The goal of this squib is to provide new empirical evidence in support of the view that there is more than one type of split construction (see (1)) and that the types should be distinguished on the basis of their Information Structure (IS). (Unless otherwise noted, examples are from Russian.)
The evidence is based on a reinterpretation of the observation that P(reposition)-doubling is possible (or even obligatory) in some split PPs and infelicitous in others. These facts are reported in Fanselow and Ćavar 2002 for Croatian and German and in Féry, Paslawska, and Fanselow 2007 for Ukrainian. I show that similar observations can be made about Russian and argue that the (un)availability of P-doubling at least in Russian correlates with different ISs and does not directly depend on the order of split parts (contrary to what is proposed in Fanselow and Ćavar 2002) or their prosody (contrary to what is proposed in Féry, Paslawska, and Fanselow 2007).
The squib is organized as follows. In section 2, I identify two types of split constructions—contrastive splits (c-splits) and topicalization splits (t-splits)—and show that they have distinct ISs, intonation patterns, and syntactic behavior. In section 3, I present new data concerning P-doubling in discontinuous PPs in Russian. The generalization that emerges is that c-splits disallow P-doubling, whereas in tsplits, P-doubling is possible (and in some cases, obligatory). In section 4, I argue that the two previous accounts of P-doubling in split constructions in Slavic that I am aware of—in Fanselow and Ćavar 2002 and Féry, Paslawska, and Fanselow 2007—cannot accommodate the Russian facts. In section 5, I offer some concluding remarks.
2 C-Splits vs. T-Splits
The goal of this section is to show that c-splits and t-splits differ from the IS point of view and have distinct prosody and syntactic behavior. A c-split is a construction in which one of the split parts has contrastive interpretation (contrastive focus or contrastive topic).1 A t-split is a construction in which the two split parts have different information status: one is a topic, the other is a new information focus.2 In addition, the data below show that c-splits and t-splits can have both preserved and inverted word order, which will be important for the later discussion.
(2) is an example of a c-split in which the adjective (here, malinovogo ‘raspberry.N.SG.GEN’) is preposed and bears contrastive stress; that is to say, it is marked with a falling intonational contour, dubbed IK-2 in the traditional literature on Russian (e.g., Bryzgunova 1981). In the examples, the split parts are underlined and each prosody mark follows the word to which it applies.3 Words that bear contrastive stress appear in capital letters in the translation.
Malinovogo\2 ona mne prislala varen’ja (a ne raspberry.N.SG.GEN she to.me sent jam.N.GEN and not klubničnogo).
‘She sent me RASPBERRY jam (and not strawberry).’
Djadja\2 staralsja moj ukrotit’ ee.
uncle.NOM tried my.M.SG.NOM to.reason her
‘My UNCLE (as opposed to other family members) tried to reason with her.’
An important syntactic property of c-splits is that splitting is not necessary. As correctly noted in Zubatow and Junghanns 1995 and Pereltsvaig 2008, the contrastive interpretation can be achieved by stress alone or by preposing the adjective to different positions; see (4) for the example in (2).
(✓Malinovogo\2) ona (✓malinovogo\2) mne
raspberry.N.SG.GEN she raspberry.N.SG.GEN to.me
(✓malinovogo\2) prislala ✓malinovogo\2
raspberry.N.SG.GEN sent raspberry.N.SG.GEN
varen’ja (a ne klubničnogo).
jam.N.GEN and not strawberry.N.SG.GEN
‘She sent me RASPBERRY jam (and not strawberry).’
T-splits differ significantly from c-splits in this respect, as we will see. (5) is an example of a t-split in which the topic noun (here, dobyča ‘booty.F.NOM’) is fronted and marked with a rising intonation characteristic of topics in Russian and the adjective (here, xorošaja ‘good.F.SG.NOM’) expresses new information and has a falling intonation IK-1, which marks new information focus in Russian and is less intense and lower in tone than IK-2 (Bryzgunova 1981).
(S takim entuziazmom ljudi ispokon vekov i vkalyvajut na svoju sobstvennuju vygodu, konkretnuju dolju v dobyče . . . )
(With such enthusiasm people have worked hard for ages for their own profit, for a concrete share of the booty . . . ) Dobyča/ im dostalas’ xorošaja\1 . . . booty.F.NOM to.them passed good.F.SG.NOM
‘They got a good booty.’ = ‘As for the booty, what they got was good.’
