DEFINING GRAMMATICAL COMPLEXIFICATION

There are various ways in which grammars can become more or less complex. One way is through changes that affect categorical phenomena: increases or decreases in complexity can result from the addition or loss of linguistic structures or categories (e.g., Ferguson, Reference Ferguson, Obler and Menn1982:59–63; Trudgill, Reference Trudgill2011:15–32), as in the incorporation of case-marking into the Arawak language Tariana as a result of contact with Tucano (Aikhenvald, Reference Aikhenvald2003:3). A second way in which grammars can become more or less complex is through changes that affect variable phenomena. To detect this latter type of change, patterns of structured variation are compared across groups of speakers (e.g., Poplack & Levey, Reference Poplack, Levey, Auer and Schmidt2010; Poplack & Meechan, Reference Poplack and Meechan1998; Tagliamonte, Reference Tagliamonte2012:163ff; Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2011).Footnote ¹

How do we determine whether changes in patterns of structured variation are instances of complexification? An answer can be derived from Dahl's (Reference Dahl2004:24) method of measuring linguistic complexity: the “length of the specification [or description] . . . of the totality of patterns it contains.” Patterns that take longer to describe are more complex than patterns that are shorter to describe. Dahl's (Reference Dahl2004:46–48) analogy of a three-course dinner elucidates this idea. Like variation in grammar, the set of dinner choices is probabilistic and predictable, as choices may be influenced by factors such as hunger and health. If I am hungry, I might choose steak as my second course, which may in turn prompt me to choose a light appetizer. Pattern complexity increases with more choices and with more factors impacting the choices (hunger, diet, etc.). As choices and factors increase in number, so will the length of the description required to account for the dinner choices.

Returning to grammar, relative complexity in grammatical patterns can be assessed as follows: (a) the more choices of variants, the more complex the pattern; and (b) the more factors constraining the choice between variants, the more complex the pattern. Thus, I posit the following operationalization of simplification and complexification:

For example, assume that the choice to express or omit a subject pronoun is conditioned by factors x and y, but not z among first-generation US Spanish speakers, while this choice is conditioned by x, y, and z among second-generation speakers. In this case, second-generation Spanish has a grammatical constraint (z) that was absent in the Spanish spoken by the first generation. This operationalization of complexification fits well with Dahl's (Reference Dahl2004) definition because the gain of z increases the length of the description of the pattern (“conditioned by x, y, and z” as opposed to “conditioned by x and y”).

Complexification, as just defined, pertains to linguistic constraints guiding variation. So, if second-generation Spanish includes constraint x, but first-generation Spanish does not, then this is one way that the former is more complex than the latter. But, at the same time, second-generation Spanish can be lacking a constraint that is present in first-generation Spanish. That is, second-generation Spanish can be more complex than first-generation Spanish with respect to one constraint and, at the same time, be simpler with respect to a different constraint. In other words, grammatical simplification and complexification can proceed in tandem. In the current article, I assume this to be the case, that is, that the complexification in él/ella expression manifested as the emergence of sensitivity to TMA accompanies (and perhaps is triggered by) simplification of other constraints on él/ella expression.

VARIATIONIST RESEARCH ON SPANISH SUBJECT PRONOUN USE

The extensive sociolinguistic research on Spanish subject pronoun expression provides a wealth of information regarding patterns of use across varieties spoken in Latin America, Spain, and the United States. Six variables that constrain pronoun use are: reference, priming, semantic class, clause type, reflexive, and TMA.Footnote ²

Reference: Pronouns are expressed more often when the referent of two consecutive grammatical subjects in discourse is different (switch-reference) than when the subjects share the same referent (same-reference) (e.g., Cameron, Reference Cameron1993, Reference Cameron1995; Carvalho & Bessett, in press; Holmquist, Reference Holmquist2012; Otheguy & Zentella, Reference Otheguy and Zentella2012; Shin & Otheguy, Reference Shin, Otheguy, Lacorte and Leeman2009; Silva-Corvalán, Reference Silva-Corvalán1994; Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2011).

Priming: Subject pronouns are more likely to be expressed when the previous mention of the referent also appears as an expressed subject pronoun. The result is a clustering of the [pronoun + verb] construction, illustrated by (1).

1. (1) él se hace . . . él se mete en su dieta él mismo, no . . . él no toma cerveza. [311C]Footnote ³

   ‘he makes himself . . . . . . he puts himself on his diet, no . . . . . . he doesn't drink beer.’

The finding that pronoun use triggers further pronoun use has been interpreted as a priming or perseveration effect (Cameron & Flores-Ferrán, Reference Cameron and Flores-Ferrán2004; Carvalho & Bessett, in press; Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2011; Travis, Reference Travis2007).

