Cognitive Construction Grammar

Cognitive Construction Grammar (CCxG) is particularly fit to model sociolinguistic data for a number of reasons. First, it is a usage-based syntactic theory, which means that language is taken to represent variability and change at any given moment in time and that knowledge of language is assumed to include knowledge of the quantitative patterns in this variation (Bybee & Beckner, Reference Bybee, Beckner, Heine and Narrog2010:830). Second, CCxG assumes an encyclopedic/experiential view of semantics, which means that recurrent aspects of the usage events in which a particular structure is used are recorded as part of its meaning (Bybee & Beckner, Reference Bybee, Beckner, Heine and Narrog2010:846; Goldberg, Reference Goldberg2006:10; Langacker, Reference Langacker1987:63). Third, this theoretical approach includes a commitment to producing corpus (see, e.g., Goldberg, Reference Goldberg2011), psycholinguistic (see, e.g., Bencini & Goldberg, Reference Bencini and Golberg2000; Goldberg, Reference Goldberg2006:69–92; Goldberg, Casenhiser, & Sethuraman, Reference Goldberg, Casenhiser and Sethuraman2004), and even neurological (see, e.g., Allen, Pereira, Botvinick, & Goldberg, Reference Allen, Pereira, Botvinick and Goldberg2012) evidence for the constructs the framework depends upon, which ensures that the model is psychologically plausible.

CCxG claims that language consists entirely of form-meaning pairs of different degrees of abstractness, called “constructions” (Goldberg, Reference Goldberg1995:7, Reference Goldberg2006:45). As a result, no differences are posited between rules/generalizations (e.g., grammatical patterns such as transitivity) and lexical items/idiosyncratic patterns (e.g., idioms), because both are captured with the same principles (Goldberg, Reference Goldberg2009:94; Langacker, Reference Langacker1987:28). In other words, in this framework, the grammar is depicted as a network of interconnected construction schemas that offers speakers symbols to express conceptualizations (Langacker, Reference Langacker1987:409).

For example, in the short utterance given in (4), no less than five constructions can be distinguished: Juan, coger ‘take’, el ‘the’, vaso ‘glass’, and the transitive pattern <Subject Verb Object>.Footnote ³

1. (4) Juan coge el vaso.

   ‘John takes the glass.’

Contrary to low-level constructions (or words) such as Juan, coger, el, and vaso, the transitive pattern possesses a more schematic meaning, which, according to Langacker (Reference Langacker1991:302) relates to the conceptualization of a transfer of energy from the agent/subject to the patient/object participant. Based on this abstract meaning, the argument-structure construction determines that there will be two profiled argument roles (i.e., without these two participants, events of this type cannot take place): agent and patient. In addition, the schema determines the mapping of these roles onto syntactic functions and specifies the information structure of the clause. In sum, in this framework, argument-structure constructions, rather than individual verbs, are taken to be the determinants of clause structure.

A Cognitive Construction Grammar approach to the pluralization of haber

Within this theoretical setting, I would like to propose the hypothesis that the pluralization of haber results from a competition within the grammarFootnote ⁴ between two variants of the presentational construction with this verb, which allows speakers to position themselves in the social landscape. Formally, both are characterized by an adverbial phrase (AdvP) and an NP argument. In the impersonal argument-structure construction, the nominal is a direct object; in the personal variant, the NP functions as a subject.

When it comes to semantics, both construction schemas encode the POINTING-OUT event type proposed by Lakoff (Reference Lakoff1987:490):

Because both variants refer to the same event type, the constructions also assign the same semantic role to the NP slot. That is, as the referent of the NP is simply present in the scene that is being presented through the construction, it is probably safe to assume that it is assigned the “zero” semantic role (Langacker, Reference Langacker1991:288).

