Extraction and Pronoun Preposing in Scandinavian

: It has been noted that examples with extractions out of relative clauses that have been attested in Danish, Norwegian and Swedish are judged to be unacceptable in Icelandic and Faroese. We hypothesize that this may reflect whether or not speakers tend to prepose unstressed object pronouns as a way of establishing a coherent discourse. In this article we investigate to what extent pronoun preposing is used in Swedish, Icelandic and Faroese and whether there is any correlation with the acceptabilty of extractions from relative clauses. We show that Icelandic speakers use pronoun preposing to a very limited extent whereas Faroese speakers often prepose the VP or sentential anaphor tað . In both languages extraction from relative clauses is mainly judged to be unacceptable, with Faroese speakers being somewhat more accepting of extraction from presentational relatives. A crucial factor seems to be whether preposing is associated with a marked, contrastive interpretation or not.


Introduction
At least since Ross' dissertation (Ross 1967), linguists have been trying to characterize the restrictions on extractions in natural languages. Ross showed that many types of extractions could be subsumed under a structural constraint, the so called Complex NP Constraint CNPC, which prohibited extraction out of a clause dominated by a noun phrase. The CNPC was soon subsumed under subjacency (Chomsky 1973) and in the following years, a number of proposals were made for what the relevant bounding nodes were in different languages in order to account for the cross-linguistic variation that was found (see, e.g., Rizzi 1990;Taraldsen 1981). However, soon Scandinavian speaking linguists such as Erteschik-Shir (1973); Andersson (1975) and Allwood (1976) pointed out that certain extractions out of relative clauses were possible in the mainland Scandinavian languages Danish, Norwegian and Swedish. In the following years several linguists approached this topic from different perspectives (see, e.g., the papers in Engdahl and Ejerhed 1982). When written and spoken corpora became available, studies based on spontaneously produced written or spoken extractions started to appear see, e.g., (Engdahl 1997;Jensen 2002;Lindahl 2010Lindahl , 2011Lindahl , 2017b. The documented spontaneous extractions mainly come from mainland Scandinavian whereas the insular Scandinavian languages Faroese and Icelandic appear to behave more similar to English when it comes to extractions from relative clauses (Platzack 2014;Thráinsson 2007;Thráinsson et al. 2004;Zaenen 1985). There have been occasional reports of spontaneous extractions from relative clauses in English (see, e.g., Chung and McCloskey 1983;McCawley 1981) 1 and similar examples are reported by Cinque (2013) to be at least marginally acceptable in Italian, French, and Spanish. However, they do not seem to be used productively the way they are in mainland Scandinavian, see the recent corpus investigations in Kush et al. (2021) and Müller and Eggers (2022). This suggests that the mainland Scandinavian languages have some common property which facilitates long extractions. In this article we explore the hypothesis that the common property that sets the mainland Scandinavian languages apart is the tendency to use preposing (topicalization) of unstressed pronouns as a way of connecting utterances in spoken language as well as sentences in texts. 2 We test this hypothesis by comparing how pronoun preposing is used in spoken Swedish, a mainland Scandinavian language, with similar data from the insular languages Icelandic and Faroese. In this article we bring together results from previous studies that we have carried out on individual languages and apply an explicit comparative perspective.
We start in Section 2 by looking at what characterizes spontaneous extractions from relative clauses in Swedish. Since a large proportion of them involve pronouns, we turn, in Section 3, to the question how pronoun preposing is used more generally in Swedish and what similarities there are between local preposing (within a clause), long preposing (from an embedded clause) and what is often called extraction (long preposing from a relative clause). In Section 4, we look at preposing and extraction in Icelandic and in Section 5 we present relevant data from Faroese. In Section 6, we look briefly at the ways preposing is used in English, German and Dutch before evaluating our hypothesis in Section 7.

What Do Spontaneous Extractions Look Like?
The literature on extractions from relative clauses is largely based on constructed examples-not surprising given that most of the studies deal with languages where such extractions are not used in ordinary conversations. It is clearly a good idea to start from the types of extractions from relative clauses that are actually used. Engdahl (1997) looked at 30 naturally occurring extractions in Swedish, both written and spoken. Some examples from this article are given in (1). We mark the relative clause and the gap site of the extracted element.
det '*an unromantic name which I did not know anyone at the time who was called.' (novel 1996) In (1a,b) the pronoun det 'it' has been preposed from a relative clause which modifies the indefinite pivot in an existential sentence; we refer to this construction as a presentational relative, see Lambrecht (1988) and Chaves and Putnam (2020, p. 27). 3 In (1c) a short deictic adverb där 'there' has been preposed from a have-construction which is very similar to presentational sentences (Keenan 1987;McCawley 1981). (1d) is a cleft construction where the object tapeterna 'the wall paper' has been preposed from a cleft clause; the clefted constituent Sven is definite. (1e) is a relative clause where the relativized item namn 'name' has been extracted from a relative clause which modifies the object of the lexical verb känna 'be acquainted with'. These types of constructions, preposing out of presentational relatives or cleft constructions and preposing or relativization from an embedded relative, were most common in Engdahl's (1997) sample. 4 There were a few examples involving preposing of a wh-phrase; they all had the form in (2) which resembles a cleft.
( In (2), the clefted constituent remains in situ and vem is linked to a position inside the cleft clause. 5 The relative rareness of spontaneous examples with preposed wh-phrases fits well with the experimental findings in Kush et al. (2018); Kush et al. (2019) and the corpus studies in Müller and Eggers (2022 'It was only dad that I could ask about maths.' In some of the examples the preposed phrase was a lexical DP, as in (1d) and (4), but a large proportion of Engdahl's examples involved either an anaphoric pronoun, as in (1a,b), or a light adverb, as in (1c). Engdahl showed that preposing of unstressed pronouns in Swedish is used as a way of establishing a coherent discourse and suggested that this might be what distinguishes the mainland Scandinavian languages from other languages, essentially the hypothesis that we are testing in this article. In the next section we look first at local preposing in Swedish and then return to extractions from relative clauses.

