Functional Variation of German Also across Registers and Speaker Groups

This study investigates the variation of the polyfunctional linguistic item also across registers and language contact setting s. I present findings from a corpus study using the corpus of the Research Unit Emerging Grammars, RUEG for short (Wiese et al., 201 9), which provides comparable, register-differentiated data of bilingual and mono-lingual speakers of German in Germany and bilingual heritage speakers of German in the US . The data suggest that functional variation of a specific lexical item reflects the use of functional features in specific communicative situation s. The data further indicate an impact of the societal status that a language occupies in the larger society (majority v s. heritage language) on the distribution of such function s.

The present study ties in with this by focusing on the German polyfunctional linguistic item also.Recent functions of also in the literature vary from its use as a consecutive adverbial connector, as a metacommunicative connector or repair marker to uses with a mere discourse function.However, the denomination of also in view of certain specific functions remains a point of discussion (e.g.Alm, 2007).
Characteristics of the communicative situation such as mode (written vs. spoken) and (in)formality are relevant parameters for register variation (cf.Biber and Conrad, 2009).They are assumed to influence the use of specific functional features (cf.Pichler, 2010).For instance, Redeker (1990) found evidence for a higher number of discourse-pragmatic markers in informal situations (see also Crible, 2018;Sánchez-Muñoz, 2007).Therefore, polyfunctional linguistic items are an interesting case for investigating the functional variation across registers.
Discourse-pragmatic markers are a typical feature of spoken language (e.g.Brinton, 1996).Studies that investigate them are therefore often based solely on spoken language data (e.g.Fuller, 2003;Crible, 2018;Redeker, 1990;Imo, 2017 on the need to broaden DM studies) and only a few take variation across written registers into account (see Ravetto and Ballestracci, 2013 on also in novels).In a previous study on also, Konerding (2004) relates different functions to the conceptually spoken and conceptually written continuum of Koch and Oesterreicher (1985).1However, this attribution is based on examples without providing a quantitative analysis of the distribution of functions across registers.A detailed analysis of the functional distribution, considering potential influencing factors such as mode, is still a desideratum (Konerding, 2004: 234;Fernandéz-Villanueva, 2007: 102).
1 Koch and Oesterreicher (1985) model mode (written vs. spoken) in juxtaposition to a conceptual dimension which represents aspects of 'proximity and distance' (cf.Hennig and Feilke, 2016).Following this, there are texts that are written but conceptually oral (e.g., printed interviews) and others that are spoken but conceptually written (e.g., a sermon) (see Koch and Oesterreicher, 1985: 18 ff.).Another line of research focuses on discourse-pragmatic markers in language contact situations (e.g.various studies in Maschler, 2000;Matras, 2020;Sankoff et al., 1997).However, they often focus on code-switched or borrowed elements in bilinguals.Only a few studies look at functional variation of such elements in one named language and its relation to societal status differences of that language, namely as a majority language spoken by the larger society vs. as a heritage language with a minoritised status, in addition to register differences (see Sánchez-Muñoz, 2007 on Spanish heritage speakers in the US).Fernandéz-Villanueva (2007) compares usages of also in German monolinguals in contrast to late language learners of German in semi-informal data.However, late language learners are in many respects different from heritage speakers.Only the latter are part of the native speaker continuum (Rothman and Treffers-Daller, 2014;Wiese et al., 2021).To the best of my knowledge, there has been no quantitative analysis of the functional variation of also across registers and contact situations in the literature so far.This article aims to fill this gap.
Based on comparable corpus data from the German subcorpus of the 'Research Unit Emerging Grammars' , short RUEG (see section 3), elicited in different communicative situations, I investigate the distribution of the different functions of also across 1) registers, broadly understood as the linguistic manifestation related to the characteristics of the communicative situation and 2) speaker groups, namely, with early bi-and monolingual speakers of German in Germany and bilingual heritage speakers (HS s) of German in the US.These three speaker groups are an interesting basis for the investigation because they present different contact situations in which German has different status: either as a majority or as a minoritised heritage language.
HSs are bilinguals who, in early childhood, acquire a language within the family that is not the language of the larger society (e.g.Rothman, 2009).In line with many studies (e.g.ibid.), the former is the heritage language and the latter is referred to as the majority language.Speakers acquire these languages in many different ways with various factors that influence the acquisition process.One important fact is that formal school education usually takes place in the majority language.Especially in countries with a strong monolingual habi-tus2 like Germany and the US (Wiese et al., 2021;Gogolin, 1994), this leads to a shift towards the majority language becoming dominant for most of the HSs (cf.Flores et al., 2020;Polinsky and Scontras, 2020).While early bilinguals in Labrenz Contrastive PragmaticS 4 (2023) 289-320 Germany receive their formal school (and further) education in German and deal with many communicative situations in German in everyday life, the use of German by HS s in the US is mostly restricted to family members.Therefore, societal language status might have an impact on the use of functional features in specific situations.
Against this background, I focus on the following general research questions: RQ1 Do the functions of also differ across registers in terms of frequencies?RQ2 Do bi-and monolingual speakers of German in Germany use the functions of also with similar frequencies?RQ3 Do HS s of German use the functions of also with similar frequencies compared to speakers of German (bi-and monolingual) in Germany?
In pursuit of these questions, I first provide an overview of functional classifications of also in previous work (Section 2).Based on this, I present a categorisation scheme in Section 3.2.Section 3 presents the database and the procedure.In Section 4, I report the results of the quantitative analysis; Section 5 contains a discussion of the results and a conclusion.