As these constructions are used to introduce referents into discourse, the nominal argument cannot be “given” with respect to the hearer's beliefs, consciousness, or world-knowledge (i.e., “hearer-old” in Prince's [Reference Prince, Mann and Thompson1992] terminology). This is not to say that it is impossible to use definite NPs in haber clauses. Rather, Ward and Birner (Reference Ward and Birner1995) showed that the nominal argument of presentational constructions can be definite and/or discourse-old as long as it can somehow be construed as hearer-new. This possibility is exemplified in (5), where the “reminder” (Lakoff, Reference Lakoff1987:561–562; Ward & Birner, Reference Ward and Birner1995:730–731) presentational reading emerges when the speaker treats condominios ‘condominiums’ as hearer-new, although this referent has already appeared in discourse and the interview takes place in such a building.

1. (5) Inv: ¿Este, cuando se mudó aquí, habían cosas a la que tuvo que acostumbrarse?

   Subj: ¿Tales cómo?

   Inv: No sé por ejemplo. E, este, no sé, cosas.

   Subj: Este, bueno, no, me tuve que acostumbrarme a vivir en un condominio cuando yo viví en una casa. . . .

   Inv: ¿Y usted recuerda como la ciudad era antes? O sea, cuando era niña.

   Subj: Cuando yo era niña, sí. No habían estos condominios, desde luego (SJ01M22/ SJ07).

   Inv: ‘Eh, when you moved here, were there things to which you had to get used to?’

   Subj: ‘Such as?’

   Inv: ‘I don't know, for example. E, eh, I don't know, things.’

   Subj: Eh, well, no, I had to get used to living in a condominium, when I had lived in a house. . . .

   Inv: And do you remember how the city was like before? That is to say, when you were a child.

   Subj: ‘When I was a girl, yes. There weren't _Pers these condominiums, of course.’

Concerning the adverbial phrase that frequently appears with presentational expressions (e.g., Lyons, Reference Lyons1967), Lakoff (Reference Lakoff1987:542–543) argues that in the English presentational there construction, the adverbial denotes the nature of the “mental space,” that is, “partial assemblies constructed as we think and talk for purposes of local understanding and action” (Fauconnier, Reference Fauconnier, Geeraerts and Cuyckens2007:351), that is set up by the expletive (the “space builder”). Examples (6) to (8) give reason to believe that in haber clauses the adverbial phrase fulfills a similar function. However, the difference between English and Spanish presentationals appears to be that, as the haber construction does not involve an expletive, the adverbial does not only specify the nature of the mental space, it also functions as the space builder (i.e., it sets up the mental space).

1. (6) En mi salónhabían treinta o treinta y un niños, sí (SJ12M12/SJ1404).

   ‘In my classroomthere were _Pers thirty or thirty-one children, yes.’

In contrast, without the adverbial phrase, as in (7), the expression localizes the referent of the NP in the current “base space,” that is, “the mutually known world of the interlocutors” (Croft & Cruse, Reference Croft and Cruse2004:33), or a previously built mental space, such as the time when the speaker was in high school in (8).

1. (7) Hay veces que tú sales a la calle y ves personas jóvenes hablando con palabras feas (SJ01M22/SJ60).

   ‘There are _Imp times that you go out on the streets and you see young people talking with ugly words.’

2. (8) E, pero yo diría que, que las electivas, las, los cursos electivos probablemente en la escuela superior eran los más que me gustaban. . . . Podrían haber días en que yo tenía dos horas libres entremedio (SJ13H11/SJ1566).

   ‘E, but I would say that, that the electives, the, the elective courses, probably in high school were what I liked most. . . . There could be _Pers days that I had two hours of free time in between.’

However, the grammatical status of the adverbial phrase does not imply that it is completely optional, because its absence makes the expression uninformative when the NP argument cannot be located in the base space or a previously evoked mental space (Hernández-Díaz, Reference Hernández-Díaz and Company-Company2006:1130–1132; Lyons, Reference Lyons1967; Meulleman & Roegiest, Reference Meulleman and Roegiest2012). If we consider, for example, the last utterance of (8) in isolation of the evocation of the “high school” mental space, we cannot help but wonder against which setting we have to interpret the assertion. Therefore, the adverbial as well is most accurately described as profiled.