Semantic class of verb: Pronoun rates tend to be highest with verbs of cognition, such as creer ‘to believe,’ and least common with nonmental activity verbs (Bentivoglio, Reference Bentivoglio1987:52, 60; Enríquez, Reference Enríquez1984; Erker & Guy, Reference Erker and Guy2012:541; Flores-Ferrán, Reference Flores-Ferrán2002; Otheguy & Zentella, Reference Otheguy and Zentella2012; Otheguy, Zentella, & Livert, Reference Otheguy, Zentella and Livert2007; Posio, Reference Posio2011; de Prada Pérez, in press; Silva-Corvalán, Reference Silva-Corvalán1994; Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2011:252; Travis, Reference Travis2007).

Clause type: Several studies have found that pronoun expression is most common in main clauses, less common in subordinate clauses, and least common in coordinate clauses (Abreu, Reference Abreu2009:125; Enríquez, Reference Enríquez1984:256-258, Otheguy & Zentella, Reference Otheguy and Zentella2012).

Reflexive: Pronoun omission is likelier with verbs that occur with a reflexive pronoun. For instance, yo is less likely to occur with a verb such as me baño ‘I bathe myself’ than with a verb such as bailo ‘I dance’ (Bayley & Pease-Álvarez, Reference Bayley and Pease-Álvarez1997; Michnowicz, in press; Otheguy & Zentella, Reference Otheguy and Zentella2012).

Tense/mood/aspect morphology: Certain verb forms, such as the imperfect (imperfective past), increase the likelihood of pronoun expression, whereas other forms, such as the preterit (perfective past) decrease it. Consider the singular forms of the three most common verb paradigms, illustrated with the verb bailar ‘to dance’.

As shown in Table 1, bailaba can be first-person singular (1sg) or a 3sg, but no such ambiguity exists in the simple present or the preterit.Footnote ⁴ Numerous studies have found that morphologically ambiguous forms as in the imperfect promote pronoun expression, whereas unambiguous forms as in the preterit do not (Abreu, Reference Abreu2009; Bentivoglio, Reference Bentivoglio1987:45; Carvalho & Bessett, in press; Flores-Ferrán, Reference Flores-Ferrán2002; Hochberg, Reference Hochberg1986; Hurtado, Reference Hurtado2005; Lastra & Martín Butragueño, in press; Michnowicz, in press; Orozco, in press; Otheguy & Zentella, Reference Otheguy and Zentella2012; de Prada Pérez, in press; Travis, Reference Travis2007). There is also evidence that this TMA effect is significant in switch-reference but not same-reference contexts (Cameron, Reference Cameron and Beals1994:34–37). Previous research, however, has not focused on one grammatical person in particular (except Travis, Reference Travis2007; see also Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2011, who study 1sg), thus it is unknown how TMA may affect the different grammatical persons.

Table 1.

1sg, 2sg, and 3sg forms in the simple present, preterit, and imperfect indicative

RESULTS: THE IMPACT OF SIX PREDICTORS ON ÉL/ELLA EXPRESSION

Tables 3 and 4 present variable and constraint hierarchies based on two logistic regressionsFootnote ¹¹—one for newcomers and one for NYRs—investigating the influence of reference, priming, clause, semantic class, reflexive, and TMA on 3sg pronoun expression. I use variable hierarchies to refer to the rankings of predictor variables such as reference or TMA, and constraint hierarchies to refer to the rankings of factors within each variable, such as main clause or preterit tense.Footnote ¹²

As to interactions between the predictor variables, for both groups of speakers, reference by clause was significant (newcomers: p < .0001; NYRs: p = .001) and reference by TMA approached significance among NYRs for preterit (p = .066) and imperfect verbs (p = .066). Because previous research indicates that the TMA effect impacts switch-reference but not same-reference contexts (Cameron, Reference Cameron and Beals1994), the reference by TMA variable was included in the regressions. Another issue that was taken into consideration was the possibility of multicollinearity, which does not appear to be a problem in the current study, as demonstrated by the measures in Appendix B.

