Syntactic alternations and socio-stylistic constraints: the case of asyndetic complementation in the history of Spanish

2.1 Linguistic predictors

Previous historical accounts of Spanish complementation patterns maintain that asyndetic complementation started spreading in the 15th century and became particularly frequent in the 16th and 17th century (Blas Arroyo and Porcar Miralles 2016; Pountain 2015). Mazzola et al. (2022) show that, until the end of the 18th century, the construction is still very popular and closely competing with the syndetic variant, which is in line with other studies that document the phenomenon as late as the beginning of the 19th century (Girón Alconchel 2005; Herrero Ruiz de Loizaga 2014). García Aguiar (2020), who studies a corpus of municipal documents, even registers an increase of the asyndetic variant between the second half of the 18th and at the beginning of the 19th century.

To date, several studies suggest that the alternation between syndetic and asyndetic clauses is regulated by a number of language-internal factors, such as the semantics of the main verb, the mood of the complement clause, the distance between the verbs, and the subordination of the main verb in a relative clause (cf. i.a. Blas Arroyo and Porcar Miralles 2018; Delbecque and Lamiroy 1999; Folgar Fariña 1997; García Aguiar 2020; Herrero Ruiz de Loizaga 2014; Pountain 2015; Real Academia Española and Asociación de Academias de la Lengua Española 2009).

In Mazzola et al. (2022), we used Szmrecsanyi’s (2013) diachronic probabilistic grammar to investigate the syntactic alternation on the basis of documents produced between the 15th and 18th century. Our study largely confirmed previous claims in the literature, but also highlighted new aspects of the alternation between these complement types. On the basis of an analysis of a dataset containing five verbs from different semantic classes (fitted as random effects), we identified a number of language-internal constraints relevant to the alternation between syndetic and asyndetic clauses. Asyndetic clauses are more likely predicted by complex sentences like (3), where: (i) the indirect object of the main clause (nos ‘to us’) is coreferential with the subject of the subordinate clause (tornemos ‘we return’); (ii) the complement subject is implicit (i.e., morphologically coded by the first person plural inflection of the verb, but not expressed by an explicit NP or pronoun); (iii) the complement verb is inflected in subjunctive mood. These morphosyntactic characteristics largely coincide with the syntactic schema of what Givón (2001: 44) called “non-implicative manipulation verbs”, typically instantiated by verbs of request and order. The non-implicative manipulation schema is characterized by a high semantic and syntactic integration between main and subordinate clause. We therefore concluded that asyndetic clauses signal an advanced syntactic integration of finite clauses.

Additionally, our paper evinces that usage of the asyndeton is strongly favored by the presence of an impersonal se-construction, the adjacency between the predicates and when the main clause is embedded in a relative clause. According to our (2022) study, an example such as (3), that contains all the mentioned features, is prototypical for asyndetic complementation in the classical period.

More importantly, our study showed that the relevance of some of these constraints increase diachronically. Asyndeton became more strongly associated with some specific schemas (coreferentiality of the indirect object, se-constructions and proximity of verbs), which indicates that, in spite of the fact that we did not find a demise of the construction until the end of the 18th century, the construction started to be used in more restricted and specific contextual environments, which might have eventually led to the less productive and more marked use of the asyndeton in Present-day Spanish.

2.2 Stylistic distribution of asyndetic complementation

It is often claimed that the asyndetic pattern developed in Spanish around the 15th century as a syntactic calque, triggered by the prestige of Classical Latin and mediated through the model of Italian, where the conjunction che was omitted not only from complement clauses, but also from relative clauses and comparative constructions (Meszler and Samu 2009; Wanner 1981). This hypothesis is corroborated by Pountain (2015: 79), who notices that, already from the start, the Spanish asyndeton appeared in more contexts than in Classical Latin, where it typically occurred with some specific volitional predicates (4), and some impersonal verbs (5) (cf. Del Rey Quesada 2017; Schulte 2007).