Swedish
The type of preposing of pronouns that we have seen in (1a,b,c) is of course not only found in extraction sentences. Local preposing, in the same clause, of object pronouns is quite common in spoken and written Swedish, as well as in Danish and Norwegian. In this section, we present data from Swedish but similar data are found in Danish and Norwegian, see Engdahl and Lindahl (2014).

Pronoun Preposing
Starting an utterance with a pronoun is very common in spoken language and Swedish is no exception to this. In most cases the pronoun functions as the subject of the sentence but in the mainland Scandinavian languages it is quite common to start with an object pronoun, or a subject pronoun from an embedded clause. 6 In order to find out when this word order is used, we have conducted several studies using the Nordic Dialect Corpus (NDC) (Johannessen et al. 2009) which contains recordings and aligned transcriptions of informal conversations in all the Nordic languages (Lindahl and Engdahl Forthcoming). 7 The Swedish part of the NDC consists of 361 184 words, produced by 133 speakers in 37 locations. Consider the following examples, taken from the NDC. We identify the interviewer by int and the speakers by s1, s2, etc. The antecedent of the pronoun is underlined and the preposed pronoun is italicized. (.) marks a short pause, and = that the speaker continues on the next line. In both (5) and (6), the preposed element is the third person singular neuter pronoun det 'it'. In (5) the pronoun acts as a VP anaphor which refers to the action gå i skolan 'go to school', expressed as a VP in the interviewer's question. 8 In (6) det is a propositional anaphor, referring to the event just described by the sentence de hade slagits 'they had been fighting'. Here det has been preposed from an embedded clause. In both utterances, det was unstressed.
Preposing of VP and propositional anaphors, as in (5) and (6), is very common in the NDC but there are also examples with a preposed pronoun which refers to a recently mentioned entity, as in (7)  Here the pronoun den agrees in gender and number with the antecedent bil, which is the only likely referent. The pronoun is unstressed. If it had been stressed, it would have been understood as a demonstrative, that, but this interpretation would not have been plausible in this context. Unstressed den and det are interpreted as personal pronouns whereas the stressed versions function as distal demonstrative pronouns (Faarlund 2019, p. 27). 9 We gloss the unstressed occurrences as 'it' and the stressed ones as 'that'. As we will show in Section 6.1, this type of non-contrastive preposing is hardly used in English.
In (Lindahl and Engdahl Forthcoming) we investigated how preposing of pronouns is used in the Swedish part of the NDC. We found that in a large majority of cases the preposed pronoun referred back to something that was newly introduced, an action as in (5), an event as in (6), or an entity, as in (8a). Following Erteschik-Shir (2007) we refer to this pattern as focus chaining since the antecedent is (part of) the relational focus of the preceding utterance, i.e., the new information that is asserted or questioned about the topic (Gundel and Fretheim 2004, p. 177). There were also cases where the antecedent of the preposed pronoun was the topic of the preceding utterance, often realized as a subject. This kind of topic chaining was mainly found when the antecedent was an entity, as in (8b) (we just provide the English version of the interviewer's question).
(8) int: Are you in touch with anyone who did their military service with you? The first occurrence of dom 'they' in (8a) is an instance of focus chaining. Once dom is realized as a subject, it becomes the topic of that utterance. Consequently the second dom in (8b) is in a topic chain.
In the examples shown above the preposed pronouns are unstressed. There are also examples where an initial pronoun is stressed and receives a contrastive interpretation, as in (9). By stressing the pronoun and negating the utterance, the speaker emphasizes that he definitely does not want contact lenses. We found that preposing of pronouns is primarily used in contexts where the antecedent and the pronoun formed a focus chain or a topic chain. In 85% of the investigated cases, the pronoun was locally preposed and in 15% we had long preposing from a subordinate clause, as in (6). One effect of the preposing is that the initial pronoun is interpreted as the aboutness topic of the utterance (Reinhart 1981) which we can show using Reinhart's rewriting test. We here apply the test to s1's utterance in (8b).
(10) a. s1 said about them (the two paramedics) that he had pretty good contacts with them.
The paraphrase (10) works well and we have a good indication that the preposed dom in (8b) acts as the aboutness topic for that utterance.
When it comes to relative frequency of object preposing in spoken Swedish, Jörgensen (1976) showed that this varies with the type of spoken interaction. Based on the material in Talbanken 10 he found 14% object initial sentences in interviews, and 9% in conversations and debates (Jörgensen 1976, p. 103). In a more recent study the second author investigated the word order patterns in the Swedish part of the NDC, Lindahl (under review). To obtain an unbiased sample, 1000 sentences were extracted from the corpus with the only criterion being that the sentence should have a finite verb. The sentences were manually investigated and all declarative main clauses, 712 in total, were further analyzed. In total, 46 of the sentences in that sample, or 6.5%, had a preposed object and of these, 36 were pronominal (see Table 1 in Section 4.1).
Studies of spoken Swedish show that starting a sentence with det is the most common pattern and in most cases it functions as an expletive subject (Allwood 1999;Engdahl 2012). In Engdahl and Lindahl (2014) where we specifically looked for preposed object pronouns (including subject pronouns from embedded clauses), 95% of the hits were det. The reason for this is that preposing of VP anaphors or propositional anaphors is by far more common than preposing of pronouns with entity antecedents. Mikkelsen (2015) claims that preposing of VP anaphors is actually obligatory in Danish whereas we argue that there is a strong preference for preposing in Swedish but not a grammatical constraint, Lindahl and Engdahl (Forthcoming). According to Bentzen and Anderssen (2019), preposing of VP and propositional anaphors is always an option in Norwegian. 11

