Allostructions and stancetaking: a corpus study of the German discourse management constructions Wo/wenn wir gerade/schon dabei sind

2.1 Construction Grammar and Construction Discourse

The theoretical approach I adopt in this paper is Construction Grammar (CxG, e.g., Beckner et al. 2009; Fillmore et al. 1988; Goldberg 1995, 2006, 2019; Hilpert 2014). A central tenet of CxG is that all our linguistic knowledge may be captured in terms of constructions, i.e., conventional form-meaning-pairings. While most constructionalist studies focus on constructions that operate on units no larger than the sentence level, Croft’s early model of the structure of constructions already comprises discourse-functional properties (Croft 2001: 18 cf. Figure 1).

Figure 1:

The symbolic structure of a construction (Croft 2001: 18).

Östmann (e.g., 2005 and 2015) has argued that CxG should not be limited to examining constructions at sentence level, but should also include further structures:

CxG methodology can be fruitfully extended to account for discourse phenomena, and in doing so will enhance our understanding of how discourse works, and will allow us to explicate discourse structures and processes in a more systematic manner. (Östmann 2005: 125)

He refers to constructions that emerge out of this enterprise as “Construction Discourse” (Östmann 2005, 2015). Similarly, Traugott advocates paying more attention to pragmatic markers such as discourse markers in CxG since they “are an important part of speakers’ knowledge of language.”, […] having “conventional pragmatic meanings” (Traugott 2021: 1). Since then, CxG has successfully been applied to several structures fulfilling discourse functions (e.g., Traugott 2020, 2021; Ziem 2015^[2]).

Works from the realm of interactionally oriented CxG (e.g., Günthner and Bücker 2009; Günthner and Imo 2006; Imo 2010) have especially focused on constructions operating at discourse level. These include discourse markers and projecting phrases, which resemble one another in that they are often syntactically disintegrated, usually filling the position in front of an independent main clause (so-called pre-prefield position in German ‘Vor-Vorfeldposition’). Often, they have units larger than a sentence in their scope and serve to manage discourse. By fulfilling pragmatic functions, they are non-obligatory markers, typically used in conceptually spoken language.

Such pragmatic markers are good examples of constructions because they often consist of sedimented patterns which are characterized by a low degree of variability and are memorized in their entirety. It is central to usage-based approaches and CxG that constructions display a certain degree of fixation. Besides fixed position, they contain open schematic slots that often tend to collocate with a set of specific slot filler items (e.g., Fillmore et al. 1988). These so-called collexemes are crucial to our understanding of constructions because the preference of certain lexical items is often motivated by functional features. That means that lexemes that are attracted to the open slots of a construction often provide some clues to its function. For example, Goldberg observes that “[g]ive, in fact, is the most prototypical ditransitive verb because its lexical semantics is identical with the construction’s semantics. […] [Therefore,] it is the most frequent word in this construction” (Goldberg 1992: 54). By comparing the collocates of two seemingly synonymous constructions, we are able to determine how the constructions differ from one another and in which ways they are related. The underlying assumption is that co-occurrence and distributional patterns mirror functional properties of constructions and provide evidence for differences in meaning. In what follows, I will argue that differences in collocations point to pragmatic differences, in particular the stances the patterns conventionally convey. Due to the high overlap of the patterns examined in this paper, I will conclude that they are allostructions – a concept that will be elaborated on in the subsequent section.

2.2 Allostructions

The concept of allostructions has sparked debate within the realm of CxG. The term was originally coined by Cappelle (2006) in reference to other allo-relationships such as allophones or allomorphs and subsequently further developed by Perek (2012, 2015; for a recent comprehensive overview see Zehentner 2023). Similar to allophones or allomorphs, allostructions are variants of the same construction. While being highly similar or even equivalent in terms of their function, they exhibit formal differences. However, in most studies, it is assumed that some level of formal similarity is necessary for two variants to be classified as allostructions (Zehentner 2023: 4–5). It is still under debate to what extent the variants need to resemble one another formally and functionally to count as allostructions (see Zehentner 2023).

According to Perek (2012), allostructions are vertically related to an overarching construction that unites the formal and functional properties which are shared by both allostructions. This so-called constructeme is more schematic than the allostructions because it only contains features that are shared by both allostructions and is not specified with respect to the slots the allostructions differ in. In addition, it encompasses shared semantic features of both allostructions, while excluding those that are unique to only one of them. Figure 2 demonstrates how this can be visualized within a constructional network, based on Cappelle’s (2006) allostructions of verb particle placement (e.g., She turned off the TV/She turned the TV off).