In line with the idea that the asyndetic construction in Spanish is a learned calque, it has been described as a stylistic marker of formal register, e.g., typical of juridical and administrative genres, as well as literary genres (Barraza Carbajal 2014; Benot 1991; Espinosa Elorza 2014; Girón Alconchel 2005). Yet, some other scholars argue that asyndetic complementation originated as a feature of the language of communicative immediacy, typical of colloquial and private epistolary communications (Blas Arroyo and Porcar Miralles 2016; Fernández Alcaide 2009; Herrero Ruiz de Loizaga 2005; Keniston 1937a). According to Pountain (2015), asyndetic complements with the verb rogar ‘to beg/pray’ appear more often in dialogues and religious works. Furthermore, it has also been considered an idiolectal stylistic feature of some authors (Folgar Fariña 1997; Keniston 1937b; Octavio de Toledo y Huerta 2011; Pountain 2015).

It is worth mentioning that most of the literature reviewed here shows methodological limitations on several levels. Most studies address the topic by adopting a merely descriptive approach (Barraza Carbajal 2014; Espinosa Elorza 2014; Pountain 2015), or do not provide any statistics at all (Girón Alconchel 2005; Herrero Ruiz de Loizaga 2005). Other analyses only look at one textual genre (e.g., Blas Arroyo and Porcar Miralles 2016; Fernández Alcaide 2009; García Aguiar 2020) or are based on only one main predicate (rogar ‘to pray’ in Pountain 2015; creer ‘to believe’ in Blas Arroyo and Porcar Miralles 2016).

In terms of methodology, Blas Arroyo and Porcar Miralles’s (2016, 2018 present a convincing multidimensional approach to the study of complementation, combining language-internal and -external factors. Blas Arroyo and Porcar Miralles (2016) is the only study that takes into account sociolinguistic variables, such as the solidarity between sender and recipient of the letter, and the social rank of the writers. Their results show that the rise of asyndetic complements can be considered a “change from below”, which spread from the lower ranks of the society towards the higher social groups. Moreover, they examine the type of audience of the letters, showing that asyndetic constructions with creer are mainly used in epistolary exchanges between equals. These results are in sharp contrast with the idea that asyndetic complements are a “learned” syntactic calque due to the influence of Latin and Italian. Probably, the conclusions of Blas Arroyo and Porcar Miralles (2016) are imputable to the composition of their dataset, which, as mentioned, includes only private letters and complement clauses dependent from the epistemic verb creer ‘to believe’.

In this paper, we maintain that in order to study this syntactic alternation and to clarify the role of sociolinguistic variables we need to analyze a broader dataset, which includes different classes of main verbs and a wider range of textual types.

3.1 The limitations of a traditional approach to textual variation

Many corpus-based studies of linguistic (especially grammatical) alternations have shown that the selection of alternative grammatical variants is determined by probabilistic language-internal constraints (e.g., Cuyckens et al. 2014; Mazzola et al. 2022; Rosemeyer 2014; Shank et al. 2016), but it has also been demonstrated that “language users do indeed adjust their choice-making processes to the situational context” (Szmrecsanyi 2019: 15–16), i.e., grammatical variants are often selected due to the influence of language-external factors, such as register, genre and style.

A classical textual approach to grammatical alternations such as the syndetic/asyndetic alternation is to extract a sample of the relevant variable contexts (finite complement clauses with or without complementizer) and to analyze their distribution according to textual categories. With this approach, one would measure the proportional selection of one or the other variant within each textual type. For our case study, such an approach would yield the percentages of syndetic and asyndetic complements, calculated from the total of the complements in each textual type, similarly to the analyses proposed, for instance, in Pountain (2015) and Yoon (2015).

One of the first problems that the analyst encounters in such a study, however, is how to define the different textual categories. Typically, variationist studies make use of the concepts of register, genre and style (RGS), but very often the distinction between the three concepts is blurred (cf. discussion in Rosemeyer in press; Szmrecsanyi 2019). In the end, regardless of terms and definitions, most corpus-based studies analyze RGS variation by relying on the categories offered by the existing corpora. For instance, consider a corpus frequently employed in the study of Spanish historical linguistics such as the CODEA+2015 (GITHE 2015). In this corpus the documents are classified according to three parameters:

1. Setting (ámbito) in which the document was produced: Diplomatic, Juridical, Municipal, Ecclesiastic and Private.

2. Textual typology, established by the CHARTA research group (Corpus Hispánico y Americano en la Red: Textos Antiguos), which consists of Deeds and Declarations, Trade Letters and Contracts, Private Letters, Certificates, Charters, Reports, Notes and Briefs,^[1] Wills and Inventories, and Laws.