The two variable hierarchies presented in Table 3 are constructed based on logistic regressions run in SPSS and Rbrul. SPSS provides a Wald statistic for each variable, which tests the significance of individual coefficients (Menard, Reference Menard2002:43, Otheguy & Zentella, Reference Otheguy and Zentella2012:158–161). In Table 3, the variables are ranked in order of highest to lowest Wald value. Statistical significance values are listed under the column heading p. The significance cutoff value for these analyses is .01 due to the Bonferroni adjustment needed when including six predictor variables (Johnson, Reference Johnson2009a:363). Reported at the bottom of the table are (a) Nagelkerke R², which indicates the amount of variability explained by the logistic regression model as a whole; (b) % correct (baseline), which is the percentage of correctly predicted cases as compared with the baseline prediction and indicates the predictive efficiency of the model; (c) deviance, which measures how well the model fits the data (the smaller the deviance, the better the fit); (d) intercept, which provides a baseline for building the model predictions and is similar to the input value provided by Goldvarb; and (e) centered input probability, which is the inverse logit of the model intercept and reports the overall prediction of the model. (For information regarding measures a and b, see Menard, Reference Menard2002:18–41, and Szmrecsanyi, Reference Szmrecsanyi2005:119; for measures c, d, and e, see Johnson, Reference Johnson2009b, and Tagliamonte, Reference Tagliamonte2012:140–143).

The primary point to take away from Table 3 is that TMA predicts él/ella expression among NYRs but not newcomers. At the same time, the speaker groups are similar with respect to the other five variables: él/ella is conditioned most strongly by priming/reference, followed by clause (and reference interacts with clause for both groups). Reflexive does not reach significance, and semantic class does not condition él/ella for either group.Footnote ¹³ The speaker groups are also similar in terms of how much variability is predicted by the regression model, approximately 15% for each group, per the Nagelkerke R² values. The deviance and centered input probability values indicate that the model is a better fit for the NYRs than for the newcomers. To summarize, the hierarchies in Table 3 tell us that newcomers and NYRs are similar, because several variables are significant for both groups. At the same time, there is also evidence for complexification because TMA is significant among NYRs but not among newcomers.

Table 4 presents the results from the same regressions reported in Table 3, but this time with more detailed information for each factor. Each variable is listed in boldface with its factors underneath. For example, priming includes three factors: (i) previous mention of referent was subject pronoun él or ella, which is denoted by ‘él/ella >’; (ii) previous mention was an omitted subject pronoun, denoted by ‘Ø >’; and (iii) other, which were cases not considered relevant for priming. Also included in Table 4 are factor weights, log odds, p-values, and range. Factor weights (FW) indicate the probability that the dependent variable will occur. The closer to 1.0, the greater the probability that él/ella will be expressed; the closer to 0, the greater the probability that él/ella will be omitted. Log odds are coefficients that, like FWs, express the probability that the dependent variable will occur. Positive log odds indicate that a factor favors él/ella expression; negative log odds indicate that a factor favors omission. The p-value(s) indicate the statistical significance associated with each factor. Range is calculated by subtracting the lowest factor weight from the highest. Generally speaking, the greater the range, the stronger the variable (Tagliamonte, Reference Tagliamonte2012:123).

The constraint hierarchies in Table 4, like the variable hierarchies in Table 3, show similarities between newcomers and NYRs in terms of how priming, reference, and clause impact él/ella expression. For both groups, the same factors are significant and the direction of the effects is the same. Expressed él/ella promotes a subsequent expressed él/ella, while omitted él/ella promotes a subsequent omitted él/ella, which can be interpreted as evidence of priming. Also, for both groups, switch-reference promotes él/ella expression, same-reference promotes él/ella omission, and main clauses promote él/ella expression. Nevertheless, and of primary importance for the current article, the results in Table 4 underscore the evidence for complexification: NYRs are sensitive to TMA, while newcomers are not. Moreover, the results pinpoint the complexification: NYRs' él/ella expression is constrained by TMA-imperfect and TMA-preterit, while for newcomers, neither imperfect nor preterit verbs significantly predict él/ella expression. This difference between newcomers and NYRs is also clear in the rates of él/ella expression in each TMA category. Newcomers' rates are nearly identical across TMA categories (34%, 33%, 33% for imperfect, present, and preterit, respectively; [X ² (1,2) = .12, p = .94 (ns)]), whereas NYRs express significantly higher él/ella rates with imperfect (51%) than with present (47%) or preterit (41%) verbs (X ² (1,2) = 10.67, p = .005).

EXPLAINING THE EMERGING TMA EFFECT IN SPANISH IN NYC

Before exploring reasons for why complexification arises in the grammar of second-generation speakers, we must first understand the TMA effect itself. Why do speakers choose to express pronouns with verbs in the imperfect and to omit pronouns with verbs in the preterit? There are currently two explanations in the literature, one focusing on the avoidance of ambiguity and the other on the foregrounding/backgrounding functions of the verb tenses. In the following sections, I discuss each of these explanations.