Figure 2:

The transitive verb-particle constructeme and its two allostructions (Cappelle 2006: 18).

Figure 2 illustrates that Capelle captures the similarity between constructions not only through their vertical relationship to an overarching constructeme but also through the horizontal links among the allostructions themselves, reflecting their formal and functional proximity. Despite their strong resemblance, the alternating constructions may differ in their pragmatic or discourse-functional meaning (De Vaere et al. 2020) or in terms of other external characteristics (Zehentner 2023: 6).

To identify allostructions, it is necessary to measure the degree of formal and functional overlap of the potential allostructions. Maekelberghe (2022) shows that the degree of functional and formal overlap and distinction may be measured by using Hierarchical Configural Frequency Analysis (HCFA), a method that has been employed to assess the similarity of constructions (Maekelberghe 2022) or to identify different types of constructions (e.g., Hilpert 2013: 55–66). In this study, I will analyze the (dis)similarity of German discourse organizing constructions by using HCFA as well. Before outlining the empirical study in more detail, the subquent section sketches what we know about these constructions from previous research.

3.1 Information structure of causal wo and wenn clauses

Before delving into the patterns that are of primary interest in this paper (examples in (2)), I will first address some characteristics of the causal connectors they contain, namely wo and wenn.

I will especially focus on their information-structural properties because they will be relevant throughout this paper. While wo and wenn are used to introduce a causal clause in (3), this is not their core function. Wo is a genuinely spatial relative adverb meaning ‘where’, wenn ‘if, when’ mainly serves to introduce conditional or temporal clauses. However, both connectors are known to establish causal relations in specific contexts, especially in spoken language (Gohl 2000; Günthner 2003, 2005; Pasch 1999), see ((3)).

It is noteworthy that causal clauses with wo and wenn share pragmatic properties that set them apart from clauses introduced by weil ‘because’, the most commonly used causal connector in German. Several studies have reported that the proposition of causal wo and wenn clauses is usually “evident” or “interactionally given” (Frohning 2007: 130–135; Gohl 2000: 3–4; Günthner 2005: 169). That means that the content of the causal clause either builds on shared knowledge or has previously occurred in the discourse. It is thus familiar to the interlocutors or it is at least presented as being presupposed. Weil, in contrast, may introduce a reason or cause that contains information that is new to the interlocutors.^[4]

However, the ways in which wenn and wo relate to familiar content are not entirely equivalent. This particularly applies to the sources that are implied for the presented content. Causal wenn clauses provide information on the source a statement is based on. By using causal wenn clauses, the speaker indicates that the evidence that supports their statement is not based on their own knowledge or an utterance made by themselves, but on an utterance previously made by another interlocutor, often the addressee (Gohl 2000: 17–18). That means the content presented by wenn clauses does not draw on speaker-based knowledge, rather, it is based on a different source. This is illustrated in example (4) cited from Gohl (2000).

In example ((4)), a telephone counselor recommends that the client consider long-term counseling. He offers this advice based on the client’s earlier comment regarding her siblings’ support for their mother (lines 06–07). Even though the counselor slightly modifies the statement (lines 39–40 if/as your siblings cannot help you), he makes it clear by using the connector wenn ‘if/as’ that his reasoning and conclusions are based on his addressee’s previous utterance. This is an example of how causal wenn clauses draw on non-speaker based common knowledge, implying that the source of the content presented is not the speaker but another interlocutor. Such causal uses of wenn can be categorized as an “evidential strategy” in the sense of Aikhenvald (2006; also see Diewald and Smirnova 2010: 40–46). According to Aikhenvald (2006) it is typical for conditionals to be used as “evidential strategies” (Aikhenvald 2006: 321): “Conditionals and other none-declarative moods may acquire overtones of uncertain information obtained from some other source, for which the speaker does not take any responsibility […]” (Aikhenvald 2006: 321). Dancygier and Sweetser describe a similar use of English if, which “may just acknowledge that this claim is not the speaker’s own” (Dancygier and Sweetser 2000: 130).

