Stage-level and Individual-level distinction in morphological variation: An example with variable haber agreement

This work examines the role of the stage-level (SL)/individual-level (IL) distinction applied to nouns in a case of morphosyntactic regularization in Spanish: variable reanalysis of the NP argument as subject in the presentational haber construction (había/habían perros). We conduct variationist, quantitative analyses on all instances of existential haber with a plural NP in corpora of spoken Puerto Rican Spanish (>500,000 words) to determine the linguistic factor groups that promote reanalysis and, hence, pluralized forms. Results of variable rule analyses reveal that the SL-IL distinction constrains the regularization. IL predicates significantly favor haber regularization (e.g., habían muchas personas de las Antillas ‘there were a lot of people from the Antillas’) whereas SL predicates significantly disfavor pluralized forms (este año hubo menos tiros que en años pasados ‘this year there were fewer shots fired than previous years’). These results are interpreted from within a usage-based framework in which the status of the noun introduced in the [haber + NP] construction, as either a likely or unlikely subject for haber, influences the analogical leveling. IL predicates are more prototypical nouns than SL predicates because the former are temporally persistent. IL predicates promote nouns’ candidacy as subjects over direct objects because prototypical subjects present two temporally-persistent characteristics: independent existence and referentiality. As a result, IL predicates increase the likelihood of reanalyzing the direct object as subject, thus triggering agreement of the verbal form with plural NPs. SL predicates, on the other hand, because they display low temporal stability, inhibit regularization.


Introduction
In its original formulation (Carlson 1977), the semantic distinction between Individual-Level (IL) and Stage-Level (SL) predicates describes a cross-linguistic commonality in which some predicates express more permanent, immutable properties than others in which the properties expressed are more transient or changeable. This conceptual distinction has repercussions at the morphosyntactic level. Carlson (1977) shows that two predicates such as naked and intelligent provide different degrees of acceptability when combined with perception verbs, as is illustrated in examples (1) and (2).
(1) He saw John naked (2) *He saw John intelligent As can be seen in example (1), naked is acceptable after the object of a perception verb (John), while intelligent yields an unacceptable sequence in the same syntactic slot in (2). This difference in grammaticality judgments is argued to be a result of the SL-IL distinction. 1 The adjective naked expresses a transient property that is spatio-temporally located even in the absence of explicit locative and temporal adverbials, and thus SL, while intelligent refers to a more permanent property, and thus IL.
In addition to secondary predications, cross-linguistic research has shown that the SL-IL distinction is relevant to account for the distribution of alienable and inalienable possession (Ogawa 2001), auxiliaries (Benedicto 2002;Koontz-Gaborden 2012), and copulas (Sulger 2011). Indeed, in Spanish this distinction has been used to explain grammatical phenomena such as small clauses 2 (Fernández Leborans 1995;Hernanz and Suñer 1999), secondary predicates (Demonte and Masullo 1999), and the ser-estar ('to be') contrast in copulative constructions (Clements 1988;Fernández Leborans 1999;Escandell-Vidal and Leonetti 2002;Marín 2004;Camacho 2012;Gumiel-Molina and Pérez Jiménez 2012). Ser typically combines with IL predicates while estar introduces SL predicates. However, as Camacho (2012: 471) notes, certain IL predicates can be contextually coerced 3 into expressing a SL meaning. For example, guapo 'handsome' typically expresses an IL property and thus often combines with ser. However, a SL meaning can be coerced out of this form to express a transient property with estar (i.e.; Paco no es guapo pero está guapo hoy 'Paco is not handsome, but he looks handsome today').
McNally (1994,1998) and Escandell-Vidal and Leonetti (2002) argue the permanent/transitory distinction that is generally associated with the SL-IL distinction (Carlson 1977) is actually a pragmatic inference. Rather than permanent properties, Escandell-Vidal and Leonetti (2002) maintain that IL predicates denote classificatory properties, whereas SL predicates denote episodic states. By classificatory properties, these authors mean "properties that are used to categorise individuals as belonging to a specific class" (Escandell-Vidal and Leonetti 2002: 160). Properties that are associated with a class tend to be long-lasting, but they are not necessarily permanent. For example, young is an IL predicate, since it refers to a classificatory property, but it is not a permanent property. In addition, Kratzer (1995), Chierchia (1995), and Maienborn (2004) note that locative and temporal adjuncts, which are typically associated with SL predicates, may also combine in discourse with IL predicates. For example, blonde is an IL predicate. However, it may be anchored in time, as in the following example: when I was a child, I was blonde (but then my hair darkened).
