Analogy as driving force of language change: a usage-based approach to wo and da clauses in 17th and 18th century German

2.1 Da and wo clauses in present-day German

This section addresses and compares the functions of da and wo clauses in present-day German. Since the paper focuses on the historical development, the contemporary uses will only be briefly sketched (for more details see Breindl and Walter 2009; Günthner 2002, 2005; Pasch 1999; Pittner 2004).

When used as a connector, German da clauses fulfil a causal function^[2], in which it competes with the more frequent causal connector weil ‘because’, see (1).

Gillmann (2020) shows that da clauses are preferred in formal and written registers. Wo, in contrast, appears as a relative adverb in the German standard variety and is restricted to the spatial function, see (2).

Only in less formal, spoken registers does wo introduce “temporal, causal and even concessive clauses” (Günthner 2002: 310) which exist alongside the spatial use. Examples (3)–(5), cited from Günthner’s (2005) study on wo clauses, illustrate uses that belong to these categories.

According to Günthner “[t]he meaning and function of wo-constructions are neither static nor fully determined, but wo-constructions have a broad functionality which can invite particular inferences depending upon the context at hand.” (Günthner 2002: 179). Thus, da and wo clauses only overlap in their causal uses.^[3]

In addition, the connectors differ in that they belong to different registers. These variational differences may be described with Koch and Oesterreicher’s (e.g., 2007) model of language of immediacy and language of distance or conceptionally spoken and written language (also see Ágel and Hennig 2006a, 2006b; Raible 1994).^[4] As for the connectors this paper is dealing with, it can be stated that da is associated with conceptionally written language, i.e., formal registers. Adverbial wo clauses, in contrast, are associated with conceptionally spoken language, i.e., informal registers.

In what follows, I will show that da and wo clauses resembled one another even more in historical stages of German, which can be traced back to the fact that they arose out of similar source constructions.

2.2 Historical sources of da and wo clauses

Both da and wo clauses developed in similar ways: a relative adverb was grammaticalized as a connector (MHG Grammar 1998: 413–414, 459). However, the development of da was somewhat more complicated since it emerged from two distinct relative adverbs, which initially had different forms and meaning, but converged in Early New High German.

The first source construction of da clauses were temporal relatives introduced by Old High German thō ‘then’ (>MHG dô > ENHG da(/do); Dal and Eroms 2014: 246–247); this use is illustrated in (6) by an example from the Middle High German Iwein.

The second source construction were spatial relatives introduced by the Old High German adverb thār ‘there, where’ (>MHG dâ > ENHG da; Dal and Eroms 2014: 246–242), as illustrated by the Middle High German example (7).

While being formally distinct in Middle High German, dô and dâ converged in Early New High German, leading to the emergence of a polysemous marker; the resulting da encoded both spatial and temporal meaning.

The connector wo arose from a functionally similar source, namely the Old High German relative adverb construction sô (h)wâr (sô), which combined the construction sô adverb (sô) with the interrogative adverb wâr ‘where’ (DWB 1854–1961: 911; MHG Grammar 1998: 413–414) and had a generalizing meaning corresponding to English ‘wherever’.^[5] In Middle High German, sô (h)wâr (sô) was formally reduced to swâ.

Thus, Middle High German dâ and swâ were similar in that both indicated a spatial relation, but differed in that dâ referred to specific spaces, see (7), while swâ denoted generalizing ones (DWB 1854–1961: 911). The latter is illustrated by example (8).

After being subject to several sound changes, swâ continued to exist as wo in Early New High German. Simultaneously, it lost its generalizing function and spread to specific contexts where it started to gradually replace the adverb da (ENHG Grammar 1993: 447–448). As a result, da and wo were interchangeable when referring to spaces in Early New High German.

Subordinating da, in addition, was a frequent marker of temporal clauses in Middle High German see (9) until the connector als ‘when’ came about in Early New High German and started to replace da in its temporal uses (Axel-Tober 2012: 269).

