american norwegian discourse marking: convergence, detachability, pragmatic change

This paper provides a first picture of discourse marking in American Norwegian, drawing on word order data and native speaker judgments. Like many others since Salmons (1990), we see convergence between the systems of bilinguals, increasing similarities at the expense of differences. Matras (1988) and Fuller (2001) argue that more ‘pragmatically detachable’ material is more easily borrowed, like well / vel vs. you know / vet du. Our results partially align with this expectation, but with a wrinkle: less pragmatically detachable material appears to be borrowed where the two languages have similar pre-existing markers (namely vet du / you know). We find more convergence, regardless of detachability, at points of similarity than of difference (e.g. English anyhow, which has no close Norwegian parallel.) This suggests that convergence may play the bigger role here, so that less detachable forms may be more borrowable in converging systems.

[1] introduction 1 Bilingual discourse marking has been widely discussed in heritage language linguistics (Polinsky 2018), but American Norwegian presents a surprising gap. As Johannessen & Salmons (forthcoming) argue, 'Since the Germanic languages share many grammatical features but vary on others, Germanic heritage languages provide a natural laboratory for micro-comparison', suggesting that such comparison could inform the field. We pursue that here by examining discourse markers (DMs) in American Norwegian. Our data comes primarily from the Corpus of American Nordic Speech, CANS, a project led by Janne Bondi Johannessen (e.g. 2015), and we compare what we find especially to DMs in [1] We are grateful to the editors for this opportunity to express our deep appreciation for Janne Bondi Johannessen's contributions to our field and our own work. An early version of this paper was presented at WILA10, Halden, Norway, October 2019. In addition to much valuable feedback from that audience, we thank Monica Macaulay and an anonymous reader for comments that led us to substantially revise this paper. [304] moquin & salmons OSLa volume 11(2), 2020 Pennsylvania Dutch. Defining 'discourse markers' is a longstanding problem, as laid out in detail by Maschler (1994) and many scholars have adopted stripped down definitions, as such Schiffrin's 'sequentially dependent elements which bracket units of talk ' (1987, p. 31). We adopt that view here, seeing DMs as elements that help organize discourse, while keeping in mind what Fuller notes (2001, p. 3): 'In the present research, the operational definition of DMs is that they are neither a semantic nor a syntactic phenomenon (i.e. they do not change the truth condition of a sentence, nor do they participate in the syntactic frame) but serve to frame an utterance in terms of its pragmatic interpretation.' Our first goal is simply descriptive, to provide basic data on American Norwegian discourse marking, including a new (as far as we know) use of judgment tasks to get at the naturalness of bilingual patterns. Initial searches showed that corresponding forms in Norwegian and English like well / vel, you know / vet du / veit du, så / so, and ja / yeah occur frequently, but with the Norwegian forms occurring more often. 2 Not every token in the initial search functioned as a DM, but the numbers give us an initial idea of patterning within the corpus. More frequent occurrence of the Norwegian forms could in part be influenced by the recording situation where participants are encouraged to speak as much Norwegian as possible, and therefore avoid code-switching to some degree. However, a preference for the Norwegian form, even if it is functioning more like its English counterpart, could provide evidence for convergence. Distinctively English DMs occur at a much lower rate. Considering our first goal and initial observations, we make two hypotheses based on previous work.
First, we hypothesize that bilingual discourse marking systems will show convergence. Salmons (1990) and others since have shown that bilingual discourse systems tend to converge over time. In his data, markers like well, you know, and of course are widespread in American German, while native German markerse.g. 'modal particles' like doch, ja, aber -have been partially displaced. Our results suggest that this hypothesis holds here.
Second, we hypothesize that 'pragmatic detachability' correlates with borrowability, as argued by Matras (1988) andFuller (2001). More detachable markers have several characteristics, two of which we focus on: (1) they are less 'lexical' in their semantics (e.g. well is less lexical than you know) and (2) they are more 'operational', e.g. tied to turn-taking (e.g. well is more operational than but) (Fuller 2001, p. 355-356, and see the more detailed discussion of detachability in §1). Here, initial data on English markers in Norwegian look more complex: some [2] The affirmative particle ja has become part of English in the American Upper Midwest and beyond. With our focus on Norwegian, we leave it aside here but see Peterson (2018).

