2.1 The V2 phenomenon

As is well known, the so-called verb second (V2) phenomenon is a central trait of the Germanic languages other than English. This can be illustrated by main clause examples like the following from Icelandic (for detailed overviews see e.g. Holmberg Reference Holmberg, Kiss and Alexiadou2015, Angantýsson Reference Angantýsson2020).Footnote ¹

In all these examples the finite verb comes in second position, be it a main verb as in (1a, b) or an auxiliary verb as in (1c, d), regardless of the grammatical function of the initial constituent. In the following we will refer to this phenomenon as the V2 constraint for convenience, without implying anything about its theoretical or descriptive status by calling it a ‘constraint’.

But although the V2 order illustrated here typically holds for main clauses in the Germanic V2 languages, there is a well-known set of exceptions to it, as discussed in considerable detail by Holmberg (Reference Holmberg, Kiss and Alexiadou2015; see also Angantýsson Reference Angantýsson2020).Footnote ² These will be reviewed in the next subsections, using Icelandic examples for illustration as before (for a more detailed overview see Angantýsson et al. Reference Angantýsson, Edda Nowenstein and Thráinsson2021).

2.2 V1

The finite verb (apparently) sits in initial position in a number of constructions. These include the following.

As has often been pointed out in the literature, it is entirely possible that many of these examples are best analyzed as containing a non-overt preverbal element of some sort. Since we will not be concerned with (apparent) V1 constructions in this paper, we will not comment any further on these examples.

2.3 Non-subject-initial V3

A second class of well-known exceptions to V2 is characterized by initial non-subject constituents and the verb apparently in third position.

Closer inspection reveals that there are several differences between the non-subject-initial V3 examples in (3), and they are not all equally well known or straightforwardly analyzed. But it is fairly obvious that the initial constituent in left dislocation in examples like (3a) is in some sense outside the main clause (base generated there or externally merged). One piece of evidence is the fact that case-marked left-dislocated constituents show up in the nominative case whereas their pronominal copy is appropriately case-marked (Acc in (3a) above). By contrast, the case of the initial constituent in the contrastive dislocation construction in (3b) is determined by the relevant case assigner in the main clause. This suggests a closer relationship between the initial constituent and the rest of the sentence (see e.g. Thráinsson Reference Thráinsson1979:59 ff., Reference Thráinsson2007:358–359; see also Zaenen Reference Zaenen1980 and Holmberg Reference Holmberg, Kiss and Alexiadou2015:section 2.3.3).Footnote ³ Note also that in the contrastive dislocation example in (3b), the pronominal copy has been fronted and the subject follows the verb.

In examples like (3c) it appears that the finite verb is preceded by a number of adverbial constituents but it is followed by the subject. In this case the adverbial constituents are all of the same nature (temporal), so an adjunction analysis, where each adverbial is adjoined to the next one, would seem feasible. If so, examples of this sort do not represent a violation of the V2 constraint (see Holmberg Reference Holmberg, Kiss and Alexiadou2015:section 2.3.2). Another possibility is that a cartographic analysis along the lines of Rizzi (Reference Rizzi and Haegeman1997, and later work, especially Benincà & Poletto Reference Benincà, Poletto and Rizzi2004) is relevant in this context. In that kind of approach the ‘left periphery’ of sentences has a more complex structure than assumed in early generative structural accounts, which makes room for more than one preverbal constituent. But any analysis of the V2 constraint has to account for the fact that although examples like (3c) are perfectly fine, it is normally not the case that more than one constituent can precede the finite verb, as will be further discussed below.

The so-called XP-þá construction in (3d) has been analyzed in considerable detail by Jónsson (Reference Jónsson, Ramshøj Christensen, Jørgensen and Wood2019), who takes a cartographic approach and contrasts this Icelandic construction to the superficially similar XP-så-construction discussed by Eide (Reference Eide2011), for instance. Meklenborg (Reference Meklenborg2020) explores resumptive adverbials of this type in all modern V2 languages and shows that the possibility of such resumption in Icelandic is relatively restricted in comparison to the other North-Germanic languages. According to her thorough overview, the most felicitous environment for resumptive adverbials in modern Icelandic is after an initial adverbial clause and after an initial temporal PP or adverbial.

Finally, the ‘subjunctive exception’ in (3e) is often mentioned as a violation of the V2 constraint. The fact that the initial kannski (historically related to kann ske ʻmay happen’) may optionally be followed by the complementizer að ʻthat’ and the following verb shows up in the subjunctive (komi in (3e)) indicates that the structure of examples of this sort may not be that of a simple main clause. Interestingly, sentences like (4) contrast with (3e) in various ways, illustrated as follows (see e.g. Thráinsson Reference Thráinsson, Haider and Prinzhorn1986:187–188).Footnote ⁴

So if the verb following kannski is in the indicative (which is, of course, the default main clause mood), it has to show up in second position and cannot be preceded by the complementizer að. We have nothing new to say about the difference between examples like (3e) and (4).