The SL-IL distinction has been described as belonging to the category of inherent lexical aspect or Aktionsart (Leonetti 2004;Marín 2004Marín , 2009Camacho 2012), which can characterize lexical items such as adjectives, verbs and common nouns. SL predicates are bounded, that is to say, they refer to episodes with a beginning and an ending, whereas IL predicates are unbounded, i.e., they are temporally-persistent. From this perspective, the goal of this study is to explore the relevance of the SL-IL distinction in cases of morphosyntactic variation in Spanish and its potential role in language change. In this way, this work contributes to the body of research on the semantic SL-IL distinction through novel application to a case of synchronic variation. In line with Ogawa (2001), we will apply the SL-IL distinction to nominal predications, as they occur in the existential construction with haber in Spanish. In many varieties of Spanish, existential constructions with haber present two variants: in one, haber does not display verbal agreement with its noun phrase (NP) argument, whereas in the other, haber agrees in number and person with its NP argument. We will discuss this phenomenon of morphosyntactic variation using Puerto Rican data.
This work is organized as follows; Section 2 outlines salient points of the linguistic problem we investigate in the current work, namely, the variable pluralization of haber with plural NPs. In section 3, we detail our data and methods, followed by section 4 in which we present the results of our statistical analyses and explain the role of the SL-IL distinction in this case of synchronic variation. Lastly in section 5 we present our conclusions.

Background
Prescriptively, Spanish presentational haber 'there (be)' is used in the singular regardless of the number of the NP it introduces, as can be observed in examples (3) and (4)  However, this construction co-occurs in many varieties of Spanish with another in which plural NPs may trigger the use of a plural form of haber, as can be seen in examples (5) and (6): Cortés-Torres, Interview 11, 37 (5) J: ...Sabes, hubieron tiendas que cerraron a las nueve 'J:...You know, there were stores that closed at nine' Cortés-Torres, Interview 15,157 (6) No, pero habían muchas hormigas y esas hormigas son de las que pican bueno 'No, but there were a lot of ants, and of the type that really bite' As is shown by DeMello (1991: 468), variable haber agreement is a pervasive phenomenon in contemporary spoken Spanish. In the Puerto Rican data we use for this study, plural NPs combine with plural forms of haber (habían, hubieron, hayan) in 44% (N=83) of the examples, and with singular forms of haber in 56% (N=107) of the examples. In fact, some NPs may occur both with singular and plural forms of haber, as the following examples illustrate with the noun personas 'people': Davies, Habla Culta San Juan y entonces allí no había protección, porque no había nadie, había pocas personas, y allí fue el... el incidente grande, donde por poco queman esta gente 'and back then there was no protection there, because there wasn't anybody, there were few people, and the... the big incident was there, where they almost burnt those people ' Davies,Habla Culta San Juan (8) Habían muchas personas de las Antillas. 'There were a lot of people from the Antilles' The process of haber regularization entails a reanalysis of the grammatical relation of the NP argument (Waltereit and Detges 2008). The NP argument of haber is traditionally regarded as a direct object because, as is shown in examples (3) and (4) above, it does not trigger verbal concord, that is to say, haber always appears in the singular regardless of whether the NP occurs in the singular or in the plural (e.g., hubo un problema / problemas 'there was a problem / there were problems'). In addition, like with any other Spanish direct object, the NP argument of haber may, in some contexts, be replaced by a clitic in the accusative case, as in the following example: Davies, Habla Culta San Juan (9) …hay unas constantes, igual que las hubo en la dramaturgia griega 'there are some constant themes, as there were in Greek drama' In (9), there are two constructions with haber. In the first construction, haber is in the present tense (hay) and its NP argument is unas constantes 'some constant themes'. In the second construction, haber is in the preterit (hubo) and its NP argument is the feminine singular accusative clitic las, which is coreferential with unas constantes 'some constant themes' in the first clause.