TMA effect as a strategy to avoid ambiguous reference

The most common explanation for the TMA effect on pronoun usage focuses on verb ambiguity; pronouns are expressed more often with the imperfect than with the preterit due to the greater amount of morphological ambiguity in the former than in the latter (see Table 1). There are two lines of evidence indicating that morphological ambiguity triggers pronoun expression primarily in contexts where reference-tracking is hindered by other variables. The first is that the TMA effect applies in contexts of switch- reference but not same-reference. The second is that the tendency to express él/ella is related to contextual ambiguity and the potential for such ambiguity is greater with imperfect verbs.

The first line of evidence that the TMA effect serves to disambiguate reference comes from the link between TMA and reference. As mentioned earlier, Cameron (Reference Cameron and Beals1994:34–37) found that the TMA effect was significant in switch-reference but not same-reference contexts. In the current study, TMA interacts with reference among the NYRs in the same way: in the NYRs' regression analysis, the TMA*reference variable approached significance for preterit (p = .066) and imperfect verbs (p = .066). The interaction is more clearly illustrated by bivariate analyses comparing same-reference and switch-reference contexts, which are presented in Table 5.

Table 5.

Rates expressed él/ella by TMA, same-reference and switch-reference contexts

Table 5 shows that NYRs are highly sensitive to TMA for switch-reference (X ² (1,2) = 15.57, p < .0001), but not same-reference (X ² (1,2) = 3.46, p = .18 [ns]). In switch-reference contexts, NYRs' TMA effect is robust: the él/ella expression rate reaches 69% with imperfect verbs, a rate that is 18 percentage points higher than the rate with preterit verbs. In contrast, TMA did not impact newcomers' use of él/ella, even when separating same-reference (X ² (1,2) = .82, p = .66 [ns]) from switch-reference (X ² (1,2) = .53, p = .77) contexts. The finding that TMA matters especially when switching reference tells us that NYRs pay attention to TMA when it is helpful to do so; when speakers switch reference, the referent becomes less accessible and more information must be encoded in the subject noun phrase in order to identify the referent (e.g., Ariel, Reference Ariel1990; Diver, Reference Diver, Huffman and David2011:257).

The second line of evidence showing that the TMA effect is related to avoiding ambiguity comes from research showing that speakers increase their use of pronouns in contexts of “contextual ambiguity,” that is, when the discourse context hinders reference-tracking due to, for example, competing referents (Hurtado, Reference Hurtado2005:344–346; Ranson, Reference Ranson1991:140–148). To illustrate contextual ambiguity, consider example (3), in which a Colombian woman describes how she learned English upon arriving in the United States at the age of 3. Example (3) has been altered from the original version; all subject pronouns have been removed. The original version, in which the subject pronouns have been reinserted, is presented in (4). Relevant contexts are underlined and the symbol “?” is used in the English version to denote ambiguous reference.

1. (3) No . . . ^aØ no tuve mucho . . . ah . . . problema con el inglés, porque como mi hermano . . . ^bØ tenía un hermano mayor, y eh . . . E ^cØ iba a la casa y ^dØ hablaba y así mismo ^eØ me enseñaba el inglés. Más, ^fØ hablaba con el otro hermano mío, entonces ^gØ aprendí el inglés, más o menos.

   No . . . ^a(I) didn't have much . . . ah . . . problem with English because my brother . . . ^b(I) had an older brother, and uh, . . . And ^c(?) went to the house, and ^d(?) talked and just like that ^e(?) taught me/myself English. Actually, ^f(?) talked with my other brother, so ^g(I) learned English, more or less.

The morphology of some pronounless verbs in (3) uniquely marks the person and number of the referent. The verbs in contexts a and g are unambiguous because they are morphologically marked for 1sg. But c, d, e, and f are both morphologically and contextually ambiguous.Footnote ¹⁴ Who “went to the house, and talked”? Who “taught” the participant English? Who “talked with [her] other brother”? Ambiguity arises primarily because there are competing referents (the speaker and her older brother) for the imperfect verbs in c, d, e, and f. Now consider (4), which is what the speaker really said.

1. (4) No . . . ^aYo no tuve mucho . . . ah . . . problema con el inglés, porque como mi hermano . . . ^byo tenía un hermano mayor, y eh . . . . Y ^cél iba a la casa y ^dél hablaba y así mismo ^eme enseñaba el inglés. Más, ^fél hablaba con el otro hermano mío, entonces ^gyo aprendí el inglés, más o menos.