3. Diplomatic typology: the traditional philological classification of historical documents (cartas plomadas, cartas de poder, notas de abandon, privilegios rodados, etc. …).

The variable of setting is considered by the developers of CODEA+2015 to be a proxy of formality (thus corresponding roughly to the concept of register), where the highest degree of formality is represented by the diplomatic setting and the lowest by the private setting, as illustrated in (6).

The variable of textual typology represents the concept of genre, i.e., “the conventional structures used to construct a complete text” (Biber and Conrad 2009: 2). The third variable, diplomatic typology, is a finer and more descriptive classification of the genre of the documents, compared to the textual typology, with many categories and subcategories, which is useful to have a more precise idea of the nature of the documents, but it lacks internal homogeneity and it is too fragmented to be used to measure proportions meaningfully.

The relative distribution of the syndetic/asyndetic alternation in terms of setting suggests that asyndetic complements are selected more often in informal settings (Figure 1).^[2] In line with the formality scale presented in (6), and the assumption that the syndetic-asyndetic alternation is governed by formality, we would expect the Ecclesiastical setting to be more similar to the Private setting. In contrast, asyndetic frequencies in this setting are found to approximate the Diplomatic setting. In other words, while the distribution of these complementation types is quite clear and easily interpretable for the two opposite poles of the formality proxy (i.e., Private and Diplomatic), intermediate levels show mixed results, which are less easily explained. In addition, the formality scale as such can be criticized. One might wonder why the documents produced in the Ecclesiastical setting are necessarily less or more formal than the ones produced, for instance, in the Municipal setting, given that the formality does not only depend on the setting of production (register), but also on the content and the purpose of the communication, as well as on the relationship between the writer and the addressee (genre and style).^[3]

Figure 1:

Percentages of que versus Ø according to the variable of setting. ³

The variable textual typology presents a similar problem (see Figure 2). In Notes and Private letters, asyndetic complements outnumber syndetic ones. In contrast, asyndetic complements only occur in 10% of the cases in Laws and Trade letters. The rest of the textual types exhibit an in-between pattern of selection, where usage of asyndetic complement clauses averages 25.6%. These textual types are mostly administrative documents, which are used to regulate relationships between private individuals (e.g., Wills), or between individuals and institutions.

Figure 2:

Percentages of que versus Ø according to the variable of textual typology. ⁴

On the basis of these descriptive results, we are in a position to claim that the use of asyndetic complements is favored by private communications and especially epistolary texts (in line with the conclusions Blas Arroyo and Porcar Miralles 2016). However, such an analysis would fail to account for all those textual categories and document settings which show a relevant, but less prominent, amount of asyndetic complements, such as administrative documents or texts associated with the Municipal and Juridical setting.

The problems regarding the interpretation of these results highlight the need for the analysis of explanatory variables that can account for the distribution of the two complementation patterns, as well as a more fine-grained methodological approach towards the description of the influence of these variables. Thus, we expect that the unclear or insufficient results found in the literature (and in this paragraph) are due to the use of macro-textual categories, which are too generic to capture the essential factors that determine variation. Crucially, in this paper we will present other relevant variables that are “diluted” in these macro-textual categories, and measure their effects on the syntactic alternation from a panchronic and a diachronic perspective. We will focus particularly on stylistic and social factors and their interactions, under the light of the Audience/Speaker Design framework, and the concept of Discourse Traditions.^[4]

3.2 Syndetic/asyndetic alternation and audience design

Having shown the limitations of traditional textual classifications, in this section we formulate the hypothesis that the type of audience of the text is a better stylistic variable for the syndetic/asyndetic alternation.