3.2 Discourse management projecting constructions with wo and wenn

Due to their information-structural properties, causal wo and wenn clauses are especially suitable for the constructions that are examined in this paper. While there has been no study dedicated to those patterns in particular, they have been previously mentioned in research, especially in papers where causal wenn and causal wo clauses in general are examined (Gohl 2000; Günthner 2003, 2005). It has been shown that the patterns are used to create discourse coherence and introduce a new topic, clarifying how the following utterance is related to the previous discourse. That means they figure as discourse management markers, in particular as “topic orientation markers” in the sense of Fraser (2009: 893). It is not surprising that they favor causal wenn and wo over weil since a central function is to relate to the preceding context and, in doing so, reference familiar content (see Section 3.1). Examples ((5)) and ((6)) provide some usage examples cited from previous research.

In example ((5)), the speaker refers back to an earlier point in the conversation. Having previously debated Marx, she alludes to this by the phrase wemmer vorhin scho bei marx warn ‘since we have already talked about Marx before’ (line 05), which she uses to introduce a famous Marxist quotation: Religion is the opium of the people. In example ((6)), the speaker refers directly to the previous topic of school enrollment (line 01). She takes the opportunity to ask the addressee about her view on the best age for a child to start school. Both clauses operate on the speech act domain. That means they offer a reason as to why the following speech act is uttered. By doing so, they anticipate the subsequent utterance. In addition, they are used to manage the topic shift, showing how it is related to the preceding discourse.

The examples ((5)) and ((6)) suggest that, the patterns exhibit a recurring structure which combines a subordinate causal connector wo or wenn with a subject (usually a personal pronoun) followed by an adverb and a prepositional phrase that serves as the object of a clause-final finite verb (see (7)). Despite their subordinate structure, the clauses are not integrated into a matrix clause, but occur independently, anticipating the speech act of the subsequent clause.^[5]

This structure is associated with specific pragmatic functions. As typical of metatextual discourse markers (Traugott 2020: 5), they usually operate on the speech act level. They serve as “framing devices and provide the reason or background for the following activity (e.g., for uttering the following suggestion, for asking the following question, etc.)” (Günthner 2005: 163). In this use, the clauses serve as projecting constructions which anticipate the subsequent speech act (Günthner 2008). That is why the wenn and wo clauses occur in a syntactically unintegrated position: As illustrated in examples ((5)) and ((6)), they are not embedded in one of the matrix clauses’s internal positions but occur in the so-called pre-prefield (‘Vor-Vorfeld’), immediately preceding and anticipating the following matrix clause. The disintegration is thus reflected by the fact that the preposed clause, unlike syntactically integrated clauses, does not trigger inversion of subject and predicate in the matrix clause; instead, the main clause preserves its Verb-2-structure (see example (5), lines 5–6). In spoken language, these usage patterns constitute independent prosodic units with falling pitch and a pause after the subordinate clause (Günthner 1999: 11).

Overall, the patterns introduced by wenn and wo overlap in their discourse organizing functions and in their recurring structure. Although the high resemblance of the patterns is evident, they have up to now not been jointly examined. Considering their high overlap, questions arise as to whether they are synonymous and are thus formal variants of the same construction or whether and, if so, in which ways they are distinct from one another. This is particularly relevant because, as I will show in the following study, they differ in pragmatic features (Finkbeiner 2019), in particular the stance they conventionally tend to convey. Before delving into the empirical study, I will therefore briefly outline the terms stance/stancetaking and positioning.

6.1 Syntactic placement

Corpus examples illustrating syntactic placement are given in (10). In contrast to what the research literature might lead one to expect, there is a noteworthy degree of variability in syntactic placement. As mentioned in previous research (see Section 3), the phrases are structurally disintegrated in most of the cases and followed by an independent speech act (e.g., (10)a). However, the adverbial clause can also be syntactically integrated into the matrix clause (e.g., (10)b,c). In these cases, the main clause usually contains a verbum dicendi, explicitly designating a specific type of speech act (e.g., nachfragen ‘inquire’ in (10)b), which is then followed by indirect speech.^[12] Interestingly, there are individual cases where the phrases are syntactically integrated without the use of verba dicendi (e.g., (10)c). In these cases, the subsequent main clause typically contains requests, pieces of advice, or proposals (as exemplified by the imperative erwähnt auch … ‘also mention…’ in (10)c).