   No . . . ^aI didn't have much . . . ah . . . problem with English because my brother . . . ^bI had an older brother, and uh, . . . And ^che went to the house, and ^dhe talked and just like that ^e(I? he?) taught me/myself English. Actually, ^fhe talked with my other brother, so ^g(I) learned English, more or less.

In (4), the inclusion of él identifies the older brother as the subject of contexts c, d, and f.Footnote ¹⁵ By saying él, the speaker increases the likelihood of communicative efficacy.

Another example of the greater potential for ambiguous reference with verbs in the imperfect is presented in (5). “P” indicates the turn of the participant; “I” indicates that of the interviewer.'

1. (5)

   P: . . . el día que yo me gradúe y le enseñé a mi mamá que . . . Por el día del orgullo de mi mamá de verme graduar de xxxx.

   I: Sí.

   P: Ella siempre quería eso.

   P: . . . the day that I graduated and I showed my mom that . . . For the day of my mom's pride from seeing me graduate from xxxx.

   I: Yes.

   P: She always wanted that.

Had the speaker not included ella ‘she’, the likeliest subject referent of quería would have been yo ‘I’. Both (4) and (5) show that communication can be hindered due to the fact that imperfect morphology does not distinguish between 1sg and 3sg. Expressed pronouns, then, help disambiguate reference. In contrast, potential ambiguity between 1sg and 3sg does not arise with preterit verbs due to distinctive person morphology.

To investigate the link between the TMA effect and contextual ambiguity, I coded 3sg verbs in the preterit or imperfect for potential contextual ambiguity. Drawing from Ranson (Reference Ranson1991:142–145), potential contextual ambiguity was operationalized as no potential ambiguity and potentially ambiguous.

No potential ambiguity

The 3sg verbs are not potentially ambiguous if, even when expressed subject pronouns are removed, reference can be clearly established because the referent was identified (anywhere)Footnote ¹⁶ in the preceding discourse and there were no pragmatically viable competing referents,Footnote ¹⁷ as in (6) and (7).

1. (6)

   P: Estuvo casi, así como seis meses, por un tiempo internada, y por otro tiempo, como otros cuatro meses, así. . . .

   I: ¿Qué tenía?

   P: Ø Tenía . . . um . . . en realidad no saben bien que sea . . . que sea la causa. Pero lo que Ø tenía, eran convulsiones, como si fuera epilepsia. [315M]

   P: (She) was almost, like that for almost six months, admitted for a time, and another time, like another four months, like that . . .

   I: What did (she) have?

   P: (She) had . . . um . . . in reality (they) don't really know what is . . . what is the cause. But what (she) had, were convulsions, as if it were epilepsy.

In (6), the participant is talking about his sister who had been hospitalized. There is no potential ambiguity upon encountering the imperfect verb tenía ‘had’, even though morphologically it could be 1sg or 3sg. The establishment of the sister as the main referent in the discourse, coupled with the general knowledge that a person who was hospitalized is the person who ‘had something’, renders reference unambiguous.

Example (7) is also unambiguous; had the speaker omitted ella, nació would still refer to the daughter, not the father.

1. (7) . . . el papá de ella murió como a los ocho meses que ella nació. [311C]

   ‘ . . . her father died about eight months after she was born.’

The lack of potential ambiguity in (7) is due to general knowledge; if the father had died at the age of eight months, he would not have been a father. Notice that there were always two steps involved in coding for potential ambiguity in cases with expressed pronouns: first, the pronoun was omitted, and second, upon rereading the context, it was determined whether the referent was still clear. If the referent was still clear, the context was coded as not potentially ambiguous.

Potentially ambiguous

The 3sg verbs were coded as potentially ambiguous if reference could not be established via contextual cues. In these cases, reference could not be established after removing subject pronouns. Potential contextual ambiguity generally resulted from competition between 1sg and 3sg referents, as in (5) and (8).

1. (8)

   I: ¿Y tú eras la única americana en la escuela?

   P: Había otra muchacha no más.

   I: ¿Tú te relacionabas bastante con esa muchacha?

   P: No, porque ella era . . . ya mayor, ya era en octavo. [181C]

   I: and you were the only American in the school?

   P: There was only one other girl.

   I: And you hung out a lot with that girl?

   P: No, because she was . . . already older, (she) was already in eighth.