Our analysis starts from the assumption that asyndetic complements are a marked variant, and probably an innovation “from above”, a Latinate calque perceived as a prestigious variant (cf. Section 2.2). Octavio de Toledo y Huerta (2011) demonstrates that this is the case in Santa Teresa de Ávila’s Camino de perfección (of which the first autograph text was written around 1562).^[5] Santa Teresa edited a second version of the text in 1569, in order to address it to a broader, more prestigious and less familiar audience. In this attempt of intensive linguistic elaboration, she made some linguistic modifications to the manuscript. Among these, she eliminated many instances of the complementizer que. Octavio de Toledo y Huerta’s results lead us to expect asyndetic complements to typically appear in texts characterized by an elegant and prestigious language style. This expectation contrasts with Blas Arroyo and Porcar Miralles’s (2016) claim that the construction came about as a change “from below”, and that it was mainly used in communications among equals about private issues, as mentioned earlier (Section 2.2). In spite of their conflicting claims, both works suggest that the type of audience might be a relevant explanatory variable for the selection of a specific complement type.^[6] This is what we will further explore in this article.

Building on this idea, we assume that the selection of alternative grammatical options is a strategy that language users employ to fine-tune the style of their communication, in order to adapt the components of linguistic production to the conditions of communication (cf. Koch and Oesterreicher 2012). In this view, by virtue of their selection of different grammatical and lexical variants, speakers and writers are “potential agent[s] of choice” (Johnstone 2000: 417), who select linguistic variants and varieties in order to convey a particular type of identity or social meaning through a linguistic performance. This approach to speaker’s agentivity and stylistic variation is tightly bound to recent accounts of the concept of style in sociolinguistics, especially in the framework of Audience and Speaker Design.

Audience Design (Bell 1984, 2001), a concept inspired by Speech Accommodation Theory (Giles and St. Clair 1979), defines style as what an individual does in relation to other people (Bell 2001: 141). Style marks interpersonal and intergroup relations, in so far as style-shifting is mainly designed in response to an audience (i.e., it is “reactive”). Bell’s work highlights that audience-driven style-shift affects grammatical, lexical and pragmatic options, and even language and dialect shifts. Moreover, audience design effects can be observed through a quantitative style-shift: “a speaker’s style will be amenable to counting the relative frequency of certain variants” (Bell 2001: 166).

The model of Speaker Design builds on this idea and expands the theory of stylistic variation by conceiving style as a multidimensional phenomenon. In this view, stylistic variation and style-shifting include a wide range of co-occurring internal (purpose, key, frame, etc.) and external (audience, topic, setting, age, familiarity, etc.) parameters which customize people’s speech (cf. Coupland 2007; Hernández-Campoy 2019; Johnstone 2000; Schilling-Estes 2013). This framework does not exclude the validity of the Audience design approach, but insists on the importance of speakers’ identity construction by means of their linguistic choices. Thus, style-shifting is not only about reacting to an audience: it is a “proactive” rather than a merely reactive phenomenon, which considers speech as performance, as a way for the speaker to project different roles and identities according to the circumstances (cf. Hernández-Campoy 2019: 11).

In line with these frameworks, and similarly to what has been observed by Octavio de Toledo y Huerta (2011) about Santa Teresa’s work, we can formulate the hypothesis that the choice of a marked stylistic variant (such as the asyndetic complement clauses) is due to the type of audience addressed, and the type of role and social identity that writers want to represent. In particular, we will focus on the role played by the mutual relationship between writer and addressee in terms of social power (addressee status, explained in detail in Section 4.3),^[7] and the communicative goals of the writer.

3.3 Macro and micro discourse traditions

As mentioned earlier, the models of Audience and Speaker Design suggest that style is multidimensional, and that each of the individual factors of variation is highly intertwined with other factors (e.g., purpose, frame, topic …). With the purpose of capturing this complexity, this paper examines the stylistic distribution of the asyndetic complement clauses in 15th–18th century Spanish by considering the concept of Discourse Traditions (Kabatek 2005; Koch 1997; Schlieben-Lange 1983).

The central idea behind the concept of Discourse Traditions (henceforth DTs) is that language is not a monolithic homogenous object, but an aggregate of subsystems, i.e., the texts with their organizations and their traditions. As stated in Kabatek (2018: Ch. 7), it is virtually impossible to say anything about the “state of a language X” in a given period. Rather, what an analyst should do is to account for the wide spectrum of textual possibilities of that language in that period. In this view, to investigate language variation and language change, the discourse-textual dimension must be considered, as language change takes place and spreads through DTs (Kabatek 2013, 2018: Ch. 7).