AMERICAN NORWEGIAN DISCOURSE MARKING
[305] OSLa volume 11(2), 2020 highly detachable markers (well, so) appear less frequent than some less detachable ones (you know), though the latter form appears more widely borrowed than the former. These hypotheses seem to make conflicting predictions, but we resolve this by exploring the relationship between convergence and detachability: the forms in question are instances where both languages have pre-existing overlap, namely similar forms and meanings, like vel and well or ja and yeah in contrast to forms like English anyhow, which lack close parallels in Norwegian. In her data, Fuller finds somewhat similar complexity in semantically and functionally similar English you know and Pennsylvania German weescht. In CANS data, vel sometimes functions like English well in American Norwegian, as in (1).
(1) ulen_MN_03gm ja vel # det tok et par år 'yes well # it took a couple years' The same phrase can, in European Norwegian, express a variety of other things beyond what the English does, including reluctant agreement, surprise, acknowledgement, etc., as in (2). Forms like this can be ambiguous, allowing for reanalysis and spread of markers.
(2) hatton_ND_02gm ja [laughter] ja vel 'yes [laughter] I see' Where detachability does not predict borrowing patterns precisely, this suggests that the process of convergence may play the bigger role, that the expansion of similarities at the expense of differences is more important than detachability. In other words, the existence of similarities in both Pattern and Matter (to use Sakel's 2007 terms, see also Matras 2009 and §2 below) may override pragmatic detachability. The paper is structured as follows: In §1, we introduce the notions of convergence and pragmatic detachability, while §2 provides background on Norwegian-American DMs. In §3, we present corpus evidence, including European and American Norwegian, and in §4, we report on Norwegian native speaker judgments. We conclude in §5.
[2] convergence and pragmatic detachability The term 'convergence' has been widely used in the study of language contact, defined by Weinreich (1954, p. 395) as 'partial similarities increasing at the expense of differences' and by Romaine (1988, p. 79) in terms of 'mutual restructuring'. These simple definitions suffice for our purposes. In discourse marking, convergence could involve increasing similarity in function, where forms from the two languages align more closely than they would in monolingual speech, so that vet du might appear in uses more similar to English you know. It could also involve an increase in similarity of form, e.g. with English well preserving its English function, but taking on the phonetic realization of Norwegian vel.
'Pragmatic detachability', according to Matras (1998) and Fuller (2001), describes the borrowability of a DM based on three scales: (1) a scale that determines vulnerability to borrowing, (2) a category sensitivity scale, and (3) an operational scale that looks at whether or not a DM negotiates turn-taking. Fuller uses these scales to explore DMs in Pennsylvania Dutch (PD)-English bilingual settings, where PD is the 'recipient language' and English the 'donor'. She ranks DMs from both languages as high, medium, or low in a hierarchy of pragmatic detachability. Fuller found that highly detachable DMs (well, so) are easily borrowed, especially if they fill a gap in the recipient language. With no competition from the recipient language, these DMs are high frequency. They also facilitate turn-taking, and are typically nonlexical.
While mid-range DMs (y'know) are also borrowed, the PD equivalent (weescht) is still used at a similar frequency and in similar places -sometimes even together side by side. These DMs are highly or only somewhat lexical, and can be related to either content or turn-taking. Low-range native origin DMs (ja, mol) without an English equivalent fade away. These DMs may be the first to go from a recipient language since they cannot be easily detached and utilized within an increasingly donor-dominant DM system.
Fuller (2001, p. 17) concludes that 'it appears that for a DM to be easily borrowed or lost in language contact, it must fit the criteria on two of the scales.' This is an important insight, and we build on it to suggest that in the case of similar DMs (either in form or function) in both languages, convergence could happen more rapidly, and represent an exception to tendencies of pragmatic detachability.
[3] background on norw egian -american discourse marking The picture of DMs in American Norwegian we find in CANS matches well overall with the patterns Fuller found in PD, but with some differences. When we first looked at frequency of certain DMs, we found the following (including så / so, which has wide use in other contexts and will not be a focus in this paper): These initial numbers give us a snapshot of how common these strings are in the corpus, but they do not tell us anything about actual frequency of functions.