However, there is an alternative construction in which haber agrees in number with its NP argument (e.g., hubieron problemas 'there were problems'). This shows that the NP has undergone a process of reanalysis from direct object to subject, as is evidenced by verbal agreement. This reanalysis is promoted by the existence of constructions such as (10): Constructed example (10) Hubo un problema 'There was a problem' In (10), the verbal form hubo 'there was' occurs in the singular form. This fact allows the speaker to reinterpret the NP as the subject of the construction. The occurrence of examples such as hubieron problemas 'there were problems', in which both the NP (problemas) and the verbal form hubieron occur in the third person plural, may be used as evidence that reanalysis has taken place. In such cases, the plural verbal form agrees in number with the plural NP referent, indicating the noun is no longer viewed by the speaker as the object, but rather the subject. Indeed, recent work (Brown and Rivas 2012) demonstrates that nouns most often used with subject function generally are precisely the nouns to most strongly trigger haber pluralization. Thus, the noun's propensity and suitability as subject plays a pivotal role in a speaker's likelihood to regularize haber with plural NPs.
The phenomenon of variable haber agreement has received a lot of attention in the literature ( (Bentivoglio and Sedano 1989;DeMello 1991;Montes de Oca-Sicilia 1994;Domínguez, Guzmán, Moros, Pabón and Vilaín 1998;Díaz Campos 1999-2000, 2003Freites Barros 2004;D'Aquino Ruiz 2004Castillo-Trelles 2007; Quintanilla Aguilar 2009). However, as is noted by Bentivoglio and Sedano (2012), very little consensus has been reached, since studies differ in the significance of linguistic factors when accounting for this morpho-syntactic variation. In what follows, we will provide a summary of the findings reported in previous studies regarding this morphosyntactic variation. We will restrict our outline to linguistic factors.
One of the linguistic factors that has been argued to account for variable haber agreement is the human vs. non-human referent of the NP. Using data from Venezuelan varieties, Bentivoglio and Sedano (1989), Domínguez, Guzmán, Moros, Pabón and Vilaín (1998) and Díaz-Campos (1999-2000 report that an NP with a human referent tends to occur with pluralized forms of haber, whereas NPs with nonhuman referents typically occur with haber in the singular. According to DeMello (1991: 462) the same result also applies to San Juan, Santiago and Bogotá. However, more recent studies based on multi-regression analyses using Varbrul (Díaz-Campos 2003;D'Aquino Ruiz 2004) argue that the factor group +/-human NP does not significantly constrain variable haber agreement in Venezuelan Spanish. Similar results are reported for other varieties of Spanish such as México (Castillo-Trelles 2007) and El Salvador (Quintanilla-Aguilar 2009). 4 Additionally, DeMello (1991) shows that in La Paz and Lima, non-human NPs pluralize more often than human NPs.
In addition to the factor +/-human NP, Bentivoglio and Sedano (1989) report that haber pluralization also correlates with the reinforcement of plurality markers (indefinites, quantifiers and coordinated NPs) in the NP argument, especially when the NP is inanimate. However, other authors such as Domínguez, Guzmán, Moros, Pabón and Vilaín (1998), and Díaz-Campos (1999-2000, 2003 do not find this factor to have an impact on the use of plural forms of haber in their Venezuelan corpora, and nor does Quintanilla-Aguilar (2009) in his corpus of casual conversation from El Salvador. In contrast, Castillo-Trelles (2007) indicates that the absence of quantifiers favors pluralization of haber in her corpus of Mexican Spanish. D'Aquino Ruiz (2004) shows that polarity is also a conditioning factor for variable haber agreement in Spanish. Her Varbrul results, which are based on an oral corpus of Venezuelan Spanish, show that affirmative clauses favor the use of a plural form of haber, whereas negative clauses disfavor it. This linguistic factor has not been used in other studies with the exception of Quintanilla-Aguilar (2009), who reports that Varbrul does not select polarity as significant in his analysis of Spanish from El Salvador.
