Since 2003 and for a period of about ten years a network of ten research groups in the five Nordic countries worked within the Scandinavian Dialect Syntax project (ScanDiaSyn) towards mapping syntactic variation across the North Germanic dialect continuum. Two major research tools grew out of the collaboration: the Nordic Dialect Corpus (NDC) and the Nordic Syntax Database (NSD), see Section 2.

In 2014 The Nordic Atlas of Language Structures Online (NALS) was launched with some 50 papers based on data from the two databases organised under the following six general topics: (i) Noun phrases, (ii) Verb phrase: Argument structure and verb particles, (iii) Verb placement, (iv) Middle field/TP: Subject placement, object shift, auxiliaries and tense marking, (v) Left and right periphery: complementisers, questions, extractions etc., and (vi) Binding and co-reference. NALS is formally organised as a journal, and has continued afterwards to publish papers that map variation across the North Germanic dialect continuum.

In this paper we will discuss two syntactic phenomena on the basis of data available from the dual NDC/NSD research infrastructure: (i) the relative placement of adverbs and infinitive markers (±split infinitive), and (ii) non-V2 in matrix wh-questions across Norwegian dialects. For each of these we discuss possibilities, limitations and achievements posed by the research infrastructure as promised by the title of the paper. Section 2 presents the two types of research infrastructure. Section 3 presents the investigation of placement of adverbs and infinitives, and also illustrates how the corpus and database can be used to find the empirical evidence needed. Section 4 contains a thorough presentation of the variation of word order in wh-questions, while Section 5 concludes the paper.

The ScanDiaSyn network of research groups consisted of a number of smaller and bigger projects funded by a variety of sources. In effect, ScanDiaSyn was therefore a project umbrella and there was also funding for the network itself from the Nordic research bodies NordForsk and NOS-HS. In the various countries national funding was obtained to carry out basic and systematic data collection in the projects DanDiaSyn, FinDiaSyn, IceDiaSyn, NorDiaSyn, and SweDiaSyn. The Norwegian national project was responsible for the technical solutions, including building the Nordic Dialect Corpus (Johannessen et al. 2009, 2014) and the Nordic Syntax Database (Lindstad et al. 2009) as well as getting the annotations (double transcriptions and tagging) done (Johannessen 2017). Furthermore, between 2005 and 2010 the ScanDiaSyn umbrella also included substantial funding for the Nordic Center of Excellence in Microcomparative Syntax (NORMS), which financed a number of postdoctoral visiting fellows, cross-institutional thematic research groups, fieldwork trips to targeted areas in the Nordic countries, as well as seminars and conferences. Between 2005 and 2010 a project blog was kept up with entries written both in Scandinavian and English (mainly), including quite extensive reports from the NORMS field trips, and these reports can be found at https://tinyurl.com/NORMSfieldwork.

In total, data from 228 different locations across the Nordic countries were collected in the project, and the data were of two types: (i) recordings of spontaneous speech from conversations between and interviews with dialect speakers, and (ii) judgment data on a list of prepared test sentences that probe a number of different syntactic constructions. These data formed the basis for the two different infrastructures in the project: the Nordic Dialect Corpus and the Nordic Syntax Database.

The Nordic Dialect Corpus (NDC) consists of spontaneous speech. Ideally, each location in the corpus should be represented by four speakers (two age groups and both genders), and each speaker would be recorded both in an interview and a free conversation. However, in practice, the different funding situations in the different countries meant that this could not always be achieved. For example, in Sweden, there was less funding, but fortunately the project was given access to interview data from a recently completed project, SweDia 2000. On the other end of the scale was Norway, where funding covered extensive fieldwork across the whole country in addition to the development of the technical research infrastructure. The NDC corpus contains transcribed recordings with over 3 Million words, by 823 dialect speakers from 228 locations in six countries (Denmark, Faroe Islands, Finland, Iceland, Norway and Sweden), mainly sampled between 2005 and 2010.

The Nordic Syntax Database (NSD) contains judgment data for a number of test sentences (140–240, depending on country) probing variation in various (morpho)syntactic phenomena from more than 900 informants – many of the same ones as in the NDC. The sentences were carefully selected to represent grammatical phenomena that we knew varied across dialects in one or more of the five languages. A major goal was to present sentences even in locations where a phenomenon was thought not to exist. This way we would obtain negative as well as positive data, making it possible to draw new syntactic isoglosses across the North Germanic dialect continuum.