2
Categorisation of Also Pasch et al. (2003) and Breindl et al. (2014) -a reference work on connectors in German (vol.I and II) -assign also to the syntactic class of adverbial connectors that are not positionally restricted.In addition, this handbook assigns it to two semantic classes: 1) causal/consecutive connectors (Volodina, 2014: 899) and 2) metacommunicative connectors (Breindl, 2014(Breindl, : 1166) ) (see Fernandéz-Villanueva, 2007 for a literature review focusing on descriptions in grammaticography/lexicography).Further discourse-pragmatic functions have been discussed in a range of studies (Alm, 2007;Auer, 1996;Blühdorn, 2003;Deppermann and Helmer, 2013;Dittmar, 2002Dittmar, , 2011;;Fernandéz-Villanueva, 2007;Konerding, 2004;Ravetto and Ballestracci, 2013).In these studies, both positions and functions and their interaction are considered, with syntactic (dis)integration being a particularly important aspect.First, I briefly outline some aspects of the German syntax necessary for understanding syntactic (dis)integration.Subsequently, the functions and positions of also are discussed based on previous findings using examples from the RUEG corpus (2.3).Section 2.4 briefly presents what is known on also across registers.

2.1
Topological Aspects of German Syntax Syntactic (dis)integration in German can best be captured by the topological model.According to the topological field model, going back to Drach (1937) (cf.also Pafel, 2009;Wöllstein, 2014), sentences are organised by two "sentence brackets" (linke Satzklammer, LK; rechte Satzklammer, RK) that delineate three fields: prefield (Vorfeld, VF), middlefield (Mittelfeld, MF) and postfield (Nachfeld, NF).In main clauses, the left sentence bracket (LK) is occupied by the finite verb, while infinite verbs and verb particles can occupy the right sentence bracket.In standard German, the prefield of main clauses contains exactly one constituent, leading to what is typically assumed to be a rigid verbsecond constraint (but see e.g.Wiese and Müller, 2018 for a critical discussion).Elements in the prefield and the middlefield are syntactically integrated.In addition, some elements can occur before the prefield, that is, before the sentence proper -also called the pre-prefield (Vor-Vorfeld, VVF) (e.g.Auer, 1996).These elements are not part of the core syntactic structure, and are thus not syntactically integrated.An illustration of the topological fields is given in Figure 1, including also in the pre-prefield position.