Additionally, Kabatek (2018: 158) notes that even the textual dimension itself is not monolithic, as each text can contain multiple DTs internally at various levels (so-called “horizontal heterogeneity”). The fact that texts can allow for multiple global or internal classifications (Kabatek 2018: 156) explains why global a priori classifications, such as textual types and genre, are very often not sufficiently powerful to explain language variation (as demonstrated in Section 3.1). Rosemeyer (in press) approaches to the idea of the multi-layered nature of DTs, arguing that most of the current studies only deal with “macro” DTs (which roughly correspond to the notion of RGS), while largely neglecting the importance of “micro” DTs, because of the difficult operationalization of multi-layered textual categorization. Here, we would like to account for DTs by taking into account both the macro and the micro level.

The selection of alternative grammatical options by the speaker/writer is obviously a matter of choice-making. In fact, according to the DTs framework: “[…] speakers will make their choices of grammatical variants according to the textual tradition they wish to represent” (Kabatek et al. 2010: 7). This view of speaker’s agentivity is, hence, compatible with the above-mentioned framework of Audience and Speaker Design.

Moreover, the complementation variants analyzed here are particularly interesting to study variation across DTs, as clause linkage techniques are known to be indicators of the position of a text along a continuum between the poles of Communicative Immediacy and Distance^[8] (Kabatek et al. 2010; Raible 2001). Thus, we investigate the hypothesis of whether the selection of syndetic versus asyndetic complementation is influenced by the characteristics of different DTs.

What is more, we will distinguish between macro and micro DTs, in order to investigate whether stylistic effects are active at the global textual level or rather at the sentential level. One possibility is that the use of an asyndetic clause can be explained by stylistic constraints at the textual level, e.g., it is frequently found in any type of sentence within documents that were addressed to a superior addressee. Another possibility is that the presence of an asyndetic clause is not strictly predicted by the audience of the document as a whole, but rather by the audience of a specific sentence, for instance if it is only found in sentences where the issuer is directly addressing the recipient. The former scenario would represent an effect at the level of macro DTs, while the latter would mean that stylistic constraints that regulate the syndetic/asyndetic alternation are active at the level of micro DTs. With this aim in mind, we will introduce the variables of Main Verb (see Section 4.5 ) and its morphological Person (see Section 4.4 ), to analyze DTs also at the sentential level.

4.1 Data

The data used for this study are drawn from a dataset originally extracted in a pilot study by Cornillie and Rosemeyer (2018), from the corpus CODEA+2015 (GITHE 2015). This corpus contains archival documents, from a wide range of geographical areas, and it is a particularly trustful source from a philological point of view, as it is a “primary” corpus, which means that the same research group (GITHE) is responsible for the edition of the documents and for building the corpus.

The dataset encompasses a time range stretching from the 15th to the end of the 18th century and includes n = 1,090 documents produced in Spain. The texts belong to the document types and settings indicated in Section 3.1. We extracted 46 Complement-taking Predicates (CTPs) identified as amenable to asyndetic complementation in Pountain (2015) by means of regular expression. The dataset was manually cleaned to reduce the sample to the variable context: CTPs + (que/Ø) + finite complement clause. The resulting sample was manually coded at the sentential level so as to detect semantic and morphosyntactic characteristics of the complex clause (cf. Mazzola et al. 2022).

For the present study, we included all observations with a main verb occurring at least 40 times in the sample, followed by an object complement clause. These elimination procedures yielded a final sample of n = 3,706 complement clauses, n = 2,715 syndetic and n = 991 asyndetic. As shown in Table 1, the final dataset used for this study contains n = 967 documents belonging to nine textual types (seen in Figure 2), which adds up to a total of n = 723,779 words. Each of the documents was coded in order to annotate the relationship between writer and speakers (see Section 4.3).

4.2 Mixed-effect logistic regression

Given that we were interested in the alternation between two grammatical variants, we employed mixed-effect logistic regression,^[9] which models the relationship between the binary dependent variable asyndetic (que vs. Ø) and multiple explanatory variables (or predictors). In particular, these models predict the probability for an asyndetic complement (Ø) to be chosen instead of a syndetic complement (que), depending on the value of the predictor variables, addressee status, person and main verb (see Sections 4.3–4.5). Additionally, mixed-effect models allow fitting random effects, which are used to model varying intercepts for a variable, while evaluating the fixed effects of the predictor variables. The random effect in our models is textual typology, ^[10] and inclusion of this random effect thus enabled us to control for variation due to this parameter.

