1. INTRODUCTION
In the Scandinavian languages, extraction from embedded clauses without the verb second word order (henceforth V2) is generally possible, whereas embedded V2 clauses have been observed to display restrictions on extraction to varying degrees; see Holmberg (Reference Holmberg1986), Vikner (Reference Vikner1995), and more recently Holmberg (to appear) and references therein. The contrast is exemplified by Swedish in (1).Footnote 1
(1)
This short communication presents a broad overview of wh-extraction from V2 clauses in Danish, Faroese, Icelandic, Norwegian and Swedish. We will show that V2 clauses with a fronted non-subject are islands for extraction in all of these Scandinavian varieties, whereas subject-initial V2 clauses differ with regard to the possibility of extraction, also within the Mainland Scandinavian languages (i.e. Danish, Norwegian and Swedish). From a theoretical perspective the extraction facts presented here favor an elaborate structure of the left periphery (Rizzi Reference Rizzi and Haegeman1997) and suggest that there is no single verb second position in the clause (see Hróarsdóttir et al. Reference Hróarsdóttir, Wiklund, Bentzen and Hrafnbjargarson2007; Hrafnbjargarson & Wiklund Reference Hrafnbjargarson and Wiklund2010; Migdalski Reference Migdalski2010). Moreover, they suggest that the verb second phenomenon per se is not responsible for island effects unless we distinguish between different types of verb second (see Bentzen to appear). Before we turn to the empirical facts, a couple of issues that are crucial to the interpretation of the data need to be mentioned. These concern subject-initial V2 in Insular Scandinavian and the that-trace effect.
2. SUBJECT-INITIAL V2 IN INSULAR SCANDINAVIAN
The Icelandic example in (2) illustrates extraction from an embedded clause that may be analyzed either as a subject-initial V2 clause with the subject and the verb in Spec-CP and C, respectively, or as involving V-to-I movement of the verb.
(2)
For theory-internal reasons, Vikner (Reference Vikner1995:108ff.) assumes that extraction is impossible from V2 clauses and therefore that an analysis involving verb movement to the inflectional domain of the clause is the only one available for the embedded clause in (2). However, there is no a priori reason to assume that all V2 clauses are islands for extraction. As we will see, subject-initial V2 clauses allow argument extraction also in Norwegian varieties that do not display independent verb movement to the inflectional domain of the clause and where a V-to-I analysis of these clauses is therefore unavailable. Moreover, we will show that under certain circumstances subject-initial V2 clauses disallow extraction also in Icelandic. For these reasons, extraction facts do not easily lend themselves as arguments in favor of a V-to-I over a V-to-C analysis of Icelandic verb movement.
In what follows, we assume that all verb movement out of vP in Icelandic targets the CP domain of the clause and that embedded clauses like that in (2) above involve subject-initial V2 with both the subject and the verb in the left peripheral domain of the clause (CP). For detailed arguments in favor of this analysis and against the traditional V-to-I movement analysis, see Wiklund et al. (Reference Wiklund, Hrafnbjargarson, Bentzen and Hróarsdóttir2007) and more recently Hrafnbjargarson & Wiklund (Reference Hrafnbjargarson and Wiklund2010). For an attempt to rescue a V-to-I analysis of Icelandic verb movement, see Thráinsson (Reference Thráinsson2010).Footnote 2
3. THE THAT-TRACE EFFECT
Danish, Faroese, Swedish and some varieties of Norwegian display the that-trace effect, whereas Icelandic and other varieties of Norwegian (including Northern Norwegian) do not; see Taraldsen (Reference Taraldsen1980), Thráinsson et al. (Reference Thráinsson, Petersen, Jacobsen and Hansen2004), and more recently Lohndal (Reference Lohndal2009) for discussion of the that-trace effect in the Scandinavian languages and English. In brief, the that-trace effect means that it is impossible to extract a subject across the complementizer. For this reason, the Danish, Faroese and Swedish examples in (3) cannot tell us anything about the possibility of extraction from V2 clauses. They would be ungrammatical in any case.
