Towards a constructional approach to discourse-level phenomena: The case of the Spanish interpersonal epistemic stance construction

2.1 Why a constructional approach?

Before proceeding to the data analysis, the added value of adopting a constructional model for the diachronic study of IES expressions needs to be addressed. ^[4] First and foremost, the constructional approach has been put forward as a highly adequate model to account for language change. However, it is important to bear in mind that a constructional approach is not intended as a replacement of more widespread accounts of pragmatic markers in terms of grammaticalization. Indeed it has been argued that the two approaches, grammaticalization and constructional change (or constructionalization), are not mutually exclusive but in fact complementary models, which, besides an important overlap in their theoretical and methodological assumptions, partially foreground different questions (Hilpert 2013; Traugott and Trousdale 2013; Trousdale 2014). ^[5] Whereas grammaticalization has focused mainly on the development of grammatical categories, also often taking into account contextual factors (e.g. Himmelmann 2004) and the regularity of changes at different levels of language (e.g. Company 2006), the constructional approach focuses on how the internal and external configuration of sets of expressions changes. With regard to the two alternative traditions within which grammaticalization is defined – either as a process of reduction and increased dependency or as a phenomenon of contextual expansion – the constructional approach argues in favor of a combination of both views: a constructional change can involve expansion in the range of uses of a construction which, at the same time, can become formally more fixed and reduced and appear as a chunk (Traugott and Trousdale 2013: 147).

In addition, a particularly appealing implication of the constructional view, which is not absent but less prominent in grammaticalization accounts of pragmatic markers, is that the perception of relationships between patterns within a paradigm is part of a competent speaker’s knowledge. As a consequence, the usage of one particular item cannot be seen as independent of the functioning of others related to it. This entails that linguistic change does not apply to an individual linguistic item alone, but always involves its network family members. A construction can thus be assumed to have its own history which is, however, often constrained by, or has an impact on, the history of related patterns or schemas in the network. This interconnectedness can take very different forms and has consequences for both the internal (a) and external (b) configuration of a network (Hilpert 2013; Traugott and Trousdale 2013):

1. Within a particular network, new patterns can arise, and older ones can disappear; a pattern can become more frequent and productive and can invade the domain of another, or, conversely, it can lose its vitality, become obsolescent and leave a functional gap.

2. As a consequence, the configuration of the network itself can remain stable, or can come to be reorganized, grow, or shrink.

The view taken in this study squares with the recent and growing interest in applying constructional models and insights to the study of discourse-level phenomena. Until the beginning of this century, the focus of Construction Grammar (CxG) was mainly on argument structure constructions, but more recently, lower levels (i.e. morphology and phonology) as well as higher levels have been more explicitly integrated into the model. This development is completely in line with the all-inclusive principle of CxG, which declines to distinguish between core, peripheral, or even ungrammatical patterns (Brinton 2008).

Also non-propositional material typically found in spoken discourse, such as particles (Fried and Östman 2005), pragmatic markers (Traugott forthcoming), parentheticals (Hilpert 2013), and comment clauses (Kaltenböck 2013; Van Bogaert 2009; Van Bogaert 2010), has been argued to be part of grammar and of a speaker’s constructional knowledge. The central idea is that different expressions with analogous and overlapping functions and formal characteristics are instances of an overarching construction. Thus one of the main advantages of the CxG approach is that it reveals generalizations across particular items. Fischer (2015) shows, for instance, that the formal behavior of the English particle oh (defined by the structural contexts in which it occurs) interacts with its meaning (consisting of a series of different communicative tasks), and argues that these form-meaning pairings count as schematic structures and can thus apply to other particles as well. In a recent paper, Traugott (forthcoming: 9–10) argues in favor of a general pragmatic-marker macro-schema, defined by a very general form (“position typically precedes or follows the clausal construction it accompanies, or the finite verb of a clausal construction”) and meaning (“pragmatic, non-contentful meaning, and scope over the clausal construction it accompanies”) components. At the schema level, metatextual, epistemic, and interpersonal pragmatic-marker constructions are distinguished, which are then further concretized in different discourse-marker subschemas.