Adverbial Connector vs. Discourse Marker
The fact that also is positionally not restricted and thus can be integrated into the sentence structure on the one hand, but on the other hand can also occur non-integrated in the periphery of the sentence, has led to a variety of approaches as to how this item is to be understood.In research on German discourse markers3 (DM s), syntactic disintegration is proposed as one criterion for DM status (cf.Auer, 1996;Gohl and Günthner, 1999;Imo, 2013Imo, , 2017 amongst many others) amongst many others).At the same time, the position at the periphery outside the sentence proper is considered to favour discourse-pragmatic functions (Auer, 1996;Diewald and Fischer, 1998: 88;Diewald, 2006).Another criterion for DM status stated by some scholars in German linguistics is that they "no longer link statements on a propositional level, but project actions on a pragmatic level, such as justifications, admissions, and narratives" (Imo, 2017: 50, author's translation).In line with this, previous studies discussed a number of lexical items under the notion of DM, when they occur syntactically disintegrated (e.g.weil in Gohl and Günthner, 1999; also in Dittmar, 2002; see also Imo, 2013: 180).
According to Imo, the two criteria help to delimit DM uses from other uses in German, but not so easily in other languages (Imo, 2017: 51).This is mainly due to the rigid verb-second constraint in German (see 2.1) that facilitates identification syntactic disintegration in contrast to other languages that do not have such a constraint (Ravetto and Ballestracci, 2013: 354).This arguably led to a tendency in German linguistics to distinguish between (adverbial) connectors and DM s on the basis of syntactic (dis)integration (cf.Imo, 2017: 60;Dittmar, 2002).However, Dittmar (2011) shows that in the case of also the distinction between its discourse-pragmatic use and other uses is not as straightforward.
The ambiguity results from the fact that also in non-integrated position can either 1) relate to its core meaning of inferencing (e.g.Konerding, 2004) or 2) be used turn-initially or signal the beginning of a new ideational discourse sequence (e.g.Blühdorn, 2003;Deppermann and Helmer, 2013).Whereas some researchers classify also as a DM only in the latter function (e.g.Konerding, 2004), others assume a broader functional spectrum in this category (e.g.Alm, 2007;Pfeiffer, 2017;Deppermann and Helmer, 2013).The first group assigns also to the category of conclusive/consecutive connectors or connective particles anytime a process of inferencing is involved, irrespective of the type of inference (Konerding, 2004).Nevertheless, these inferences can be drawn on different levels.Let us consider the following examples from the RUEG corpus:4 (1) Ich war nicht mehr benötigt also bin ich dann gegangen.'I was no longer needed so I left' .
Contrastive PragmaticS 4 (2023) 289-320 (2) also ähm meine Einschätzung wär auf jeden Fall ähm dass dort keiner Schuld hat 'well um my impression is actually um that it is nobody's fault' (DEbi05FR_fsD) In Example 1, the inference is drawn on a semantic level, that is, the second proposition is a logical consequence of the previously stated one.Also indicates a rule of inference in the sense of a modus ponens, that is, if the premises p → q and p are fulfilled, we can conclude that p so that q (cf.Ravetto and Ballestracci, 2013: 336;Breindl, 2014: 829).In this example also is situated in the prefield and is thus syntactically integrated.Corresponding expressions in English are, for example, thus, so, as a result (see Fernandéz-Villanueva 2007: 107 for possible translations of also in different functions).In Example 2 also indicates an inference, but on a pragmatical level.The speaker gives a subjective evaluation of the preceding proposition(s).The prefield position is occupied by meine einschätzung.Also is situated in the pre-prefield and is thus not syntactically integrated, which further supports a pragmatic reading.Similar to what Deppermann and Helmer (2013: 10) describe as an "explication of an intersubjective inference" (author's translation), in dialogic data, Example 2 is a subjective inference which is not a logical consequence of the preceding proposition.A corresponding expression in English is well.Rather than constituting strictly distinct categories, inferences might be situated on a scale from a more logical-semantic to a more pragmatical level (Fernandéz-Villanueva, 2007;Konerding, 2004; see also Müller, 2005: 74 on the English so for a similar issue).However, the pre-prefield position imposes a pragmatic reading to the reader/hearer (Ravetto and Ballestracci, 2013: 351) and generally favours discourse-pragmatic functions (see above).Therefore, I assume, in line with Alm (2007: 147), that a wide spectrum of functions is possible for also in the pre-prefield position, ranging from functions that are similar to functions in other positions (inference-related) to unique functions in the pre-prefield.
In accordance with the literature on German, Examples 3 and 4 are cases of DM classification.In Example 3 also initialises a narrative sequence and is not syntactically integrated.In Example 4 also is followed by a speech act and initialises the entire narration.

2.3
Repair Marker Other functions discussed in the literature of also can be subsumed as repair markers.These are self-repairs such as elaborations/specifications, corrections, restarts and the use as a filler, although with diverging terminology (cf.Alm, 2007;Auer, 1996;Konerding, 2004;Pfeiffer, 2017).Inspired by Crible and Pascuan (2020), I follow Levelt's (1983Levelt's ( , 1989) categorisation of self-repairs (see also Pfeiffer, 2015 on an extensive account of self-repairs in German).They state: In Levelt's (1983: 44) terminology, self-repairs comprise four main parts, a reparandum ("item to be repaired"), a moment of interruption ("the point at which the flow of speech is interrupted"), an editing phase […] with an optional editing term, and a "repair" (also called reparans, i.e. the repairing segment).Crible and Pascuan (2020: 56) Hereafter, also can function as an editing term in different types of self-repairs that are addressed subsequently.