Finally, verb tense is the linguistic factor that has been almost consistently mentioned as conditioning regularization of haber. 5 The imperfect tense (había) is argued to favor pluralization in different varieties of Spanish (Bentivoglio and Sedano 1989;DeMello 1991;Domínguez, Guzmán, Moros, Pabón and Vilaín 1998;Díaz-Campos 1999-2000, 2003; Quintanilla-Aguilar 2009), whereas preterit (hubo) and present (hay) tenses highly disfavor pluralized forms. Explanations for this crossdialectal tendency are varied. Waltereit and Detges (2008) argue that regularization takes place in tenses with a low token frequency because high token frequency forms are more conservative due to their increased lexical strength and, therefore, less prone to participate in regularization processes (Bybee and Thompson 1997). According to Waltereit and Detges (2008), the imperfect is the less frequent tense in the past, and hence regularization is more frequent with imperfect than with the preterit. Differences in paradigmatic relationships have also been implicated in this process of haber regularization (Del Rosario 1970;Bentivoglio and Sedano 1989;Hernández Díaz 2006). The difference in morphological shape of singular vs. plural in third person is more subtle in the imperfect (había~habían) than in the preterit (hubo~hubieron) in which considerably more phonetic material is added, and thus pluralization is said to be more easily applied to this imperfect paradigm.
Both of the previous arguments (role of word frequency, paradigmatic relationships) are supported by the lack of pluralized forms in the present indicative (hay) in varieties of Spanish with plural haber forms. On the one hand, the present tense form hay has a very high token frequency relative to the other forms, and such strong lexical entrenchment does not favor analogical extensions via regularization. Further, the form hay is already irregular (García 1986) within not just the haber paradigm, but within the inflectional morphology generally, and thus hay lacks an obvious plural counterpart. 6 Nevertheless, the frequency explanation loses strength considering two facts. It is not at all clear that the preterit has a higher token frequency than the imperfect in Spanish. In fact, according to the Corpus del español (Davies 2002-), in oral Spanish the imperfect actually has a higher textual frequency (11,174 per million) than the preterit (10,576 per million). Indeed, if we only consider uses of haber as an existential verb (and not as an auxiliary), greater textual frequency of the imperfect over the preterit is even more apparent. In the Cortés-Torres corpus of Puerto Rican Spanish, the token frequency of the imperfect is considerably higher (619 per million) than the token frequency of the preterit (176 per million). In this same line, the paradigmatic explanation also presents some weaknesses since it has not been demonstrated that analogical changes are necessarily shaped in any predictable way by the addition of phonetic material. What, then, could account for the recurrent pattern of significant differences in rates of regularization across tenses?
The present work addresses this question by exploring the potential role of the SL-IL distinction in relation to tense in this phenomenon of morphosyntactic variation using naturally occurring data from Puerto Rican Spanish. We will show that the tense difference noted in previous research also holds in our data, but that it is indeed masking an underlying, more general, SL-IL distinction in the nominal predicates. We will show that IL nominal predicates favor pluralization. This is due to the fact that pluralization entails a change from direct object to subject grammatical relation, and IL predicates are better candidates to act as subjects. For this reason, IL predicates tend to trigger 'subject-verb' agreement (eg. Habían gatos 'there were cats') more often than SL predicates, as will be shown in more detail in our discussion. We outline our data and methods below.

Data and Methods
To elucidate this SL-IL distinction in the analogical extension of haber, we take a variationist approach (Poplack and Tagliamonte 2001) and, owing to the low textual frequency of presentational haber with plural nouns, use two oral corpora of Puerto Rican Spanish. One corpus is the online Corpus del español (Davies 2002-). Our examples are all taken from the oral section of Puerto Rican Spanish, which amounts to 200,000 words. The other corpus we employ (Cortés-Torres 2005) is made up of casual, recorded conversations between native speakers of Puerto Rican Spanish and totals approximately 370,000 words representing 27 hours of speech. From both corpora, we extract all the examples of haber with a plural noun phrase (i.e.; those nouns that could possibly trigger pluralization). We exclude instances of the present indicative form hay, because it does not display variation. Altogether, this yields 190 examples of haber used with a plural NP.