2.4 V3 adverbials

A well-known class of subject-initial exceptions to the V2 contains adverbials of a certain type that can occur between the subject and the finite verb, as originally pointed out by Sigurðsson (Reference Sigurðsson, Dahl and Holmberg1986:144–145) and Thráinsson (Reference Thráinsson, Haider and Prinzhorn1986:175–176). In the following we will refer to these adverbials as V3 adverbials, and they are illustrated in (5).

As (5c) shows, the relevant adverbial can have a complex structure, and as (5d) indicates (based on examples in Sigurðsson Reference Sigurðsson, Dahl and Holmberg1986:145), the relevant adverbs can be combined or stacked. Interestingly, a similar class of adverbials can ‘trigger’ V2 violations in Norwegian and Swedish, for instance (see especially the detailed discussion by Brandtler & Håkansson Reference Brandtler and Håkansson2017, Julien Reference Julien2018, and Lundquist Reference Lundquist2018). This is exemplified in (6).

As example (6b) indicates, the ‘V3 adverbials’ in Mainland Scandinavian need not be single words and they can also be stacked (compare the Icelandic examples in (5c, d)). We will return to the V3 adverbials below.

As the preceding overview shows, the V2 constraint is not entirely categorical in Icelandic. It can be said to ‘leak’, and comparable leakage is found in other Germanic V2 languages although they vary in detail in this respect. One question is, then, whether this leakage makes the V2 constraint vulnerable. If so, we might expect to find exceptions to it that could be considered extensions of this leakage, but it would be interesting to see where this leak occurs and what could trigger it. We will keep this in mind in the following sections.

3.1 Why study V2 in heritage Icelandic?

As English is not a V2 language, it is often assumed that observed violations of the V2 constraint in Germanic V2 languages could be due to influence from English because of its omnipresence in modern European societies and cultures. As for North-American Icelandic (NAI), it is well known that certain characteristics of that language are in part the result of intense language contact with English (for overviews see Arnbjörnsdóttir Reference Arnbjörnsdóttir2006, Arnbjörnsdóttir & Thráinsson Reference Arnbjörnsdóttir and Thráinsson2018, Arnbjörnsdóttir, Thráinsson & Bragason Reference Arnbjörnsdóttir, Thráinsson, Bragason, Arnbjörnsóttir, Thráinsson and Bragason2023). Before describing patterns in NAI speakers’ preferences for V2/V3 orders, based on the results from Arnbjörnsdóttir, Thráinsson & Nowenstein (Reference Arnbjörnsdóttir and Thráinsson2018), it is necessary to give some background information on NAI and its speakers.

Speakers of NAI are heritage (language) speakers who align well with, for example, Polinsky’s (Reference Polinsky2018:9) definition of a ‘simultaneous or sequential (successive) bilingual whose weaker language corresponds to the minority language of their society and whose stronger language is the dominant language of that society’. NAI is therefore a heritage language, ‘a language spoken at home or otherwise readily available to young children, and crucially this language is not a dominant language of the larger (national) society’ (Rothman Reference Rothman2009:156). NAI is preserved in third and fourth generation Canadians and Americans of Icelandic descent. It is now mainly spoken in Canada, more precisely in the Interlake Region north of Winnipeg in Manitoba and in Northern Saskatchewan, and parts of the United States. The bulk of the emigration of Icelandic speakers took place between 1873 and 1914, when over 14,000 Icelanders, out of 75,000 inhabitants, moved to North America. But few emigrated from Iceland after 1914, resulting in almost no renewal of speakers since then. This has resulted in a dramatic decline of the number of persons claiming to speak Icelandic in North America, with few speakers under the age of 75. The data presented in Section 3.2 were collected between 2013 and 2015 within the pluridisciplinary project ‘Heritage Language, Linguistic Change and Cultural Identity’ (PIs Höskuldur Thráinsson and Birna Arnbjörnsdóttir). 126 Western Icelanders, as they are typically called, participated in the project, and about half of them participated in data collection targeting the V2 constraint.