2.3.1
Elaboration/Specification This function corresponds to Levelt's repair type of appropriateness.A need for specifying or elaboration (as a broader term in Pfeiffer, 2017) can result out of "potential ambiguity given the context" (Levelt, 1983: 52), for example, with "referentially ambiguous deictic expressions" (ibid.)or when referring to ambiguous referents (for example, when two objects of a similar kind are involved).Further reasons for a specifying repair can be the use of more appropriate terminology and coherence with previously used expressions (ibid.).In Example 5, also signals a specification of the preceding nominal phrase (the man) by adding a modification (see modified repetitions in Crible and Pascuan, 2020: 4).Example 6 gives a more appropriate description of the situation or additional information for someone who does not know the situation.In these cases, also is the editing term between reparandum and reparans, and always precedes the reparans.Those instances (editing term and reparans) can occur in the prefield, in the middlefield and in the postfield.In contrast to other types of overt repairs, elaborations are realised after a syntactic closure (in postfield position) or can be inserted in a syntactic structure (prefield, middlefield) with Contrastive PragmaticS 4 (2023) 289-320 subsequent continuation typically with scope over one constituent (see Kaiser 2016 on a distinction).Additionally, also signalling an elaboration can occur in the pre-prefield followed by an elaboration on sentence level, as in Example 7. Pfeiffer (2017) takes this as an argument to classify repair markers as a subtype of DM s.Also in Example 7 can be interpreted as indicating a more precise and appropriate description given the situational context or as indicating a conclusion of the speaker informed of the circumstances.The inferential core meaning still shines through, but again rather on a discourse-pragmatic level.
(5) ich glaube es war schon der Ball den der Mann also dieser junge Mann hingeworfen hat 'I think it was indeed the ball that the man, that is, this young man threw down' (DEbi01MR_fsD) (6) Dieser rollte nun auf die Straße in Richtung der Frau mit den Einkäufen also einmal quer über die Straße 'It rolled on to the road in the direction of the woman with the purchases, that is/thus straight across the road' (DEbi03FG_fwD) (7) [zwei Autos fuhren sehr nah aneinander] also sie hatten nisch wirklich viel Sicherheitsabstand '[two cars were driving very close to each other] so/well they didn't really have much of a safety distance' (DEbi51MT_fsD) 2.3.2Restart Restarts correspond to also signalling a reformulation without a clearly identifiable previous error.In contrast to elaborations, the point of interruption is not after a fully uttered constituent and/or syntactic structure, but occurs before finishing it (Example 8).It is related to planning issues in the sense of "what idea to express first and in which way".As such, it overlaps with Levelt's delay repairs including fresh starts (Levelt, 1989: 490).
(8) und dann keine Ahnung hat der Ball irgendwie also der ist irgendwie auf den Fuß von dem Typen gefallen 'and then I don't know the ball has somehow well it fell somehow on the guy's foot' (DEmo71FD_isD) 2.3.3Correction As a correction marker also indicates a repair in the sense of an overt error repair or an explicit correction at the lexical, syntactical, or phonetic level (Levelt, 1983: 54).Corrections at the syntactic level are reduced to corrections of word order (cf.Levelt, 1989: 462).In Example 9, the interruption is followed by a correction at the lexical level.
(9) da war so eine Frau mit dem Einkaufswagen also beziehungsweise mit den äh Supermarkttaschen 'There was a woman with a shopping trolley, I mean with the um supermarket bags' (DEbi05FG_isD) 2.3.4Filler Fillers are covert repairs in the sense of Levelt (1983).In this function also can occur as an editing term in an interruption or within repetitions, as in Example 10 (cf.Levelt, 1983: 55).Konerding (2004) refers to such cases as hesitation particles.
(10) und dann ähm ist der ist also is der hund auf den ball zugerannt 'and then um was the was well the dog was running towards the ball' (DEbi08FT_isD) The categorisation scheme for also in Section 3.2 is based on the previous description.In this paper, I use category as an overarching term for a set of functions.In the following, I briefly describe what is known about also across registers.

2.4
Also across Registers Works on also are based on a variety of data.On one hand, descriptions in grammars that are usually orientated towards a written standard variety mainly focus on its syntactical distributions and its use on a semantic level as a conclusive adverbial connector (cf.Dittmar, 2011: 122;Fernandéz-Villanueva, 2007: 102).Similarly, based on data from novels (i.e.written data), Ravetto and Ballestracci show a preference for use as an adverbial connector rather than as a DM (Ravetto and Ballestracci, 2013: 345).5 On the other hand, studies on spoken data show its predominant use in non-integrated position (Fernandéz-Villanueva, 2007 on semi-informal oral data; Dittmar, 2011 on narrative oral data; Deppermann and Helmer, 2013 on conversational data).Konerding (2004) considers written as well as spoken data.According to him, also occurs as a consecutive connective particle and as a specification marker on a non-sentential level (as an apposition) in written and spoken mode, while the functions of correction, filler and discourse marker are reserved for conceptually spoken productions.

3
Corpus Study

Database
The database for this study is the German subcorpus of the RUEG corpus, snapshot version RUEG-DE 1.06 (Wiese et al., 2019).The total number of tokens and speakers of each group is listed in Table 1.The corpus is freely available and searchable in Annis7 (Krause and Zeldes, 2016).

3.1.1
Elicitation Procedure The corpus contains comparable data that was elicited using the Language Situations set-up (Wiese, 2020).Participants watch a video of a minor car accident.They are then asked to act as if they had just experienced the situation as bystanders and tell the story in four different communicative situations: 1) write a text-message; 2) send a spoken message via an instant messenger programme to a friend; 3) type a written witness report for the police; 4) leave a voice message at an imagined police-witness-line on the phone.This way, each participant produced 4 texts related to different registers: 1) informal-written, 2) informal-spoken, 3) formal-written and 4) formal-spoken.8This set-up has shown to yield naturalistic, register-differentiated language productions (e.g.Wiese and Bracke, 2021;Wiese et al., 2021).