Whereas the idea of a constructional approach to pragmatic markers and associated phenomena is in itself perhaps not so novel, to date not much work has been done on their history within the constructional change framework. This, of course, leads us directly to the main research question of this paper and the aim of accounting for the apparently contradictory tendencies observed in the literature (see Section 1), namely that: (a) sabes increasingly acts as a (grammaticalized) pragmatic marker but (b) at the same time still alternates with a series of complex alternative expressions.

2.2 Parameters of analysis

It has been observed that constructional changes involve different phenomena. Traugott and Trousdale (2013) emphasize the expansion of the productivity and schematicity of constructions and the reduction of their compositionality. Expansion of productivity mainly refers to an increase in the range of collocates that co-occur with a grammaticalizing construction. This collocational expansion is also known as host-class expansion (Himmelmann 2004), or the increase of the type frequency of a construction. Indeed, a higher degree of constructional vitality coincides with a higher probability of the constructional schema generating new instances. This can be seen as a correlate of the increasing degree of schematicity of a construction. When constructions are used more frequently, speakers generalize over the individual types and develop more abstract schemas. In their turn, these abstract schemas attract new types, which again contributes to the productivity at the schema level (Barðdal 2008). In addition, a grammaticalizing construction suffers a gradual decrease in compositionality; that is, the meaning of the individual parts no longer coincides with that of the construction as a whole. Additionally this can lead to formal (morphological and phonological) reduction.

Hilpert (2013) adds that, in order to understand how constructions have merged and developed, it is crucial to observe possible changes in the distribution of different structural variables within a more general schema. These changes can point towards changes in the configuration of a network (see Section 2.1 above), such as the creation or decline of particular structures, or the internal reorganization of a network.

These diagnostics or correlates of constructional change can guide us through the diachronic analysis of the IESCx. However, to my knowledge, no previous study has attempted to empirically operationalize these principles for the analysis of pragmatic markers and related constructions. As a result, a number of methodological (and terminological) questions are raised during the analysis. Among these are: How can we account for the productivity of the IESCx? What is its type frequency? How can its degree of compositionality and formal reduction be accounted for? Indeed, although an increasing number of scholars have argued in favor of a constructional approach to discourse-related phenomena, its implementation still needs to be worked out in more detail. This study can be considered as a first step in that direction.

3.1 Data collection

This study starts from the assumption that language change is not restricted to drastic long-term shifts affecting the language system as a whole; it can also consist of a more subtle variation in language use that comes about in the course of just a few decades. The study of these current changes defined as “changes in the language that have taken place over relatively short spans of time, over decades rather than centuries” focuses on changes in the recent past, and often also in contemporary language (Aarts et al. 2013:1). As is the case for any data-driven investigation, good sampling of data is crucial – but, alas, not easy.

First, the monitoring of the use and development of the IESCx can only be achieved on the basis of a spoken data corpus, not only because spoken language is known to be “at the forefront of linguistic innovation” (Kaltenböck 2013: 287), but mainly because the construction under scrutiny tends to be exclusively used in informal conversational language. But, for obvious reasons, it is difficult to have access to a non-contemporary native speaker’s linguistic intuition (Jucker 1995). Second, short-term patterns of change can only been detected when differences in the corpus cannot be attributed to factors of influence other than the time variable.

Both restrictions constitute a methodological challenge for the present purpose, given that spoken corpora for Spanish are not as widespread as for other languages such as English. ^[6] However, I have striven for a maximally large corpus including reliably transcribed texts, and, most importantly, with data sampled from different decades. I have resorted to a micro-diachronic analysis of the development of the IESCx over four decades, starting with the period for which the first spoken corpus for peninsular Spanish is available, namely the 1970s.

The database was compiled from eight existing spoken-language corpora. It includes interpersonal epistemic stance (IES) expressions which all appear in a dialogic context of spontaneous conversations. These are face-to-face conversations (e.g. COLAm, CORLEC, C-Oral-Rom), telephone conversations (e.g. CREA oral), interviews in television and radio shows (CREA oral), and semi-directed interviews (e.g. Habla culta de Madrid, COSER). The tokens were selected through a lexical query for the string sabe(s), and were then manually sorted in order to discard full lexical uses of the verb. These include all instances of sabes followed by a Direct Object with full lexical meaning (1a–1b), in which the verb directly refers to the knowledge of the interlocutor, without negotiating any point of view towards the contents of the message. The DO can then be an NP (as in 1a) or a subordinate clause (as in 1b).