In 21 of the instances, the constructions occur interposed, meaning they appear as syntactically disintegrated parentheses. Their disintegration is either marked by brackets or dashes. These interposed instances contain metapragmatic comments on why a specific term or wording is used (e.g., Aggression ‘aggression’ in example (10)d). In 41 cases, the constructions occur postponed, offering a retrospective justification for their previous utterance (e.g., (10)e).

According to Traugott, the observed syntactic variability is typical of pragmatic markers which “may appear in several positions in the clause” (Traugott 2021: 5). However, the majority of the cases (1,062 of 1,177) are preposed and disintegrated such as ((10)a). During the qualitative analysis of the data, the variants were found to be very similar even in different syntactic positions. This means that their basic discourse-management function (namely to provide a reason for a topic shift) does not change depending on the sentence’s syntactic placement. Therefore, all instances were included in the study.

The next two sections are dedicated to identifying overlaps as well as differences between the patterns, determining the extent to which they diverge from one another, and assessing the degrees of their similarity.

6.2 Hierarchical configural frequency analysis

Statistical tests employed in this paper aim to detect collocational preferences of the patterns under scrutiny. The underlying assumption is that different collocational preferences, i.e., the frequent co-occurrence of lexical elements, reflect functional differences (see Section 2.1).

In order to identify different collocation preferences, I ran a Hierarchical Configural Frequency Analysis (HCFA) (Gries 2004; Posit team 2023). HCFA is a multivariate exploratory method (Gries 2009: 248–252; Hilpert 2013: 55–66; Maekelberghe 2022) that is used to “detect data structures in a bottom-up fashion” (Hilpert 2013: 56). It tests whether specific combinations of variables, so-called configurations, occur more frequently than would be expected if the variables were distributed by chance (Gries 2009: 24). When studying constructions, the variables correspond to slot fillers, i.e., lexemes that may occur in specific positions of a construction; the test examines whether specific fillers of different slots are combined more frequently than would be expected if they were distributed by chance. Thus, HCFA can answer the question “whether the tokens of a grammatical construction can be characterized in terms of one or more types […] [it] can determine the prototype structure of a construction and its variants” (Hilpert 2013: 57). By doing so, groups in a population that share similar features can be identified. I use HCFA to detect the prototype structure of the discourse management constructions introduced above. Six different slots are prevalent in the dataset: Connector, subject, adverb, preposition, noun phrase and predicate (cf. Table 1).

Figure 5 illustrates the token frequencies of the individual connectors among the projecting constructions. Wenn and wo are almost equal in terms of their frequency, proving that both are commonly used in the construction.

Figure 5:

Frequency of connectors in the discourse managment construction (n = 1,177).

In most of the cases (1,173 of 1,177 instances), the subject slot contains a personal pronoun such as ich ‘I’, du ‘you’ or wir ‘we’, and is followed by a temporal adverb such as gerade ‘just’ or schon ‘already’ (1,069 of 1,177 instances). In addition, the examples always include a prepositional phrase, which either consists of a preposition containing a nominal phrase or a so-called pronominal adverb (e.g., dabei ‘on this’). As is common in German subordinate clauses, the finite verb is placed at the end of the clause.

The input variables for the HCFA comprise almost all the constructional slots presented in Table 1. The prepositional phrase (including the preposition and the embedded noun phrase) had to be excluded because it contained too many different filler items to be handled in the analysis. That is why preposition and nominal phrase are represented in light grey colors in Table 1. In order to determine which nouns occurred above chance, a Distinctive Collexeme Analysis was run separately (see Section 6.4).

To keep the HCFA feasible, the number of filler items had to be reduced. Formal spelling variants were unified with the variants that conform to the codified standard (e.g., gerade, grad, grade > gerade ‘just’); combinations of two adverbs that occurred with less than five instances were reduced to the more frequent one (e.g., dann schon ‘then already’, hier schon ‘here already’ > schon ‘already’).

Below, HCFA will be used to examine whether the specific slot fillers co-occur above chance. Slot fillers that are combined more frequently than expected are referred to as types or configurations. In addition, HCFA reveals whether certain slot fillers co-occur less frequently than would be expected if they were distributed by chance. That means the combination of these lexical filler items is dispreferred. These instances are therefore called antitypes.

6.3 Results obtained by the hierarchical configural frequency analysis

As mentioned above, four different variables were included in the HCFA: Connector, subject, adverb, and predicate. The analysis yields 4 types,^[13] all of which are highly significant (column Dec in Table 2). The results are ordered according to effect size (Q). The column Freq contains the observed token frequency of each combination. Since all cases provided in Table 2 are significant, these observed frequencies are always higher than the expected ones (column Exp).