Extraction from Relative Clauses
As already mentioned, the first author's investigation was limited to 30 authentic examples (Engdahl 1997). For her dissertation, the second author carried out a larger empirical study of extraction from relative clauses in Swedish based on a collection of 270 spontaneous examples, gathered between 2011 and 2016 (Lindahl 2017b). The collection contained 101 spoken examples from everyday conversations, 60 from radio and television, and 109 written examples. 12 In a representative subset of the spoken preposing examples (Sample B), slightly over half of the examples, 56%, involved preposed pronouns. Just as in the examples with local preposing, the pronouns extracted from relative clauses typically refer back to something which has just been uttered, either by the same speaker or by an interlocutor. The extracted pronouns are thus part of a focus or a topic chain which connects the utterance to the preceding context. A few examples are given in (11)-(13). For space reasons, the preceding context is sometimes given only in English. 'It was the nobility that had the jurisdiction over the largest part of the population, the farmers.' (Radio Sweden, 2015) (Lindahl 2017b, p. 91) These examples resemble the ones shown in Section 2 from Engdahl (1997). Regarding (11) and (12), these are examples of presentational relatives, cf. (1abc), and (13) is a cleft construction, cf. (1d). This example also illustrates a common use of pronoun preposing in Swedish, namely in left dislocation where the initial topic is resumed by a preposed agreeing pronoun, see Holmberg (2020) and Lindahl and Engdahl (Forthcoming) for discussion.
The next two examples involve the lexical main verbs veta 'know' and störa sig på 'be annoyed at'. In (14) 'People who say that annoy me.' (Lindahl 2017b, p. 89) In Lindahl's Sample B, 74% were presentational relatives, 8% were clefts such as (13) and 13% were constructions with lexical verbs. 13 In many of the presentational relatives, the relative clauses are short and contain just a finite auxiliary verb, as in (11), or the support verb göra 'do', as in (14). These short relative clauses help to identify or restrict the head of the relative clause. Together with the matrix, the head DP and the relative clause must form a coherent comment on the extracted item which is relevant in the sense of Grice (1975). 14 We end this overview of extraction from relative clauses by noting that Swedish speakers have been using such constructions for a long time. In a guide to Proper Swedish, Wellander (1939)  'I had never met anyone who had done that.' (Lindahl 2017b, p. 28)

Summary
In this section, we have shown that the use conditions for extraction of pronouns from relative clauses are the same as for local and non-local pronoun preposing in Swedish. 16 The pronouns are typically part of focus or topic chains; they refer to an event, a property or an entity that has either just been introduced or is already established as a topic. This resembles a well-known strategy for cohesion in discourse, namely to start with a subject pronoun which is part of an anaphoric chain with some element in the preceding utterance (Daneš 1974;Erteschik-Shir 2007). What seems to be special about speakers of Swedish and the other mainland Scandinavian languages is that they use this strategy also for non-subject pronouns, as becomes clear when we look at spontaneous conversations. By preposing a pronoun in a focus or topic chain, the speaker ensures that it will be understood as the aboutness topic for the upcoming utterance. This may in turn explain why preposing the VP or propositional anaphor det, as in (5) and (6), is especially common. By doing so the speaker signals that det's antecedent is what the sentence is about.
We noted earlier that preposed pronouns in Swedish are often unstressed; when a pronoun is stressed, it often invokes contrast and the presence of alternatives, see (9). Lexical DPs can also be preposed or extracted (44% in Lindahl's Sample B). These always carry stress and normally invoke alternatives. 17

Icelandic
When Icelandic is mentioned in connection with extractions from relative clauses, it is mainly to establish that such extractions are ungrammatical (Thráinsson 2007;Zaenen 1985). There are no reports of spontaneous extractions in Icelandic in the literature. According to our hypothesis, this might indicate that preposing is not used as a method for establishing cohesion in the discourse. We start by investigating local pronoun preposing in spoken Icelandic and then turn to extractions from relative clauses.