The V2 constraint was one of many linguistic variables which were tested. It is of interest in the context of heritage languages more generally, in part because syntactic phenomena have been thought to be more ‘resilient’ than, for example, inflectional morphology in heritage languages, although differences in the development of complex syntax (in comparison with non-heritage speakers), that is, syntactic constructions involving the higher projections of the CP domain, have also been observed (Benmamoun et al. Reference Benmamoun, Montrul and Polinsky2013).Footnote ⁵ In this context, seemingly contradictory findings about V2 in non-heritage Icelandic and other languages are relevant. As mentioned above, V2 is generally robust in non-heritage Icelandic, despite certain leakage, and it is acquired early (Sigurjónsdóttir Reference Sigurjónsdóttir1991; see also the discussion of this point in Arnbjörnsdóttir et al. Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018). At the same time, V2 is typologically rare, potentially difficult to acquire by adults in a second language (e.g. Håkansson et al. Reference Håkansson, Pienemann and Sayehli2002, Walkden Reference Walkden2017) and the relevant cues involve non-subject-initial clauses (Westergaard Reference Westergaard2009), i.e. complex syntax. Considering this, should we expect Icelandic V2 to be preserved in a heritage language situation? If we consider the fact that V2 is robust and acquired early in non-heritage Icelandic, as well as the ‘resilience’ of syntax in heritage languages, we would expect V2 conservation in NAI. But the possible difficulties of acquiring it in a second language,Footnote ⁶ the fact that cues depend on complex syntax, and the leaks outlined in previous sections, point in the opposite direction. Previous research does too. Indeed, there is clear cross-linguistic evidence of V3 patterns (which would be ungrammatical in the non-heritage language) in English-dominant heritage language situations. This has been shown for German (Schmid Reference Schmid2002), Danish (Kühl & Heegård Petersen Reference Kühl and Heegård Petersen2018), Swedish (Larsson & Johannessen Reference Larsson, Bondi Johannessen, Richard Page and Putnam2015), and Norwegian (Eide & Hjelde Reference Eide, Hjelde, Richard Page and Putnam2015, Johannessen Reference Johannessen2015, Westergaard et al. Reference Westergaard, Lohndal and Lundquist2021). Additionally, it is relevant that the stronger language, English, has SVO and residual V2. This leads to the question of whether the preservation of the V2 constraint in heritage languages, or lack thereof, is conditioned by the word order in the dominant language, i.e. English.

In Section 3.2, we use the data from Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018) to investigate North-American speakers’ preferences for V2 or V3. Specifically, we investigate the contexts in which heritage speakers of NAI are most likely to select V3 orders. We expect the speakers’ choices to be conditioned by the exceptions documented for non-heritage Icelandic but maybe more so by the word order of the dominant language. We therefore hypothesized that the rate of V3 would be greater in non-subject-initial clauses (as opposed to subject-initial clauses), where Icelandic is different from the dominant English in that it requires V2. On the other hand, because English is SVO, the subject-initial clauses often have the same word order in the two languages. Looking at subject-initial clauses more closely, we also hypothesized that the rate of V3 would be higher when the finite verb is the lexical verb rather than an auxiliary. The reason is that lexical verbs follow sentential adverbs in English more frequently than auxiliary verbs do, and it has also been reported that children acquiring Nordic languages seem to have a stronger tendency to move auxiliaries than lexical verbs early on (Westergaard Reference Westergaard2009). Finally, we assumed that V3 adverbs would entail higher rates of V3, as opposed to V2 adverbs and negation (which also triggers V2 orders in English). Here it is important to note that we do not exclude the possibility that loss of V2 represents an internal change in heritage Icelandic instead of being contact-induced, as will be elaborated on in later sections.

3.2 The results of the study

The data were collected during three visits to North America in 2013–2014. The regions visited were Alberta, British Columbia, Manitoba, Saskatchewan, North Dakota, and Washington state. V2 was tested in 60 participants aged 26–98, with a mean age of 77. As described in Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018), all the participants were exposed to Icelandic from birth, but 50.9% of them were exposed to English from birth as well, with the rest not encountering English until school. V2/V3 orders were mainly tested as part of a larger forced-choice task battery where participants chose between two or more options. A total of 28 sentences were tested for the V2/V3 variable, 8 non-subject-initial and 20 subject-initial. A typical example of the testing set-up is shown in (7).

As shown here, the testing set-up contained a context sentence followed by the actual test sentence. The test sentence in (7) is a non-subject-initial example and the first option in this case is the V2 variant (skulum við ‘shall we’) and the second is the V3 version (the order of the variants was randomized in the test). In a corresponding English example only the V3 version would be grammatical. Further examples of the V2/V3 sentences tested are given in a simplified fashion in (8) with the context sentences omitted for space-saving reasons.

The subject-initial test sentence in (8a) contains the sentence adverb alltaf ‘always’, which would normally follow the finite verb in non-heritage Icelandic, observing the V2 constraint. The same is true of (8b). In English it would be more natural for most speakers to have the adverb always precede the finite main verb in a sentence corresponding to (8a), whereas it would be more common to have the finite auxiliary precede the adverb always in an example corresponding to (8b).

The participants were presented with the context and test sentences orally but could also read them on a tablet screen before selecting one or both options. This type of relative judgment contrasts with the absolute judgment task used in various other studies, including Maling & Sigurjónsdóttir (Reference Maling and Sigurjónsdóttir2002), since both options are apparent, the targeted construction clear, and the attention to speech therefore arguably higher, prompting more standard forms. As is common in the fieldwork setting, there was extensive variation in the amount of presented sentences, making statistical analyses more complicated, but mixed effects logistic regression models were run when possible (lme4 in R, see Bates et al. Reference Bates, Mächler, Bolker and Walker2015; full modeling results are provided in the supplementary material). In addition to the forced-choice task, elicitation data were collected through storytelling tasks but have only been partially analyzed.Footnote ⁷

Before presenting the results of the study, it is important to note that on average the NAI subjects selected V2 more often than the V3 alternative in both subject-initial and non-subject-initial constructions (see Arnbjörnsdóttir et al. Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018). As already pointed out, V2 would be expected in both constructions in non-heritage Icelandic but not in English. The fact that V2 is preferred over V3 would be consistent with the assumption that a V2 grammar is the target grammar during the acquisition of NAI. Hence that is how we will frame the following discussion.