The first lines represent the combinations of slot fillers that co-occur with the highest effect size. In all significant cases, the predicate slot contains the copula sein ‘to be’. In the first two lines, there is an additional overlap in the subject slot: Both configurations favor the pronoun wir ‘we’ as their subject. The fact that the subject and the verb slots attract the same filler items suggests that there is a strong functional overlap: The referent as well as the denoted event appear to be the same. This supports the assumption that the different types are cognitively linked and inherit from one dominating macro-construction.

However, Table 2 indicates that there is a certain degree of variability, especially concerning the connector and adverb slot as well as the subject slot. The first two lines differ regarding the collocational preferences among the connector and the adverb slot. The type introduced by wenn attracts the adverb schon ‘already’ while the type introduced by wo collocates with the adverb gerade ‘just, currently’. Proceeding to the next two rows (lines 3 and 4), these collocational preferences continue to exist, but we find additional differences in the subject slot. While the collocation wo … gerade attracts the first-person singular pronoun ich ‘I’, wenn … schon tends to occur with the second person singular pronoun du ‘you’. We will come back to these collocational preferences in Section 6.5, which will be explained by differences in how the types relate to the preceding discourse.

The antitypes that were obtained by means of the statistical analysis corroborate the results achieved for the types. Table 3 gives an overview of the combinations that are statistically dispreferred:

It confirms that the constructions introduced by wo and wenn mainly differ in the subject and the adverb slot. While the connector wenn disprefers the subject ich ‘I’ (line 4) and the adverb gerade ‘just, currently’ (lines 1 and 4), wo statistically repels the subject du (line 3) and the adverb schon (lines 3 and 4). This does not mean that the combinations do not occur at all; they are simply statistically highly dispreferred. There are for example 32 cases of wenn wir gerade PP sind ‘as we are just PP’ opposed to 288 cases of wenn wir schon PP sind ‘as we are already PP’. Interestingly, the combination wenn ich gerade PP bin ‘as I am just PP’ does not occur at all in the dataset.

Overall, HCFA has revealed that there is a considerable overlap in the slot fillers that co-occur in the structures: The same subject pronoun (wir ‘we’) as well as predicate (sein ‘to be’) are preferred with both connectors. This reinforces the interpretation that we are not dealing with entirely different constructions but rather with variants of a single construction, that is allostructions. The different collocational preferences between the connector and the adverb slot and, to a lesser degree, the subject slot, however, suggests that these allostructions encompass distinct sedimented patterns, which are summarized in ((11)). Subsequently, I will refer to these allostructions as wo-wir/ich-gerade-PP-sind/bin and wenn-wir/du-schon-PP-sind/bist.

The next section presents the results of a Distinctive Collexeme Analysis that was run in order to test which (kind of) nouns are attracted to either of these allostructions.

6.4 Measuring the distinctiveness of the discourse organizing constructions: distinctive collexeme analysis

As mentioned in Section 6.2, the preposition and NP slots could not be included in HCFA because they contain too many different types. I assumed that especially the NP, which mainly contains content words, might provide some additional clues regarding the constructions’ functions. For this reason, I ran a distinctive collexeme analysis to see if certain nouns are attracted by any of the constructions identified above (Gries and Stefanowitsch 2004; Levshina 2015: 241–251).

“Collostructional analysis” (Gries and Stefanowitsch 2004; Stefanowitsch and Gries 2003) is a statistical method based on the Fisher-Yates exact test, which has been widely used in CxG (e.g., Flick 2016; Hartmann 2018; Hoffmann et al. 2019). It determines “the strength of the relationship between word constructions and the grammatical structures they occur in” (Stefanowitsch 2012: 290). Collostructional analysis, thus, helps us to assess how strongly a grammatical construction is associated with certain lexemes.

Gries and Stefanowitsch have extended the method in order to compare the collocation strength of certain lexemes in different constructions (Gries and Stefanowitsch 2004). This so-called “distinctive-collexeme analysis” calculates the dissimilarity of “near-synonymous (or functionally near-equivalent)” (Gries and Stefanowitsch 2004: 101) constructions on the basis of lexemes that occur in a particular slot of the construction. Distinctive collexeme analysis aims to study how semantically similar constructions differ in terms of their collocational preferences. Again, the assumption is that lexical collocates can help to identify the meaning of constructions.