Removal of the pronoun ella in (8) obscures the intended referent. Had the speaker said No porque era . . . ya mayor ‘No, because Ø was . . . already older’, era ‘was’ would most likely have been interpreted as having a 1sg subject (yo era ‘I was’), and there is nothing in the preceding discourse to contradict that interpretation. It is the combination of ambiguous morphology (era can be 1sg or 3sg) and the existence of competing referents (also 1sg and 3sg) that render era potentially ambiguous. To review, cases (5) and (8) were coded as potentially ambiguous; cases like (6) and (7) were coded as not potentially ambiguous. Because we know that the difference between the imperfect and the preterit affects cases of switch-reference but not same-reference among NYRs (see Table 5), I focused the analyses of potential contextual ambiguity on 3sg preterit or imperfect verbs in contexts of switch-reference only.

Table 6 illustrates how potential contextual ambiguity impacts 3sg pronoun expression with preterit and imperfect verbs. Looking across the row labeled “Ambiguous,” we see that there were 63 and 104 cases of potential contextual ambiguity among newcomers and NYRs, respectively. The majority of these potentially ambiguous cases were the result of competing referents for both newcomers and NYRs alike (n = 97% and 91%, respectively).

Table 6.

Potential contextual ambiguity by él/ella expression, 3sg preterit and imperfect verbs in switch-reference contexts

Note: ^aThe percentage-point difference is calculated by subtracting % él/ella in unambiguous contexts from % él/ella in ambiguous contexts.

For both groups, imperfect verbs occurred with expressed él/ella at significantly higher rates in contextually ambiguous contexts than in unambiguous contexts (90% vs. 39% for newcomers [X ² (1,1) = 35.42, p < .0001]; and 91% vs. 51% for NYRs [X ² (1,1) = 28.54, p < .0001]). Newcomers also demonstrated sensitivity to contextual ambiguity with verbs in the preterit (X ² (1,1) = 24.06, p < .0001), but this effect was not significant among NYRs (X ² (1,1) = 3.05, p = .08). Overall the results indicate that newcomers express él/ella in order to avoid ambiguous reference regardless of verb tense, whereas NYRs' use of él/ella to avoid ambiguity manifests more strongly with imperfect than with preterit verbs. These results suggest that newcomers attune to discursive ambiguity, whereas NYRs attune to both discursive and morphological ambiguity. NYRs' higher rate of pronouns with imperfect verbs thus reflects a routinized, but not automatic or obligatory, strategy aimed at clear communication of the subject referent.

One objection to the ambiguity account, raised by Bayley and Pease-Álvarez (Reference Bayley and Pease-Álvarez1997:363–364), is that first-person plural nosotros ‘we’ and third-person plural ellos/ellas ‘they’ are also expressed more often with imperfect than with preterit verbs. Expressing these pronouns does not serve to avoid ambiguity, as there is distinctive person morphology for plural verb forms, including in the imperfect. But plural pronouns are rarely expressed in general. Diver (Reference Diver, Huffman and David2011:252) argued that “arbitrary [patterns] are usually dependent on non-arbitrary [patterns].” It is possible that patterns of usage that apply to singular pronouns, which are nonarbitrary in terms of function, are mapped onto their plural counterparts. The TMA effect with plural pronouns, then, may be an arbitrary pattern that arises based on the nonarbitrary pattern involving singular pronouns.

TMA and pronoun expression: Backgrounding and foregrounding events

There is robust evidence that in narratives perfective forms tend to foreground events, while imperfect forms signal background information (e.g., Bardovi-Harlig, Reference Bardovi-Harlig1985; Comajoan, Reference Comajoan, Salaberry and Comajoan2013; Dowty, Reference Dowty1986; Hopper, Reference Hopper1979; Longacre, Reference Longacre1996; Reid, Reference Reid, Kegel, Nash and Zaenen1977; Schiffrin, Reference Schiffrin1981; Tomlin, Reference Tomlin1985). This relationship between perfectivity and foregrounding on the one hand and imperfectivity and backgrounding on the other has been documented in Spanish (Flores-Ferrán, Reference Flores-Ferrán, Potowski and Cameron2007:303; Ozete, Reference Ozete1988; Silva-Corvalán, Reference Silva-Corvalán1983:765).

How might the foregrounding/backgrounding functions of verbs be related to subject pronoun expression? Silva-Corvalán (Reference Silva-Corvalán and Fernández1997:127, Reference Silva-Corvalán2001:161–163) proposed that the backgrounding function of the imperfect is more compatible with expressed subjects, whereas the foregrounding function of the preterit is more compatible with subject omission. Her idea is that pronoun expression draws attention to the referent of the grammatical subject and, by so doing, draws attention away from the event (the verb). If a speaker foregrounds an event (in the preterit), she might omit the subject so as not to take away attention from that foregrounded event. In contrast, a backgrounded event (in the imperfect) lends itself to the inclusion of an overt subject because there is no competition between the subject and event. Example (9) illustrates this idea.