In order to facilitate comparisons of our results with previous research, we code for linguistic factors found to be significant in some of the analyses of other varieties of Spanish mentioned above. These are the following: a) +/-human NP: Comrie (1989: 191) points out that there is a cross-linguistic tendency for NPs occupying higher positions in the animacy hierarchy to trigger verbal agreement more often than less animate NPs. Human NPs are more salient than non-human NPs, and therefore more likely to trigger agreement, since agreement acts as an index for the salient participants of the construction (Croft 1986: 43). Thus, in order to determine if this characteristic of the noun motivates the regularization of haber, we code each noun as either human or non-human. b) Definite vs. indefinite NP: Bentivoglio (1993: 222) points out that lexical subjects tend to be definite in Spanish. In this same line, Keenan (1976: 319) mentions that in some Bantu languages, such as Kinyarwanda, subjects must be definite. Since haber pluralization entails a reanalysis of the NP as subject (as outlined in section 2) we predict that definite NPs will be more likely to trigger pluralization than indefinite NPs. We thus code each of our examples as definite or indefinite. Following Du Bois (1980), we consider definiteness to be a grammatical category. NPs preceded by the definite article, a demonstrative or a possessive pronoun are definite NPs. All the other NPs are indefinite. c) +/-numeral or indefinite quantifier in the NP: If haber pluralization is favored by increased saliency of plurality, the presence of plural markers such as numerals or indefinite quantifiers such as algunos 'some' would favor regularization. Conversely, the absence of such surface-level plural markers would not increase the probability of a pluralized form. For this reason we code each example for the presence or absence of numeric or indefinite quantifiers. d) Tense of haber (preterit vs. others): we code each example for the tense of the verbal form of haber. For the quantitative analyses, we distinguish preterit vs. others, on the basis of their aspectual meanings. Preterit forms convey perfective aspect, whereas all the other tenses (the vast majority of which are imperfect forms) convey imperfective aspect. e) Polarity: Du Bois (1980) indicates that negative clauses contain non-referential NPs, that is to say, NPs used to talk about the noun as a class and not as an object. Non-referential NPs are not sensitive to the singular/plural distinction. Therefore, in line with D'Aquino Ruiz (2004), we predict that affirmative clauses are more likely to trigger haber pluralization than negative clauses. For this reason, we code each example as affirmative or negative. f) We also code each NP for a factor best understood from within the usage-based framework from which we approach this analysis: proportion of noun use as subject. Within the exemplar model (Bybee 2001), we assume that instances of use in both production and perception influence lexical representations of words and that speakers' knowledge includes linguistic probabilities including the likelihood of a noun to be used in subject function. On this basis, we consider every noun that appears with haber in our corpora and its proportion of use with subject function generally in the language (Brown and Rivas 2012). This calculation is determined by dividing the number of occurrences of a noun with subject function by the total number of times it appears in the corpus 7 (tokens of use in subject function/total tokens of the noun). This data is discretized into three groups: high, medium and low. Tokens with values falling in the highest third are categorized as high (the noun is proportionally often used with subject function), those in the lowest third are categorized as low, and those in between are categorized as medium. Brown and Rivas (2012) show that the proportion of use as subject significantly predicts haber regularization. The more likely a noun is to be used in discourse with subject function generally, the more likely it is to trigger haber regularization. Brown and Rivas (2012) conclude the noun plays a pivotal role in the regularization of haber. g) Given the importance of the noun in constraining this variation, the present study further explores the potential role of other characteristics of the noun by considering the role of the SL-IL distinction manifested in the nouns. We limit this classification to specifically the semantic characteristics of the noun, as either bounded (SL) or unbounded (IL). For example, event nouns are SL (Ogawa 2001), because they are temporally bounded, that is to say, they refer to events that have an understood beginning and ending. Consider in this respect examples (11) and (12) with the event nouns elecciones 'elections' and ataques terroristas 'terrorist attacks': Davies, Habla Culta San Juan (11) porque fue cuando hubo, este... las elecciones 'because it was when there were, uhm... the elections' Cortés-Torres, Interview 15, 67 (12) Se cree que haya más ataques terroristas, 'People believe there are going to be more terrorist attacks' Other types of SL nouns include temporal nouns such as días 'days' and años 'years', nouns of communication or speech acts such as anuncios 'announcements', comentarios 'comments' -as in example (13) -and chismes 'gossip', as well as other deverbal nouns that also refer to bounded activities, such as contradicciones 'contradictions', as example (14)  Other nouns categorized as IL include animals such as caballos 'horses', concrete objects such as baterías 'batteries' and empanadas 'empanadas', and others such as lavanderías 'laudrymats'.