We fitted two mix-effect logistic regression models, a panchronic and a diachronic one. As described in the next two subsections, the panchronic model considers all data together, without taking into account real-time, but calculating the mutual interactions among predictor variables; the diachronic model calculates the interaction of each of the predictors with time.

4.3 Addressee status

The audience of each of the texts contained in our dataset was annotated by applying the concept of power semantics, introduced in the seminal work by Brown and Gilman (1960). Power semantics is based on the idea that in static societies, in which social power is distributed by birth right, one person can be said to have power over another, in as much as s/he can control the behavior of the other.

The textual sources we use in our study are particularly suitable to be studied from the point of view of social power, because we often can infer the mutual relationship between the issuer and the addressee, thanks to the use of names and titles, and the information provided by the regesto (a short summary of the document) integrated in the metadata of the corpus CODEA+2015 (GITHE 2015).^[12] Alongside the information of the regesto, the content of the document itself was perused for indicators of the social hierarchy between writer and addressee (profession, role in the communicative exchange, direct mentions to the mutual relationship, etc.).

If these criteria of inspection did not lead to a definitive decision, information provided by the types of forms of address used in the text was considered as a last resource. In this case, the forms of address and the titles (e.g., tú ‘you’, vos ‘you’, Vuestra Merced ‘Your Grace’, Vuestra Señoría ‘Your Lordship’, Vuestra excelencia ‘Your Excelence’, Ilustríssimo Señor ‘Illustrious Lord’, etc.) could give us a hint about the relation of the issuer with the addressee. However, we are aware that forms of address should be used with great caution, because their paradigm in Golden Age Spanish was unstable: as shown by Moreno (2002), the address system of Spanish was changing in the period under study. Moreover, the lack of evidence for reciprocal and non-reciprocal usages makes it hard to use address terms and pronouns as a reliable criterion to ascertain social power relationships.

Ultimately, the documents in the sample where coded according to the addressee status in comparison to the issuer of the document, by indicating whether or not the addressee had the power to control his/her behavior. The coding resulted in the following categories:

1. Superior: the addressee has a higher social power than the issuer (citizen to an institution, nun to aristocrat, children to parents, subjects to monarch, etc.)

2. Equal: addressee and issuer have the same status (siblings, cousins, nun to nun, friends, mayor to another mayor, etc.)

3. Inferior: the addressee has a lower social power than the issuer (king to citizens, bishop to monks, judge to defendant, husband to wife, parents to children, etc.)

4. No addressee: in some documents, especially certificates, charters and reports, we found no explicit reference to an addressee. Of course, also these documents imply an intended audience, the “referees” according to Bell’s (2001) terminology, which still influence the style even though they are not directly addressed.

According to the current literature (as discussed in Sections 2.1 and 3.2), we could expect the asyndetic variant to be predicted by superior addressees, following the results of Octavio de Toledo y Huerta (2011).^[13] On the other hand, in line with Blas Arroyo and Porcar Miralles (2016), we might find that asyndetic complements are preferred when writer and addressee are equal.

4.4 Person

As anticipated in Section 3.3, the annotation at the textual level proposed for the addressee status can only be used to investigate global textual effects related to the type of audience. But analyzing audience design only at the textual level might produce misleading results, which is why we propose to focus on the interaction between the person of the main verb and addressee status.

In fact, within a document, a speech act can be directly addressed to the audience of the document, as in (10). However, it can also refer to reported speech, as in (11), where the indirect object and the addressee of the document do not coincide.

Example (10) belongs to a document in which the governor of a town asks the king to cancel an order that prevented grazing in certain territories. This document was coded as being addressed to a superior. In this case, the actual addressee of the sentence and the addressee of the document coincide. In contrast, example (11) was found in a document from a Mother Superior to the Royal Council, therefore also coded as being addressed to a superior. Yet, this sentence is a report of an order that she received from the addressee. Hence, on the sentential level, the Mother Superior is the addressee of the order in the sentence, and therefore ‘inferior’, but, on a textual level, this occurrence would be marked as ‘superior’, as the document in general is a communication addressed to the Royal Council.

