coffee with someone, but I don’t know who.’
Potential counterexamples to the PSG have been adduced from other languages, yet further analyses reveal that they do not involve Pstranding by wh-movement (e.g., Brazilian Portuguese, French (Rodrigues, Nevins, and Vicente 2009), Serbo-Croatian (Stjepanović 2008)). At first glance, EA might seem to be one such case since it possesses two types of wh-constructions: wh-fronting (a movement construction) and wh-clefts (a nonmovement construction). For the sake of uniformity, I call the elliptical wh-construction formed by wh-fronting sluicing and the one formed by wh-clefts pseudosluicing (Merchant 2001).
In this squib, I defend four claims. First, EA allows both sluicing and pseudosluicing. Second, sluicing and pseudosluicing are distinguished by individual lexical and morphosyntactic properties, on the one hand, and the syntactic projection of the antecedent clause, on the other hand. Third, the PSG is falsified even though both sluicing and pseudosluicing are at work. Fourth, in order to preserve the original insight of the PSG, its statement should be modified so that for languages in which the P-stranding constraint is defined at the level of PF, violations can be rescued by sluicing as a result of PF deletion. That is to say, any language that parameterizes the P-stranding constraint under wh-movement as a PF condition can salvage P-stranding violations via sluicing as PF deletion.
2 Two Types of Wh-Constructions in Emirati Arabic
ʃuui ʃtər-eet ti ʔms?
what bought-2SM yesterday
‘What did you buy yesterday?’
ʃuui (hu) εlli ʃtər-eet ti ʔms?
what 3SM that bought-2SM-3SM yesterday
‘What was it that you bought (it) yesterday?’
The two wh-constructions have been documented in various Arabic dialects (e.g., Wahba 1984, Shlonsky 1997, 2002, Aoun, Benmamoun, and Choueiri 2010). Wh-fronting leaves a movement gap (i.e., ti in (4a)), whereas wh-clefts are a nonmovement type of wh-dependency that requires a resumptive pronoun (e.g., ah in (4b)) at the base position, along with the relative complementizer εlli ‘that’.2 Moreover, wh-clefts allow an optional copular pronoun hu (as in (4b)), which signals a cleft structure (Eid 1983).3Wh-fronting is more productive than wh-clefts in that the former cooccurs with any type of wh-expression: for example, wh-words and wh-phrases ((4), (5a)), including wh- PPs (5c), as well as wh-arguments and wh-adjuncts (5b). P-stranding by wh-movement is strictly ungrammatical (5d). On the other hand, wh-clefts only allow the use of bare wh-words and wh-arguments (4b). They strictly ban the use of which-NPs (6a), wh-adjuncts (6b), and wh-PPs (6c).
ʔaj kitab ʃtər-ə ʔms?
which book bought-2SM yesterday
‘Which book did you buy yesterday?’
kεɪf ʕxalʕlʕasʕ-t əl-wadӡeb how finished-2SM the-assignment
‘How did you finish the assignment?’
f-ʔaj mFkaan laag-ət John?
at-which place met-2SM John
‘At which place did you meet John?’
*ʔaj mFkaan laag-ət John fi?
which place met-2SM John at
‘Which place did you meet John at?’
*ʔaj kitab (hu) εlli ʃtər-ət-ah ʔms?
which book 3SM that bought-2SM-3SM yesterday
‘Which book is it that you bought yesterday?’
*kεɪf (hu) εlli j-xalləsʕ… John fi-ha əl-wadӡeb?
how 3SM that 3SM-finish John in-3SM the-assignment
‘How is it that John finishes the assignment?’
*f-ʔaj mʊkaan (hu) εlli laag-et John fi-h?
at-which place 3sm that met-2SM John at-3SM
‘At which place is it that you met John?’
These distinctions will be crucial in identifying the underlying source of sluicing/pseudosluicing.