The locations represented in the infrastructure were chosen to ensure a good geographical distribution and also to cover well-known local/regional dialect boundaries, and although most locations are in rural areas there are also some cities in the Norwegian and Danish list of locations. For the Swedish speaking area, the SweDia 2000 material (see above) only included material from rural and semi-rural locations, and hence no Swedish (and Finnish) cities are represented in the infrastructure. Since geographical distribution was prioritised, the locations were not balanced for population size, hence there are both “small” and “big” dialects in the sample.

The fieldworkers were a good mix of senior and junior researchers and student assistants who travelled to the locations of the informants to carry out the data collection. In most cases two fieldworkers would do the data sampling together. In some cases the fieldworkers would speak a similar dialect as the informants, but in other cases not.

The test sentences in the questionnaire were pre-recorded by a speaker of the same regional variety so that the pronunciation would be the same as or similar to that of the informants. This was done to avoid the sentences being deemed unacceptable because of pronunciation. The informants were instructed to judge the sentences on a Likert scale from 1 (bad) to 5 (good) according to their own dialect intuitions, and they would give their judgments after hearing each pre-recorded sentence. These questionnaire sessions lasted about one to one and a half hours. The recording sessions consisted of an interview of about 15–20 minutes with one of the fieldworkers and a conversation with another informant of about 30–45 minutes. Sampling data from four informants at one location normally took a full work day.

The informants were typically recruited through a local contact person according to a set of criteria targeting traditional dialect speakers. Beyond age and gender information sociolinguistic information about the informants was not recorded. The informants were not paid, but given a symbolic gift as a token of gratitude and they were also served coffee, tea and soft drinks as well as fruit and (non-crunchy) candy at the sessions. For further details about the project logistics, including methodologies, see Vangsnes (2007a; b), Johannessen et al. (2008), Lindstad et al. (2009), Johannessen et al. (2014).

The papers in the NALS Journal, which typically exploit both the the NDC (corpus) and the NSD (judgment database), show that the dialect infrastructure developed under the ScanDiaSyn umbrella does indeed allow researchers to investigate new isoglosses and dialect phenomena across the Nordic countries. The two case studies to be presented in this paper should serve to make the same point.

The differing relative placement of infinitival markers and adverbs across the Scandinavian written languages is a well-known issue (see for example Hulthén 1947; Faarlund et al. 1997). The received wisdom is that Danish requires the adverb (especially the negative adverb) to be placed before the infinitival marker, as in (1), whereas Swedish requires the adverb to follow the infinitival marker, as in (2), and Norwegian is supposed to accept both orders. The “Danish pattern” is what often is referred to as “unsplit infinitives” since the adverb does not split the infinitival marker from the verb, and conversely the “Swedish pattern” is generally referred to as “split infinitives” precisely since the adverb separates the infinitival marker from the verb. A third pattern, given in (3), is the standard word order in Icelandic infinitivals (with an infinitival marker), and this pattern was also tested for the mainland languages (and Faroese). The sentences are all given here with Bokmål Norwegian words and orthography.

The issue of “split infinitive” is well-known also from English grammar and is a source of controversy where prescriptivists favour the use of unsplit over split infinitives (see Huddleston & Pullum 2002: 581f, for references). This same prescriptivism can be witnessed also in the context of Norwegian as unsplit infinitives have traditionally been the recommended word order. Furthermore, the Norwegian reference grammar (Faarlund et al. 1997: 997) claims that while both orders are possible in Norwegian, the unsplit pattern is the most natural for many language users (Faarlund 1997: 997). We will see below that the NSD database does not support this claim.

Figure 1 is a screenshot from the NSD of sentence (1) as it was presented to the informants, and we see that the Swedish test sentence is worded differently but nevertheless probes the same word order as its Norwegian and Danish counterparts. Similar adjustments across the language-specific questionnaires were made in several cases, but each “bundle of test sentences” probing a specific phenomenon is always given a unique identity in the database, in this case the number 143.

Searching the database gives a long list with all the answers of all the informants, divided over several results pages. Each informant would grade this sentence (and all the others presented to them) on a scale from 1 to 5. These results are rendered as colour codes in the database, to enable the researcher to assess the results at a glance, see Figure 2.