(On a cold November morning, a communist parade is taking place. There is a requirement that every participant must carry something red. A news reporter: ‘‘Here comes the fourth group of people . . . )
i krasnye/ oni nesut znamena\1.
and red.N.PL.ACC they carry flags.N.ACC
‘and they are carrying red flags.’ = ‘As for red (things),
they are carrying flags.’
As mentioned above, t-splits differ from c-splits in that t-splits must be split in Russian: the topic must be fronted and the new information focus must surface clause-finally (see Neeleman and Titov 2009). This is illustrated in (7) for the example in (5).
✓ Dobyča/ im (*dobyča/ ) dostalas’ booty.F.NOM to.them booty.F.NOM passed
xorošaja\1 (*dobyča/ ).
‘They got a good booty.’ = ‘As for the booty, what they got was good.’
To summarize, in c-splits one part is contrastively stressed and optionally fronted. The two parts of t-splits must be separated, with the topic part fronted and the new information focus part appearing clause-finally. The parts in both c-splits and t-splits can preserve the normal word order or be inverted.
3 P-Doubling in Split Constructions
The new generalization about the availability of P-doubling in split constructions is that it depends on the IS of the utterance. In this section, I present data showing that P-doubling is possible (and in some cases, obligatory) with t-splits, whereas it is ill-formed with csplits. Whether P-doubling is possible or obligatory depends on the choice of preposition. I illustrate both patterns: with the preposition iz ‘from’, P-doubling is obligatory, whereas with na ‘on’, it is optional. The factors that can influence the optionality of P-doubling include the semantic heaviness of the preposition, the case it assigns, and, possibly, the structure of the predicate itself. Further research is needed to determine which of these factors (or their combination) is responsible for the optionality of P-doubling when IS is kept constant (see Yadroff 1999 and Caha 2009 for discussion of Russian prepositions and their connection with case). Examples (8a) and (9a) show that Pdoubling is possible with t-splits; examples (8b) and (9b) show that c-splits disallow P-doubling.
Iz čaški/ ja pila *(iz) krasnoj\1.
from cup.F.GEN I drank from red.F.SG.GEN
‘As for cups, I drank from a red one.’
Iz čaški\2 ja pila (*iz) krasnoj.
from cup.F.GEN I drank from red.F.SG.GEN
‘I drank from a red CUP (not from a red glass).’
Na tarelku/ položi jabloki ?(na) bol’šuju\1.
on plate.F.ACC put apples on big.F.SG.ACC
‘As for plates, put the apples on a big one.’
Na tarelku\2 položi jabloki (*na) bol’šuju.
on plate.F.ACC put apples on big.F.SG.ACC
’Put the apples on a big PLATE (not on a big box).’
(8) and (9) show PP splits in which the noun is preposed. (10) illustrates PP splits in which an adjective is preposed. The same pattern is observed in (10) as in (8) and (9): P-doubling is possible with tsplits, but not with c-splits.
(10) (A pre-wedding discussion of the tradition that a bride must have ‘‘something old, something new, something borrowed, something blue.’’)
V sinem/ ja prijdu ?(v) plat’je\1.
in blue.N.SG.LOC I will.come in dress.N.LOC
‘As for the blue (thing), I will come in a blue dress.’
V sinem\2 ja prijdu (*v) plat’je. in blue.N.SG.LOC I will.come in dress.N.LOC ‘I will come in a BLUE dress (not in a green one).’