3.1.2
Speaker Criteria Two age groups are represented in the corpus.Adolescents ranging from 14 to 18 years old and adults between 22 and 35 years old.The German subcorpus of the RUEG corpus contains data from three different groups of German speakers: (1) Bilingual speakers of German who are HS s of another language (Turkish, Greek and Russian) and have been speaking German as a majority language Labrenz Contrastive PragmaticS 4 (2023) 289-320 in Germany since early childhood (for short DEbi), (2) Monolingual speakers of German as a majority language in Germany (for short DEmo) and (3) HS s of German in the US who grew up with English as their majority language (for short USbi).The speakers of the first group can be further divided into three groups because of the potential influence of their different heritage languages.Although I did not expect such an influence because of the shared status of German as a majority language, I retained the group distinction between bilingual and monolingual majority speakers of German in order to check for the manifestation of a potential impact of the different contact languages.This is because studies have shown that language contact favours innovations and can accelerate change especially in the area of discourse-pragmatic markers (e.g.Aijmer, 2013;Matras, 2020;Pichler, 2016b;Silva-Corvalán, 1986).

3.2
Also in the RUEG Data The corpus has been tagged for Parts of Speech following the guidelines of the Stuttgart Tübingen Tagset (for short STTS 2.0; Westpfahl et al., 2017).According to these guidelines, also is either tagged as an adverb or as a DM.A hard criterion for DM s in STTS 2.0 is their position in the pre-prefield.All other occurrences of also are tagged as adverbs.A more fine-grained analysis had to be done manually.An overview of the categorisations of also in the RUEG data is presented in Table 2. Against the background presented in the previous Section 2, I identified the following positions: VF, MF, VVF, NF.Furthermore, I identified the following main functions: consequence/conclusion, evaluation, elaboration/specification, initialising a narrative or a new propositional sequence, correction, restart, and filler.In the previous literature on also, restarts have In one case, also was used to initialise a sequence of direct speech, a typical function of German so in colloquial speech that was not expected for also (labelled as OTH in Table 2).For further analysis, I use the broader categories DM s and repairs while keeping consecutive adverbial connectors (at the logical-semantic level), evaluations and elaborations as separate categories.
Evaluations are related to the postulated core meaning of also as indicating an inference (see Example 2).This is, however, not true for all instances.Example 12 is an evaluation indicating solely the speaker's epistemic stance.Therefore, this function is neither subsumed under the narrow notion of DM s nor under a general label for inferences.Note that, in line with Fernandéz-Villanueva (2007), I consider also in elliptic structures as in Example 12 as occurrences in the pre-prefield.
(12) und ähm was fällt dem erstmal ein die Einkäufe zurück in die Tüte zu packen also keine Ahnung konnt ich auch nicht nachvollziehen 'and um the first thing that occurred to him he puts the groceries back in the bag so/well no idea I couldn't understand it either' (DEbi11MR_isD) Elaborations/specifications that are discussed under the notion of repairs (e.g.Pfeiffer, 2017) are treated here separately because they are supposed to occur in the written as well as in the spoken mode, at least as appositions (Konerding, 2004), and are thus interesting for an investigation of patterns across registers.

Results
First, I give an overview of the positional distribution of also (4.1).In Section 4.2, I report the inter-rater reliability for the function annotation presented in Table 2.After that, I present the variational analysis of these categories with respect to registers (4.3) and speaker groups (4.4).

4.1
Overall Positional Distribution Overall, there are 35 instances of also in integrated position (VF, MF) and 302 in non-integrated position, that is at the left (pre-prefield) or right border (NF) of an utterance.These numbers do not comprise repairs because their status of integration is not that clear.For instance, in the case of specifications on a phrasal level (see Example 5), both the editing term and the reparans depend syntactically on a preceding constituent.The editing term is the head of the reparans and is positioned reparans-initially.In our data, such elaboration instances occur about 30% of the time in the postfield, 43% in the middlefield and 22% in the prefield.9Interestingly, the integrated position is not exclusively correlated with a logical-semantic consecutive function but contains also five occurrences of indicating subjective inferences.In addition, six instances in the middlefield indicate elaborations.In contrast, the data show only discourse-pragmatic uses in the pre-prefield.

4.2
Inter-rater Reliability Three annotators were involved in the annotation process.Two of them annotated the entire data independently.The agreement of function annotation resulted in Cohen's κ = 0.63.10According to Landis and Koch (1977), this refers to a substantial strength of agreement.Subsequently, the annotators discussed all cases of disagreement until they agreed.