We will first present the results for the selection between V2 and V3 in subject-initial and non-subject-initial examples. Figure 1 shows the difference in V3 rate between these two types of sentences.Footnote ⁸

Figure 1. Rate of V3 selection and standard error in forced-choice task by fronting type, NAI.

As can be seen, V3 is selected more frequently when fronting of a non-subject occurs.Footnote ⁹ Non-subject-initial constructions also arguably involve more complexity (or movement) than subject-initial sentences, a factor which might contribute to lack of preservation in heritage languages. This could be explained by various reasons, one of them being simply less exposure to non-subject-initial clauses. In fact, Westergaard et al. (Reference Westergaard, Lohndal and Lundquist2021) report fewer non-subject-initial clauses in heritage Norwegian and associate this loss of context with lower rates of V2. It is still interesting that across the data, V2 is more commonly selected (73%) than V3 in our results. Even in non-subject-initial sentences, the rate of V3 only reaches 41%, showing that the speakers are clearly far from aligning completely with the word order of the dominant language. This fits individual response profiles. 56.3% of speakers who were presented with more than one non-subject-initial sentence (N = 48) showed some intra-speaker variation, the rest chose V2 consistently, meaning that no speaker chose V3 consistently.

Taking a closer look at the subject-initial clauses, Figure 2 shows that the speakers align with the patterns in English and select V3 more often with lexical verbs than with auxiliaries. Figure 3 furthermore shows that the so-called V3 adverbs in non-heritage Icelandic trigger more V3 selection than negation (V2 with negation is consistent with English) and V2 adverbs. Interestingly, the statistical analysis shows that the effect of verb type disappears once we correct for adverb type.Footnote ¹⁰ This means that the contrast in Figure 2 Footnote ¹¹ is actually a reflection of the contrast between adverb types in Figure 3, with more of the sentences containing lexical verbs also having V2 adverbs or negation. We see that the adverb type triggers the sharpest contrast (Figure 3), with V3 orders only being selected in 14% of sentences with V2 adverbs and negation but reaching 42% with V3 adverbs.

Figure 2. Rate of V3 selection and standard error in forced-choice task by verb type, NAI.

Figure 3. Rate of V3 selection and standard error in forced-choice task by adverb type, NAI.

Summarizing this reanalysis of the data from Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018), it is clear that our results show only a partial preservation of V2 in NAI in comparison to non-heritage Icelandic, with speakers selecting V3 orders to varying degrees depending on context in a forced-choice task. V3 was selected more frequently in non-subject-initial sentences. Within subject-initial sentences, adverb type (and negation) also mattered. In line with the somewhat contradictory predictions which can be deduced from the literature, these results cannot be interpreted in any straightforward manner. V2 is in part present, which would be expected considering its robustness, early acquisition, and previously observed ‘resilience’ of syntactic phenomena in word order. But V3 is also very present, consistent with the dominant language, difficulties in late acquistion of V2, and possibly less exposure to the necessary cues. Additionally, the conditioning patterns point towards influence from English (Figure 1) as well as the expansion of patterns existing in non-heritage Icelandic (Figure 3). The contrast between the rate of V3 in clauses which are subject-initial and clauses with fronting of a non-subject is consistent with English word order, while the adverb conditioning in Figure 3 is consistent with patterns in non-heritage Icelandic. This suggests that various factors are crucial to the understanding of possible and existing leaks to the V2 constraint. In the following section, we explore additional pieces to this puzzle by studying more closely the exceptions to V2 found in non-heritage Icelandic.

4.1 The goal of this section

In their paper on V2 in NAI, Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018) report on a small study designed to compare the selection of V2 and V3 in NAI and non-heritage Icelandic. In this study, the following four test sentences that had been used in the study of NAI were presented to 15 speakers of non-heritage Icelandic under the exact same conditions. The participants in Iceland and the NAI speakers were evenly matched in age, all being over 70 years of age.

A comparison of the results for around 15 speakers of NAI and the 15 speakers of non-heritage Icelandic is shown in Table 1.Footnote ¹²

Table 1. Selection of V2/V3 orders in non-subject-initial structures in an interview. Number of speakers in parentheses

This table is in line with the general belief that the V2 constraint is quite robust in non-heritage Icelandic. Thus the following remark by Maling & Sigurjónsdóttir (Reference Maling and Sigurjónsdóttir2002:114n), in their well-known paper on the new impersonal (aka new passive) construction in Icelandic, came as a surprise:

What Maling & Sigurjónsdóttir are talking about in this quote is the fact that they included example (10b) in their test as a control sentence. They expected that everybody taking the test seriously would reject it since it would be a clear ‘V2 violation’ and (10a) would be the only acceptable variant.