A Cognitive Construction Grammar approach to the social distribution of alternating constructions

Before we start exploring how the social distribution of alternating constructions can be incorporated in CCxG, first, it is important to recall that constructional variation will only be interpreted in social terms if the variants refer to exactly the same event type (Lavandera, Reference Lavandera1978). With these limitations, because CCxG is a usage-based framework, it is hardly controversial to claim that speakers notice that, in the absence of other differences of meaning between two formally distinct options to refer to a given event type, individuals instantiating a particular social type use a certain alternative more often and that they store this social distribution as part of the meaning of the relevant alternating constructions (Bybee & Beckner, Reference Bybee, Beckner, Heine and Narrog2010:830, 846; Goldberg, Reference Goldberg2006:10; Langacker, Reference Langacker1987:63). Therefore, it is rather unproblematic to suppose that if two constructions refer to exactly the same event type, each will include a social distribution specification, such that the use of one of these constructions at a certain rate where the constructional semantics is observed will be interpreted as signaling social subgroup membership (“1st-order index” in Silverstein's [Reference Silverstein2003] terms) and, potentially, everything there is to know about this social subgroup (“1 + nth order index” in Silverstein's [Reference Silverstein2003] terms).

All of this knowledge will not be activated every time a particular distribution is observed in a particular context. Rather, the context will activate or background certain things language users know about a particular social group (see Langacker, Reference Langacker1987:154–166), which explains how the multiple context-dependent interpretations of sociolinguistic variables arise (e.g., Eckert, Reference Eckert2008). In other words, CCxG can readily accommodate the social meaning of alternating constructions that refer to the same event type. In this paper, I will only be concerned with the links Puerto Ricans establish between the distribution of presentational haber and (knowledge on) social groups.

Markedness of coding

Against the background of the main hypothesis, a first cognitive factor that can be expected to condition haber pluralization is the preference for unmarked coding, that is, the tendency for a “notion approximating an archetypical conception [to be] coded linguistically by a category taking that conception as its prototype” (Langacker, Reference Langacker1991:298). This leads to the first hypothesis:

This hypothesis raises the question as to which features can portray the NP of presentational haber as a more prototypical subject or object. In this regard, the typological literature suggests that prototypical subjects are agents in events (Comrie, Reference Comrie1989:66; Keenan, Reference Keenan and Li1976:321; Lakoff, Reference Lakoff1987:64; Langacker, Reference Langacker1991:294). However, as we have seen in the previous section, the nominal argument of presentational haber is clearly not an agent, because it is merely present in a static situation. Still, it is inarguably the case that some entities (say, driver) are intrinsically more likely than others (say, victim) to be agents in events. Therefore, with constructions such as presentational haber, which do not explicitly construe the nominal argument as an agent or a patient, entities such as driver may be perceived as more potential agents, and, thus, as more prototypical subjects (Langacker, Reference Langacker1991:294) than entities such as victim.

In cognitive linguistics, the semantic roles “agent” and “patient” are defined in relation to what Langacker (Reference Langacker1991:283–285) called the “canonical event model” or the “action-chain model”: the head initiates physical activity, resulting “through physical contact, in the transfer of energy to an external object” (Langacker, Reference Langacker1991:285) and an internal change of state of that entity, the tail of the chain. The semantic roles of agent and patient, in turn, are defined as, respectively, “action-chain head,” and “action-chain tail.” Additionally, events take place in a particular setting, such that the event model minimally includes three elements: action-chain head/agent, action-chain tail/patient, and setting.

To test the first hypothesis, I coded the data for the typical action-chain position of the entity indicated by the noun, for which I relied on the answers to the question in (9).

1. (9) Is the referent of the noun highly likely to cause an internal change of state to a second entity without being affected by a third entity first?