Pronoun Preposing
Observations of informal conversations between Icelanders give the impression that they use preposing of pronouns much more seldom than Danes, Norwegians or Swedes. In order to investigate whether this impression is correct, we looked at the Icelandic part of the NDC (94 338 token, 48 speakers). We found a few examples of object preposing, all of which involve the pronoun það 'it, that', as shown in (17) and (18) (5) and (6) and are both examples of focus chaining. This type of preposing of það can also be found in written Icelandic and is judged to be natural by many speakers in the acceptability study reported in  The referent Olaf is newly introduced in A's question so this is also a case of focus chaining. Note that B contrasts hann 'him' with konuna hans 'his wife'. This suggests that preposing of entity pronouns is mainly used when the referent is contrasted with some other element. 21 The impression that object preposing is not very common in Icelandic is confirmed in Lindahl (under review). Using the same method as described for Swedish in Section 3.1, Lindahl extracted 1000 utterances with a finite verb in the Icelandic part of the NDC and then investigated all declarative main clauses in the sample. 22 Lindahl categorized the initial constituents in this sample and found substantial differences with respect to the comparable Swedish sample. These are summarized in Table 1. First we note that the proportion of subject initial utterances was much larger in Icelandic, 75.2% compared to 59.1%, whereas preposing of adverbials was more common in Swedish. Verb initial (V1) declaratives, was somewhat more common in Icelandic than in Swedish. As for preposing of objects, there were actually no unambiguous examples with a preposed object in this random sample of Icelandic utterances, compared to 46 in the Swedish sample. We interpret this as an indication that object preposing is not a particularly common strategy in spoken Icelandic.

Extraction from Relative Clauses
As mentioned in the Introduction, extraction from relative clauses in Icelandic is considered to be ungrammatical (Thráinsson 2007;Zaenen 1985). However, since these articles only mention examples with extracted lexical DPs, an interesting question is whether Icelandic speakers accept the types of extractions that are found in Swedish (see Section 3.2) even if they do not produce them spontaneously. The second author reports on such a study in . In two parallel acceptability studies, speakers of Swedish and Icelandic were asked to rate sentences with extractions, using examples similar to the spontaneously produced ones from mainland Scandinavian we have seen in this article. The extracted phrases used in all of the test sentences were það in Icelandic and det in Swedish, and they all had sentential or VP antecedents. The participants were asked to rate examples such as (20),  In brief, Lindahl found that extraction from relative clauses was rated very poorly in the Icelandic part of the study, and clearly much worse than the parallel Swedish examples. Local preposing was judged to be natural by two thirds of the participants. Long preposing from an að-clause received mixed ratings, but was rated worse than good fillers. In Swedish, preposing from att-clauses was rated on a par with good fillers. As for extraction from relative clauses, the Icelandic speakers found extraction unnatural in all of the test sentences, regardless of the type of relative clause. The Swedish speakers rated test items with a presentational relative clause as natural sounding more often than those items which contained a lexical verb. The study thus confirms previous reports that extraction from relative clauses is not acceptable in Icelandic.

Summary
Preposing of the VP or propositional anaphor það is sometimes used in Icelandic, primarily in local clauses, in a way that resembles the Swedish pattern, albeit much less frequently. However, when pronouns with entity antecedents are preposed, they reportedly receive a contrastive interpretation, just as in English (see Section 6.1). The low occurrence of preposed pronominal objects in Icelandic suggests that preposing is not used as a natural way of connecting utterances in the same way as in mainland Scandinavian. Since extraction from relative clauses seems to be unacceptable, we think this supports our hypothesis that there is a connection between using pronoun preposing as a way of connecting utterances and a willingness to accept long extractions. The study reported in  confirms that there is indeed a difference between Icelandic and Swedish in how long preposing of anaphoric pronouns is judged-such preposings are rated as natural more often in Swedish than in Icelandic, in line with our hypothesis. In addition, Lindahl found a clear effect of clause type in Icelandic. While a substantial part of the Icelandic participants rated long preposing from að-clauses as natural, even though such preposings are not common in spoken Icelandic, they did not do so for extraction from relative clauses. This suggests that there is also a structural constraint which blocks extraction from relative clauses in Icelandic.

Faroese
Turning to Faroese, very little is known about possible uses of pronoun preposing in Faroese and extraction from relative clauses has not been discussed much in previous research, apart from when it is pointed out that it is unacceptable (Platzack 2014;Thráinsson et al. 2004).

Pronoun Preposing
In the Faroese part of the NDC (64803 tokens from 20 speakers) there are more than a hundred examples with a preposed tað 'it, that', pronounced /tae/, including the two following examples. In (21) tað is a VP anaphor and in (22) tað refers back to a proposition. There are notably fewer preposed pronouns with entity antecedents, as in (23). Since there is so little data available on spoken Faroese, the second author carried out a controlled production study in Faroese. 23 The study took advantage of the fact that practically all Faroese speakers know Danish well; Danish is taught in schools from third grade and used frequently in the society. In the study, 91 native Faroese speakers were asked to translate Danish sentences into Faroese. 82 of the participants were high school students, and 9 were between 38 and 69 years at the time. The Danish sentences were chosen so as to resemble spoken dialogue. An example of the task is shown in (24). Each sentence was given in a context, also in Danish, which the participants could choose to translate if they wanted to. In the examples below, we have underlined the antecedent and italicized the pronoun, but this was not performed in the experiment. In total, 92% of the Faroese participants translated the Danish sentence using a preposing strategy, either a local or long preposing of tað or relativization with sum. However, when the Danish sentence contained a preposed entity pronoun as in (28), the Faroese participants preferred a different strategy.