In (10a) the finite verb is in second position, as expected for a V2 language like Icelandic, but in (10b) it is in third position. So why did a considerable number of Maling & Sigurjónsdóttir’s subjects accept this variant? Three possibilities come to mind.

We will now report on two studies that were designed to study these possibilities.

4.2 Non-subject-initial V3 examples

4.2.1 Results from the MoLiCoDiLaCo project

Non-subject-initial V3 matrix clauses were tested within the MoLiCoDiLaCo project (short for Modeling the Linguistic Consequences of Digital Language Contact) at the University of Iceland (PIs Sigríður Sigurjónsdóttir and Eiríkur Rögnvaldsson). The data were collected between 2017 and 2019, and consisted of a large-scale online survey as well as follow-up interviews and testing sessions (Sigurjónsdóttir & Rögnvaldsson Reference Sigurjónsdóttir and Rögnvaldsson2018, Nowenstein & Sigurjónsdóttir Reference Nowenstein and Sigurjónsdóttir2021, Sigurjónsdóttir & Nowenstein Reference Sigurjónsdóttir and Edda Nowenstein2021). The main goal of the project was to document the presence of English in the Icelandic language community and investigate the possible effects of English input and use on Icelandic. Amongst other things, data on the distribution of English and Icelandic in participants’ language environment were collected and a judgment task administered. 1,615 speakers aged 13–98 participated in the online questionnaire (completed independently by participants) and 126 did the follow-up sessions.Footnote ¹³ Sentences were rated on a five-point Likert scale, ranging from ‘unacceptable’ to ‘completely natural’, and each V3 sentence appeared with a context sentence. Each participant rated only one V3 sentence online (four versions were assigned at random, all with preposed adverbials and a pronominal subject) and two sentences in a comparable survey administered during the interviews (one with a pronominal subject and one with a full NP). Additionally, the participants who had accepted V3 sentences in the online survey (ratings 4 and 5) also rated four recorded sentences consisting of two minimal pairs where the variables subject type (pronoun/full NP) and a clear intonational breakFootnote ¹⁴ after the preposed element (break/no break) were manipulated. For the purpose of this paper, we ran regression analyses and nested mixed model comparisons to find out whether contact with English predicted V3 acceptance.

The examples in (12) show the overall acceptance rate of non-subject-initial V3 in the large-scale online survey described above, where 1,504 participants (others did not complete the task) rated one of four sentences targeting the variable.Footnote ¹⁵

This indicates that the mean acceptance rate was 27.5%, which is unexpectedly high. The question is then what the reason could be.

In (11) we listed three possible reasons for acceptance of V3 examples and these were all tested for in the analysis of the data. First, an English contact measure developed within the MoLiCoDiLaCo project, based on participants’ answers on their English input and use, did not have a significant effect on their V3 acceptance. Second, V3 acceptance did not show a clear relationship with age, as one might have expected if this V3 acceptance were an indication of an ongoing expansion of existing V3 patterns in the language. This was also surprising because syntactic variation in Icelandic generally correlates with age rather than other social factors (see e.g. the papers in Thráinsson et al. (eds.) Reference Thráinsson, Angantýsson and Freyr Sigurðsson2015). The regression model with the contact measure, including age as well, explained only a minor part of the variation (adjusted R-squared, 0.004017, F = 2.904, DF = 1,413, p < 0.05; full modeling results are provided in the supplementary material). This leaves the third possibility, namely that the V3 acceptance is due to a task effect, i.e. some kind of stable processing effect where participants parse the sentence as V2 even though it is V3, possibly only in a reading task.

To investigate this (and other variables) further, 126 MoLiCoDiLaCo participants came to the University of Iceland for more extensive testing, with 35 of them (28%) having accepted V3 in the online survey. In this test, the acceptance rate of non-subject-initial V3 dropped to 10.3% for light pronominal subjects and 5.6% for full NP subjects, as shown in (13). This result further points towards a processing task effect, but the difference between acceptance of pronominal and non-pronominal subjects is interesting and we will come back to this.

Although the set-up of the in-person questionnaire was identical to the online survey, the participants were in a different testing environment (university setting, investigator present) which might have prompted another approach to the task. In the testing interviews, more precise information about participants’ English input and use were collected. These measures (average English use, input, and proportion) as well as the age of the participant, rating for V3 in the online survey and subject type (pronoun/full NP), were used in a nested mixed effects model comparison with the V3 rating of the in-person survey as the outcome variable.Footnote ¹⁶ The only variable which improved the fit of the base model significantly was V3 acceptance in the online survey (χ2 (1) = 4.61, p < 0.05). As can be seen in Figure 4, speakers who accepted V3 in the online survey rated the V3 sentences in the in-person survey slightly higher.