Since the configurations identified by HCFA all differ in terms of the connector slot, I compared how often one of the connectors wenn and wo co-occurred with which filler item of the NP slot. I measured if any of the connectors are combined with particular lemmas above chance when compared to the nouns co-occurring with the other connector. To measure this attraction, I used a function^[14] that is included in the Rling package (Levshina 2015: 232). Table 4 presents the nouns that are attracted to the pattern with wo. According to Stefanowitsch and Gries, “absolute values exceeding 1.30103 are significant at the level of 5 % […] and values exceeding 2 and 3 are significant at the levels of 1 % and 0.1 % respectively” (Stefanowitsch and Gries 2005: 7).^[15] By this definition, only the first two instances are significant. However, looking at the patterns in the meaning of the most frequently occurring collexemes may still help us to identify differences in meaning, regardless of their being conventionalized as part of the construction or not, especially if some shared semantic features can be identified.

All these slot fillers attracted to the variant with wo have in common that they provide neutral ways of relating to the preceding context, without conveying any evaluative meaning. Table 4 shows that wo-wir/ich-gerade-PP-sind/bin mainly attracts highly general collexemes, the most frequent of which being the anaphoric element da-, which occurs in pronominal adverbs such as dabei or darüber. This suggests that pronominal adverbs, which anaphorically refer to previous statements, are strongly associated with wo-wir/ich-gerade-PP-sind/bin.^[16] In addition to da-, the noun Begriff ‘term’ is significantly attracted to the variant with wo. Also, lemmas such as Thema ‘topic’, Point of View or Punkt ‘point’ occur more often with wo than with wenn even though the difference is not significant.

Preceding to the pattern with initial wenn, we find that none of the collexemes is significant (cf. Table 5). In total, 237 different types occur in the NP slot, most of which are hapagax or dis legomena. The ones which were found to be most strongly associated occur with a maximum frequency of five tokens. This suggests that wenn-wir/du-schon-PP-sind/bist is less entrenched with a specific filler item in the noun slot than wo-wir/ich-gerade-PP-sind/bin.

Although no collocations of wenn-wir/du-schon-PP-sind/bist could be identified, the lemmas occurring in the NP slot give us some clues to its meaning, especially because the nouns seem to be semantically related to one another. We can thus make generalizations by considering the collocations in aggregate rather than individually. It is striking that the construction often contains nouns which are semantically related to assessing the reliability of information and sources such as Beleg(quell)en ‘evidence, sources’, Literatur(liste) ‘reference(list)’, Spekulieren/Spekulation ‘speculating/speculation’ or Wahrheit ‘truth’. This suggests that the construction is often used when the reliability or accuracy of a statement is called into question. Example ((12)) illustrates this usage.

In this example, the editors Gdo and Informationswiedergutmachung debate which citizenship the violinist Rachel Schmidt possesses. Gdo asks whether Informationswiedergutmachung has any proof of his claim that the violinist is indeed an Israeli citizen (lines 02–03). When Informationswiedergutmachung replies that his characterization is based on Schmidt’s place of birth and the so-called ius soli (line 04), Gdo responds that this is not sufficient evidence since not everyone born in Israel can be assumed to have Israeli citizenship (line 05).^[17] His critical stance on Informationswiedergutmachung’s statement is visible in the choice of the expression Es ist mitnichten ersichtlich, dass ‘It is by no means evident that’ (line 05). To provide a counter argument, he mentions that it might also be possible that Schmidt has adopted German citizenship (lines 06–07). Using the phrase wenn wir schon am Spekulieren sind ‘as we are already speculating’ (line 06), he acknowledges his own speculations, which he justifies by highlighting that Informationswiedergutmachung’s reasoning has been highly speculative, as well. Note that the construction here references a contribution previously made by another discussant, which will be relevant in the following discussion (see Sections 6.5 and 7). Gdo justifies his own speculating, highlighting Informationswiedergutmachung’s previous speculations in order to prove that his reasoning is inadequate. This example illustrates how wenn-wir/du-schon-PP-sind/bist is used to call into question the reliability of a statement made by another editor. The use of the allostruction provides a way of disaligning with the addressee and to invoke a negative evaluation of their reasoning.