What we see suggests that DMs can be highly detachable AND have competition from forms in the recipient language. The Norwegian forms outnumber the English ones, but with English you know occurring far more frequently than well or so. The recipient language DMs may retain their forms while also taking on the functions of similar DMs in the donor language. Where DMs have more nuanced differences, donor language DMs may camouflage themselves, if only slightly, as recipient language DMs, thereby becoming similar enough to not seem out of place to bilinguals, but still retaining their original functions. Here, it becomes valuable to distinguish Matter (MAT) vs. Pattern (PAT) following Sakel (2007). MAT borrowing happens where 'morphological material and its phonological shape from one language is replicated in another language' and PAT borrowing happens where 'only the patterns of the other language are replicated' (Sakel 2007, p. 1 [4] comparative corpus data To compare our CANS data with European Norwegian usage, we examined data from four other corpora, the Lege-pasient-korpus, BigBrother-korpus, Nordisk dialektkorpus, and LIA norsk. All are contemporary corpora except for LIA norsk which contains older recordings . In this initial exploration, we focus on word order, where Norwegian and English sometimes offer telling contrasts.  p. 356). It is also used in consideration, and in summarization. Norwegian vel works similarly but has some unique characteristics. According to Det norske akademis ordbok, vel is a modal particle that 'softens' impact (similar to English surely), occurs as part of interjections like nåvel, and in particle pairings with ironic sentiment: ja vel, nei vel, jo vel.
Considering the overlapping characteristics of the DMs vel and well, we looked for ways that English use of this DM differed from European Norwegian usage. English well, in its capacity as introducer of new or unexpected information and in its capacity as a summarizer, is often located at the beginning of utterances. European Norwegian vel, on the other hand, is most often sentence medial. We hypothesized that if we were to compare the location of well and vel in Norwegian American speech, it would contrast with the position of vel in European Norwegian.
Searches for well and vel in CANS reveal that they often occur sentence initially. We took a random selection of 200 well tokens and 200 vel tokens from CANS, and read through the transcripts to determine how many from each sample occurred sentence-initially. 75% of well tokens were sentence initial, and 44% of vel tokens were sentence initial. We also listened to each of the vel and well tokens in CANS, and determined that at least 76% of the sentence initial vel tokens were pronounced with a clear [v] or [ʋ] rather than [w] and at least 81% of well tokens were spoken with a clear [w]. The transcripts generally match the audio, though some tokens were unclear. We compared the frequency of sentence initial vel in CANS to the usage of vel in European Norwegian corpora in Table 2.

Corpus
Sentence initial vel tokens Sentence initial percentage CANS 87 44% Lege-pasient-korpuset 1 Less than 1 % BigBrother-korpuset 1 Less than 1 % Nordisk dialektkorpus 0 0% LIA norsk 1 Less than 1 % Fuller characterizes English you know as 'a plea for understanding' that presents information as shared, thus creating common ground. It is lexical, somewhat content oriented, and without syntactic constraints, giving it the flexibility to appear almost anywhere. From this, Fuller rates English you know as 'medium' on the scale of pragmatic detachability. Our sense is that Norwegian vet du or veit du plays a slightly stronger role where it is used to insist on the truth of a statement, so creating common ground, but as an assertion of common knowledge rather than a 'plea'. Like English you know, Norwegian vet du and veit du are lexical and somewhat content oriented. As above, we treat syntactic and semantic-pragmatic patterns as indicators of whether a DM is behaving more Norwegian-like or more English-like. Clues like this are critical when the parallel DMs in the recipient and donor languages are similar in form. Such pre-existing similarity may promote convergence and borrowing more than pragmatic detachability.
Using a random selection of 200 tokens of vet du and veit du in CANS, and in the various European Norwegian corpora, we looked for how these DMs occurred outside of the typical European Norwegian syntactic position, namely sentence final. We read through each transcript to determine the word order, and listened to examples for clarification as needed, and where possible. Tokens that began a question or were the main subject and verb of a declarative sentence make up the Non-DM category.