Figure 4. V3 rating and standard error in the in-person questionnaire by rejection/acceptance in the online survey, N = 126.

This is still only a small effect, and the participants who accepted V3 in person did not necessarily do so online. Finally, the 35 subjects who accepted V3 in the online survey were asked to judge V3 orders by listening to a recording. As shown in (14), it does not really matter whether there is an intonation break (represented by the comma in (14b)) after the preposed constituent when the subject is pronominal.

When there is a full NP subject as in (15), the example improves if an intonation break is inserted, as in example (15b), but this variable did not significantly improve the fit of a mixed effects model with a subject type/intonation break interaction.

If this effect is in fact present in the judgments of speakers who accept V3, the effect in the general population is probably too small to be detected with N = 35. In general, we can see that of the 35 speakers who accepted V3 in the online survey, only 17.1% of them accepted the sentences when listening to a recording containing the construction. This indicates that a task effect is indeed present, perhaps indicating that participants read the V3 order as V2, possibly in part due to the increased processing cost of parsing a V2 violation (documented by Sayehli et al. Reference Sayehli, Gullberg, Newman and Andersson2022 for Swedish).

Although a large part of the observed V3 matrix clause acceptance in the MoLiCoDiLaCo study seems to be due to task effects for the most part, there is still evidence for some leaks from what has traditionally been viewed as robust and categorical. This includes the fact that V3 examples with pronominal subjects get a higher acceptance rate than those with full NP subjects, but an intonation break improves the latter. A further study of the nature of these leaks is presented in the next subsection.

4.2.2 Results from a new online study

The questionnaire data presented in this subsection were collected online by Jónsdóttir (Reference Jónsdóttir2021) in March 2021 (159 participants of various ages).Footnote ¹⁷ The questionnaire included 28 minimal pairs contrasting subject-initial and non-subject-initial V2/V3 orders in matrix clauses. For each test sentence, there were five possible responses, exactly the same as in the online survey in the MoLiCoDiLaCo project:

Seven out of the 28 minimal pairs were introduced with a context sentence in order to help the participants get the intended reading of the test sentences. For a direct comparison, the four relevant examples from the MoLiCoDiLaCo project (see (12)) were used in the questionnaire, and most of the other test sentences were either identical to or modeled after examples from Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018). It should be emphasized here that in Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018) the participants were asked to select between alternatives, while in Jónsdóttir’s (Reference Jónsdóttir2021) survey the speakers were asked to evaluate every example.

Based on the previously presented results from MoLiCoDiLaCo and the empirical observation made by Maling and Sigurjónsdóttir (Reference Maling and Sigurjónsdóttir2002), it is interesting to see to what extent the participants in Jónsdóttir’s (Reference Jónsdóttir2021) survey accepted sentences of this type. The evaluation of each non-subject-initial V3 example is shown in (17). Examples (17b) and (17d, e, f) are from MoLiCoDiLaCo; the other examples are modeled after Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018). As can be seen, they are arranged here in the order from the most frequently to the least frequently accepted. 159 speakers participated in this survey and the acceptance rate is the percentage of speakers.

Note that 5 out of the 8 non-subject-initial test sentences presented here receive an acceptance rate of 15% or higher. In the most commonly accepted sentences the fronted element is an adverbial time frame-setter and the subject is a light pronoun.Footnote ¹⁸ We will return to this issue in the discussion section. For the moment we will just note that a couple of examples of non-subject-initial V3 are accepted by more than 20% of the participants (around 30 out of 159 speakers). This can hardly be attributed to data noise.

4.3 Subject-initial V3 examples

The examples in (18) show the overall acceptance rate by 159 speakers of some selected subject-initial V3 sentences with potential V3 adverbials, arranged as before from the most frequently accepted to the least frequently accepted examples. The test sentences in (18b, c, f, h) are taken from Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018).

With the exception of (18e) and (18g), there is a speaker-oriented adverb intervening between the subject and the finite verb. Beforehand, one would expect such sentences to be acceptable for the most part because they contain the typical V3 adverbs mentioned above. However, some of them receive quite low scores. A possible explanation is that in all cases the participants were also asked to evaluate an equivalent example with subject-initial V2. Another possibility is that the participants were not thinking of the appropriate intonation since they were reading the examples online rather than listening to them. The relatively high acceptance rate of the English-like word order in (18e) is somewhat surprising as well.

In (19) there is a central sentence adverb intervening between the subject and the finite verb; the examples are all from Arnbjörnsdóttir et al. (Reference Arnbjörnsdóttir, Thráinsson and Edda Nowenstein2018). A priori, one would not expect them to receive high scores. They were presented to the same 159 speakers as the examples in (18) and the acceptance rate was calculated in the same way as before.