The negative evaluation appears to be a common characteristic of the allostruction and is also reflected by its other collexemes (cf. Table 5), most of which bear negative connotations: Spekulieren/Spekulation ‘speculating/speculation’,^[18] for instance, suggests that someone made claims without having sufficient evidence, or Korinthenkacken/-produktion ‘nit-picking’ refers to someone being overly correct. The next section further elborates on these evaluative properties, showing that the types identified using the HCFA typically differ in terms of conventionalized pragmatic features.

6.5 Qualitative analysis of the results

This section focuses on a qualitative analysis of the structural types that were identified using HCFA in order to explore their functional commonalities and differences. After a qualitative examination of all the data, examples were extracted that have proven to be most representative of the two configurations that were identified using HCFA. In what follows, I will provide prototypical examples of those configurations and will describe their functions.

Example ((13)) illustrates a prototypical usage context of wo-wir/ich-gerade-PP-sind/bin. In the example, the editor Topspeed first proposes incorporating the article Audi R18 to the article Audi R18 TDI. After several editors have agreed to relocate the article, Topspeed states that he has now moved the article (line 06). He proceeds to discussing a new topic, suggesting to also relocate the article Peugeot 908 (Rennprototyp) to Peugeot 908 (line 08). The topics are clearly related – in both cases the editors suggest to move an article to another section. Topspeed uses wo-wir/ich-gerade-PP-sind/bin (line 07) to link the new topic to the previous one by highlighting how both are similar (the similaritty is also indicated by the repetition of the verb verschieben ‘to move’)^[19] (Wo wir gerade beim Verschieben sind ‘Speaking of moving’). Doing so, the construction establishes topical relevance for the subsequent statement and is used to slightly shift the topic.

In (13), wo-wir/ich-gerade-PP-sind/bin introduces a topic shift that links a novel topic to the one immediately preceding it. This may explain the collocational preferences observed in Section 6.4: The textual proximity of the linked topics is the reason why anaphoric elements and nouns referring to discourse topics in general, without naming any specific one, show the strongest attraction to wo-wir/ich-gerade-PP-sind/bin. The previous topic typically stems from the same editor or has been collaboratively developed by a group of discussants, including the editor, which explains why the first-person pronouns ich ‘I’ or wir ‘we’ are statistically attracted to the subject slot (see Section 6.3). In such contexts, the allostruction usually does not invoke any evaluation and remains neutral in terms of stance.

This is different when it comes to wenn-wir/du-schon-PP-sind/bist. Although its text management functions closely resemble those described for wo-wir/ich-gerade-PP-sind/bin, it diverges in how it relates to the previous discourse and, as a result, in terms of the stance conveyed, often expressing disalignment.

Example ((14)) illustrates a prototypical usage context. The editor Lefanu critically notes that the literature that has been cited in the article Alchemie ‘Alchemy’ does not meet the requirements of Wikipedia articles (line 02). In her subsequent statement, Henriette relates to his objection using wenn-wir/du-schon-PP-sind/bist (Wenn wir schon bei Zweifeln zur Eignung von Büchern sind ‘While we’re on the subject of books and their suitability’, line 06). Again, the construction is used to introduce a new topic by showing how it relates to the previous one.

Typically (also see example (12)), wenn-wir/du-schon-PP-sind/bist refers to content or statements introduced by an interlocutor other than the speaker, and it is used to express disagreement or to disalign with their point. The discussion in ((14)), for example, starts with Lefanu critizing the selected literature (line 02). Subsequently, Henriette, who is not the editor responsible for the critized literature, inquires on what his critical stance on the books is based (lines 03–04), indicating that his criticism might not be well-enough substained. By adding the mitigating and face-saving statement that she herself has little confidence in books with such titles, she concludes her contribution. However, 5 minutes later she adds a new thread introduced by P.S., which marks her contribution as an addendum to her previous statement, followed by the phrase Wenn wir schon bei Zweifeln zur Eignung von Büchern sind ‘While we’re on doubts about books and their suitability’ (line 6). The allostruction here is used to add a similar criticism, this time, however, it applies to references that were added by Lefanu himself. Using the construction, Henriette relates to their previous discussion and ultimately, to the criticism of the reference selection initiated by Lefanu (line 03). Here, a crucial difference becomes evident in how the two allostructions relate to the preceding discourse. Wenn-wir/du-schon-PP-sind/bist usually refers back to a topic that has been introduced not by the speaker but by the addressee. It introduces a critical comment and is used to express disalignment.