The resulting picture does not indicate that veit du and vet du tokens from CANS occur in much different surface order or frequencies than in the European Norwegian corpora. The final column shows the percentage of DMs which were initial or medial, and places CANS in the middle of results for the whole set of corpora.
[310]  Word order does not tell us much here, but the distinction between English and Norwegian may rely more on the kind of content leading up to the vet du or veit du. Where English you know serves as a 'plea for understanding', and Norwegian vet du and veit du as an assertion of common knowledge, we see differences in the semantic and pragmatic roles of these DMs. Some instances of vet du, for example, are used where common knowledge is lacking and must be created with a 'plea' making the token more semantically English-like. Further research is needed to make any conclusions about this aspect, but it would be well worth pursuing. The one striking difference is the frequency of Non-DM tokens in Table 3, but that may reflect elicitation techniques that rely heavily on asking questions, which results in respondents mostly giving answers. The three interview-based corpora -CANS, Nordisk dialektkorpus, and LIA norsk -had very few examples of questions that began with 'Vet du …' or inverted declarative structures like 'Da vet du at …'. This could be an artifact of interview-based corpora for these purposes, and warrants future consideration.
Both well / vel and you know / vet du ~ veit du show patterns of convergence, strikingly in the word order patterns of the former. While you know is less detachable than well, the English form you know is used relatively more frequently, contrary to our initial expectations.
[5] native s peaker judgments The use of pragmatic markers like these is remarkably nuanced and difficult to judge, so we wanted to ascertain whether the American Norwegian patterns were perceived as natural by European speakers. We surveyed colleagues using the online Qualtrics survey software. 3 We constructed the survey from real examples from CANS, but modified the original transcripts to eliminate potentially disruptive elements for European speakers of Norwegian, including loanwords. For example in the original CANS sentence, 'Så, jeg levde over seksti år i huset', there are two potentially distracting features: (1) the verb å leve 'to live' in European Norwegian is exclusively used for the state of being alive, not for residing in a place or home, (2) the time adverbial 'over seksti år' lacks the European Norwegian preposition of duration i and has potentially confusing placement. The sentence was changed to 'Så, jeg bodde i huset i over seksti år;' where levde is exchanged for bodde 'resided', and i is added to the time adverbial and placed after i huset for clarity. Most sentences did not require this level of reworking, but this illustrates the issues involved. We adapted the orthography as well, adding commas to indicate pauses, for example. We also included a set of distractor items in the survey. Respondents were asked to rate, on a scale of 1 to 7, from high to low in terms of how natural the statements sounded to them. We gathered 18 responses, including from two L2 Norwegian speakers, who appear to behave like our L1 speakers on this task. Below, we highlight results from a selection of examples.

[5.1] Judgments on use of vel
The CANS sentences using vel on the judgment test included two sentence initial uses, three that were part of an initial ja, vel phrase, and two that were sentence medial. Examples of sentence initial vel were, to our surprise, the most natural with means of 2.44 and 2.33. The initial phrase ja, vel was rated to be slightly less natural, but with a broader range of means; 2.89, 4.06, 4.17. The two sentence medial examples differed; a mid-range 2.67 and the least natural rated vel example with a mean of 4.94. Below we give figures of respondent data from the most natural (3) and the least natural (4) rated sentences.
(3) blair_WI_04gk Vel, jeg vet ikke om han var født i Norge. 'Well, I don't know if he was born in Norway.' (4) flom_MN_01gm De hadde ei vel hytte oppi Tydal. 'They had a, well, cabin up in Tydal.' [3] We understand this as consultation with colleagues and not human subject research, based on discussion with our Institutional Review Board on very similar past projects. These results illustrate items that scored at both ends of the 'naturalness' spectrum, and both reveal variation in judgments. There is a more evenly distributed variation between the judgments on the vel examples with middle-range means. The location of vel in the syntactic frame appears to correlate with how natural a sentence is judged to be, but this evidence is not conclusive.
Some respondents commented on their judgments of vel. One respondent insisted, more generally, that they 'could sort of "smell" the English usage there.' Another commented more specifically to the occurrence of vel and så: Jeg syntes mange setninger var veldig preget av engelsk. Spesielt med setninger som begynte med 'så' eller 'vel'. 'I thought many sentences were very characteristically English. Especially with sentences that began with "så" or "vel."' This comment is particularly interesting considering that, at least within our small set of samples, sentence-initial vel was judged to be the most natural of the options.