4.3
Functions across Registers With respect to the first research question, I made the following hypothesis: H1 The frequency distribution of the functions of also differs across registers, such that the use of also: -as a consecutive adverbial connector is expected to occur in the written and the spoken mode -as a repair marker is expected to occur only in the spoken mode -as a discourse marker and as indicating evaluations is expected to occur more frequently in the spoken mode and in informal communicative situations; and -as marking elaborations is expected to occur more frequently in formal settings.9 Three instances could not be clearly classified.10 Such degree of agreement is achieved when corrections, restarts and fillers are subsumed under the category repairs, and a former distinction of specifications on phrasal level and elaborations on clausal level is suspended.To test the above hypothesis, I report the raw and normalised counts of the categories of also in the respective registers, show the distribution of normalised frequencies in individual texts11 across registers (Figure 2), and report the results of generalised linear mixed effects models (glmer in R) for each category.I included the individual speaker as random effect in the model.Table 3 and the relative frequencies of also per individual text in the scatterplot (Figure 2) show that also is used substantially more in the spoken than in the written mode.Figure 2 further shows a large inter-individual variation in frequencies.This is a common finding in studies on discourse-pragmatic features (see, for instance, Aijmer, 2011;Sánchez-Muñoz, 2007).A second tendency is that also is more frequently used for indicating an evaluation in informal than in formal settings.
In the following, I present the results of the statistical analysis for each category and their distribution patterns across registers.

11
An individual text is one production by one speaker; this includes informal/formal and spoken/written productions.Thus, each speaker produced four texts.

4.3.1
Adverbial Connector This category is the least used overall in our data (Table 3).However, a general linear mixed effects model shows a significant effect for mode (Std.error = 1.0860; z = 2.880; p < 0.01), suggesting a preference for using also as an adverbial connector in the written mode.A Kruskal-Wallis test based on medians of normalised frequencies taking variance into account supports a significant difference between modes (chi-squared = 3.6857, p-value = 0.05).

4.3.2
Evaluation A significant effect for setting (Std.error = 0.2558; z = 4.005; p < 0.01) predicts that this function is preferentially used in the informal setting.This is consistent with the relative frequencies in Table 3.

4.3.3
Discourse Marker There are no significant effects for also as a discourse marker if we consider all speaker groups together.However, there is no use of this function in the formal-written register, while a few occurrences are attested in the informal-written texts (Table 3).

Elaboration
The model shows a significant effect for setting (Std.error 0.1858; z = −2.891;p < 0.01), suggesting a preference for the formal situation.This most likely reflects the nature of the register of a spoken witness report, where precise and detailed descriptions are important and thus there is a need for specification and appropriateness.

4.3.5
Repair Markers The model demonstrates a significant effect for mode (Std.error 0.7466; z = −3.103;p < 0.01).Not surprisingly, the effect indicates that repair markers are used more frequently in the spoken mode.This reflects the nature of online speech production as opposed to written production where there is time for editing.
Taken together, the findings support Hypothesis 1: The functions of also are systematically distributed along the parameters of mode and setting, reflecting the circumstances and goals of language production.

4.4
Functions across Speaker Groups In pursuit of the Research Questions 2 and 3, I focused on the following hypotheses: H2 The distribution of the functions of also is similar for bi-and monolingual speakers of German in Germany.H3 The distribution of the functions of also is different for HS s of German compared to speakers of German (bi-and monolingual) in Germany, such that a more frequent use of repair markers is expected for HS s.
To address these hypotheses, I first compare functional variation patterns across the groups (4.4.1).In Section 4.4.2,I report the results from the mixed effects models (glmer in R) for each category.Furthermore, I show normalised frequencies for each category across speaker groups and registers.The normalised frequencies were calculated based on the total number of tokens for each register in each speaker group individually.I considered only the texts of speakers that used also in at least one of its functions.If these normalised frequencies seemed to differ across the groups and the model did not show significant effects, I employed a Kruskal-Wallis test (kruskal.test,pairwise.wilcox.test in R) to further explore significant differences.proportional use of the distinct functions related to all uses of also in a given group.As hypothesised, the patterns of bilingual and monolingual speakers of German in Germany look very similar in the spoken mode (Figure 3), whereas the proportional distribution is different for HS s of German.This holds for both settings.Interestingly, the patterns look rather different between the three groups with respect to the written mode (Figure 4).Although the overall occurrences are not very high (see Table 3 for raw numbers), and thus the findings have to be treated with caution, a tendency towards greater functional variation can be observed for bilingual speakers in Germany, who use a wider spectrum of functions in informal-written texts than monolinguals.Additionally, both bilingual speaker groups (HS s and majority speakers) use also in formal-written registers, whereas there is no occurrence of also in the monolingual group.Figure 3 and Figure 4 reflect functional preferences of the speaker groups on the basis of proportional uses.This allows for a comparison of variational patterns.In order to compare frequencies across speaker groups (and/or registers), though, normalised frequencies need to be addressed (see Biber, 2012 on the difference between variationist and text-linguistic approaches).These are presented subsequently together with the results of the statistical analysis for each functional category.Adverbial Connector The model shows no significant effect for also as an adverbial connector between the different groups, regardless of which group is used as the baseline.However, the normalised numbers in Table 4 and the medians represented in the boxplot in Figure 5 point to an unexpected tendency.That is, the bi-and monolingual speakers in Germany use also as an adverbial connector very rarely across registers compared to HS s of German in the US, but with a larger intra-group variation in the latter.The dots in Figure 5 represent normalised frequencies of individual texts.In the HS group, two productions contain a normalised frequency of more than three occurrences per 100 tokens, whereas the median is slightly over 1.This variation in HS s might be the reason for the non-significant result in the model.However, a Kruskal-Wallis test (chi-squared = 7.8397, p = 0.05) shows a significant difference between groups.(p < 0.1) and monolinguals (p < 0.01).As evident in Figure 7, HS German speakers use also as a DM most frequently and German-Turkish bilinguals use it the least often in this function.The medians of the other speaker groups from Germany are closer to the German-Turkish bilinguals with no significant difference.The normalised frequencies in Table 6 suggest that the DM function is more frequent in the formal-spoken productions of HS s compared to the other speaker groups.Elaboration The model shows a highly significant effect for speaker group that points to a preference of bi-and monolingual speakers in Germany to use also for indicating elaborations compared to HS s in the US (DEbi: Std.error = 0.4218, z = 3.896, p < 0.01; DEmo: Std.error = 0.4577, z = 3.861, p < 0.01).The Kruskal-Wallis test, however, does not confirm this, although the normalised frequencies in Table 7 show the same tendency.Table 7 further indicates different patterns across the formal and informal settings.Whereas monolingual speakers seem to use this function more in the informal setting, HS s show a more frequent use in the formal setting.Bilingual speakers in Germany instead show no differences across settings.