The changes described above, especially the replacement of da by wo relatives were part of a more general ongoing reorganization process that affected the coding of relative clauses in general (Fleischmann 1973: 115–121). From Old to Early New High German, there were two kinds of relative adverbs which differed in terms of their onset: the ones having d- in their onset and the ones with w- in their onset. In addition to da and wo, this included darum/warum ‘for that/what reason’, deshalb/weshalb ‘therefor, wherefor’, deswegen/weswegen ‘because of this/which is why’, darauf/worauf ‘thereupon, whereupon’ or dann/wann ‘then, when’. The d-adverbs were ambiguous in that they were also used as demonstrative adverbs in main clauses. Over the course of the 18th century, a new division of labor arose when d-adverbs were gradually restricted to their function as demonstratives; now, only w-adverbs were used to encode relative clauses. In the wake of this reorganization, the adverb da was gradually replaced by wo in relatives and was limited to demonstrative contexts such as (10).

However, da was not eliminated in subordinate clauses, but evolved into a connector that introduces adverbial clauses. Being restricted to causal uses in present-day German, it fulfilled a range of functions in 17th and 18th century German, which will be described in the following section.

2.3 Polysemous da clauses in 17th and 18th century German

Although da initially had a broader range of meanings than wo, it was being replaced in both the temporal and spatial domains from Early New High German on. However, the connector did not go extinct, but was reinterpreted spreading into new semantic domains by metaphorically mapping the spatial and/or temporal sense to more abstract ones. In a study conducted using corpora of 17th and 18th century German, Gillmann (2020) shows that da clauses fulfilled a range of semantic functions, including spatial, temporal, causal, conditional and even contrast uses (also see Arndt 1960). This suggests that da clauses were characterized by a tremendous polysemy (or semantic vagueness). The different readings of da are illustrated in (11)–(14) (for a more detailed description see Gillmann 2020).

Over the course of the 18th century, the polysemy of da clauses reduced and the causal uses were gaining ground (Brooks 2006: 180–182; Gillmann 2020).

It is striking that in 17th and 18th century texts, da clauses fulfilled a similar set of functions as has been reported for wo clauses in present-day spoken varieties of German (see Section 2.1). Like present-day wo clauses, da introduced spatial, temporal, causal and contrast clauses and, in addition, established conditional relations. The observation of these similarities gave rise to questions of how wo clauses were used in historical texts of the same period and how both connectors interacted. The existing research on Early New High German mentions conditional and temporal uses of wo clauses (DWB 1854–1961: 916; ENHG Grammar 1993: 447–448, 461). The first causal instances of wo clauses cited in the DWB date back to the 19th century (DWB 1854–1961: 916), suggesting that causal uses have emerged only recently. However, empirical research on the diachrony of wo clauses is lacking. For this reason, this paper seeks to establish the historical semantic functions of wo clauses using a corpus of 17th and 18th century texts. I examine the diachronic development of wo clauses, as well as their distribution across different genres. These findings are compared to the results obtained for da clauses in Gillmann (2020).

3.1 Hypotheses

As discussed, it is worthwhile to compare the development of the two connectors because of the observed similarities between da in historical language stages and wo in present-day German. Today, only the causal uses of da and wo clauses share semantic and pragmatic qualities (see Section 2.1). While existing studies deal with the semantics of the connectors in present-day German, we are lacking empirical studies on their development at the beginning of the New High German period. To fill this gap, this paper examines all instances of da and wo that introduce a subordinate clause, based on a corpus of written German texts from the middle of the 17th century to the end of the 18th century. The following hypotheses will be explored:

The observation that da fulfilled a similar set of functions as the connector wo today gives reasons to believe that both connectors had overlapping functions in the relevant period and were linguistic variants (Hypothesis 1). It is thus likely that their distribution was determined by language-external factors. Since both connectors are associated with different registers today, I assume that the same holds for the period under scrutiny (Hypothesis 2). The question is whether we find distributional properties that either show that these differences have already existed or give clues as to why they might have arisen. Since it cannot be excluded that dialect region had an impact, this will be tested as well.