Examples (10) and (11) demonstrate the need to distinguish between the textual and the sentential dimension (i.e., between macro and micro DTs). In order to capture this distinction, we created the variable Person by annotating the morphological person of the main verb.^[14]

First, in terms of general effects, according to what was previously observed in the literature, absence of a complementizer should be correlated with first-person morphology. In English, this correlation has been observed multiple times: for instance, according to Torres Cacoullos and Walker (2009), the combination of first-person singular morphology with a verb of cognition produces a near-categorical absence of that, to the point that these combinations can be considered as discourse formulas (cf. also Tagliamonte and Smith 2005). As for Spanish, the claim for a more frequent association of asyndetic complements with first-person morphology is also found in Blas Arroyo and Porcar Miralles’s (2018) study on asyndetic complements with the verb creer ‘to believe’.

More importantly, we are interested in distinguishing between main verbs with first person and verbs with second/third person morphology, in order to set apart sentences, in which the author of the text is the performer of the speech act (first-person morphology), from all the other sentences (second/third-person morphology) which lack this author-relevant performative dimension.^[15]

Given that our hypothesis is that asyndeton should be favored by superior addresses (as mentioned in Section 3.2), we are interested in whether such a stylistic constraint would be active at the textual or at the sentential level. If asyndetic complements are favored by the textual dimension (macro DTs), they are expected to be used in any type of complex sentence within documents addressed to superior addressees, i.e., as a generic marker of style. If they are favored by the sentential dimension (micro DTs), asyndeton should be more likely when the issuer is directly addressing someone superior, i.e., when the main verb has a first-person morphology. In the latter case then, the effect is due to the pragmatic nature of the speech act, rather than to the general stylistic features of the document as a whole.

4.5 Main verb

This variable is a bridge between the language-internal and the language-external dimension, which is useful to follow-up on the results of the previous study (Mazzola et al. 2022), by making a pragmatic distinction within the category of manipulation verbs (order vs. request). The analysis of the interactions of the main verb types with the other variables (addressee status and person), further allows for a more in-depth understanding of the selection of the syntactic variant depending on the type of speech-act.

Table 2 describes the sample of main verbs included in this study. We categorized these main verbs into five semantic classes: communication verbs, orders, requests, volition and cognition/perception. The distinction between pure verbs of request and verbs of order was motivated by the fact that order verbs entail a high degree of imposition on the hearer, whereas the imposition of verbs of request is milder, or at least mitigated (cf. Iglesias Recuero 2016). Therefore, we expect this fundamental pragmatic meaning to affect the selection of a stylistically marked syntactic variant.

Note that the verb decir can have two meanings: we distinguished its communicative uses (‘to say’) from its uses with the subjunctive in the subclause, as in the latter case it conveys a command reading (‘to tell someone to do something’), as illustrated in example (12) in which both instances of the verb decir express commands.

5.1 The social power of the addressee

The results concerning the variable addressee status ( Figure 3 ) illustrate that the likelihood of use of an asyndetic clause is higher in documents addressed to a superior addressee, i.e., someone with more social power than the issuer of the document. To a smaller extent, documents with no explicit addressee and documents addressed to equals also favor absence of the complementizer. In contrast, asyndetic complementation is least likely to be used in documents addressed to recipients with less social power than the writer (inferior).

Figure 3:

Main effect of addressee status – Panchronic Model.

This result points to the relevance of the social power distinction established in Section 4.3. The effect of Addressee Status follows a cline of ascending/descending social power: the probability of asyndetic complementation increases as the social power of the addressee increases.

Figure 4 visualizes the effect of the variable addressee status from a diachronic perspective. First of all, we can observe a general sharp increase of the likelihood of use of the asyndetic variant between the 16th and 17th century, followed by a mild decrease in the course of the 18th century. In the first stage, between 1400 and 1500, the likelihood of asyndetic complements is very low, and seems to be only weakly affected by the social power cline (even though asyndetic complements are mostly associated with superior addressees already in this initial phase). The probabilistic effect of social status increases after the second half of the 16th century, where the different lines referring the addressee status diverge. In this time period, the relative impact of the different categories of social status on selection of complementation patterns follow the trend found in the panchronic model (see Figure 3). However, this picture changes after the beginning of the 17th century. In particular, we document a marked increase of the likelihood of use of asyndetic complements with addressees that have the same social status as the writer. According to our model, likelihood of use of asyndetic complementation is highest in these usage contexts after 1750, even surpassing texts directed to an addressee of superior social status. This is due to the fact that texts addressed to equals partake to much lesser degree in the general decrease in the likelihood of asyndetic complementation after 1750.