These uses contrast sharply with the intersubjective stance meaning of the IES expressions included in the database. In (2a), sabes is followed by the semantically vague NP una cosa, and the expression as a whole has the main function of calling upon the interlocutor’s attention, just like the pragmatic marker sabes further in the context does. The same applies to (2b) in which sabes followed by the subordinate DO que pasa fulfils a mainly phatic function (see also Sections 1 and 3.2 for a more detailed definition of the IESCx).

The result is a large database of 3279 tokens, representing speech of both male and female speakers, of mostly intermediate sociocultural level (although higher and lower levels are also included), and of different generations (e.g. adolescents in COLAm, elderly in COSER) living in Spain. ^[7]

Table 1 gives an overview of the different corpora that have been consulted for each decade and the number of IESCxs attested in each. ^[8] In a last phase, these hits have been manually annotated for a series of functional and formal properties which are detailed and explained throughout this article.

One of the main characteristics of a construction is its non-compositional nature (Traugott and Trousdale 2013). A construction appears as a chunk, following a mismatch between the meaning of the individual elements that make up the construction, and the meaning of the construction as a whole. This characterization applies in varying degrees to the following expressions found in the database, ordered from most to least frequent: sabes (n=2385) ^[9], sabes lo que V (251), ya sabes (132) / sabes ya (1), sabes que+clause (69), tú sabes lo que V (55) / sabes tú lo que V (16), tú sabes+cl (52) / sabes tú+cl (3), tú sabes (46) / sabes tú (5), ya sabes+cl (42), no sabes (39), como tú sabes (18) / como sabes tú (2), no sabes lo que V (19), sabes qué (18), ya sabes lo que V (15), como sabes (13), no lo sabes (11), ya lo sabes (10), lo sabes (10), tú no sabes (10), tú no sabes lo que V (6), tú lo sabes (6) / lo sabes tú (2), ya sabes tú (4) / tú ya sabes (2) / ya tú sabes (1), tú ya sabes+cl (4) / ya sabes tú+cl (4), tú no sabes+cl (4), como sabes+cl (3), sabes (una) cosa (3), tú sabes+NP (2), ya sabes+cl (2), como sabes lo que V (1), te lo sabes (1), como ya sabes+cl (1), tú ya sabes lo que V (1) / ya sabes tú lo que V (1) / ya tú sabes lo que V (1), no lo sabes tú (1), tú sabes lo que V (1), no sabes+cl (1), no sabes tú (1), no sabes tú lo que V (1), no sabes tú+cl (1), sabes cómo V (1), tú ya sabes algo (1).

These instances can be seen as (part of) the “cloud of exemplars” (Bybee 2010: 14) that represent the IESCx in the Spanish speaker’s mind. ^[10] This cloud, in fact, represents constructions which can be placed on a continuum of schematicity, going from the most abstract macro-construction to its most concrete instantiations or constructs.

3.2 Functional definition of the interpersonal epistemic stance construction

As shown by the above-mentioned list, speakers have at their disposal many different expressions in order to convey personal feelings, value judgements, assessments or attitudes towards things, beings, processes or facts (Biber et al. 1999). With regard to 2nd person expressions with nucleus sabes, the complex notion of stance can be reduced (following the terminology of Brinton 2008) to three main functions, namely an interpersonal, a modal, and a textual one.

In order to avoid an intuition-based classification, it is crucial to come to an operational definition of these pragmatic functions, open to objectification (as has been argued for especially in recent studies on discourse-related phenomena such as Torres Cacoullos and Schwenter 2007; Hernández and Baldazo 2013 or Tanghe 2016a, 2016b, but also in earlier approaches such as Schwenter 1996). This goes in line with the general tendency in contemporary linguistic research to strive for maximally empirical and quantifiable methods in the research on semantic and pragmatic meaning (e.g. Enghels and Jansegers 2013; Glynn and Fischer 2010). However, I agree with Geeraerts (2010), who states that the best way to proceed is to combine different methodologies (which then engage in the so-called empirical cycle), taking into account that not all aspects of the meaning have a direct reflection on the morphosyntactic or distributional behavior of linguistic items. Moreover, although they frequently go hand in hand, objectifiable criteria do not necessarily equal quantifiable criteria, as for instance a particular pragmatic function can be observed through a particular co-occurrence pattern (as is shown below), without it being possible to integrate that pattern as a (quantifiable) variable in a more general model.