(3)
The ungrammaticality of the sentences in (3) disappears if the complementizer is omitted but with different results in the three languages:
(4)
Whereas (4b) involves subject extraction from a V2 clause, the only available reading for (4a and c) is a parenthetical reading (for the same proposal for German, see Reis Reference Reis, Lutz and Pafel1996). The first piece of evidence for this difference concerns object extraction. As we will see, object extraction is possible from a subject-initial V2 clause in Faroese but not in Danish and Swedish. This difference already gives us a hint that subject extraction from subject-initial V2 clauses may be available in varieties of Faroese but unavailable in Danish and Swedish.Footnote 3
Another piece of evidence comes from binding. The examples in (5) show that a quantifier in the matrix clause can bind a pronoun in the embedded V2 clause in Danish, Faroese and Swedish; the reading where the possessive pronoun covaries with the matrix quantifier is available in both varieties. This property sets Scandinavian embedded V2 clauses apart from their counterparts in e.g. West Frisian (see de Haan Reference de Haan2001), in that the V2 clause is really embedded under the matrix clause in Scandinavian.
(5)
In contexts like (4) above, however, the possibility of binding is restricted to Faroese. Neither Swedish nor Danish permits binding in this context, see (6).Footnote 4
(6)
We take these data to show that Faroese allows subject extraction from subject-initial V2 clauses, whereas Danish and Swedish do not. Rather, what looks like subject extraction from a V2 clause involves a parenthetical construction in the latter languages, corresponding to His mother, said every man, hadn't read the book. Our investigation of Norwegian is confined to varieties that do not display the that-trace effect.
4. EXTRACTING FROM V2
We have confined our survey to the tests below, all involving extraction of wh-elements corresponding to English who, what and why. We leave for future research the possibility that the type and complexity of the extracted element may affect extraction possibilities, see Szabolcsi (Reference Szabolcsi, Everaert and Riemsdijk2006) for an overview.
(7) Extraction tests
(i) wh-subject extraction from non-subject-initial V2 clauses
(ii) wh-object extraction from non-subject-initial V2 clauses
(iii) wh-subject extraction from subject-initial V2 clauses
(iv) wh-object extraction from subject-initial V2 clauses
(v) wh-adjunct extraction from subject-initial V2 clauses
Furthermore, all examples in what follows involve counterparts of the English verb say, an assertive predicate which has been observed to allow root phenomena cross-linguistically (see Hooper & Thompson Reference Hooper and Thompson1973 for an early discussion) and which may embed both subject-initial and non-subject-initial V2 clauses in all of the Scandinavian languages; see Julien (Reference Julien2007) and Wiklund et al. (Reference Wiklund, Bentzen, Hrafnbjargarson and Hróarsdóttir2009).Footnote 5
4.1 Argument extraction from non-subject-initial V2 clauses
The following tests identify the possibility of argument extraction from V2 clauses involving a fronted non-subject:
- (8)
a. wh-subjectemb said he that [non-subject Vfin_neg]
b. wh-objectemb said he that [non-subject Vfin subject neg_]
As the examples in (9)–(13) illustrate, none of the five Scandinavian varieties allow extraction from clauses involving non-subject topicalization (see Holmberg Reference Holmberg1986 on Swedish). Judgments do not change if the negation is left out. The target meanings of the (a) and (b) examples below are, respectively, ‘Who did he say that had not given these books to Trine?’ and ‘Whom did he say that he had not given these books to?’
(9)
(10)
(11)
(12)
(13)
4.2 Argument extraction from subject-initial V2 clauses
The possibility of argument extraction from subject-initial V2 clauses is identified by the following tests:
- (14)
a. wh-subjectemb said he (that) [_ Vfinneg]
b. wh-objectemb said he that [subject Vfin neg_]
The intended (or idiomatic) translations of the (a) and (b) examples below are, respectively, ‘Who did he say could not sing this song?’ and ‘What did he say that he could not sing?’.
(15)
(16)
(17)
(18)
(19)
Danish and Swedish are the only varieties investigated that completely disallow argument extraction from subject-initial V2 clauses (see Holmberg Reference Holmberg1986), bearing in mind the facts presented in Section 3 and the fact that we have restricted our study of Norwegian to the varieties that do not display the that-trace effect. Although Swedish and Norwegian pattern together regarding the availability of embedded V2 (Wiklund et al. (Reference Wiklund, Bentzen, Hrafnbjargarson and Hróarsdóttir2009), they apparently differ with respect to extraction from V2 clauses. While Swedish does not allow wh-argument extraction from subject-initial V2 clauses, Norwegian does (see Lohndal Reference Lohndal2009). In this sense, the relevant varieties of Norwegian pattern with Faroese and Icelandic with respect to argument extraction.
4.3 Adjunct extraction from subject-initial V2 clauses
The possibility of adjunct extraction is identified by the following test:
(20) wh-adjunctemb said he that [subject Vfinneg object _]
By incorporating the test into question–answer pairs, the available readings appear clearer. The question is:
(21) Why did you say that you had not met the queen?