3 Emirati Arabic Sluicing and Pseudosluicing
Given the two types of wh-constructions, we can distinguish between the use of a wh-sluice (as derived by wh-fronting) and the use of a wh-pseudosluice (as derived by wh-clefts). The examples in (7) show that any type of wh-expression can form a bare wh-sluice. The use of a wh-pseudosluice, signaled by the copular pronoun hu, is grammatical in limited cases (e.g., (7a)).4
John ʃərab ʃaj, bəs maa ʕərf [ʃuu (hu)].
John drank something but not 1.know what 3SM
‘John drank something, but I don’t know what.’
John laaga Mary, bəs maa ʕərf [kεɪf (*hu)].
John met Mary but not 1.know how 3SM
‘John met Mary, but I don’t know how.’
John jəʃrəb xamər, bəs maa ʕərf [ʔaj nooʕ (*hu)].
John drink alcohol but not 1.know which kind 3SM
‘John drinks alcohol, but I don’t know which kind.’
John jəʃrəb xamər, bəs maa ʕərf [wɪjja mənu (*hu)].
John drink alcohol but not 1.know with who 3SM
‘John drinks alcohol, but I don’t know with who.’
Given the distinctive properties of wh-fronting and wh-clefts as listed above, we can immediately identify the underlying source of some of the instances of sluicing in (7). For instance, since wh-adjuncts (e.g., kεɪf ‘how’), which-NPs (e.g., ʔaj nooʕ ‘which kind’), and wh-PPs (e.g., wɪjja mənu ‘with who’) can only be used in wh-fronting (see (5a–c)), the underlying source of (7b–d) must be wh-fronting. This is further confirmed by the ungrammaticality of the use of a wh-pseudosluice in expressions such as *kεɪf hu? ‘How is it?’, *ʔaj nooʕ hu? ‘Which kind is it?’, and *wɪjja mənu hu? ‘With who is it?’. The use of the bare wh-expression ʃuu ‘what’ in (7a), however, does not provide a clear indication of its underlying source. (7a) can be either a wh-sluice, or a wh-pseudosluice in which the copular pronoun hu is deleted, as shown in (8).
John ʃərab ʃaj, bəs maa ʕərf [ʃuu
John drank something but not 1.know what 3SM
‘John drank something, but I don’t know what it is.’
As a result, argument wh-NPs such as ʃuu ‘what’ or mənu ‘who’ do not allow a clear identification of the sluicing source.
4 Falsifying the Preposition Stranding Generalization
Given the distinction between sluicing (formed by wh-fronting) and pseudosluicing (formed by wh-clefting), let us look again at the PSG. Example (3) neither supports nor falsifies the PSG since bare wh-NPs (e.g., ʃuu ‘what’ and mənu ‘who’) can be cases of either wh-sluicing or wh-pseudosluicing. Now consider the examples in (9), which are more convincing.
John ʃərab gahwa [wɪjja sʕadiq], bəs maa ʕərf [ʔaj sʕadiq].
John drank coffee with friend but not 1.know which friend
‘John drank coffee with a friend, but I don’t know which friend.’
John kətab baħθ-ah [f-kʊmbjutər], bəs maa ʕərf [ʔaj kʊmbjutər].
John wrote research-his in-computer but not 1.know which computer
‘John wrote his research on a computer, but I don’t know which computer.’
Recall that wh-sluices formed by which-NPs must be derived by whfronting. 5 In (9a–b), the use of which-NPs in the second clause in the absence of a preposition is fully grammatical, suggesting that (9a–b) can only be the result of wh-movement of the which-NP, which strands a preposition, followed by IP-deletion. This is shown in (10).
John ʃərab gahwa [wɪjja sʕadiq], bəs maa ʕərf [[ʔaj sʕadiq]i
John ʃərab gahwa [PP wɪjja ti]].
John drank coffee with friend but not 1.know which friend John drank coffee with
‘John drank coffee with a friend, but I don’t know which friend.’
The observation can be stated as in (11), which runs counter to the PSG.
(11) Emirati Arabic allows P-stranding under sluicing but not under regular wh-movement.
One issue is whether the PSG is falsified because of (9), or whether the PSG should be modified to allow for this counterexample.