This evaluative function is also illustrated by example ((15)), which is taken from the Talk page of the article about the movie Der Baader-Meinhof-Komplex ‘The Baader Meinhof Complex’. The discussion basically evolves between two editors Filoump and Hybscher. Filoump has started the conversation by stating that a reference to a detail about Che Guevara’s death is irrelevant to the article while Hybscher, who has added the respective piece of information, considers it as an important detail and argues in favor of keeping it. After a heated discussion extending over more than one week and involving also other editors, Filoump and Hybscher start swearing at one another ((15)).

Filoump concludes his posting by prompting Hybscher to now refrain from further discussions and by calling him a discussion troll. Hybscher, who apparently feels offended, starts calling Filoump names as well (Korinthenkacker ‘nitpicker’ and Rechthaber ‘know-it-all’). He introduces his swears using Wenn wir mal schon bei gegenseitigen Angiften sind ‘As we’re already swearing at one another’. Once again, the construction refers to a prior contribution made by the addressee. In contrast to example ((14)), however, it does not pertain to previous content but rather to a speech act carried out by this other editor. Hybscher justifies his swearings by pointing out that Filoump previously called him names as well.

Both examples illustrate that wenn-wir/du-schon-PP-sind/bist often introduces utterances in which the speaker takes a critical stance or invokes negative evaluation: In ((14)), Henriette uses the construction to introduce a statement that is critical on the literature selected by Lefanu (Die Erstausgabe des von Dir eingesetzten Schmieder stammt aus dem Jahr 1832 ‘The first edition of the Schmieder you used dates from 1832’, line 04). In ((15)), the construction introduces the utterance of plain insults. After a qualitative examination of several instances, this usage seems to reflect a general tendency of the construction to introduce an episode that marks disalignment (also see example (12)).

The different ways of referring back to the preceding discourse provide an explanation for the collocations identified using HCFA: The observation that wenn-wir/du-schon-PP-sind/bist typically references content introduced or previously discussed by the addressee, not the writer, explains why the allostruction often collocates with the second-person pronoun du ‘you’ (see Section 6.3). The subject position of the construction contains the discussants who have introduced the topic or were involved in the discussion (wenn wir/du beim Thema sind/bist ‘since we/you are on the topic’). As the examples ((14)) and ((15)) have demonstrated, the allostruction with initial wenn often refers to contributions that originate from another discussant even if the subject slot does not contain a second-person pronoun, but the first-person plural pronoun wir ‘we’. In these cases, wir is to be understood as an inclusive pronoun that encompasses the addressee. The allostruction is used to reference a statement made by the addressee within the shared discourse.

Interestingly, the statements of other discussants are not only referenced but also negatively evaluated. The allostruction is used to express disalignment with the addressee, which is reflected by nouns collocating with wenn wir/du schon PP sind/bist (see Section 6.4). It appears that there is a connection between the subject referent and the stance that is being conveyed: Traugott notes that “when the subjectification is orientation toward a negative evaluation, the locus of change will be comments about non-first persons, especially third persons (‘They’ are negatively evaluated in some way)” (Traugott 2010: 58). As for the projecting construction discussed here, it is the addressee or an utterance made by them that is being negatively evaluated. This may be at least partially explained by the argumentative context of Wikipedia Talk pages where ways of collaboratively creating and improving articles are discussed. In these contexts, discussants often position themselves towards statements made by other editors.

The question arises as to why a construction that tends to mark disalignment and negative evaluation collocates with the adverb schon ‘already’. According to the DWDS dictionary,^[20] schon ‘already’ expresses that an event occurred earlier than expected, thus often conveying surprise, astonishment or even reluctance. It is especially because of these latter evaluative aspects of meaning that the adverb seems to go well with the construction and its tendency to mark a negative speaker’s stance and disalignment. The adverb gerade ‘just, right at this moment’ instead is more neutral, referring to the current moment of utterance.

Due to its tendency to express negative evaluation or disalignment, wenn-wir/du-schon-PP-sind/bist is highly suitable for texts, such as the Talk pages analyzed in this paper, where different positions on how to improve an article are discussed. It references an often offensive or inaccurate statement in order to introduce a related argument that the speaker uses against other interlocutors. Wo-wir/ich-gerade-PP-sind/bin, in turn, is rather used to manage topic shifts in neutral ways. The following section offers a proposal of how the two allostructions could be analyzed within a constructional network.