The diachronic development of da and wo may be explained in terms of analogy (Hypothesis 3). Due to the strong resemblance and the similar distribution of their sources, da and wo were cognitively associated and mutually aligned to one another, which is why they developed overlapping functions. Towards contemporary German, however, the connectors’ functions diverged. This process of differentiation might be due to a change in (a) related high-level construction(s) and may thus be explained by means of a broader reorganisation within the constructional network (e.g., De Smet et al. 2018). The following section describes the corpus used in the present study to test these hypotheses.

3.2 Corpus

The analysis in this paper is based on data from the GerManC corpus.^[6] The publicly accessible corpus contains 676,508 words across seven different genres, and arranged into three periods comprising 50 years each (Durrell et al. 2012). An advantage of this corpus is that it is balanced, enabling the comparison of the token frequencies. For the present study six genres were used (see Table 1).

The goal was to determine the function and token frequencies of wo clauses, and to compare the results to the results obtained for da clauses in the same period (Gillmann 2020). To this end, all occurrences of wo clauses were analyzed manually. The data were double-checked by two independent annotators. In order to evaluate the reliability and consistency of the annotations, the agreement among the two annotators was determined by a statistical measure of inter-annotator agreement (Artstein and Poesio 2008). The annotation was created in an iterative process, in which it was repeatedly compared and doubtful cases were discussed. In this fashion, a data-driven annotation scheme was developed and refined. Whenever the annotations of the two annotators did not match, the cases in question were categorized as ambiguous. The Kappa statistics (Fleiß’s Kappa) show an agreement of 0.77 for wo clauses, which corresponds to a substantial agreement according to Landis and Koch (1977).^[7]

According to Durrell et al. (2012), the corpus contains different genres representing “orally oriented registers” (Durrell et al. 2012: 1) as well as “more print-oriented registers” (Durrell et al. 2012: 1), which correspond to Koch and Oesterreicher’s spoken and written conception. However, in many cases a precise classification is difficult. Dramatic texts most likely come closest to the conceptionally spoken language of the time. These dialogic texts can be considered ‘fictitious orality’, but, of course, they do not represent authentic sources of conceptionally spoken language since they are literary texts. The genre of legal texts are most representative of conceptionally written language and are known as the prestigious variety of the 17th century (e.g., Schwitalla 2002: 387). In the corpus, this genre contains texts pertaining to chancery, administration and law. The other genres take an intermediate position in terms of their conceptionality.^[8] A problem we often face when working with historical data is that we are dealing with medially written texts, which makes it generally difficult to draw conclusions especially on spoken language. This should be kept in mind when analyzing the data.

4.1 Token frequency of da and wo clauses

Before delving into the analysis of the semantic functions of wo clauses, I will briefly give an overview of the token frequencies of da and wo clauses in the periods and genres under scrutiny. Figure 1 traces the diachronic changes in frequency. At first glance, it is evident that da clauses were more frequent than wo clauses in all periods. However, in the third period, the incidence of da clauses dropped, resulting in nearly identical frequencies of both connectors.

Figure 1:

Token frequency of da and wo in the three periods of GerManC (n = 1,413).

Each period comprises about 225,000 tokens.

Inspecting the frequencies across individual genres reveals that they differed dramatically, especially in the first period (Figure 2). While legal texts clearly preferred da to wo clauses in the first period, wo clauses greatly outnumbered da clauses in dramas. This observation seems to be in line with Hypothesis 2: The connectors seem to have already been associated with different registers (for more details see Sections 4.2 and 5). However, diachronically, we witness a decrease in da clauses, which reduced the gap between da and wo in all genres (Figure 2).

Figure 2:

Token frequency of da and wo for each period.

By 1800 the frequency of da and wo clauses became more similar in all genres. These tendencies can partly be explained by the semantic functions of the connectors, which will be addressed in the following section.