We also find in the corpus some nouns that are not easily classifiable as IL or SL. Examples include: cambios 'changes' (N=2), agencias de gobierno 'government offices' and puestos 'positions'. These nouns only constitute 2% of the data (N=4) and were excluded for this category from the quantitative analyses. The following section outlines the results of our quantitative and multiple regression analyses conducted on the haber + plural NP examples extracted from the two Puerto Rican corpora.

Results and Discussion
In order to determine which linguistic factor groups may favor pluralization of haber in Puerto Rican Spanish, we submit our data to a variable rule analysis using Varbrul (Rand and Sankoff 2001). This enables us to determine if a factor group makes an independent contribution to the analysis while controlling for all the other independent variables (Guy 1993). Through this analysis, we are able to determine the independent statistical significance of each factor group -determined by both a 'p' value and by the log likelihood (Sankoff 1988). Further, Varbrul enables us to determine the relative strength of each factor group. The greater the range of the factor group, the greater the magnitude of effect. The factor group with the greatest range, therefore, is the group that contributes most significantly to constraining the occurrence of a pluralized form of haber. Lastly, we can determine a constraint hierarchy through the Varbrul analyses. Within each factor group, the individual factors are ranked according to their factor weight. These weights reflect the degree to which they favor (> .50) or disfavor (< .50) the application of the dependent variable (in this case, pluralization of haber). Log likelihood -118.414, χ 2 per cell = 1.0328, p= 0.032 As is evident in Table 1, of the factors considered in the analysis (+/-human, definite vs. indefinite NP, +/-numeral or indefinite quantifier in the NP, tense, polarity, proportion of use as subject, SL-IL distinction), Varbrul selects as significant just two: proportion of noun use as subject and the SL-IL distinction. The usage-based factor group, proportion of noun use as subject, significantly constrains pluralization in our data. A noun frequently used as subject (e.g.; estudiantes 'students') favors pluralization at 56% and has a factor weight of .60, whereas nouns falling in the low range of percent use as subject (e.g.; fraternidades 'fraternities') highly disfavor pluralization (30%) with a factor weight of .36. Those nouns used with haber that fall in the mid-range (e.g.; juguetes 'toys') pluralize at a rate of 45% and only very slightly (factor weight .54) favor regularization. This result, which reflects syntactic probabilities, suggests the lexical representation of nouns contains grammatical relation probabilities in addition to other semantic information such as gender and number. This probabilistic knowledge of grammatical relations impacts haber regularization. The theoretical implications of this finding are discussed in Brown and Rivas (2012).
Importantly, the only other factor group that Varbrul selects as significant is the SL-IL distinction. The rate of pluralization of IL predicates is 53%, whereas SL predicates only pluralize in 33% of the cases. IL predicates favor pluralization with a factor weight of .59, whereas SL predicates disfavor pluralization with a factor weight of .40. Why would this be the case?
Our quantitative results show that the reanalysis of the NP from direct object to subject is promoted by IL nouns. The reason, we argue, is that IL nouns are more prototypical nouns than SL nouns. As is noted by Givón (2000), prototypical nouns have temporal stability, that is to say, "the properties of prototypical nouns change only little over repeated perceptual scans" (Givón 2000: 51). This characteristic is shared by IL nouns. IL nouns are also temporally stable, since they exist prior to and after the point of reference of the predication. In contrast, SL nouns are less prototypical nouns, since they show a low degree of temporal stability; they correspond with entities of relatively short duration.