The overall numbers are obviously much lower than we saw for subject-initial V3 with potential V3 adverbs. Interestingly, however, some 10% of the speakers accepted the English-like word order in (19a). This might be linked to the fact that the adverb is very short.Footnote ¹⁹ But if the possibility of subject-initial V3 in sentences like (19b) and (19c) was somehow linked to English influence (auxiliaries, unlike main verbs, usually do not follow adverbs of this type in English), one would expect (19c) to receive a higher score than (19b), but that is not the case. However, since the scores are so low for both sentences, nothing can be concluded from this comparison.Footnote ²⁰

4.4 An interim conclusion for V2 violations in non-heritage Icelandic

The main results for the subject-initial V3 examples are summarized in (20).

Considering the non-subject-initial V3 constructions, the most interesting results described in this section include the following.

These results bear a certain resemblance to the V3 data reported for Germanic urban dialects by Walkden (Reference Walkden2017), in particular the preference for preposed adjunct over preposed arguments and for pronominal subjects over non-pronominal ones. The highest-rated non-subject-initital V3 examples also have similar clause-initial properties as the most natural resumptive V3 structures in Icelandic, described in Meklenborg (Reference Meklenborg2020) and Van Kemenade & Meklenborg (Reference Kemenade, Meklenborg, Wolfe and Meklenborg2021). We will come back to these resemblances in the discussion section at the end of the paper.

6.1 Summary of the results

First, let us summarize some of the results from the study of V2/V3 in heritage Icelandic (NAI) outlined in Section 3.

We argued that while (26a) suggests influence from English (V3 is the common order in non-subject-initial sentences in English), (26b) is consistent with the assumption that the target grammar of the NAI speakers is an Icelandic V2 grammar.

To verify the difference between the acceptance of V2 violations in NAI and non-heritage Icelandic, part of the V2 experiment that had been used in North America was administered to an equally large group of speakers in Iceland who were evenly matched with respect to age (the mean age of the participants was over 70 in both groups). The difference was very clear: the speakers of non-heritage Icelandic did not accept (or select) the V3 variants (see the discussion around Table 1 in Section 4.1 above). But because there had been some evidence that certain V2 violations were accepted by some Icelandic speakers, the extent and nature of such violations were examined in two studies, partly to check for evidence from English and partly to see if younger speakers were more likely to accept such violations. Somewhat surprisingly, the acceptance of V3 showed no significant relationship with the participants’ age or any other sociological variables. Equally surprisingly perhaps, there was no significant relationship between acceptance of V3 and the participants’ ‘exposure to English’ (as measured by their answers to questions about their English input and use), making influence from English unlikely. An investigation of the acceptance of subject-initial V3 examples also revealed that these were typically rejected unless they contained the well-known ‘V3 adverbs’ (e.g. ‘just’, ‘simply’, ‘obviously’). Examples with other adverbs intervening between the subject and the finite verbs, such as ‘sometimes’ and ‘always’, generally received a very low score. Since comparable examples are typically fine in English, this is perhaps not what we would have expected if V3 orders in Icelandic were the result of influence from English.

Nevertheless, some of the non-subject-initial examples were accepted by an unexpectedly large proportion of the participants (mean acceptance 27.5% of the population in one of the online studies). The fact that the acceptance rate for V3 dropped significantly in interviews indicates that this high acceptance rate can partly be explained as a task effect, that is, the participants were actually processing some of the V3 examples as V2 constructions when presented with them visually (in reading). But a group of speakers persisted in their acceptance of V3, even when presented with audio recordings of the sentences. This suggests that the online acceptance cannot be explained solely as a task effect (perhaps based on processing cost; see Sayehli et al. Reference Sayehli, Gullberg, Newman and Andersson2022) as originally suggested by Maling & Sigurjónsdóttir (Reference Maling and Sigurjónsdóttir2002).

A further investigation of the nature of the V2 violations studied revealed interesting facts. First, the highest-rated non-subject-initial V3 examples had a fronted adjunct (a time or space adverbial) and a pronominal subject. Comparable non-subject-initial examples with non-pronominal subjects received a lower score, as did examples with fronted arguments and clauses. Second, the insertion of an intonation break between fronted elements and a non-pronominal subject led to somewhat increased acceptance (see the discussion around the examples in (15)).