4.2 Semantic functions and their frequencies in different genres

The next aspect of the diachronic changes of da and wo clauses to consider is their function. The study conducted in Gillmann (2020) reveals that the connector da was characterized by a tremendous polysemy in 17th and 18th century corpora, exhibiting spatial, temporal, causal, conditional, and contrast meaning (see Section 2.3).^[9] Interestingly, we find an only partially different set of functions for wo clauses in the data. Wo clauses occur with spatial, temporal, conditional and contrast reading; only causal wo clauses are not attested in the data. Examples (15)–(18) provide representative instances of each reading. Note that for each semantic function, there are always instances which do not allow for only one single reading; they will be included in the analyses and figures below as ambiguous instances.

Hypothesis 1 is thus confirmed: The data shows that da and wo clauses were polysemous in similar ways. The overlapping functions suggest that the connectors were variants which were interchangeable to some degree.

In order to test which factors determine the choice between da and wo in the corpus, a mixed-effects regression model was computed in R using lme4 (Bates et al. 2015; R Core Team 2020), with the binary dependent variable connector (occurrence of da or wo), the independent variables dialect region, period, genre, and semantic function. I examined whether one of the connectors was preferred in specific language regions or whether significant changes occurred over the periods under scrutiny (also see Section 4.1). Moreover, it was tested whether one of the connectors was preferred in a specific genre, which might suggest that it was associated with a specific register. As mentioned in Section 2.1, da is linked to conceptionally written language (or language of distance) in present-day German, whereas adverbial wo clauses are preferred in conceptionally spoken language (or language of immediacy). The question is whether similar tendencies can be observed in historical stages of German. In addition to these language-external factors, the regression model looks at the semantic functions and examines whether one of the connectors displays a significant preference of one or more of these functions.

The variable individual text was included as a random effect. It was assumed that in addition to the independent variables, there might be additional variables that could influence the dependent variable. However, these are not of immediate interest as explanatory variables. For example, one individual text might contain many instances of the connector wo since the author was prone to use this connector. This is why, the variable individual text was run as a random effect in the mixed-effects regression model. This means that data not only from individual, but from multiple texts is required to draw conclusions about the impact of the independent variables. The final model was selected based on anova tests. Since the model with and without dialect region revealed no significant differences, this variable was not included in the final model. The same holds for interactions.^[10]

The mixed-effects regression analysis shows that while dialect regions do not differ significantly, the difference between genres and semantic functions is significant (Table 2). In Table 2, variables that turned out to be significant are in bold and marked by asterisks according to their level of significance. Additional shading indicates that da is statistically preferred, which is reflected in the negative z value. The significant factors that are in bold but not highlighted by shading trigger the usage of wo.

Especially legal texts and sermons^[11] prefer da over wo, which suggests that the connector was already associated with language of distance (Koch and Oesterreicher 2007). Diachronically, however, the share of wo clauses increased in all genres, which is reflected in the significant result yielded for period 3 (1750–1800). As for semantics, the model reveals clear differences between both connectors. While da is preferred in temporal and causal usage, wo tends to be used with spatial, conditional and contrast readings. The statistical model shows that although da and wo overlap in their functions, both connectors show slightly different tendencies towards different sets of semantic functions.

To gain a better understanding of these differences, the results for the semantic categories and their token frequencies in different genres will now be addressed in turn. I will restrict myself to comparing the spatial, temporal and conditional uses of wo and da since they are most revealing as to how the functions evolved and how the connectors competed with one another.

The statistical analysis has revealed that the spatial function is significantly associated with wo clauses in the data. Unlike in Middle High German, spatial relatives are not restricted to generalizing contexts, but may refer to specific spaces as well ((15)a). This proves that wo had, in fact, spread into the domain of da, which is supported by its high token frequency.

Figure 3:

Spatial uses of wo clauses in the periods under scrutiny (period 1: 180, period 2: 174, period 3: 192 tokens).