Although both subjects and direct objects are typically realized by nouns (or NPs), temporally-stable (i.e., prototypical) nouns are more likely to occur in subject than in direct object function. As is noted by Keenan (1976: 312-313), one of the characteristics of prototypical subjects is independent existence, that is to say, subjects exist prior to the process expressed by the verb. In this respect, they differ from direct objects, which may be created by means of the activity expressed by the verb (e.g., Pedro hizo un pastel 'Peter baked a cake'. A cake exists as a consequence of the activity of baking). In addition to this, discourse-based studies such as Thompson and Hopper (2001: 33) show that in casual conversation many nouns in direct object position are actually part of V-O compounds, that is to say, a lexicalized combination of a low-content verb and a non-referential noun. Examples are have fun, get sleep, make sense, have time, have a green card, etc. 8 As Du Bois (1980: 209) points out, non-referential nouns make reference to "the quality defined by the noun" and are in opposition to referential nouns, i.e., nouns that are "used to speak about an object as an object, with continuous identity over time." Therefore, unlike nonreferential nouns, referential nouns are also temporally-stable. And, as is noted by Keenan (1976: 319), prototypical subjects are referential nouns.
Two of the characteristics of prototypical subjects, namely, independent existence and referentiality, are concerned with temporal stability. Since temporal stability is the defining criterion of IL nouns, IL nouns are therefore better candidates to act as subjects than SL nouns. As a result, when IL nouns occur in the plural, they are more likely to trigger pluralization of haber, because verbal agreement is one of the defining characteristics of subjects in Spanish. Plural SL nouns, on the other hand, will tend to occur with singular forms of haber.
Unlike in previous studies, our analyses do not select tense as significantly constraining variation of haber. Although results regarding the significance of tense are to a certain extent different according to the study, 9 the following are recurrent results: i) the imperfect is the tense that is most frequently reported to favor pluralization cross-dialectally (Bentivoglio and Sedano 1989;DeMello 1991;Díaz-Campos 2003;Quintanilla-Aguilar 2009), ii) the preterit is the tense that is most frequently reported to disfavor pluralization, especially in Venezuelan varieties (Díaz Campos 2003;D'Aquino-Ruiz 2004). 10 In line with previous studies, our results also suggest that the imperfect tends to occur frequently in the plural (habían), whereas the preterit tends to occur in the singular (hubo). As Table 2 illustrates, the imperfect occurs in the plural in 56% of 108 examples, whereas the preterit occurs in the singular in 70% of the 37 examples. However, as shown in Table 1, the logistic regression which measures the contribution of independent factors while controlling the effect of the other factor groups does not select tense as significant. Why would this be the case? We suggest that tense correlates with the SL-IL distinction. Studies that report a significant effect of tense may, in fact, be reporting a masked SL-IL effect. For instance, 62% of the preterit forms of haber (N = 37) combine with SL nouns, while only 33% of imperfect forms (N = 109) occur with SL nouns. Thus a study making only a tense distinction will detect a difference in rates of haber pluralization that may, in fact, be reflecting the semantic distinction instead. To test this proposal we conduct separate Varbrul analyses; one excluding the SL-IL factor group from analysis and the other including the SL-IL factor group but excluding tense. We present the results of these two separate Varbrul analyses in Table 3. As is evident on the right-hand side of Table 3, when the SL-IL factor group is excluded, tense is selected as significantly constraining the variation. Preterit forms strongly disfavor plural haber with a factor weight of .34 and low rates of pluralization (30%). Other tenses analyzed together (imperfect, present subjunctive, future) favor pluralization with a factor weight of .54 and higher rates of pluralization (47%). However, as is evident on the left-hand side of Table 3, when the data are analyzed including the SL-IL distinction, but to the exclusion of tense, the same finding is apparent as listed in Table 1. A comparison of the log-likelihood of the analyses (Paolillo 2002: 89-91) demonstrates that the semantic distinction (the SL-IL analysis on the left) yields a log-likelihood closer to zero (-118.587) than the analysis including tense (-120.892). Thus, the SL-IL analysis provides a better fit for the data. The SL-IL distinction better predicts the appearance of pluralized forms than the tense distinction in line with our current proposal (as is evident in Table 1).