1. (9) Yo venía de . . . no sé exactamente de dónde yo venía, yo sé que era en el tren E, . . . yo vivía aquí en la 47, me quedé en la estación de la 50, iba caminando así, como a las 10 de la mañana era y sentí una . . . un grito. [042]

   I was coming from . . . (I) don't know exactly where I was coming from, I know that it was on the E train, . . . I lived here on 47th, (I) got off at the 50th (street) station, (I) was walking like this, at about 10 in the morning it was and (I) heard a . . . a scream.

In (9), imperfect verbs venía ‘was coming’, vivía ‘lived’, iba caminando ‘was walking’ all describe the scene and denote background information. These imperfect verbs all occur with subject pronoun yo: yo venía, yo venía, yo vivía ‘I was coming’, ‘I was coming’, ‘I lived’. Foregrounded events are those that make up the skeleton of the story: me quedé en la estación de la 50 ‘I got off at the 50th street station’ and sentí un grito ‘I heard a scream’. The two foregrounded events, ‘got off at the station’ and ‘heard a scream’, are denoted by preterit verbs that occur without a subject pronoun.

While Silva-Corvalán's foregrounding/backgrounding explanation of the TMA effect on pronoun expression cleverly draws on the well-established tendency to manipulate TMA morphology in order to highlight crucial parts of a narrative, it has not previously been tested empirically. Therefore, I coded 3sg preterit and imperfect verbs in contexts of switch-reference to test the hypothesis that pronouns are less likely to occur with verbs that denote foregrounded events and more likely to occur with verbs denoting backgrounded information. First, narrative discourse had to be distinguished from non-narrative discourse. Following Comajoan (Reference Comajoan, Salaberry and Comajoan2013:311), the main criterion used to distinguish between the two types of discourse was temporal sequencing. Narratives tell a story where the main events unfold in a particular order. All 3sg preterit and imperfect verbs that occurred within narratives were coded for whether they denoted foreground or background events. In order to avoid circularity, and again drawing on temporal sequencing, an objective criterion was adopted to differentiate between the two event types: foregrounded events are temporally sequential; if the timing sequence is changed, the story does not make sense. In contrast, backgrounded events are not temporally sequenced, even though they are part of narrative discourse (Bardovi-Harlig, Reference Bardovi-Harlig1985:265; Comajoan, Reference Comajoan2001, Reference Comajoan, Ayoun and Salaberry2005, Reference Comajoan, Salaberry and Comajoan2013; Dry, Reference Dry1983; Schiffrin, Reference Schiffrin1981:47). An example is given in (10), in which the underlined verbs are the foreground of a narrative about being robbed.

1. (10) Entonces, yo veía que estaba así, así, y yo le dije, “¿Qué te pasa?” Y cuando miró así, me metió la mano en el bolsillo y salió corriendo con la cartera. [005U]

   So, I saw that (he) was like this, like this, and I said to him, “What's up with you?” And when he looked that way, he put his hand in my pocket and (he) went out running with my wallet.

In (10), the verbs dije ‘said’, miró ‘looked’,Footnote ¹⁸metió ‘put’, salió ‘went out’ are temporally sequenced. The speaker says something to the thief, then the thief gives a look, then he puts his hand in the speaker's pocket, and finally he runs away with the wallet. The story would not make sense if the thief ran away with the wallet before he put his hand in the speaker's pocket. In contrast, the two verbs in yo veía que estaba así ‘I saw that (he) was like that’ are not temporally sequenced. The story would still make sense if the order was ‘he was like that and I saw him’. Verbs that were not temporally sequenced but were in narratives were coded as backgrounded.

Rates of expressed él/ella with foregrounded and backgrounded preterit and imperfect verbs are presented in Table 7. There were 232 contexts (25% of the data) that were not in narratives; these could not be coded for backgrounding or foregrounding. Among the imperfect verbs that were in narratives, 297 (95%) denoted backgrounded events, which supports previous research linking morphological imperfectivity with the discursive function of backgrounding information. The preterit, on the other hand, appears to be more flexible: The majority of preterits in narratives were used for foregrounding, but 16% of newcomers' preterits and 30% of NYRs' preterits denoted background events.

Table 7.