4.4.2.5
Repair Marker The model shows no effect of speaker group.However, the relative frequencies in Table 8 point to a more frequent use of also as a repair marker in HS s of German in the US.The Kruskal-Wallis test (chi-squared = 10.988,p < 0.01) confirm the group difference and a post-hoc pairwise Wilcoxon-test shows that HS s differ significantly from both, bi-and monolingual speakers from Germany.On the basis of the above analysis, the null hypothesis of H2, namely that the distribution of functions among bilinguals and monolinguals in Germany is different, can be rejected.The analysis supports the hypothesis that the two groups behave similarly in the use of the different functions of also.The null hypothesis of H3, namely that the distribution of functions is the same for HS compared to bilingual and monolingual speakers of German in Germany, can also be rejected.HS s tend to use also to indicate evaluations, with discourse marking functions and as a repair marker more frequently than bi-and monolingual majority speakers, while the latter seem to use it more often for indicating elaborations compared to HS s.Moreover, HS s appear to use also as an adverbial connector more frequently.
Overall, the findings suggest that societal status differences, that is, German as a minoritised heritage language spoken mainly in the family vs. as the majority language of the larger society, have an impact on the functional variation of discourse-pragmatic markers.

Discussion and Conclusion
This study contributes to our understanding of the variation of discoursepragmatic functions by considering a specific polyfunctional lexical item.Including different registers and language contact settings in the analysis allowed for the identification of distributional patterns and the loci of that variation.The systematic analysis of comparable data across written and spoken modes shows that also as a consecutive adverbial connector is 1) very rare in our data and 2) more typical for written texts.The first is in line with previous findings on spoken data, the second reflects the descriptions in grammaticography/lexicography that are usually based on written data (see Dittmar, 2011;Fernandéz-Villanueva, 2007 on both aspects).Furthermore, it supports the well-attested preference for discourse-pragmatic functions in spoken productions.Interestingly, the discourse-pragmatic functions that were investigated here vary across settings.Whereas evaluations have shown to be more typical for informal communicative situations, elaborations are more typical in formal settings.This can be explained in view of Biber and Conrad's first dimension in the multidimensional approach on register variation, that is, involved vs. informational productions (Biber and Conrad, 2009: 215 ff.;see also Sánchez Muñoz, 2007).Subjective evaluations are often accompanied by expressions of the epistemic stance of the speaker and thus reflect a rather involved production with a friend.In contrast, a witness report to the police demands a high level of precision, which favours elaborations and leads to a rather informational production (see Fernandéz-Villanueva 2007 on a similar finding).However, the normalised frequencies by speaker group and register (Table 8) indicate that monolinguals use elaborations more frequently in the informal setting.This is rather surprising and might be an effect of grouping clausal and phrasal elaborations into one category.Furthermore, the repair functions (correction, restart and filler) show no effect of setting.This finding supports that repairs are a typical feature of the spoken mode in general that reflects the production circumstances of oral online vs. written production, which allows for editing (see also Crible, 2018;Pfeiffer, 2017).Methodologically, the findings on register differences support the Language Situations approach (Wiese, 2020) as a reliable elicitation method for investigating language use in different communicative situations.The perspective on language contact settings points to specific strategies of using specific markers in different communicative situations by speakers with different language backgrounds, especially in view of societal status differences or the degree of involvement in a vital speech community (cf.Sankoff et al., 1997 on the latter).That is, 1) HS s might mark elaborations differently than biand monolingual speakers of German in Germany, 2) they commonly use also for planning issues in online speech production and 3) they use it to indicate a consequence on a semantic level comparatively more often.The frequent use of discourse-pragmatic markers has been shown to be "the hallmark of the fluent speaker" (Sankoff et al., 1997: 204) and an indicator for efficient planning strategies (cf.Crible and Pascuan, 2020).In view of 2), this further supports the perspective on HS s as part of the native speaker continuum.3) might point to a cross-linguistic influence of English so in the sense of functional convergence (cf.Matras, 2020).English so and German also share the function of a connector indicating a consequence or result (see Müller, 2005 on English so).Indicating an elaboration instead is not in the functional spectrum of English so.This (non-)overlap seems to be reflected in the frequencies of use.Although this is rather speculative, it aligns with Sánchez-Muñoz's (2007) findings that indicate the same direction of cross-linguistic influence: from the majority into the heritage language.