17%
Only 4% of the translations replicated the preposing of an entity pronoun. The Faroese participants clearly preferred to leave the pronoun in the verb phrase or leave it out. Given the different strategies used by the Faroese participants, it seems that preposing of the VP or propositional anaphor tað is a common strategy in Faroese, but it does not extend to entity pronouns. There are both advantages and disadvantages with using a translation method. It can be a useful method for finding examples that are hard to locate in corpora and it can elicit natural sounding Faroese exchanges, especially if the materials to be translated come from natural dialogue. One disadvantage is that participants may transfer some aspects of the Danish originals into Faroese. However, when the Faroese translations deviate from the Danish originals, this is most likely an indication that the participants do not recognize this as an acceptable word order in Faroese. For example, only a few of the translations retained the preposed word order when the pronoun referred to an entity, see (29a) and (31a).
As for the frequency of object preposing, we can get at least an idea about it from a smaller sample of 300 sentences from the Faroese part of the NDC. The sample contained 192 declarative main clauses, and out of these, 15 started with a preposed object, in 12 cases tað, which was the only pronoun found. Although the material is smaller, the rate of object preposing, 7.8%, is comparable to what Lindahl found using the same type of search in Swedish, 6.5%, see Table 1. Object preposing thus seems to be fairly common in spoken Faroese and definitely more common than in Icelandic.

Extraction from Relative Clauses
According to the literature, extraction from relative clauses is not acceptable in Faroese. Platzack (2014) gives the following example, which is a translation of a Swedish example in Allwood (1976 Only two of the Faroese participants (2%) replicated the preposing in the Danish original and produced the versions in (36a,b). The rest of them used a variety of strategies to convey the content of the Danish sentence. 25 In (36c), tað is fronted, but the sentence has been reformulated using an at-clause instead of a relative clause. (36d,e,f) use slight reformulations leaving out the pronoun, and (36g) keeps the relative clause, but leaves the pronoun tí (the dative form of tað) in situ. For the two sentences where the Danish original was a presentational relative, as in (37), the Faroese speakers did produce some extractions from relative clauses in their translations, see (38) A total of 17 participants (19% of the replies) retained the Danish structure, as shown in (38a,b). One participant used the relativizer ið as in (38a) and 16 participants used the relativizer sum. The other 74 informants reformulated the sentence either by not using a relative clause in the translation, by leaving the object in situ, as shown in (38c-g), or both, or using other reformulations, as in (38h). In the two examples where the matrix verb was vera 'be', 10-20% of the translations retained the Danish extraction structure, compared to 2-4% when the verb was vita 'know' or kende 'be acquainted with '. 26 This suggests that this particular type of extraction sentence is less degraded in Faroese, and may even be acceptable for some speakers, while extraction from other relative clauses is unacceptable. Faroese could thus potentially be a language that shows evidence for the "pseudo-relative" hypothesis put forth by McCawley (1981). He proposed that the relative clause in existential sentences in English is not a true relative clause; instead it has a somewhat reduced structure. This could explain why extraction in such an environment is sometimes marginally acceptable in English (see also Chung and McCloskey 1983;Kush et al. 2013).

Summary
There are clearly similarities between Faroese and Icelandic. In both languages the VP or propositional anaphor það,tað can be preposed locally and from að/at-clauses, as shown in the acceptability and translations studies summarized above. However, the impression is that this is more common in Faroese. Preposing of entity pronouns seems not to be a natural strategy given the alternative translations provided in (29) and (31). Given that preposing of tað is fairly common in Faroese, one might expect speakers to use preposing of tað from relative clauses as well. The translation study showed that this is not the case, except for preposing in presentational relative clauses, as in (38). This is also the type of preposing that Faroese speakers are most likely to hear and read since this is the most common type in Danish (Müller and Eggers 2022). In this respect, the Faroese speakers differ from Icelandic speakers who judge extraction from presentational clauses to be unnatural to the same extent as extractions from other types of relative clauses, as discussed in Section 4.2.

Comparisons with Other Languages
We have suggested that the frequent use of preposing of unstressed anaphoric pronouns in Swedish is a crucial factor for explaining the presence of spontaneous extractions from relative clauses in this language. Since extraction from relative clauses is not found in English and the continental Germanic languages German and Dutch, it becomes relevant to investigate whether local and long preposing is used in these languages.