Even more visually illustrative are the maps that can be generated from the result page. The results from the sentence evaluations performed across the three countries Denmark, Norway and Sweden show very clear and different results, see Maps 1, 2 and 3. A white marker means that a sentence has a mean score of 4 or higher at that geographical location, whereas a black marker means that it has a mean score of 2 or lower. In other words, a white marker indicates acceptability, a black one non-acceptability.

There is a clear acceptance of the unsplit, “Danish pattern” in Denmark. Furthermore, in Sweden the unsplit pattern is quite clearly not accepted, but more strikingly, at the great majority of Norwegian locations the pattern is also dismissed. The split, “Swedish pattern”, on the other hand, is accepted not just in Sweden, but also in most of Norway, although there is an enclave in the central part of the country (the Trøndelag area) where the split pattern seems to be rejected at several locations. In turn, as Map 3 makes evident, the “Icelandic pattern” with the verb preceding negation is not accepted by anyone in the mainland countries (and not in the Faroe Islands either).

A striking result from the database data, evident from the maps, is that the “Danish pattern” is hardly accepted at all by the informants from the Norwegian measure points. Only at ten locations in Norway have the informants accepted it, and these are spread quite evenly across the country.

The general picture thus is very clear. Norwegian dialects generally follow the “Swedish pattern”, with negation between the infinitival marker and the infinitive. The “Danish pattern” is rejected in most cases, and there is only one place in which the “Danish pattern” gets a higher score than the Swedish one. The claim by Faarlund et al. (1997: 997) that the “Danish pattern” is more natural for many Norwegian language users is therefore severely weakened by the data in the NSD.

A note on the data from Denmark is in order. First, at one location in the North of Jutland (“Vendsyssel”) both the split and the unsplit patterns are accepted. Pedersen (2017: 44ff) has looked more closely at these data, and she finds that four informants at this location accepts the test sentence. Furthermore, she also points out that there is at least one informant at all of the other locations in Jutland that accepts the sentence, and at the measure point Eastern Jutland (“Østjylland”) three informants do so. Second, Pedersen (op. cit.) shows that the existence of the “Swedish pattern” has been mentioned and documented in the dialectological literature also for the insular parts of Denmark. In other words, even in Danish dialects, the “Danish pattern” does not seem to be as obligatory as one might think.

On the basis of these considerations, it is worth investigating to what extent data from the corpus of spontaneous speech corroborate the results from the database. Defining a search for the “Danish pattern” is, however, not trivial since a negation preceding the infinitival marker may belong to the matrix predicate rather than to the infinitival clause: the example in (4) is ambiguous between a high and low attachment for the negation. This is not the case with the “Swedish pattern”, see (5), repeated from (2).

A search for the string [negation]+[infinitival marker], tailored to the “Danish pattern” is formulated as in Figure 3. The search specifies that the first word should not be a verb in the past or present tense (to try to avoid the ambiguous pattern exemplified in (4)), followed by the negation word and the infinitival marker.

This search, when limited to just the Norwegian part of the corpus, gives 13 relevant results. All of them turn out to involve the idiomatic phrase for ikke å ‘in order not to/to not even’, i.e. only with this preposition and only in this meaning. An example is given in (6).

Other than these 13, there are zero hits for unsplit infinitives (the “Danish pattern”) in Norwegian dialects.

The search for the split “Swedish pattern”, on the other hand, gives 60 hits from 41 different locations across all of Norway. All of the hits are relevant, and an example is given in (7).

The NDC has a map function which allows the user to generate a map to show which locations the hits are from, and Map 4 shows the distribution of the 60 split infinitives found in Norwegian dialects.

There are some important differences between the maps generated by the NSD, Maps 1–3, and the NDC, Map 4. The former only generates maps on the basis of sentence evaluations, while the latter generates hits from spontaneously produced speech. This means that while the locations in Map 4 show where the split infinitive (the “Swedish pattern”) has been attested, we cannot draw the conclusion that the missing points on the map are places where it could not occur. We already know that the unsplit infinitive (the “Danish pattern”) has only been attested for a sub-construction of all logically possible syntactic possibilities, and therefore that it is not one that would cover the unmarked places on the map. When a corpus does not attest a certain usage, it may be because the informants simply did not use that construction during the recorded conversation session.

The maps illustrate clearly why language research benefits from both a database and a corpus kind of infrastructure. The database data from the NSD has many more hits than the corpus data from NDC. However, a database based on evaluations of sentences can only answer questions that have been asked, and databases will therefore contain only a subset of variations of a construction. A corpus, on the other hand, where informants speak freely, will exemplify many different constructions, even ones that the researcher has not thought of beforehand. At the same time, it is to some extent arbitrary what constructions conversation partners use in recordings. This is exemplified in Map 4, which has far fewer locations than Maps 1, 2, 3.