One option is to reformulate the PSG as follows, making reference to the level at which P-stranding violations are defined:6
(12) A language L will allow preposition stranding under sluicing, even though it may not allow it under wh-movement, iff in L, preposition stranding violations are determined at PF.
In fact, EA may not be the first language that shows that P-stranding violations can be rescued by sluicing. Sato (2011) claims that sluicing can also rescue P-stranding violations in Indonesian. He argues that Indonesian obligatorily requires the [+wh] feature to percolate to PP (e.g., ‘to whom’), thus banning P-stranding by wh-movement. He suggests that violation of P-stranding in Indonesian is defined at the level of PF and can be repaired by sluicing.7
5 The Effect of the Antecedent Correlate on the Choice of the Sluicing Source
So far, we have looked into cases in which wh-fronting and wh-clefts differ in terms of structural descriptions and numeration. That is to say, the use of either wh-construction is legitimate as long as individual morphosyntactic properties are met (e.g., (5)–(7)). However, this does not suffice to account for the distinction between the formation of whsluices and the formation of wh-pseudosluices, if the antecedent clause is taken into account. In (13a–b), the antecedent clause contains an implicit argument (indicated by the underscore).8 The use of either wh-construction in the second clause is grammatical.
John j-suug ____, bəs maa ʕərf [ʃuu j-suug].
John 3SM-drive but not 1.know what 3SM-drive
‘John drives, but I don’t know what he drives.’
John j-suug ____, bəs maa ʕərf [ʃuu (hu) εlli j-suug-ah].
John 3SM-drive but not 1.know what 3SM that 3SM-drive-3SM
‘John drives, but I don’t know what it is that he drives (it).’
(14) shows that the use of the wh-sluice ʃuu ‘what’ is grammatical. The wh-pseudosluice indicated by ʃuu hu ‘what it is’, however, is strictly banned.
John j-suug ____, bəs maa ʕərf [ʃuu (*hu)].
John 3SM-drive but not 1.know what 3SM
‘John drives, but I don’t know what (it is).’
The contrast between (13) and (14), on the one hand, and between (8) and (14), on the other hand, is puzzling if we assume that a wh-sluice/wh-pseudosluice is the elided outcome of wh-fronting/whclefts. Comparing (14) with (8), in which both wh-sluicing and wh-pseudosluicing are available options, the antecedent correlate is implicit in (14), whereas it is overtly present in (8). That is to say, whpseudosluicing is immediately precluded if the antecedent correlate is implicit. Combining this with the aforementioned observations, we can summarize the following restrictions on the use of wh-pseudosluicing: 9
(15) In Emirati Arabic, wh-pseudosluicing is banned if (a) the antecedent correlate is implicit, (b) it is a wh-adjunct, or (c) it is a wh-phrase (e.g., which-NP, wh-PP).
In this squib, I have demonstrated that in Emirati Arabic, the bare whwords in a sluicing clause can be derived from two sources, wh-fronting and wh-clefts; the bare wh-words are called the wh-sluice and the wh-pseudosluice, respectively. The choice of sluicing source is conditioned by (a) the morphosyntactic properties and conditions imposed on individual wh-constructions and (b) whether or not the antecedent correlate is implicit. A wh-sluice can be used freely for any type of wh-expression, and regardless of the syntactic projection (or not) of the antecedent correlate. The use of a wh-pseudosluice, on the other hand, is limited to bare wh-arguments (e.g., ʃuu ‘what’ and mənu ‘who’, but not wh-phrases such as which-NP or wh-PP) and is strictly banned if the antecedent correlate is implicit. I have provided evidence showing that EA is a counterexample to Merchant’s (2001) Preposition Stranding Generalization (PSG). That is to say, P-stranding is possible under sluicing, even though it is strictly banned in the case of whmovement. The PSG can be preserved by stating that the constraint on P-stranding can be defined as a PF condition in languages such as EA. Such a move is consistent with the PF deletion approach to sluicing: sluicing a result of PF deletion can rescue P-stranding violations as defined at the level of PF.