Two potential answers to the question are A1 and A2:
- (22)
a. A1: I said it because I thought you should know it.
b. A2: She did not have time to meet me.
The A2 answer is possible only if the wh-adjunct (why) originates from within the embedded clause, thus identifying adjunct extraction from the embedded V2 clause (see also Haegeman Reference Haegeman2003 and Bhatt & Pancheva Reference Bhatt, Pancheva, Everaert and van Riemsdijk2006 for a discussion of adjunct movement of this type).
(23)
(24)
(25)
(26)
(27)
Because both A1 and A2 are possible answers to the question in Faroese and Icelandic (see (24) and (25)), the question is ambiguous in these languages. It can either ask for the reason why you said something, or the reason why you had not met the queen. The possibility of A2 identifies adjunct extraction from subject-initial V2 clauses in these languages. In contrast, this does not hold for Danish, Norwegian, and Swedish, see (23), (26) and (27). In the presence of V2 in the embedded clause, the adjunct may only be read as originating from within the matrix clause. Thus, Norwegian patterns with Danish and Swedish rather than with Faroese and Icelandic regarding adjunct extraction.
5. CONCLUDING REMARKS
Conforming to observations in the literature, our investigation has shown that none of the Scandinavian languages allow wh-extraction from V2 clauses where a non-subject has been fronted. With regard to subject-initial V2 clauses on the other hand, we see more variation. Based on our data, the Scandinavian languages can be divided into three classes. Faroese and Icelandic allow both wh-argument and wh-adjunct extraction from subject-initial V2 clauses, whereas Danish and Swedish allow neither, and Norwegian falls between these two classes in that it allows wh-argument extraction but not wh-adjunct extraction.
These facts are important for a deeper understanding of the V2 phenomena because they indicate that such phenomena are not likely to be attributable to a uniform syntactic trigger but may target different positions in the left peripheral domain of the clause, see Hróarsdóttir et al. (Reference Hróarsdóttir, Wiklund, Bentzen and Hrafnbjargarson2007), Hrafnbjargarson & Wiklund (Reference Hrafnbjargarson and Wiklund2010) and also Migdalski (Reference Migdalski2010). Moreover, the facts from Norwegian indicate that verb movement to the CP domain as such does not necessarily induce islands for extraction. Following the generally accepted assumption that Norwegian does not display independent V-to-I movement of the type traditionally attributed to Icelandic (save in Northern Norwegian, see Bentzen Reference Bentzen2005), our data add an additional piece of evidence that the availability of extraction from subject-initial V2 clauses in Icelandic cannot be used as an argument for analyzing such clauses as involving V-to-I rather than V-to-C. Another piece of evidence comes from the observation in Hrafnbjargarson & Wiklund (Reference Hrafnbjargarson and Wiklund2010) that argument extraction as in (17a) above is unacceptable even in Icelandic when discourse-oriented material is inserted into the clause (see Wiklund Reference Wiklund2010), as is shown in (28) below. In other words, subject-initial V2 clauses are not always devoid of island effects in Icelandic. Whether they are or not arguably depends on the target of the V2 movement(s) involved, supposedly lower in (17a), higher in (28). In the absence of extraction, (28) is of course fine, see (29).
(28)
(29)
We leave the details of the ‘cartography of V2’ and potential intervention effects induced by a subset of the features that trigger V2 phenomena for future research.
ACKNOWLEDGEMENTS
Kristine Bentzen is a native speaker of Norwegian, Gunnar Hrafn Hrafnbjargarson a native speaker of Icelandic, and Anna-Lena Wiklund a native speaker of Swedish. We are indebted to Victoria Absalonsen, Kirsti Hansen, and especially Zakaris Hansen for providing us with data from Faroese, to Björn Lundquist for additional judgments on the Swedish examples, to Christine Bjerkan Østbø, Øystein Alexander Vangsnes, and Merete Anderssen for additional judgments on the Norwegian examples, and to Robert Zola Christensen, Pernille Folkmann, and Kjeld Kristensen for judgments on Danish examples. For comments and discussion, we would like to thank Caroline Heycock, Terje Lohndal, Christer Platzack, Peter Svenonius, and the audience at the CASTL colloquium (Tromsø, March 2007). Last, but not least, we would like to thank Þorbjörg Hróarsdóttir for her collaboration on a previous version of this paper (Bentzen et al. Reference Bentzen, Hrafnbjargarson, Hróarsdóttir and Wiklund2007).