Is there a connection between tense and the SL-IL distinction analyzed in this analogical variation? We suggest that the preterit vs. other distinction is the morphological counterpart of the SL-IL distinction. 11 Although, as is shown in Table  4, we find examples of other tenses in our corpora (present subjunctive, perfect tenses, future, conditional), the majority of our examples are in the imperfect (57%) or preterit (20%). In other words, 77% of the examples we find in our corpora are concerned with the preterit vs. imperfect distinction, which, rather than a tense distinction, is actually an aspect distinction, namely, perfective vs. imperfective. As is noted by Comrie (1976: 16), "perfectivity indicates a view of a situation as a single whole, without distinction of the various separate phases that make up the situation, while the imperfective pays essential attention to the internal structure of the situation." In this same vein, as mentioned above, the SL-IL distinction has been included within the category of inherent lexical aspect or Aktionsart. Both the imperfect and IL predicates are unbounded, the imperfect, by focusing on internal structure of the situation and IL predicates, by referring to temporally-stable entities. In contrast, both the preterit and SL predicates are bounded, because the former focuses on the situation as a whole and the latter are short-duration entities, and thus, temporally delimited.
Our quantitative results show there is a strong connection between preterit and SL predicates, on the one hand, and other tenses (especially imperfect) and IL predicates on the other. The preterit forms of haber predominantly (62%) combine with an SL predicate. As noted previously, SL predicates have fewer characteristics of prototypical nouns (e.g., they are non-temporally persistent), and thus do not highlight the noun as a possible subject candidate. Thus, we suggest that the disfavoring effect of preterit forms noted for haber regularization cross-dialectally could reflect the predominance of SL predicates in combination with this tense. Similarly, the nouns with an IL designation mostly combine with tenses other than preterit. Indeed, 77% of the imperfect forms occur with IL nouns, which, as has been shown, favor pluralization. Thus, the tendency for imperfect forms to favor pluralization that has been repeated in multiple studies may be a reflection of this correlation with IL predicates. As such, the incorporation of the SL-IL distinction provides an explanation of the tendencies described in previous analyses regarding tense.

Summary and Conclusion
This work addresses a widely studied phenomenon in Hispanic Linguistics from a new perspective, examining haber pluralization through the lens of the semantic stage-level ~ individual-level distinction. Such an approach enables us to identify a correlation between IL nominal predicates and pluralization. We examined all the tokens of haber with a plural NP extracted from two corpora of naturally occurring oral Puerto Rican Spanish (Davies 2002-;Cortés-Torres 2005) and submitted these tokens to variable rule analyses using Varbrul.
We find that haber pluralization in Puerto Rican Spanish is favored by two linguistic factors: proportion of noun use as subject, and the IL-SL distinction. Nouns that are frequently used as subjects generally in the language strongly favor pluralization of haber, as has been shown in Brown and Rivas (2012). The other linguistic factor to significantly constrain this morphosyntactic variation, the IL-SL distinction, has not been previously considered in a case of synchronic variation of this type. Results show that IL predications favor regularization. Unlike other accounts, tense (in particular, preterit) was not found to predict pluralization when considered in conjunction with this semantic distinction. Although the correlation between imperfect and IL predicates on the one hand, and preterit and SL predicates on the other, may partially explain the significance of tense in other studies, this present analysis finds that the SL-IL distinction is a more powerful predictor of this phenomenon of regularization.
We argue this significant result can be explained by appealing to the semantic differences between SL-IL nominal predicates. IL predicates are more prototypical nouns than SL predicates because the former are temporally persistent. IL predicates promote nouns' candidacy as subjects over direct objects because prototypical subjects present two temporally-persistent characteristics: independent existence and referentiality. As a result, IL predicates increase the likelihood of reanalyzing the direct object as subject, thus triggering agreement of the verbal form with plural NPs. SL predicates, on the other hand, because they display low temporal stability, inhibit regularization. Future analyses may be able to determine whether this distinction also holds for haber regularization in varieties other than Puerto Rico and for other cases of morpho-syntactic variation generally.