In concrete, in order to define the macro functions of the IES expressions, each occurrence has been subjected to a detailed analysis of its contextual behavior. The classification of the functions has been decided on by taking into account three complementary parameters:

1. the pragmatics and meaning of the message, which can be neutral (mainly with the interpersonal function), or can be marked by an explicit positive or negative prosody (mainly with the modal function, as illustrated below). A series of paraphrase tests help to define this semantic prosody in a more precise way;

2. recurrent co-occurrence patterns observed in the data, which range from explicit references that the speaker makes to him or herself (modal function), to particular reactions on behalf of the interlocutor (interpersonal function). While the observation of these patterns certainly provides an independent basis for the functional classification, the patterns themselves can hardly be quantified. That is, the absence of a particular collocational pattern does not automatically preclude a particular functional classification. ^[11]

3. the position of the stance expression, which can be related to its pragmatic function on a quantifiable basis (following previous accounts of the form-function relationship of pragmatic markers, such as Beeching and Detges 2014; Degand and Fagard 2011; House 2009; Tanghe 2016a, 2016b amongst many others). Given that the numerous studies recently conducted on the relationship between the position of pragmatic markers and their functions have analyzed diverse kinds or markers in various languages, they have sometimes led to different conclusions. However, all sources agree on a form-function correlation, although not on a one-to-one basis. The analysis of the position of the IES expression relies on an economized version of the Val.Es.Co model of discourse segmentation (Briz Gómez and Bordería 2010), and distinguishes between the position of the expression at the level of the interventions or turns of the speakers, and the turn-internal level. With regard to both discourse units, a pragmatic expression can be placed at the beginning, in the middle or at the end; or, it can appear in an independent position.

In the dataset, the three macro functions performed by IES expressions are rather evenly distributed: interpersonal function (35.3 % or 1157/3279), modal function (38 % or 1246/3279), and textual function (26.7 % or 876/3279). Let us now turn to the more detailed definition of these three domains.

To begin with, an IES expression can convey an interpersonal function, mainly centered on the interaction the speaker wants to establish or maintain with the interlocutor. This function mainly coincides with the expression being placed at the extreme left (3b–c) or right (3d) periphery of a speaker’s turn (in 47.5 % of all cases, 550/1157), or at least at the left (4a–b) or right (4a) periphery of a particular utterance within a turn (another 42.5 %, 492/1157). This macro function has two subtypes, namely the epistemic and interactional functions. Firstly, as an epistemic marker, the IES expression is an interactional marker mainly directed toward the interlocutor, with whom the speaker wants to share responsibility for the correctness of the information being communicated. The paraphrases would then be the affirmative ‘I know you share this knowledge with me’ (3a, 3c–d), or interrogative ‘Do you share this knowledge with me’ (3b). This function coincides with the presence of contextual indices of interaction such as direct references to the addressee (e.g. ¿Te interesa eso? in 3b), or an affirmative or negative answer on behalf of the addressee (e.g. Gracias, pero… in (3a), ah, sí in (3b), sí in (3d)). In these contexts the construction conveys a primary topical function (3a, 3c), but it can also be used to introduce new information, presented as possibly shared between the interlocutors (e.g. the reference to the price of the object in Portugal in 3b).

Secondly, besides the epistemic meaning, the interpersonal macro function also groups numerous cases in which the IES expression is used to control the interaction between speakers. This function also comes in at least two varieties. When adopting a phatic function, the IES expression verifies whether the interlocutor is handling the abundance of information (‘are you still following me?’). This is frequently found in contexts where the speaker sets up a rather large monologue on a particular topic (4a). When adopting a conative function, the IES expression asks for collaboration or an intervention on behalf of the interlocutor (e.g. the addressee has to pass something on to the speaker in 4b). The interpersonal function is supported by the affirmative answer in (4b).