English
In English, preposed (topicalized) phrases are normally stressed and are understood to imply contrast. Exceptions to this are the light adverbs then and there which do not require stress in initial position when they serve to connect the utterance to the preceding context. Attempts to prepose an unstressed personal pronoun sound very strange. Compare the options for answering the question in (7) in English.
(39) When did you buy your first car? a. *It 1 I bought 1 in 1980.
b. I bought it in 1980.
The preposed version in (39a) is clearly unacceptable whereas the version with the pronoun in situ in (39b) is fine. Preposing a stressed demonstrative is grammatical, (39c), but hardly appropriate in this context. 27 When we look at the Swedish examples involving the VP anaphor det, we find that they are best rendered in English using VP ellipsis (Ø) (Bentzen et al. 2013;Hankamer and Sag 1976;Hardt 1999). Consider the question in (40) In Swedish, this is a context where the VP anaphor det typically would be preposed. Leaving it in situ, (40b), is not ungrammatical but the preposed version is preferred (Lindahl and Engdahl Forthcoming). 28 In English, repeating the finite verb and leaving out the rest of the VP is common and this is also what we have done in the English translations of the examples with VP anaphors, Swedish (5), Icelandic (17) and Faroese (21). Note that the VP ellipsis strategy can also be used in embedded clauses in English where Swedish uses preposing. There are clearly similarities between VP ellipsis in English and VP anaphor preposing in Swedish; in particular both require that the antecedent VP is available in the immediate context. In contexts where a preposed det in Swedish refers back to a recently expressed proposition, the English version can sometimes have the anaphor so, as in (42)   Given that VP ellipsis can also be used inside embedded clauses in English, as in (41b), one might ask whether this strategy can be used in English translations of extraction clauses. We have actually done so in (11) and repeat the translation here in (43).
(43) However, none of them are warm-blooded, there are no insects that are Ø.
This long distance VP ellipsis appears to be acceptable at least to some English speakers. However, the rest of the Swedish extraction examples, which involve preposed propositional or entity pronouns, are best translated into English with the pronoun in situ. 29

German and Dutch
German and Dutch are both verb second languages and local preposing is quite common, but there are differences compared with Swedish. Consider the German version of the car dialogue.

gekauft. bought
The unmarked answer would be with a preposed unstressed demonstrative pronoun, a so called d-pronoun, as in (44a), or an unstressed personal pronoun in situ, (44b). Preposing of the neuter pronoun es is very restricted, see the discussions in Frey (2006); van Craenenbroeck and Haegeman (2007) and Theiler and Bouma (2012). Corpus studies reveal that preposing of personal pronouns is uncommon compared to preposing of d-pronouns. In a newspaper corpus, it was much more common for object d-pronouns to be preposed than to appear in situ whereas no personal object pronouns were preposed (Bosch et al. 2007).
In Dutch, preposing of personal pronouns is also uncommon compared to preposing of demonstratives (d-pronouns) (van Kampen 2007).

gekocht. bought
As in German, the unmarked answers have a preposed d-pronoun or a personal pronoun in situ. Preposing an unstressed personal pronoun is not ungrammatical but pragmatically odd. 30 Bouma (2008, p. 112) found 4 preposed object personal pronouns in a 9 million words corpus of spoken Dutch (Corpus Gesproken Nederlands), compared to 2723 preposed d-pronouns.
When it comes to non-local preposing, both German and Dutch restrict this to contrastively stressed phrases. This has led some German linguists to propose that there are two types of preposing. Werner Frey distinguishes what he calls formal movement from true A-bar movement (Frey 2004(Frey , 2006(Frey , 2007. Formal movement involves local preposing to Spec,CP of the highest constituent in the middle field (IP). This accounts for preposing of subjects, both referential and expletive. Objects can also be preposed by formal movement when they are topics. On Frey's analysis this means that they have first moved to a topic position above the subject position in the middle field; this way they become the highest constituent in IP. True A-bar movement accounts for all other movement into Spec,CP and is linked to a marked interpretation, typically involving contrast. A similar proposal is made by Fanselow (2016) who distinguishes unrestricted V2 where there are no pragmatic effects of the preposing and restricted V2 which comes with pragmatic effects and induces contrast.
This distinction seems to capture the situation in German. All examples of spontaneously produced long preposings in German that we are aware of are contrastive, as predicted by Frey and Fanselow, see Andersson and Kvam (1984) and Lühr (1988). 31 Two examples from Andersson and Kvam (1984) are shown here, involving preposing of contrastive adverbial phrases from a daß 'that' clause and an ob 'if' clause. To some extent, the distinction between formal movement and true A-bar movement is relevant for Swedish as well since expletive subjects can only be preposed in the local clause. However, we find no evidence for a topic position above the subject in the middle field in Swedish which means that preposing of unstressed object pronouns cannot be handled by formal movement (Lindahl and Engdahl Forthcoming). Instead, it seems that A-bar movement in Swedish is not correlated with contrast or a marked interpretation but can be used in order to connect utterances involving focus and topic chains. 32 Highly proficient L2 speakers of German with Swedish L1 sometimes extend this type of preposing into spoken and written German with the result that "the preposed item is understood to be highlighted as very important even when it is not", as a native speaker of German recently commented. 33