   Yes: Typical action-chain head (i.e., more potential agent; e.g., temblor ‘earth quake’, madre ‘mother’, carro ‘car’)

   No: Typical action-chain setting or tail (i.e., more potential setting or patient; e.g., actividad ‘activity’, víctima ‘victim’, daño ‘damage’)

I also evaluated the influence of the polarity of the clause. As we will see in the section dealing with the results for markedness of coding, this factor can be connected to subject/objecthood because, with presentational haber, negative polarity originates a nonspecific indefinite reading for the NP argument, which is prototypical of direct objects (Croft, Reference Croft2003:132).

Statistical preemption

A second likely cognitive constraint is statistical preemption. That is, in a usage-based framework such as CCxG, forms that occur in more than one construction are taken to have stronger independent cognitive representations than those that only occur in one construction (Bybee, Reference Bybee, Joseph and Janda2003:619; Croft & Cruse, Reference Croft and Cruse2004:292). In particular, if a form presents high token frequency in one construction schema, but only occurs sporadically in other patterns, it is taken to be stored as a partially lexically filled instance of this construction with a much stronger representation than both the independent form and the abstract pattern (Goldberg, Reference Goldberg1995:79; Langacker, Reference Langacker1987:59–60, Reference Langacker1991:48). As expressions based on this partially prefabricated unit require less constructive effort than do expressions involving the fusing of more abstract construction patterns and words, the prefab disfavors the use of a novel expression based on a competing construction schema that shares the same pragmatic and semantic constraints (Goldberg, Reference Goldberg2006:94, Reference Goldberg2009:102–103, Reference Goldberg2011). This is called “statistical preemption.”

In other words, if certain verb forms of haber occurred mainly in the <AdvPhaber Obj> pattern before <AdvPhaber Subj> emerged as a conventional alternative for referring to POINTING-OUT, upon actuation of the change, the personal variant would not have been used frequently to refer to a conceptualization that involved this event type in the temporal setting encoded by those tenses. In subsequent generations, repetition usually ensures that a given distribution remains intact (Bybee, Reference Bybee2006:715). This leads to:

Because these hypotheses assume that the preempting effect of certain verb tenses is a function of the degree of entrenchment of a particular form in the impersonal construction, the following prediction follows quite naturally:

The testing of these hypotheses raises two questions: When did the variation that affects presentational haber emerge as a community-wide phenomenon? Which tense forms of the verb enjoyed a relatively high frequency in a variety of constructions before this happened? The answer to the first question can only be tentative, as it is virtually impossible to know for a fact when and how the variation that affects presentational haber started exactly in Puerto Rican Spanish. For Buenos Aires, Fontanella de Weinberg (Reference Fontanella de Weinberg1992b) showed that the alternations between singular and pluralized presentational haber are already present in written discourse from the 18th century onward. Because there is usually a considerable lag between the actuation of a linguistic change and its trickling down into writing, the variation probably emerged somewhere in the 17th century, which blends in nicely with research in historical linguistics that shows the most prominent features of American Spanish to stem from a koiné variety that emerged through language and dialect contact during that century (Fontanella de Weinberg, Reference Fontanella de Weinberg1992a:42–54).

Therefore, to formulate an answer to the second question, in Appendix A, I tabulated the distribution of the third-person singular forms of haber in a 16th-century Latin American subset of the Spanish Royal Academy's Corpus Diacrónico del Español (Real Academia Española, 2008). The results show that before presentational haber became involved in large-scale variation in American Spanish, the present and preterit tense forms occurred primarily in presentational clauses, which suggests that the most salient representations of these forms were <AdvPhay Obj> and <AdvPhubo Obj>. The other tense forms, on the other hand, were either used more productively (spread over more different constructions) or are restricted to a very low frequency in the corpus (n < 100), which indicates that their independent forms probably also constituted their strongest cognitive representations. This distribution suggests two relevant tense groups: synthetic expressions in present and preterit tense versus all others.