Next, IES expressions have a more subjective and modal macro function when they are used by the speaker to reinforce the expression of his/her own emotions or opinions, which the interlocutor can agree with or not. Hence the interpersonal function is not completely absent, but backgrounded. In the corpus this function is predominantly expressed by IES expressions in turn-medial position (70 % or 872/1246) (5a–5c). Specifically, they allow the speaker to reveal a personal interpretation of the information by emphasizing some part of an utterance (namely the positive evaluation of the way of life of a third person in (5a), or good way of life of the speaker in (5c)), or by expressing (dis)content (5b). In the corpus, this function coincides with the presence of exclamation marks, recurrent references to the 1st person singular (as in 5b–5c), semantically loaded positive or negative words (as for instance consuelo de tontos in 5b), or the use of intensifiers (e.g. lo bien que in 5a).

Finally, the IES expressions can also operate at the textual level when they are used as discourse connectors coordinating the individual parts of a conversation in order to maintain coherence. Just like IESs with a predominantly modal function, the expressions adopting the discourse organizing function are mostly found in the corpus in turn-medial position (83.2 % or 729/876). But what is different, is that the IESs then typically interrupt (6a) or even distort the syntax of the utterance (in 6b–6c the expressions sabes and como sabes are inserted right in the middle of an utterance). When adopting this function, they allow the speaker to introduce a new discourse unit that specifies or explains a previous one (specifying the positioning of the dog in 6a), or to reorient previously given information (6b). They can also be used as gap-fillers (to be paraphrased by other fillers such as uhm) which provide the speaker with some extra time to look for words or to plan the upcoming message while avoiding silence (6c). In the dataset, the textual function is revealed by the following diagnostics: presence of repeated words or parts of the utterance (in 6b cuando yo cuando yo), reformulation or specification of parts of an utterance (6a), the use of vague words (y etcétera, etcétera in 6c), or repeated pauses (6c).

In sum, this overview clearly demonstrates that different stance expressions share functional behavior. This motivates their common classification under the head of a more general abstract schema or construction, namely the IESCx, whose structural characteristics are further specified in Sections 4 and 5 below.

4.1 Token frequency of IESCx

Token frequency has unequivocally been defined as the frequency of occurrence of a linguistic entity (be it a morpheme, a lexeme, a string or a construction) in running text. A high and increasing token frequency of an item would be both “a concomitant of its grammaticalization” (Thompson and Mulac 1991: 319) and a trigger for upcoming formal and functional changes of the item (Bybee 2003: 605). ^[12] As a first encounter with the data, Table 2 indicates the absolute (n) and normalized (N, tokens per 10.000 words) frequencies of the IESCx per decade. These figures have been calculated by merging the results for the various corpora available per decade (see Table 1 above). ^[13]

It immediately becomes clear that over the last four to five decades the use of the IESCx has increased dramatically: from 4.8 tokens per 10,000 words in the 1970s corpus to 12.3 in the>2000-corpus, the most important leap being situated in the new century. The log-likelihood goodness of fit test gives a result that is significant at an error level of p<0.001 (G2=84.23). This suggests that the observed increase in the use of the IESCx between the 1970s and the twenty-first century is likely to represent a real increase in the population and should not merely be explained by accident. ^[14]

There may be several causes for this observed increase. A first explanation could be that it is an effect of the nature of the data gathered for each subcorpus, i.e. the more formal nature of the 1970s corpus, which includes semi-directed interviews, as opposed to informal spontaneous speech in the 2000s. However, the results from the two corpora providing data for both the 1990s and 2000s, namely COSER and Val.Es.Co (see Table 1), mean that this explanation can be, at best, only a partial one: the COSER values rise from 5.4 to 6.9, and those of Val.Es.Co from 2.5 to 5.6. Another plausible conclusion for the observed increase, then, could be that it is one of the symptoms of the increasing colloquialization of present-day Spanish. Note also that within the 2000s sample, the value for the COLAm corpus (N=27.8) is the highest. This, however, does not need to invalidate the hypothesis that in general the use of the construction has expanded, given that, in an Apparent Time Analysis (cf. footnote 7 above), youth speech represents the most recent language stage.

This drastic numerical increase points towards the conclusion that, with regard to its external dimension, the network has grown exponentially. The IESCx has settled itself more as a schema to which speakers more frequently resort in the flow of discourse, and to which, possibly, more constructs are attracted. This leads us towards a second operationalization of the notion of productivity, namely the type frequency of a construction.