Concluding Remarks
In this article, we have explored the hypothesis that frequent use of local preposing of object pronouns increases the likelihood that speakers will also use long preposing from embedded clauses as well as from relative clauses. For Swedish, it seems clear that preposing of both subject and non-subjects pronouns is a common strategy when the pronoun refers to a recently mentioned event, property or entity, that is when they are part of a focus or topic chain. This preposing strategy can be extended to subordinate clauses and to relative clauses, provided that the rest of the utterance is a relevant comment on the preposed item in the context. Acceptable long preposings and extractions are thus doubly context dependent; first, the preposed item must be part of a focus or topic chain with respect to the previous utterance and second, it must be possible to interpret the preposed item as the aboutness topic of the utterance with the rest providing a comment that meets Grice's Maxim of Relevance. The second property presumably lies behind the fact that short presentational relative clauses are most common. When both these pragmatic conditions are met, long extractions can be used and do not cause problems for the listeners/readers. 34 This does not mean that there are no syntactic constraints on extractions in Swedish. For instance, the Coordinate Structure Constraint seems to apply, see, e.g., the discussion in Lindahl (2017b).
The situation in Icelandic is clearly different. Local preposing of the VP or propositional anaphor það occurs but is used much more seldom than in Swedish. Preposing of entity pronouns is unusual and seems to have a contrastive effect instead of being used for cohesion. Furthermore the pragmatic conditions that are important in Swedish do not seem to play any role when it comes to extraction from relative clauses. Not even when the sentences were pragmatically very plausible did the Icelandic participants in  study judge them to be natural. We conclude that there must be a structural constraint operating in Icelandic which prevents such extractions and which is not affected by pragmatic conditions.
Faroese presents a more mixed picture. On the one hand, Faroese speakers often prepose the VP or sentential anaphor tað, similar to Danish. This suggests that preposing is employed as a cohesive device, but note that it does not seem to extend to entity pronouns.
On the other hand, Faroese speakers react more similar to Icelanders when it comes to extractions from relative clauses; they try to avoid using them when they translate such sentences from Danish. This suggests that extractions from relative clauses are unacceptable in Faroese, similar to Icelandic, presumably due to a structural constraint. However, there is one interesting difference between the Faroese and Icelandic speakers present in our studies. The Faroese speakers seem to accept, and occasionally produce, extractions from presentational relatives, whereas the Icelanders find these unnatural. This may be due to influence from Danish where this kind is the most common type of relative clause extraction (Müller and Eggers 2022). It may also be that these relatives have a reduced structure in Faroese which makes extraction more acceptable. More research on Faroese is clearly called for.
One important aspect of pronoun preposing in the mainland Scandinavian languages is that it often involves unstressed pronouns. If the antecedent of the pronoun has just been introduced or already is a topic in the previous utterance, it is often natural to continue with an unstressed pronoun. (This holds for both subject and non-subject pronouns). In English, object preposing tends to induce a contrastive interpretation and consequently the preposed item must be stressed; this holds both for local and non-local preposing. In Dutch and German, unstressed demonstrative pronouns can be preposed from the same clause but only contrastively stressed items can be preposed from embedded clauses. In English, one way of connecting utterances without invoking contrast is to use VP ellipsis. This often turns out to be the best translation of Swedish utterances with preposed pronouns, but there are limitations, especially with regard to relative clauses.
Our comparative investigation has shown that preposing in some languages can be used to connect the utterance to the preceding context through anaphoric chains. This means that (long) preposing by itself is not necessarily associated with a contrastive interpretation. Whether a contrastive interpretation is plausible depends rather on the context of use and the type of preposed item. This should have consequences for psycholinguistic investigations of extractions in languages such as Swedish. If, as has been common so far, the experimenter uses materials with preposed lexical DPs, then very often a contrastive reading emerges. However, if the context calls for a cohesive continuation, a focus or topic chain, then a sentence with a preposed pronoun might be more natural.
Author Contributions: E.E. is mainly responsible for writing Sections 1, 2, 3.1, 6 and 7. The remaining Sections were written together. F.L. carried out the corpus studies reported in Sections 3-5 as well as the experimental studies in Sections 4-6. All authors have read and agreed to the published version of the manuscript. Data Availability Statement: Nordic Dialect Corpus http://www.tekstlab.uio.no/nota/scandiasyn/. The Swedish data concerning extraction from relative clauses are available at https://svn.spraakbanken. gu.se/sb-arkiv/pub/lindahl/2017/. The Icelandic and Faroese data presented in this study are available on request.
Simonsen for help with the Danish, Faroese and Icelandic questionnaires and example sentences and for contacts with test participants. For the final version of the article, the comments from two anonymous reviewers were very helpful.