Each set of graphs focuses on one specific semantic function. The frequency of this particular function is indicated by the shaded region of the bars. The middle, brighter part of the bars represents ambiguous instances which allow both the semantic function illustrated in that figure and some other interpretation. The lightest and highest part of the bars represents all instances that do not allow that particular function.

Figure 4:

Spatial uses of da clauses in the periods under scrutiny (period 1: 304, period 2: 314, period 3: 248 tokens, see Gillmann 2020: 83).

In Figures 3 and 4, the shaded parts at the bottom represent unambiguously spatial uses, the lighter middle sections represent ambiguous instances that, among others, allow for a spatial interpretation. The lightest part on top represents instances in which a spatial reading can be excluded.

As illustrated in Figure 3, spatial wo clauses have existed since the first period (1650–1700). There is then an increase in their use throughout the 18th century, leading to a share of between 60 and 80% in all genres by the end of the 18th century. The expression of spatial relations had thus become the dominant function of wo clauses in the 18th century. Simultaneously, spatial uses of da clauses diminished (see Figure 4 and Gillmann 2020: 83), which proves that wo replaced da in these usage contexts over the course of the 18th century. Today, only wo may introduce spatial relatives.

Focusing now on the temporal uses, we find that wo only rarely established a temporal link. The few temporal instances are relative clauses with a temporal head in the matrix clause, e.g., Augenblick ‘moment, instant’ in example (16). Over the course of the 18th century, we observe an increase in temporal wo clauses, but on a much lower frequency level than observed for spatial uses (Figure 5). The data suggests that temporal wo developed only by the end of the 18th century, by which time they were still relatively infrequent.

Figure 5:

Temporal uses of wo clauses in the periods under scrutiny (period 1: 180, period 2: 174, period 3: 192 tokens).

Looking at the temporal uses of da, we find that they appear and appear with consistent frequency over all periods analyzed here (Figure 6).

Figure 6:

Temporal uses of da clauses in the periods under scrutiny (period 1: 304, period 2: 314, period 3: 248 tokens, see Gillmann 2020: 84).

Thus, da and wo clauses differ in that the latter less often convey temporal meaning. As mentioned above, causal readings – which became increasingly conventionalized in da clauses over the course of the 18th century (Gillmann 2020) – are not attested at all for wo clauses in the corpus.

However, we frequently come across conditional uses of wo, e.g., example (17). They either denote an implicational relation between states of affairs or a hypothetical situation and its consequences. According to Dancygier and Sweetser (2005), they set up a ‘mental space’ (in the sense of Fauconnier 1994), i.e., the conceptualization of some possible situation in the world which provides the background for the interpretation of the event in the main clause.

“that is, a space which is about a possible state of affairs in [the] world […] Within this space (further filled out by the speaker’s and hearer’s general knowledge about the situation referred to), the speaker predicts an added aspect of the content of this mental space” (Dancygier and Sweetser 2005: 16).

In (17) above, a mental space is built where either parents kill their children or children kill their parents. The matrix clause is “predictive” (Dancygier and Sweetser 2005) in giving instructions on how a perpetrator is to be punished within this mental space.

Conditional wo clauses display comparably high frequencies in all periods under scrutiny. They are attested in all genres, but are especially frequent in legal texts during the first period. This can be explained by the fact that legal texts give instructions on how to act if certain states of affairs occur, which is why they frequently contain conditional clauses (Lühr 2007: 213; Tophinke 2009; Wallmeier 2012, 40). However, by the end of the 18th century, the incidence of conditional wo dropped dramatically across genres (see Figure 7), which is related to the fact that wo was increasingly associated with the spatial function and was replaced by the connector wenn ‘if, when’ in the conditional domain.

Figure 7:

Conditional uses of wo clauses in the periods under scrutiny (period 1: 180, period 2: 174, period 3: 192 tokens).