Structural priming

Furthermore, sociolinguistic and psycholinguistic research has shown that once speakers have used or processed a constructional pattern, they tend to recycle it in the following stretches of discourse (Bock, Dell, Chang, & Onishi, Reference Bock, Dell, Chang and Onishi2007; Goldberg, Reference Goldberg2006:120–125; Labov, Reference Labov1994:550), which is known in the psycholinguistic literature as “structural priming.” Therefore, if the variation amounts to a competition between two argument-structure constructions, as the main hypothesis claims, I expect that:

To test this hypothesis, all tokens were coded for the last variant of the haber structures that was used by the speaker (production-to-production priming) and the investigator (comprehension-to-production priming). Additionally, I binned together the occurrences in 5-clause lag groups (e.g., lag: 0–4 clauses; 5–9 clauses, etc., up to a 20-clause lag) and I separated the occurrences in which speakers repeated the verb form and the argument structure from those in which they only repeated the argument structure. This resulted in a total of 17 factors for both factor groups. However, as the initial results displayed a similar priming effect until reaching a 20-clause lag, independently of whether speakers would repeat the specific verb form or not, I collapsed the factors into the following broader categories: first occurrence/distance 20+ clauses, primed with the personal presentational haber construction, primed with the impersonal presentational haber construction.

Principles of linguistic change

The claim that the variation reflects a slowly advancing language change from below also predicts that the alternations will display patterns of social and stylistic covariation typical of this type of linguistic evolution. More specifically, the “apparent-time” construct (Labov, Reference Labov1994:43–72) predicts that:

By the same token, Labov's Gender Principle (Labov, Reference Labov2001:292) predicts that:

However, because gender-differentiated behavior is also found for changes from above (Labov, Reference Labov2001:274) and because the possibility of age-graded behavior always exists for apparent-time distributions (Labov, Reference Labov1994:45), more evidence will be needed before we can confidently conclude that this alternation constitutes a linguistic change from below. In this regard, the most conclusive indication of an ongoing change from below seems to be the curvilinear pattern (Labov, Reference Labov2001:188), which entails the following description:

In addition, in change from below, the innovative variants usually show no style-shifting or increase in frequency when formality rises (Labov, Reference Labov2001:95–101), which leads to the following hypothesis:

Finally, Labov (Reference Labov1972:138) observed that highly educated speakers tend to conform to the supralocal prestige norms, which suggests that:

To test these predictions, a mixed-effect logistic regression analysis was performed. However, before we can go on to discuss the results, a few words are in order concerning the methods that were used in compiling the corpus from which this investigation draws.

The sample

The analyses are based on a 200,000-word corpus of sociolinguistic interviews (about 24 hrs. of speech) that was recorded in March and April 2011 with 24 native speakers of Puerto Rican Spanish who reside in the San Juan metropolitan area. As is shown in Table 1, the corpus is stratified by three social parameters: academic achievement (university degree vs. no university degree), age (20–35 years vs. 55+ years), and gender (male vs. female).

Table 1.

Configuration of the sample

To have access to more variable contexts and to investigate whether the variation is sensitive to style-shifting, as is predicted by Hypothesis 7, the author, who was also the fieldworker, structured the recordings into three sections:

1. 1. Interview. Speakers were interviewed for about 30 mins. on a variety of topics related to their day-to-day life. The interview format was loosely based on the conversational modules that are found in Moreno-Fernández (Reference Moreno-Fernández2003), Quintanilla-Aguilar (Reference Quintanilla-Aguilar2009), and Tagliamonte (Reference Tagliamonte2006:App. B). In addition, a set of questions containing the variable, such as (10), was included to investigate comprehension-to-production priming effects. In these questions, the variants were used randomly.

   1. (10) ¿Cuántos estudiantes podía haber/podían haber en tu época?

      ‘How many students could there have been _Imp/could there have been _Pers in your time?’

2. 2. Reading passage. After the first part, speakers were instructed to read out a two-page children's story (Juan Sin Miedo, ‘John Without Fear’) in which 31 selection contexts, as in (11), had been inserted (20 trials, 11 fillers).

   1. (11) En una pequeña aldea, había/habían un anciano padre y sus dos hijos . . .

      ‘In a small village, there were _Imp/there were _Pers an old father and his two sons . . .’