Earlier versions of this squib were presented at the 46th and 47th annual meetings of the Chicago Linguistic Society (CLS 46, 47) and the Twenty-fifth Annual Symposium on Arabic Linguistics (ALS-25). I would like to thank Fatima Al Eisaei, Mariam Kaabi, and Aamna Shemeili for the discussion of Emirati Arabic data, and Stephen Matthews, Roumyana Pancheva, and the two LI reviewers for their comments.
This squib is dedicated to the memory of my academic and life mentor, Jean-Roger Vergnaud.
2Aoun, Choueiri, and Hornstein (2001) claim that Lebanese Arabic (LA) wh-questions formed by a resumptive pronoun involve movement. In particular, they argue that resumption outside islands is movement-driven, whereas resumption inside islands is not. EA differs from LA in that the latter allows the use of resumptive pronouns in wh-fronting. In contrast, wh-fronting in EA is formed by a gap. For further arguments that support treating wh-fronting and wh-clefts in EA as involving distinct constructions, see Leung and Al Eisaei 2011.
3 Clefts typically include a copular pronoun as a pivot. For example:
hu əl-ktaab εlli ʃtər-eet-ah.
3SM the-book.M that bought-2SM-3SM
‘It is the book that you bought (it).’
4 The claim that the wh-pseudosluice is derived from wh-clefts is evident from the observation that both display the same morphosyntactic properties. First, the copular pronoun can be used in both constructions. Also, neither can be used with wh-adjuncts (e.g., *kεɪf hu? ‘How is it?’, *leɪ hu? ‘Why is it?’) or wh-PPs (e.g., *f-ʔaj mʊkaan hu? ‘In which place is it?’).
5 Observe the following contrast between wh-fronting (i) and a wh-cleft (ii) formed by which-NPs:
fi ʔaj kʊmbjutər John kətab baħθ-ah?
in which computer John wrote research-his
‘On which computer did John write his research?’
(ii) *fi ʔaj kʊmbjutər (hu) εlli John kətab baħθ-ah?
6 Thanks to an LI reviewer for the suggestion.
7Sato’s (2011) analysis rests on the notion of wh-feature percolation (Chomsky 1972). Languages are parameterized in terms of whether the PPcontaining wh-expressions (e.g., ‘who’ as in ‘to who’) can percolate their [+wh] feature to its containing PP. In the derivation of wh-questions, the interrogative C attracts the closest goal with a [+wh]. For P-stranding languages, such [+wh] feature percolation to PP is optional, and as a result overt wh-movement can either strand a preposition or not.
8 In this squib, I consider implicit arguments as semantic arguments that are not syntactically projected. For further discussion of the analysis of implicit arguments as unprojected elements versus as null pronouns, see Bhatt and Pancheva 2006.
9 Two remarks are in order here. First, one reviewer suggests that the contrast may lie in the semantic contribution of the copular pronoun hu. Carnie (1995) points out that in Hebrew, the use of a copular pronoun is obligatory in equative sentences, as in (i).
Dani *(hu) ha-more.
Dani 3SM the-teacher
‘Dani is the teacher.’
We can claim that the EA copular pronoun hu also implies an equative structure that requires an R-expression for the wh-pseudosluice, hence the requirement for an overt antecedent correlate. Native speakers also express the intuition that the copular pronoun hu that follows the wh-pseudosluice must be anaphoric to a D(iscourse)-linked antecedent. On the other hand, the wh-sluice (without a copular pronoun) does not have this requirement. This requirement for a Dlinked antecedent also explains why wh-adjuncts are banned in wh-pseudosluicing.
Second, the effect of the implicit antecedent correlate on the sluicing source is reminiscent of, though not identical to, Chung, Ladusaw, and McCloskey’s (1995) discussion of merger or sprouting. The use of implicit antecedent correlates does not suffice to falsify the PSG, however. For example:
(ii) John yfəkker, bəs maa ʕərf <b-ʃuu/*ʃuu>.
John think but not 1.know of-what/what
‘John thinks, but I don’t know *(of) what.’