[5.2] Judgments on use of vet du / veit du The CANS sentences on the judgment test with vet du or veit du included two sentence medial examples, and one sentence final. The sentence that respondents judged to be most natural (5), with a mean of 2.94, was one that occurred medially, which matches up with the occurrence of medial vet du across corpora (Table 3).
Ja, jeg ser hun Anna, vet du, i kirka. 'Yes, I see Anna, you know, at church.' However, one respondent commented specifically on this sentence: Her går det, men det er bedre om 'vet du' kommer til slutt. 'Here it works, but it's better if "you know" comes at the end.' A second example, with a mean of 3.28, elicited a similar response, from the same respondent: Her må 'vet du' komme til slutt. 'Here "you know" must come at the end.' [314]

moquin & salmons
OSLa volume 11(2), 2020 figure 3: Judgments of example (5) The least natural of the three vet du sentences (6), with a mean of 3.78, was the only one to occur sentence finally. This, again, is contrary to an expected preference for sentence final vet du, and could indicate that either (1) vet du is more flexible syntactically than previously assumed, which matches with cross-corpora findings, or (2) there is something else in the sentence that respondents judged as unnatural. In a more detailed survey, an option for the respondents to give commentary on why something seemed natural or unnatural for each item would help clarify this point. With the option to specify which elements were natural or unnatural, it may be less necessary to make modifications to original corpora sentences by eliminating potentially distracting items. Additionally, finding a way to incorporate judgments on prosody and semantics in a survey like this would give us a more detailed perspective. These judgments suggest that the American patterns are regarded as different from European ones by European Norwegian speakers, suggesting, inter alia, that there has been change to the American Norwegian system, in fact change toward English.
[6] conclusions We have provided an initial description of discourse marking in American Norwegian and how it differs from European Norwegian. Comparison to work on Pennsylvania Dutch suggests some additional nuance in terms of the relationship between convergence and pragmatic detachability.
Our first hypothesis was that we would find convergence in DM systems in American Norwegian, and we do, notably in the extension of vel to sentence initial position. As for vet du, we find less change, though it is not widely used except as a DM in our data, which may well have other explanations.
Our second hypothesis was that we expected to find a correlation among detachability, frequency, and borrowing, in line with earlier studies. We did not find this, but instead found that a less detachable marker (you know) is represented more often by English forms than a more detachable one (well). While further work is certainly needed on this, we ask whether powerful similarities in form and function of some DMs may outweigh detachability.
Overall, we observe a change in frequency of forms that were otherwise possible, including in word order. This result reflects, we argue, not so much core grammatical change, but change in frequencies of existing grammatical patterns.
The judgment tests provide a new kind of data. Respondents differed considerably about what they found natural or unnatural, but there were indications of agreement on the use of vel. We also found a mismatch between the naturalness rating of sentences and unsolicited commentary. As an anonymous reviewer suggests, frequency differences may help shape the judgments of these sentences, where perfectly grammatical but low frequency patterns may be judged differently from high frequency structures (cf. Bohnacker & Rosén 2008, for example).
While much more basic descriptive work is needed here, some bigger issues already suggest themselves. As noted in 3.1, there seem to be differences in English-and Norwegian-like prosody in vel. For veit du and vet du, we suspect that the context of preceding discourse may differ in American Norwegian from European Norwegian.
In some ways, the changes discussed here have led to new and additional richness in the discourse marking system of American Norwegian. While we hope that this brief paper opens the door to further research, for now we conclude that this is likely less borrowing of Matter than usually found, and more borrowing of Pattern. The major outcome of this first foray into American Norwegian discourse marking is that there is much more to be done.
dedication Janne has constantly been in our thoughts as we have worked on this chapter. It has given us another chance to think about how much she did for and with both of us, as colleague, collaborator, mentor, and friend. As we've talked to each other and other WILA friends and colleagues, we're reminded of the huge and lasting contribution Janne made to building this community, a vibrant and growing group of people advancing our field thanks to what she did and how she did it. We owe you, Janne.