4.2 Type frequency of IESCx

The notion of type frequency has received multiple definitions, depending on the linguistic phenomenon being observed. In its broadest definition it can be defined as “the number of items that are available to a particular class of forms” (Hopper and Traugott 2003: 127). In constructional approaches, the notion receives particular attention given that it accounts for the expansion or reduction of the linguistic material available for a general schema. The literature itself provides rather vague indications on what type frequency exactly is: Traugott and Trousdale (2013) resort to the notion of “possible collocates” of a construction; according to Bybee (2003: 605) one should count “the different lexical items with which a construction can be used”. However, for the study of the IESCx, the measure of type frequency is operationalized following Hilpert’s proposal (2013:128), namely by registering the “different instantiations of the same construction”. A continuing increase in type frequency over time would then count as evidence of the productivity of a construction (Barðdal 2008; Hilpert 2013; Traugott and Trousdale 2013).

With regard to its formal make-up, the IESCx contains four different slots which can be filled or not. Besides the verbal nucleus sabes, there is:

1. a slot making the 2nd person subject tú (‘you’) explicit;

2. two slots for adverbs specifying the mode of knowledge (como, ya or no) (e.g. como sabes ‘as you know’);

3. a slot for the DO of sabes (e.g. as in lo sabes ‘you know it’).

Each time a different item fills in one of these slots, resulting in different lexical combinations, this is counted as a different construction type. Figure 1 details the type frequency of the IESCx per decade.

Figure 1:

Type frequency of IESCx per decade.

Figure 1 tells us that since the 1970s, the type frequency of the epistemic stance construction has evolved following an S-shaped curve with (1) a gradual beginning, with even a little regression from 12 to 9 types from the 1970s to the 1980s corpora, followed by (2) a rapid expansion of the construction to 35 different types in the 1990s corpus but (3) a gradual tapering off towards 33 types in the>2000 corpus.

However, given that the type frequency heavily depends on the token frequency of a particular construction in a corpus, the above calculation must be considered a rather naïve account of productivity. More nuanced approaches have been proposed in the literature (see e.g. Zeldes 2012 for an overview) but, unfortunately, they cannot be applied straightforwardly to the Spanish corpora used here, on whose compositional nature hardly any information is available (e.g. none of the corpora listed in Table 1 is lemmatized or POS-tagged). As an intermediate solution, one can calculate the type/token ratio for each decade. The higher the resulting number, the higher the productivity of the construction. This calculation generates a different image with regard to the evolution of the construction’s productivity:

1970s: 0.167 (12 types / 72 tokens)>1980s: 0.113 (9/80)>1990s: 0.034 (35/1021)>2000s: 0.016 (33/2105)

Alternatively, as a variant of this measurement, inspired by Hilpert (2013: 128), one can count the potential productivity of the construction by dividing the number of hapaxes of the construction (thus, non-previously attested types) by the overall frequency of the IESCx. This gives the following figures per decade starting from the 1980s:

1980s: 0.025 (2 hapaxes/80 tokens)>1990s: 0.011 (11 hapaxes / 1021 tokens)>2000s: 0.004 (9 hapaxes / 2105 tokens).

Figure 2 plots the two productivity measurements for each decade.

Figure 2:

Relativized type frequency of IESCx per decade.

As a conclusion, it can be stated that the productivity of the schema itself seems to have dropped considerably. Thus, whereas the construction itself seems to have grown in terms of absolute numbers (see Section 4.1), the figures presented in Figure 2 point towards an internal reorganization within the IESCx network. It is shown in Section 5 that these frequency data indeed reveal that particular subschemas of the construction have become marginalized, whereas others have considerably gained in vitality.

5.1 Structural reduction within the epistemic stance construction

The IESCx comprises different variants that exhibit a varying degree of finiteness, and that can be ordered on a cline from simple structures (e.g. sabes) to fully finite clauses (e.g. sabes lo que pasa ‘you know what happens’). These structural variants constitute an important variable for the present diachronic empirical research. Abstracting over specific attestations (or constructs), four different patterns can be distinguished in the corpus. These patterns correspond to four different degrees of structural complexity of the IESCx, and take into account the presence or absence of different kinds of modifiers and arguments of the verbal nucleus sabes. ^[15]

The two ends of the structural complexity continuum are occupied by the bare pragmatic marker sabes and the matrix clause construction, respectively. On the one hand, the lowest degree of formal complexity (indicated as zero) is occupied by sabes without any modifiers. On the other hand, the highest degree can be attributed to those constructional variants in which there is a relationship of subordination between the expression of stance by sabes in the main clause and the host introduced by a complementizer (viz. the conjunction que ‘that’ or the relative lo que ‘what’ as in tú sabes que+clause, como sabes que+clause, ya sabes tú que+clause etc.).