Finally, we did a brief study of V2 violations in Icelandic poetry. It is well known, of course, that poetic language does not always obey the same linguistic rules as non-poetic language. This seems to be especially true of rules involving word order (see e.g. O’Neil Reference O’Neil and Kenstowicz2001, Eythórsson Reference Eythórsson2009, Fabb Reference Fabb2010, Thoms Reference Thoms2010, Magnusson Petzell & Hellberg Reference Magnusson Petzell and Hellberg2014, Häussler et al. Reference Häussler, Anne Mucha, Weskott and Wierzba2019). Linguists disagree as to whether the rules of poetic language should be seen as an extension of the ‘standard’ rules of the relevant language, or at least as rules permitted by UG in some languages (see e.g. O’Neil Reference O’Neil and Kenstowicz2001, Eythórsson Reference Eythórsson2009, Fabb Reference Fabb2010), or whether poetic language is to some extent generated by different means (Thoms Reference Thoms2010).Footnote ²² We do not have much to say about that issue at this point, but we would like to make two comments. First, many of the non-subject-initial V3 examples from poetry found in our study are similar to those accepted most frequently by Icelandic speakers in an experimental situation involving Icelandic prose: the fronted element is an adjunct and the subject is pronominal. Second, the most obvious difference between fronted constituents in the poetry we have looked at and fronted constituents in non-poetic Icelandic is the fact that there does not seem to be a clear upper limit on the number of fronted constituents in poetic language. In addition, it is not obvious that their relative order is governed by any syntactic principles (see Thoms Reference Thoms2010:43 passim for a similar observation on displacements in English poetry).

6.2 Discussion

In our study of poetic language we concentrated on lyrics that are typically sung to children. The main reason was that we wanted to know how extensive violations of the V2 constraint were in this kind of input to children who are acquiring language. It turned out that they were quite extensive. Still it has been pointed out in the literature that Icelandic children seem to acquire the V2 constraint quite early. The same is apparently true of Norwegian children (see Westergaard Reference Westergaard2009). Interestingly, Norwegian children are also ‘bombarded’ with the same kind of V2 violations in lyrics as shown by the following examples.Footnote ²³

Thus the Norwegian facts support the observation made above. Children acquiring language can disregard the potentially ‘misleading’ word order input they receive in lyrics frequently sung (or read) to them, just as they can ignore certain instances of grammatically deviant input when building their grammar. Still, they do encounter these V2 violations in their input, and we could therefore speculate that this exposure builds up a tolerance for V3 (and other violations) and an ability to reinterpret it as V2. This exercise might consequently play a role in the task effects described in previous sections.

Another theoretically interesting fact about the V2 violations discussed is the following: the fronting data from non-heritage Icelandic typically resemble the V3 data reported for Germanic urban dialects by Walkden (Reference Walkden2017), who largely based his study on data reported by Freywald et al. (Reference Freywald, Leonie Cornips, Nistov, Opsahl, Nortier and Svendsen2015) (see also Meklenborg Reference Meklenborg2020, Van Kemenade & Meklenborg Reference Kemenade, Meklenborg, Wolfe and Meklenborg2021). Walkden’s main findings can be summarized as follows.

Although V3 is still exceptional in such constructions in Icelandic, the experimental data discussed in Section 4 showed that some V3 examples that had the properties described by Walkden were typically better received than other V3 main clauses. Many of the examples cited from Icelandic and Norwegian children’s lyrics are also of this kind. This is very interesting because the V3 data presented by Walkden arise in a completely different linguistic situation. Hence the observed similarity indicates that this particular type of V2 violation is somehow more ‘natural’ or more likely to occur than others. But the data from the study of NAI show that the violations of V2 can be extended, as it were, in a heritage language, presumably in part because of influence from the dominating language of the area. Different extensions of these violations are found in the poetic language studied, although ‘Walkden-like’ examples were also common there.

Thus we can conclude that violations of V2 that have the properties described by Walkden have a special status. They are the most common type, together with the subject-initial V3 examples involving the V3 adverbs discussed above (e.g. ‘just’, ‘simply’). Other types of V2 violation are more exceptional. They are found in heritage Icelandic (NAI) and in poetic language. But in some instances such violations can be made more acceptable with certain prosodic factors. This has been demonstrated by Breitbart (Reference Breitbart2022) for German V2 violations. It is also reminiscent of the observation made by Eide (Reference Eide2011) in her study of Norwegian non-V2 declaratives. Eide argues that in the Norwegian så-construction (e.g. [I forrige uke] så sa Marit at … lit. ‘In last week then said Mary that …’, i.e. ‘Last week Marit said that …’) and the copy left dislocation (CLD) as she calls it (e.g. [Ei leiligheit] det skulle vi hatt = lit. ‘An apartment it should we had’, i.e. ‘An apartment, we should have had one’), the resumptive elements (så and the relevant pronominal copy) are necessarily light. Hence they can occupy the ‘Wackernagel position’ right before the verb whereas heavier constituents cannot. She also points out (Eide Reference Eide2011:191; see also Anderson Reference Anderson1993) that ‘the constructions we are discussing here are typically spoken phenomena … making the potential importance of prosody and stress patterns more likely’.

In any case, the data presented here show that V3 acceptance persists through various testing scenarios in a minority of non-heritage Icelandic speakers, and that it is not predicted by a possible contact scenario. We believe that further investigation of such a marginal phenomenon with well-established parallels in related languages might be informative in the context of language variation and change more broadly, but future work should also further investigate the implications for work on acceptability judgment reliability and sentence processing (e.g. Ferreira Reference Ferreira2005, Sprouse & Almeida Reference Sprouse and Almeida2012).