What this investigation shows, is that when the researcher is fortunate enough to have a database of evaluations for a particular structure, there will be hits for all the locations investigated. A corpus does not necessarily cover all locations if a construction is not among the most common ones. Still, the corpus can be used to check whether informants in the database have given answers that are indeed compatible with their own language production. Our investigation in this particular case shows that this is indeed the case. The placement of the adverb with respect to the infinitival marker in Norwegian turns out to follow the Swedish pattern, illustrated in Maps 1, 2, 3, and the production data from the corpus show the same, even more convincingly, in Map 4. And the whole infrastructure together shows that the claim made in the Norwegian reference grammar (Faarlund et al. 1997) is not supported by our empirical investigations.

We now turn to look at a different and far more complex issue, namely the lack of Verb Second in matrix wh-questions in Norwegian dialects.

In this paper we hope to have demonstrated the assets of having access to both a database of syntactic judgments and a searchable corpus of free speech when researching topics in dialect syntax. In the case of the two Nordic dialect infrastructures, the Nordic Syntax Database and the Nordic Dialect Corpus, the data have been collected from a well-distributed set of locations, and crucially both kinds of data have been collected from largely the same set of informants. The well-known shortcoming of a corpus that particular constructions or variables may be scarce is counterbalanced by the way a questionnaire can ensure information from all participants on the selected constructions/variables. On the other hand, the closed nature of a questionnaire, where topics must be decided beforehand can be contrasted to the more dynamic nature of a corpus in which data one had not thought of in advance may occur. Moreover, data from the two sources (judgment versus production) for the same individuals may confirm each other, but they may also be contradictory. The latter kind of situation may help to identify issues that need to be further investigated.

For the concrete topics that we have chosen to base our exposition on, we have seen that the data from dialect infrastructures may represent both a correction of received knowledge and a strengthened confirmation of existing knowledge. In the case of split infinitives both the judgment data and the corpus data clearly show that in spoken Norwegian placing a sentence adverb between the infinitival marker and the infinitive (split version) is much more preferred and used than placing it before the infinitival marker (unsplit version). This runs counter to the received knowledge that Norwegian allows both structures and in fact prefers the unsplit version: the data show that in Norwegian dialects the split version is both preferred and most commonly used, placing them more in line with Swedish than with Danish.

In the case of matrix wh-questions with non-V2 word order in Norwegian, the judgment data serve to confirm the rather complex pattern of variation that can be pieced together on the basis of the existing literature going back to the early 20^th century. Furthermore, the corpus data serve to complement the questionnaire data insofar that the abundance of non-V2 wh-questions in all dialect regions but Eastern Norwegian confirms that the phenomenon is widespread, and we also see an increase in the use of it as we move northwards through the country. However, the abundance of examples applies just to questions with short wh-constituents: only 5% of the non-V2 wh-questions in the corpus contain complex wh-constituents, and they are also very few compared to V2 wh-questions with the same kind of constituents. To some extent this finding squares with the insight that complex non-V2 wh-questions are judged acceptable in two rather restricted areas (northwestern dialects and southwestern dialects), and almost all of the cases are indeed produced by speakers from these areas. But even within these restricted areas the complex non-V2 cases are fewer than complex V2 cases, and this entails that weight or complexity plays a role in production even in these dialects.

One clear shortcoming with the questionnaire data regarding the issue of wh-questions and V2, is that the number of test sentences were few. This is a direct effect of the topic being part of a general questionnaire that probed a wide range of topics. When administering data collection by questionnaires there is a limit to how much time one can keep the informants’ attention and willingness to respond. Since the questionnaire and production data were collected at the same time by fieldworkers visiting the various locations this is a limitation that it is hard to come around unless one sets up ways of getting back to each individual informant at later points in time. In turn, administering a system for that is laborious and resource demanding and was not viable in the case of the ScanDiaSyn project.

In any event, it seems quite clear that the establishing of the Nordic research infrastructure for syntactic variation has moved the field of North Germanic dialect syntax several steps forward, and although others may be in a better position to judge it objectively, we also believe that the output of the research collaboration has served to revitalise the field of Nordic dialectology.