In between, we find a wide gamut of parenthetical constructions (tú sabes, lo sabes, sabes una cosa, como sabes, etc.), which are syntactically detached from their host, and cause syntactic and/or prosodic disruption. ^[16] This corresponds to the general definition of parenthetical clauses which can be summarized as follows (e.g. Thompson and Mulac 1991; Schneider 2007, 2014; Brinton 2008; Van Bogaert 2009; Kaltenböck 2013):

1. parentheticals lack a syntactic connection with their host, to which they are related only by adjacency; as a consequence, they have a very mobile position;

2. they are semantically independent from the host, which implies that they can be added or left out without influencing the grammaticality of the clause;

3. prosodically, they are often characterized by a comma intonation implying a lower pitch, preceding and following pauses and a rising terminal. Note, however, that not all parentheticals are preceded or followed by pauses as they can also be prosodically integrated in their host, with which they adopt an enclosing melody (see Dehé 2007: 270–274 on the epistemic parentheticals I think/I believe; Avanzi 2012; Schneider 2014: 280–281).

However, with regard to their structural complexity, parenthetical constructions can be placed in two different categories. In a first class, the nucleus sabes is modified by the negative adverb no, the relational adverb como, or the time adverb ya. Given that these modifiers have a rather peripheral status with regard to the argument structure of the verbal nucleus, these instances are classified as representing a low degree of formal complexity of the construction. The second class consists of those parentheticals where the verbal nucleus is accompanied by one, or both, of its arguments, with (ya sabes tú, ya lo sabes, tú no sabes etc.) or without (tú sabes, lo sabes, sabes una cosa, sabes qué etc.) other adverbial modifiers, when it is defined as displaying a medium degree of complexity.

To recap, the large list of IES expressions (provided in Section 3.1), can be sorted into four subschemas, corresponding to various degrees of structural complexity:

1. matrix clause constructions (high degree of formal complexity): sabes lo que V, sabes que+clause, tú sabes lo que V, sabes tú lo que V, tú sabes que+clause, sabes tú+clause, ya sabes que+clause, no sabes lo que V, ya sabes lo que V,…

2. parenthetical constructions with expression of arguments (medium degree): tú sabes, sabes tú, como tú sabes, como sabes tú, sabes qué, no lo sabes, ya lo sabes, tú no sabes, lo sabes, sabes (una) cosa,…

3. parenthetical constructions with adverbial modifiers (low degree): ya sabes, sabes ya, no sabes, como sabes, como ya sabes,…

4. frozen pragmatic marker sabes (zero).

Table 3 and Figure 3 detail the formal variation of the IESCx throughout the four diachronically defined subcorpora. ^[17]

Figure 3:

Evolution of structural complexity of IESCx.

The data tell us that the evolution of the IESCx is characterized by a gradual erosion of the modifiers of the verbal nucleus. In general terms, the earlier decades (1970s–1980s) display a balanced frequency distribution between high and zero complexity structural variants. From the 1990s on, the latter category has increasingly become the dominant one. More specifically, in the 1970s corpus, despite its small size, the two most divergent categories are well represented with 37.50 % of all tokens of the construction being of zero complexity and 34.72 % of them falling into the highest complexity category. The medium parenthetical construction is also rather frequently attested (25 %), while the low complexity category is virtually absent (2.78 %). In the 1980s corpus, the medium parenthetical construction appears to have been sharply reduced (10 %), while the frequency of the low (6.26 %) and zero complex categories (46.25 %) have risen. This evolution continues in the 1990s corpus, which, however, also shows a notable decrease in the high complexity type (20.18 %). Finally, in the>2000 corpus all categories have shrunk for the benefit of the zero complexity class, frozen sabes (79.57 %).