4 Theoretical Discussion
4.1 Previous Analyses of Split Constructions
Split constructions show paradoxical properties of both movement and nonmovement (see Sekerina 1997, Gouskova 2001, Fanselow and Ćavar 2002, Bošković 2005, Franks 2007, Pereltsvaig 2008, among many others). These facts argue for a nonmovement derivation of splits: (a) both split parts can be nonconstituents; (b) the split can involve PPs, inherently marked NPs, and weak islands (wh-, negation, and factive verbs); (c) the split can involve morphological ‘‘regeneration’’ (see Van Riemsdijk 1989). And these facts argue for a movement account: (a) extraction from strong islands (subjects, complex NPs, and adjuncts) is impossible; (b) split parts show morphological agreement in case, number, and gender. Both movement (Androutsopoulou 1997, Sekerina 1997, Fanselow and Ćavar 2002, Bošković 2005, Féry, Paslawska, and Fanselow 2007, Pereltsvaig 2008, Ott 2011,5) and basegeneration (Fanselow 1988) analyses of split constructions have been proposed in the literature, although movement analyses prevail. Most of these analyses focus on split NPs and do not discuss P-doubling in split PPs. To my knowledge, there are only two accounts that directly address P-doubling in split PPs: Fanselow and Ćavar 2002,6 and Féry, Paslawska, and Fanselow 2007. In the following two sections, I will look at each of these accounts and show that neither can explain the Russian data. The problem with Fanselow and Ćavar’s account is that it ties the possibility of P-doubling to the word order in split PPs: inverted (P N . . . A in languages with prenominal adjectives) vs. noninverted (P A . . . N). However, in section 3 we saw that the possibility of P-doubling in Russian is connected to the IS status of split parts, rather than their order. Féry, Paslawska, and Fanselow make Pdoubling dependent on the prosodic structure of utterances. Although I agree that prosody reflects IS, I will argue using both theoretical and empirical evidence that IS (and not prosody) is the defining factor.
4.2 Distributed Deletion: Fanselow and Ćavar 2002
Building on the conceptualization of movement as two operations, Copy and Delete (Chomsky 1995), Fanselow and Ćavar (2002) develop an analysis of split constructions in terms of distributed deletion, which takes IS into account. According to their proposal, split constructions derive from the spell-out requirement on chains in (11a). Chains are created when elements of XP bear different semantic/ pragmatic features, such as [+wh], [+focus], [+topic]. According to (11a), each part of a split must be spelled out in the specifier of the head that checks its operator feature, as illustrated in (11b).
Suppose C=〈C1, C2〉 is formed because a strong feature of H has attracted XP and suppose that H checks the operator features f1 . . . fk of XP. Then the categories bearing f1 . . . fk must be spelt out in C1.
ap[b c]q] [Hq . . . [[XP ap [b c]q] [Hp . . . [XP ap [b c]q]]]]]
Fanselow and Ćavar demonstrate that this proposal accounts for NP splits in German and Croatian and explains why NPs and PPs are not barriers replacing movement out of XP by movement of XP itself.
The problem arises when we try to extend this proposal to PP splits. Fanselow and Ćavar’s generalization is that PP splits are possible in constructions that preserve the normal word order within an NP, as illustrated by the Croatian examples in (12a–b). They also observe that in German and some dialects of Croatian, PPs can be split in an inverted order with the noun fronted and the adjective stranded, in which case the preposition is repeated; see (12c).7 According to Fanselow and Ćavar, in noninverted PP splits, P-doubling is impossible (p. 98).
Na kakvo se Ivan stablo penje? on what.kind.of SELF Ivan tree climbs ‘On what kind of tree does Ivan climb?’
*Krov je Ivan na kakav skočio? roof has Ivan on what.kind jumped
In Schlössern habe ich noch in keinen gewohnt. in castles have I yet in no lived ‘So far, I have not yet lived in any castle.’
To account for this difference, Fanselow and Ćavar propose two spellout requirements. (a) P must be ‘‘phonologically adjacent’’ to the category it selects. (b) The categorial element (i.e., P) must be overtly realized in the highest copy. The first requirement instructs that inkeinen ‘in no’ and na kakvo ‘in what kind of ’ are spelled out with prepositions. The second requirement ensures that although the noun in the German example receives its own preposition, the noun in the Croatian example does not.
This explanation is tied to the order of split parts and predicts that splits with the inverted word order will always allow P-doubling, whereas splits with the preserved word order will not. As we saw in section 3, this prediction fails for Russian. In Russian, P-doubling is possible with both inverted and noninverted splits, and the dividing line demarcates the IS status of the split elements, rather than their category.