Figure 4:

Interaction between time and addressee status – Diachronic Model.

5.2 The main verb semantics

Figure 5 illustrates the interaction effect between main verb and addressee status. Focusing on the main verb first, we observe that selection of asyndetic complements is particularly likely with verbs of request (pedir ‘to ask’, rogar and suplicar ‘to beg’), in line with the results of Mazzola et al.’s (2022) analysis. However, in contrast with the previous study, we can observe a large difference between verbs of request and order, which have a very similar syntactic configuration, both being manipulation verbs (Givón 2001: 41). Thus, asyndetic complementation is more likely with verbs of request than with verbs of order.

Figure 5:

Interaction between Main Verb and Addressee Status – Panchronic Model.

Moving to the interaction between main verb and addressee status, Figure 5 illustrates that the use of asyndetic complements is strongly associated with verbs of request within documents addressed to a superior person/institution. Orders show a similar effect, but the chances of an asyndetic complement after a verb of order are also quite high in documents without an identified addressee. Chances of asyndetic complements are a bit higher with verbs of volition when the document is addressed to an equal, with cognition/perception verbs when the documents has no identified addressee, while for communication verbs the chances of asyndeton are low regardless of the type of addressee.

With respect to the diachronic changes of the main verb effect, Figure 6 shows that asyndetic complements are mainly predicted by verbs of request already in the second half of the 15th century, and that the size of this effect increases dramatically over time. In texts dated around 1700, our model predicts a usage frequency of 80% for asyndetic complements in such contexts. Until about 1550, next to request, the only other verb class found substantially with asyndetic complements is volition verbs. However, after 1500, we document a strong increase in the likelihood of use of asyndetic complements with verbs of order grows, followed by cognition, perception and communication verbs in the 17th century. After the beginning of the 18th century, we find a general decrease in the likelihood of use of asyndetic complements with all verb classes. Interestingly, communication verbs seem to be less affected by this trend than the other verb classes.

Figure 6:

Interaction between time and main verb – Diachronic Model.

5.3 The role of morphological person

Concerning the variable person and in contrast with our expectations, Figure 7 reports that the predicted probabilities of asyndetic complements are much higher with second- or third-person morphology than with first-person morphology in the main verb. This aspect will be further discussed in Section 6.

Figure 7:

Interaction between person and addressee status – Panchronic Model.

Crucially for our purpose, Figure 7 demonstrates that the effect of social power is particularly strong when the main verb is inflected for first person. In spite of the fact that even with second- and third-person main verbs, asyndetic complements are more probable with superior addressees, the difference between inferior, equal and superior is much more evident in cases in which the main verb is inflected in first person.^[18]

As explained in Section 4.4, we use the interaction of the addressee status and the variable of person of the main verb in order to set apart textual from sentential effects (macro and micro DTs). The results indicate that the social-power cline observed in Figure 3 is more relevant when the main verb is inflected for first person, i.e., when the issuer is directly involved in the speech act and feels the need to conform to social power conditionings. Consequently, we can conclude that social power effects are mainly relevant at the sentential level.^[19]

Finally, the interaction between person and main verb visualized in Figure 8 suggests that verbs of request are the best predictors of asyndetic complements regardless of their morphology. Most other verb types display a tendency for asyndetic complements to occur with second/third-person morphology, with the notable exception of cognition/perception verbs, which show a slight preference of asyndetic complementation when the verb is inflected in first-person morphology. The latter point suggests that the findings of Blas Arroyo and Porcar Miralles (2018) concerning the main verb morphology are clearly linked to the semantics of verb analyzed in that study (creer ‘to believe’), and to the discourse traditions and environments of private letters, which favor the expression of personal views and opinions.

Figure 8:

Interactions between person and main verb – Panchronic Model.