There has thus been an important change in the relative frequencies of the constructional variants of the IESCx over the last 50 years. The quantitative data suggest that the construction as a whole has undergone a process of structural reduction. Moreover, this reduction of the IESCx seems to have followed a cascade-like model. Initially, the erosion process has left aside the adverbial modifiers and has mainly affected the arguments. Moreover, it set off with the omission of (pro)nominal arguments (e.g. lo ‘it’, una cosa ‘something’, tú ‘you’). Indeed, the medium parenthetical construction is the only construction variant that has consistently become less frequent over the four time periods; the matrix clause construction has put up stronger resistance in the earliest decades (1970s and 1980s) before also rapidly losing ground. The class of low complexity variants, including adverbial modifiers, seems to spread until the 1990s, but shrinks in present-day Spanish.

This study thus reveals that the diachronic evolution and formal changes within a more generally defined constructional schema do not come about arbitrarily; rather, they follow a morphologically and syntactically motivated order. The four different categories are still attested in present day Spanish, but it is to be expected that the zero category will gain in strength in the following years and decades.

As a consequence, the configuration of the constructional network itself seems to have evolved from a well-balanced distribution between formally different variants in the 1970s, towards a configuration in which one variant, namely the bare sabes, has become the most salient one in present-day Spanish. The formally complex structures have gradually moved towards the periphery of the network.

However, it is not enough to provide an answer to the question whether or not and in what directions the configuration of the IESCx has changed. The – perhaps even more interesting – question of why these changes have occurred should also be addressed.

5.2 Accounting for the reconfiguration of the network

The proposal by Haiman (1994), in which he argues for a parallelism between general cultural phenomena of ritualization and the process of grammaticalization, gives some good clues as to how to deal with this issue. The central idea that the frequent repetition of a certain practice entails habituation and emancipation of that practice is particularly inspiring for understanding the structural changes which the IESCx has been subject to. It shows that, in fact, these relate to more general functional changes within the domain.

Indeed, just as the frequent repetition of a cultural practice leads towards a depletion of its original force and significance (e.g. making the sign of a cross now has more than its literal religious meaning), the speech act through which, originally, a speaker asks the interlocutor to verify his or her knowledge on some upcoming or preceding information becomes less meaningful (and thus a habit) when frequently repeated. Consequently, parallel to frequently repeated practices, the expressions concretizing this speech act (such as sabes lo que pasa, no sabes, como tú sabes etc.) gain a more symbolic (emancipated) function, and lose (a part of) their original instrumental (interpersonal and epistemic) function. To put it differently, the expression of stance moves from the foreground to the background in the flow of information.

As argued by Bybee (2003: 605), repetition is crucial for understanding universal grammaticalization paths. The same mechanism seems to (partially) explain internal reconfigurations within the network of a more general schema. She continues, “the explanation for the content of what is repeated requires reference to the kinds of things human beings talk about and the way they choose to structure their communications” (Bybee 2003: 605). From this viewpoint, the answer to the question of why the interlocutor-oriented questions, i.e. the constructs of the IESCx, have been – and still are – so entrenched in language is straightforward: it comes down to recognizing the inherently interpersonal nature of human communication.

The IESCx being a high (and increasing) frequency construction, it is now easy to understand why it has been affected by a process of internal reduction. As the meaning of the construction weakens, the informational contribution of its constituent elements also decreases. In the first place, functionally redundant elements such as the expletive 2nd person pronoun tú (in Spanish morphologically indicated in the verb suffix), or semantically empty direct objects, such as lo ‘it’, una cosa ‘a thing’, or qué ‘what’, are omitted. In a second phase this reduction process also extends to the subordinate clause in the matrix clause construction, which, often, is also semantically poor (e.g. sabes lo que pasa ‘do you know what happens’, sabes lo que pienso ‘do you know what I think’). The adverbial modifiers, although not that frequent in the corpus, seem to be more resistant to this process of formal reduction. As a hypothesis it can be stated that this is due to their overt framing function, namely, the fact that they explicitly refer to the cognitive status of the interlocutor: previous knowledge in como sabes ‘as you know’ and ya sabes ‘you already know’, doubtful knowledge in ¿no sabes? ‘don’t you know?’. In a later phase, they are also subject to erosion.

In sum, once entrenchment of the interpersonal epistemic stance schema occurred, its frequency of use has had an effect on its form. The repeated variants of the construction have become more integrated and reduced, and this process has followed a morpho-syntactically (shorter (pro)nominal items>subordinate longer items), and semantically conditioned (semantically poor element>framing elements) order.