4.3 Prosodic Account: Féry, Paslawska, and Fanselow 2007
Féry, Paslawska, and Fanselow (2007) look at split NPs and PPs in Ukrainian. According to their system based on the prosodic structure of sentences, split constructions are divided into two types: ‘‘cohesive’’ and ‘‘noncohesive.’’ Cohesive splits form one i(ntonation)-phrase and result from wh-fronting or focus fronting, which is not obligatory in Ukrainian; see (13a). Noncohesive splits result from topicalization or right-dislocation and have a prosodic structure with two i-phrases; see (13b). In this system, both cohesive and noncohesive splits can have inverted and noninverted word order, which mends the problem found in Fanselow and Ćavar’s (2002) account.8
Cohesive split: [ . . . ab . . . ]i → [ . . . a . . . b . . .]i
Noncohesive split: [ . . . ab . . . ]i → [ . . . a . . .]i
[ . . . b . . .]i
Féry, Paslawska, and Fanselow tentatively propose to associate cohesive splits with a movement derivation and noncohesive splits with base generation. In their account, P-doubling is an indicator of the fact that two parts of a split are base-generated (on a par with a morphological mismatch). Thus, the absence of P-doubling in focus fronting and wh-fronting suggests that they are derived by movement, and the optionality of P-doubling in topicalization indicates that some topics are derived by movement, whereas others are base-generated.9
However, the account proposed in Féry, Paslawska, and Fanselow 2007 is not satisfactory. From the theoretical point of view, although IS and prosody are connected, given the assumed asymmetry between semantic and phonological requirements (Chomsky 2007), it would be desirable to say that IS, but not prosody, triggers the separation of two parts of a constituent (contrary to what is proposed in Féry 2007 and Féry, Paslawska, and Fanselow 2007). So we need to classify split constructions according to their IS, rather than their prosody. From the empirical point of view, three observations are important. First, the cohesive/noncohesive prosodic distinction that Féry, Paslawska, and Fanselow argue for does not align perfectly with the IS distinction between c-splits and t-splits discussed in section 2. Russian t-splits can form one i-phrase; see the examples in section 2. Many of Féry, Paslowska, and Fanselow’s examples of noncohesive splits involve additional focus movements, as shown by the fact that the stranded part is not clause-final in (14). If Ukrainian has prosody similar to Russian, as Féry, Paslawska, and Fanselow suggest, the new information focus must surface clause-finally; see above.
[Pro ptaxivTOP]i [(✓knyžku) ty (✓knyžku) pročytav about birds book you book read knyžku]i.
‘You read a book about birds.’
Second, as we saw in section 3, with some prepositions P-doubling in t-splits is obligatory. Although more research is needed, it seems to be more plausible to account for the optionality of the second preposition in t-splits in terms of a spell-out requirement (assimilating it to approximative inversion constructions; see for instance Franks 1995, Yadroff 1999), rather than (un)availability of base generation.10 Finally, in Russian both P-doubling and splits with one preposition do not (always) observe island effects; see (15). The sentences in (15) were invariably accepted by the native Russian speakers I consulted. The same speakers categorically rejected (16), which is used to argue for the movement analysis of splits.
V fil’me/ ja pomnju vremja kogda on snimalsja in film.LOC I remember time when he starred v dokumental’nom\1. in documentary.LOC
‘As for films, I remember a time when he starred in a documentary.’
V fil’me\2 ja pomnju vremja kogda on snimalsja in film.LOC I remember time when he starred dokumental’nom. documentary.LOC
‘I remember a time when he starred in a documentary FILM.’
*Samuju interesnuju ty vstretil pisatelja most interesting.F.SG.ACC you met writer kotoryj napisal knigu. who wrote book.F.ACC ‘You met the writer who wrote the most interesting book.’
The fact that grammaticality judgments for island effects are graded is not novel, and many experimental studies have shown that factors such as discourse structure, D-linking, and definiteness can increase or decrease processing difficulty, resulting in variation in grammaticality judgments; see Pesetsky 2000, Warren and Gibson 2002, and Hofmeister and Sag 2010 for discussion. Although more (experimental) studies are needed to understand this variation, it is clear that the analysis that categorically associates splits involving P-doubling with base generation and splits having one preposition with movement is not tenable, as the sentences in (15) are of the same complexity and do not differ in grammaticality.11
In this squib, I used the observation about the (un)availability of Pdoubling in some split PPs to argue that split PPs should be classified on the basis of their IS rather than their word order or prosody, as previously proposed. In a broader perspective, the new ( partial) typology of splits has the following interesting consequence: if the (un)availability of P-doubling is indeed indicative of two distinct syntactic derivations and we keep the uniform treatment of split NPs and PPs based on their identical IS and prosody, we can argue that split NPs have distinct derivations, although this is not manifested (apart from the case mismatch constructions). This includes split NPs that surface as left-branch-extraction constructions (Ross 1986; see also Bošković 2005 for a comprehensive overview).