Conflicts of Interest:
The authors declare no conflict of interest.
Notes 1 See Chaves and Putnam (2020, p. 67) for additional references. 2 We use the term preposing rather than the commonly used term topicalization in order to distinguish the syntactic positioning from any information structural effects this may have, see Ward (1985) and the discussion in (Lindahl and Engdahl Forthcoming). 3 Swedish data are discussed in Sundman (1980, p. 59) and Teleman et al. (1999, volume 3, p. 423f.). 4 Löwenadler (2015) suggests that these common types should be seen as conventionalized constructions. 5 Søfteland (2013) calls this type a presentational cleft since the clefted constituent ingen 'nobody' is indefinite, as in presentational sentences, see (1b). She distinguishes them from it-clefts where the clefted constituent is normally definite, as in (1d). 6 Swedish is a verb second language and we assume that both subjects and non-subjects are preposed to a Spec position in the C domain in declarative clauses, which we refer to as the prefield. In (Lindahl and Engdahl Forthcoming) we develop the syntactic analysis further, adopting the bottleneck hypothesis in Holmberg (2020). 7 The sound files along with transcripts with word by word translations into English can be accessed on the web page of the Text Laboratory at the University of Oslo (http://www.tekstlab.uio.no/nota/scandiasyn/ accessed on 1 March 2022). You can search for the utterances, listen to the sound files and see the examples in context. To facilitate the search, the transcripts use standard orthography which we have retained in the examples cited, but added underlining, italics and gap locations. The Norwegian and Eldfdalian transcripts in addition contain a simplified phonetic transcription. 8 On VP anaphora in Scandinavian, see, e.g., Ørsnes (2011);Lødrup (2012) and Mikkelsen (2015). 9 Personal pronouns include first, second and third person referential pronouns. The pronoun det also functions as a non-referential expletive, in which case it cannot be stressed. See Lindahl (2017b, p. 45ff.) for details about the data collection. 13 The relative frequency for lexical verbs, 13%, is much higher in Lindahls's spoken collection than what Müller and Eggers (2022) find in their corpus study of written Danish (7/940). We suspect that a similar study of written Swedish would would also find a higher proportion of presentational relatives. 14 Establishing what is a relevant comment in a particular context is difficult, see Allwood (1976); Andersson (1982) and Engdahl (1982) for some attempts. Recently Chaves and Putnam (2020, p. 120) have introduced the term Relevance Islands for contexts which are sensitive to Gricean maxims. 15 The earliest mention of this type of extraction that we are aware of is in Mikkelsen (1894), an early grammatical descriptions of Danish. See Lindahl (2017a, p. 27) for some of his examples. 16 The same probably holds for extraction from embedded interrogatives. In  acceptability study, the participants rated sentences where the VP or propositional anaphor det had been extracted from wh-clauses as natural 65.7% of the time. 17 Lindahl (2017b, p. 146f.) also discusses examples where an extracted stressed item is understood as the focus of the utterance, typically in answers to questions. 18 The propositional or VP anaphor það is usually glossed as 'that' in the literature on Icelandic and we follow this in the examples below. 19 See Sigurðsson and Wood (2020) for an analysis of the use of conjoined subjects as in við strákarnir.

20
In Object shift contexts, such as (18b), the gap could also be located before the negative adverbial, but nothing hinges on this for our analysis. 21 A reviewer pointed out that examples with preposed object pronouns can be found in the large corpus of written Icelandic, the Gigaword Corpus (https://malheildir.arnastofnun.is/, accessed on 28 April 2022). However, if these sentences had been spoken, the pronoun would have "a regular main-clause initial stress", according to the reviewer. This may be a relevant difference with Swedish. In Lindahl and Engdahl (Forthcoming) we include Praat analyses (Boersma and Weenink 2020) of Swedish examples with preposed unstressed pronouns. These analyses show that the initial pronoun often does not form a separate prosodic phrase but is incorporated into the verb, see Myrberg and Riad (2015). Similar investigations of the prefield in Icelandic are needed as well as more informant studies. 22 This involved excluding questions, imperatives, embedded clauses and tags. The Icelandic sample contained more embedded clauses and more tags such as þú veist 'you know' than the Swedish sample with the result that there were fewer relevant clauses in Icelandic. 23 A more detailed presentation of this study in Swedish is forthcoming, Lindahl (to appear). 24 The percentages do not always add up to 100, because there are a few cases where a participant did not contribute a translation. The participants were asked to use the word order they found natural for spoken or informal written Faroese. We are showing the answers exactly how they were written by the participants, including spelling errors and any informal/non-standard spelling. For example, many speakers have chosen to leave out -ð in the definite suffix, which is silent in spoken language. 25 Since we are mainly interested in the extraction cases here, we only give the number of informants who produced such translations. 26 Note the interesting translation in (36b) where the informant inserts an additional at-clause which permits him/her to reformulate it as a presentational relative, thereby avoiding having a relative clause embedded under vide as in the Danish original. 27 (39c) would have been appropriate if the question had involved a narrow focus, When did you buy your FIRST car?. 28 It is possible that the in situ order is used more frequently in Norwegian than in Danish and Swedish, cf. Bentzen and Anderssen (2019). More comparative research is required in order to establish if this is the case and why. There is also an issue whether the VP anaphor can precede negation in Object shift, see Mikkelsen (2015); Ørsnes (2013) and Engdahl and Zaenen (2020). 29 Similarities and differences between movement and ellipsis have been much discussed, see, e.g., Johnson (2001) and Aelbrecht and Haegeman (2012). 30 The examples and judgments in (45) were supplied by Gerlof Bouma. 31 The examples in Zifonun et al. (1997) are all taken from Andersson and Kvam (1984). There is considerable discussion concerning long wh-movement, especially the so called was was construction, see the articles in Lutz et al. (2000).