It is noteworthy though that conditional da clauses exhibit an even stronger tendency to appear in legal texts (see Figure 8). Here, the gap between legal texts and other genres is striking. While conditional da clauses were highly frequent in 17th and 18th century legal texts, they rarely occurred in any other genre. The high incidence of conditional da in legal texts is particularly surprising in view of the high number of conditional wo in all genres. Since conditional wo outnumbers conditional da in most of the genres, we would expect the same to hold for legal texts. This high preponderance of conditional da clauses in only one specific genre suggests that they were associated with the register of the chancery language (see Section 5).

Figure 8:

Conditional uses of da clauses in the periods under scrutiny (period 1: 304, period 2: 314, period 3: 248 tokens, see Gillmann 2020: 81).

When analyzing the usage contexts of conditional wo clauses, it seems that not only were they more common in most genres, but they had also spread to more contexts than conditional da clauses.

For both wo and da conditionals, “generic or (specific) indefinite conditionals” (Dancygier and Sweetser 2005: 95, 151–155) are the most frequent conditionals in the data, e.g., (17) and (19). These clauses usually take present tense forms since they make predictions about a conditional relationship which generally holds. They often contain indefinite noun phrases which generalize over specific groups of individuals, e.g., (17) and (19).

Unlike da, wo occurs in different types of conditionals, which confirms that conditional wo had been fairly entrenched and conventionalized and had spread to new contexts. For example, we find instances of counterfactual conditionals, e.g., (20).

In example (20), the wo clause sets up a counterfactual mental space; a priest imagines what would have happened if he had not believed in the benefits of the Lord. The state of affairs in the main clause must be interpreted in the space in which it is embedded. The irrealis verb forms mark that the situation only holds within the counterfactual space.

In addition, there are several (14) instances of what Dancygier and Sweetser (2005) refer to as “metalinguistic conditionals”. As they put it, “[a] metalinguistic space is a complex space consisting of a pairing of a content space and a language or code space” (Dancygier and Sweetser 2005: 126). In these cases, clauses express a doubt about the mapping between a term and a referent.

The data suggest that a specific sub-construction marking metalinguistic spaces had evolved. These metalinguistic constructions consist of a fixed usage pattern realized by a negated conditional construction; the phrase wo nicht ‘where/if not’ seems to be highly conventionalized for the metalinguistic use. The prefabricated usage pattern is special in that it does not only comment on the wording, but rather proposes an alternative wording that may better describe what is meant, see (21).^[12]

In (21), the writer concludes that he has written enough about determining the perpendicular of a triangle. The word genug ‘enough’ is followed by the expression wo nicht zu viel ‘if no too much’ which does not only provide an alternative wording, but also an augmentation of its evaluative meaning. The specialized function of the expression wo nicht and its frequency suggest that this is a prefabricated pattern constituting a specific sub-construction, which has evolved from the conditional uses presented above.

Hence, the qualitative analysis gives additional evidence that the conditional usage of wo was strongly entrenched and conventionalized since wo was not limited to encoding generic conditionals, but had spread to metalinguistic and, in some cases, even to counterfactual spaces. Conditional da clauses, in contrast, were restricted to generic uses. This suggests that wo was one of the central conditional connectors while da was a peripheral one.

To sum up, the data confirms that da and wo clauses overlapped in their semantic functions (Hypothesis 1). The statistical model and the analysis, however, reveal that each connector tended to occur with a somewhat different set of semantic functions. There is ample evidence of spatial and conditional uses of wo clauses in the corpus. Regarding da clauses, we witness a tendency for temporal and causal reading. Over the course of the 18th century, the polysemy of both connectors gradually reduced. Spatial uses of wo diachronically increased at the expense of conditional ones. Simultaneously, spatial and conditional uses of da diminished and the causal function was gaining ground. Temporal da appeared with consistent frequency over all periods while temporal wo emerged only at the end of the 18th century and occurred infrequently.

In terms of language-external qualities of the connectors, the conditional uses are most revealing: While conditional wo clauses were more widespread in all other genres, conditional da clauses were only frequent in legal texts. This points to a socio-symbolic meaning of the connector, which was associated with chancery language (Hypothesis 2). The following section attempts to give a theoretical explanation of the changes observed in the data.