Distinct syntactic analyses for these two constructions are thus needed. One possible analysis of the data discussed here is presented in Goncharov 2012, where c-splits have a more straightforward, extraction-based derivation, while t-splits involve a more complex ‘‘readjustment.’’ Specifically, t-splits involve an independent case assigner for the two parts, and thus two prepositions are merged. Future research should focus on how these two types of derivations are to be connected to the distinct intonation patterns characteristic of each.
I would like to thank Diane Massam for many fruitful discussions of the research reported here. I am also grateful to two anonymous reviewers for their detailed comments on earlier drafts, the editorial team of LI Squibs and Discussion, and my Russian consultants Yanina Fedosova and Iryna Osadcha. All errors are my own.
1 For reasons of space, I will not discuss contrastive-topic examples in this squib. See Pereltsvaig 2008 for arguments that contrastive-focus and contrastive-topic splits behave alike syntactically. See also Titov 2013 for syntactic and semantic arguments in favor of the conclusion that ‘‘the notion of contrastive topic should be reduced to that of contrastive focus’’ (p. 452) and that nonsplit contrastive topics and contrastive foci are ‘‘associated with identical interpretative features, which results in their identical syntactic behaviour in Russian’’ (p. 452).
2 In this squib, I discuss only phrases that split into two parts, to avoid complications raised by multiple splits. For the treatment of multiple splits, see Fanselow and Ćavar 2002, Franks 2007, Pereltsvaig 2008, and references cited in these works. An anonymous LI reviewer remarks correctly that the study of multiple splits is important for constructing a more refined classification of splits based on their IS. However, for reasons of space I leave this question for another occasion.
3 The prosody marks are as follows: \2 = contrastive focus, falling contour IK-2; \1 = new information focus, falling contour IK-1; / = topic, rising contour.
5 An anonymous LI reviewer suggests that Ott’s (2011) symmetry-breaking analysis of split constructions in German can be extended to account for the Russian data. According to Ott, two parts of a split are merged in an argument position and are split in order to break a resulting symmetry and determine the label of the phrase. One of the strongest arguments proposed for this analysis is the existence of gapless splits. However, in Russian, gapless splits are possible only with t-splits, not with c-splits. Thus, Ott’s analysis cannot be straightforwardly extended to Russian.
6 In this squib, I chose to discuss Fanselow and Cavar’s (2002) account, rather than Pereltsvaig’s (2008), which is also a version of distributed deletion, is more recent, and targets Russian directly. There are two reasons why Pereltsvaig’s account is less suitable for comparison: (a) She discusses only two subtypes of splits, contrastive foci and contrastive topics, both of which are instances of c-splits, and it is unclear what her assumptions would be for t-splits. (b) She does not discuss P-doubling, although she mentions some examples of P-doubling.
7Fanselow and Ćavar (2002) attribute this observation to Josef Bayer. They do not provide minimal pair examples in the same language; thus, the examples reported here are from two languages, Croatian and German.
8 (13) shows the schema only for noninverted splits, but is assumed to cover inverted splits as well.
9 That is to say, the alignment between prosody and syntax is not perfect. If we assume, following Féry, Paslawska, and Fanselow (2007), that topicalization necessarily involves formation of a new i-phrase, we will have to say that some noncohesive splits are derived by movement. This seems to be the view that Féry, Paslawska, and Fanselow endorse in their conclusion.
10 An anonymous LI reviewer asks about the potential counterexample in Ukrainian presented by Féry, Paslawska, and Fanselow (2007:25, (41)) and described as having obligatory P-doubling and pairwise contrastive stress on numerals and nouns. Similar sentences can be constructed in Russian. However, such constructions in fact support the idea that P-doubling is connected to tsplits, as Féry, Paslawska, and Fanselow themselves point to the fact that fronted numerals in such sentences function as topics.
11 An anonymous LI reviewer points out Polinsky and Potsdam to appear, where genitive hanging topics in Russian are contrasted with paucal genitive constructions. Polinsky and Potsdam convincingly argue that the former are derived by base generation, whereas the latter are derived by movement. Unfortunately, they do not discuss IS or prosody distinctions between the two constructions, but their translations indicate that they aim for the topic interpretation of the fronted part. Thus, their findings at best suggest that t-splits can be derived by both movement and base generation, but they do not tell us anything about c-splits.