3. 3. Questionnaire. Finally, the interviewees were instructed to read out loud a questionnaire consisting of 45 items (32 trials, 13 fillers) preceded by a description that evoked the right pragmatic context for the interpretation of the test sentence, as can be seen in (12). For both the reading passage and the questionnaire, whenever a speaker had difficulties reading, the author read the tests to him/her and asked him/her which form he/she preferred.

   1. (12) A Inés le acaban de robar el carro, que tenía aparcado en algún callejón obscuro. Aunque no es la cosa más sensata que se pueda hacer, una amiga trata de consolarla diciendo: “No es culpa tuya, es que siempre ______ unas personas malas.”

      a. habrá b. habrán

      ‘They've just stolen Ines' car, which she had parked in a dark alley. Although this is not the most intelligent thing to do, a friend tries to comfort her, saying: “It is not your fault, ______ always be a few bad people.”’

      a. There will_Imp b. There will_Pers

Poststratification: Social class

After transcription, this sample was poststratified by social class, a combination of the speaker's weighted scores for the following variables:

1. 1. Academic achievement: 0: less than high school; 0.5: high school; 1: university.

2. 2. Housing conditions: 0: house/apartment in poor condition; 0.5: house/apartment in good condition with up to two bedrooms; 1: house/apartment in good condition with three or more bedrooms.

3. 3. Profession: 0: for example, street vendors, unskilled workers; 0.25: for example, shop-holders, secretaries, mechanics; 0.5: for example, college-educated professionals, teachers; 0.75: for example, liberal professions, mid-level managers, university professors; 1: senior managers/officials (see Moreno-Fernández, Reference Moreno-Fernández2003, for the full scale).

Rather than relying on my own intuitions, in order to establish the relative importance of each factor for social prestige, after the recording sessions, I presented the informants with an additional questionnaire with the instruction to rank academic achievement, housing, and profession by their importance for social status. Subsequently, each speaker's score for a given variable was first multiplied by the average impact factor that had been established for that variable, and these weighted scores were summed together. Then, the raw numbers were converted into percentages of the maximally possible score. Finally, speakers were grouped together in three equal bins by their score on the prestige index: those who scored 0% to 33% were considered to be members of the lower class (7 speakers), those who scored 34% to 66% were considered to belong to the middle class (7 speakers), and those who scored 67% to 100% were considered to be members of the higher classes (10 speakers).

The variable context

Earlier studies did not find any variation with present-tense hay. However, my corpus provides a handful of tokens of hayn, such that the alternation between hay and hayn cannot be excluded from the scope of this investigation. In contrast, some previous surveys (e.g., DeMello, Reference DeMello1991; Quintanilla-Aguilar, Reference Quintanilla-Aguilar2009) have included cases in which haber agrees with first-person plural NPs. Nevertheless, as examples such as (13) and (14) display a clear difference in meaning, personalized first-person plural presentational haber was considered to be outside the envelope of variation.

1. (13) Y habían bastantes, bastantes estudiantes en, e, los salones de clase (constructed example).

   ‘And there were _Pers plenty, plenty of students in, e, the classrooms.’

2. (14) Y habíamos bastantes, bastantes estudiantes en, e, los salones de clase (SJ03H22).

   ‘And we were plenty, plenty of students in, e, the classrooms.’

Statistical tools

After selecting and coding all the cases of presentational haber + plural NP, I performed a mixed-effect logistic regression analysis with Rbrul (Johnson, Reference Johnson2009). In order to evaluate the hypotheses, I tested the statistical effect of 11 factor groups, which either instantiate the three cognitive constraints referred to in Hypotheses 1 to 3 (markedness of coding, statistical preemption, and structural priming) or try to establish the social distribution of haber pluralization. Additionally, the individual speakers and the lemmas of the NPs’ head nouns were included in the runs as grouping factors. However, as these were collinear, I ran parallel analyses for the grouping factors, and I only report as statistically significant those fixed effects that proved to condition the variation for all speakers and all lexical items.