With respect to these findings, we can outline some interesting aspects for further research.For 1) and 2) it would be interesting to look in more detail at elaboration marking strategies, at specific types of elaborations and at repairs, especially but not exclusively among HS s.For 3) it might be worthwhile to have a look into highly multilingual countries (such as Namibia) in order to tease apart the possible impact of societal language status vs. the monolingual habitus of a country.As many have pointed out, researchers who focus on discourse-pragmatic markers are confronted with an ongoing terminological inconsistency.This is especially challenging across languages.For instance, the understanding of DM s in German linguistics (e.g.Imo, 2012) contrasts with approaches in the Anglophone literature that regard the linking function as a central feature for a marker to be a discourse marker (e.g.Schiffrin, 1987;Fraser, 2006).
I therefore support Pichler's approach to use the more neutral term of discourse-pragmatic features (Pichler, 2016a).In line with this, Dittmar (2011) and Fernandéz-Villanueva (2007) opt for a functional classification on a scale from more logical-semantic to more discourse-pragmatic uses.However, the notoriously challenging task of interpreting discourse-pragmatic features remains (as, for example, Crible (2018) pointed out).
In summary, this paper provides empirical evidence for the systematic variation of a specific polyfunctional lexical item across spoken and written modes and informal and formal settings, in addition to different contact situations.This way, it complements research in the field of variational pragmatics and on discourse-pragmatic features more generally.

Figure 3
Figure 3 Functional variation patterns of also across speaker groups (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS s in the US) and formal (f) and informal (i) spoken registers.Functional categories: ADV = adverbial connector, DM = discourse marker, ELAB = elaboration, EVAL = evaluation, REP = repair

Figure 4
Figure 4 Functional variation patterns of also across speaker groups (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS s in the US) and formal (f) and informal (i) written registers.Functional categories: ADV = adverbial connector, DM = discourse marker, ELAB = elaboration, EVAL = evaluation, REP = repair

Figure 5
Figure 5 Median (bars), interquartile range (boxes) and individual normalised frequencies (dots) of also as an adverbial connector (ADV) across speaker groups (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS s in the US)

Figure 6
Figure 6 Median (bars), interquartile range (boxes) and individual normalised frequencies (dots) of also indicating an evaluation (EVAL) across speaker groups (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS s in the US)

Table 1
Token and speaker figures in the German subcorpus of the RUEG corpus by speaker group (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS in the US) and contact languages (h-T = heritage Turkish, h-G = heritage Greek, h-R = heritage Russian) been discussed, but they are relevant to our data (but see Aijmer, 2011 on well).Such instances were probably treated under the broad category of reformulations.Moreover, also is sometimes used to signal that something previously mentioned (but not in the immediate surrounding text) is picked up again (see BAC in Table2).Speakers also use also to indicate the end of the narration (Example 11, see END in Table2).I subsume these functions (BAC, END) and the function of initialising the narrative (Example 4, see ININ in Table2) under the category of DM s.These three functions are textual functions that serve solely to organise the discourse, with also being semantically bleached out (seeHopper and Traugott 2003: 94ff.on semantic bleaching).

Table 2
Overview of categories, functions and positions of also and their absolute frequencies in the RUEG corpus

Table 4
Normalised frequencies per 100 tokens of also as an adverbial connector (ADV) across speaker groups (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS s in the US) and registers

Table 5
Normalised frequencies per 100 tokens of also indicating an evaluation (EVAL) across speaker groups (DEbi = bilinguals in Germany, DEmo = monolinguals in Germany, USbi = bilingual HS s in the US) and registers