The impact of foregrounding/backgrounding on él/ella expression, with 3sg preterit and imperfect verbs in switch-reference contexts

Analyses of él/ella expression in narrative sections did not support Silva-Corvalán's (Reference Silva-Corvalán and Fernández1997, Reference Silva-Corvalán2001) explanation for the TMA effect. Instead, foregrounded events occurred with higher, rather than lower, rates of pronouns. Newcomers expressed él/ella significantly more often with preterit verbs that were foregrounding an event (47%) than with preterits denoting background information (26%) (X ² (1,1) = 4.91, p = .03). The same trend was found among NYRs, but the slightly higher rate of él/ella with foregrounded preterits (51%) than with backgrounded preterits (47%) was not significant (X ² (1,1) = .34, p = .56). The relationship between pronoun expression and backgrounding versus foregrounding could not be tested for imperfect verbs because so few denoted foregrounded events. In summary, Silva-Corvalán's explanation of the TMA effect finds no support in the results presented here. In fact, the results point to the opposite tendency, that is, pronouns are expressed more often with foregrounded events, which suggests that participants do not compete with actions for attention. Instead, the event is construed as a unit involving the primary participant(s) and the activity. Drawing attention to the participant involved in the event via subject pronoun expression may serve to highlight the event as a whole.

To summarize, although there is strong evidence that perfective and imperfective verb morphology correspond to foregrounding and backgrounding in narratives, the results herein show no support for the idea that expressed pronouns are incompatible with foregrounded events. We are left with one coherent explanation for the TMA effect: it reflects a drive to avoid ambiguity in reference.

EXPLAINING COMPLEXIFICATION IN NYC

Having explained the TMA effect on pronoun expression in Spanish as an ambiguity avoidance strategy, I now discuss the reasons for the emergence of this particular strategy in Spanish in NYC. Why does TMA become a significant predictor of 3sg pronoun expression in second-generation Spanish? The answer provided in this paper is guided by a functional approach to language change, which highlights communicative needs as one reason for the adoption or retention of linguistic forms and strategies (Andersen, Reference Andersen, Lambert and Freed1982:97; Christiansen & Chater, Reference Christiansen and Chater2008; Kiparsky, Reference Kiparsky1982; Martinet, Reference Martinet1952; Nettle, Reference Nettle, Darnell, Moravcsik, Noonan and Wheatley1999; Shin & Otheguy, Reference Shin, Otheguy, Lacorte and Leeman2009). I have shown that second-generation Spanish speakers' sensitivity to TMA as a predictor of él/ella expression is related to the greater amount of ambiguity in person morphology in the imperfect than in the preterit, which in turn increases the likelihood of ambiguous reference. This suggests that the differential use of él/ella is directly related to the need to identify clear referents, which is part of a more general drive to communicate clearly.

If the complexification in Spanish in NYC is triggered by communicative needs, we might assume that something else is changing in the system that is pulling in the opposite direction, that is, a change that is obscuring communication. The idea here is that complexification in second-generation Spanish arises in order to compensate for other, concomitant changes that introduce potential obstacles for reference-tracking. What changes occur that obscure reference? Although a direct relationship between increasing sensitivity to TMA as a predictor of él/ella expression and other intergenerational changes in Spanish in NYC cannot be established conclusively in the current article, I will outline several possibilities that are worth exploring.

One explanation is that increasing sensitivity to TMA is related to decreasing sensitivity to reference. Recall that Otheguy and Zentella (Reference Otheguy and Zentella2012:163, 185) found evidence of intergenerational weakening of reference as a predictor of pronoun expression in analyses in which all grammatical persons were included. In the current study, reference was a strong predictor of él/ella expression in both the first and second generations. Nevertheless, Shin and Otheguy (Reference Shin, Otheguy, Lacorte and Leeman2009) showed that desensitization to reference occurs more readily with 1sg and second-person singular (2sg) verbs than with 3sg verbs. Even though weakening of reference primarily affects first-person and second-person forms and the increase in sensitivity to TMA affects third person, it is possible that the two changes are related. If so, this would suggest a type of trade-off: as second-generation speakers become less sensitive to some discourse constraints, they become increasingly sensitive to morphological factors. The interpretation that morphological factors remain pertinent or increase in strength while discourse factors undergo weakening is compatible with a body of research suggesting that, compared to morphological or semantic constraints on syntactic structures, discourse-pragmatic constraints are particularly susceptible to change (for discussion, see Sorace, Reference Sorace2012).

Another change that might be related to increasing sensitivity to TMA has to do with coda –s deletion. Otheguy and Zentella (Reference Otheguy and Zentella2012:165, 176, 186–189, 198), in their study of Spanish in New York, find that, compared to newcomers, NYRs pay less attention to coda –s as a predictor of 2sg (tú) pronoun expression. That is, among newcomers, 2sg verbs pronounced without coda –s (as in tú cantaba ‘you sang’) favor tú expression, whereas 2sg verbs pronounced with coda –s (as in tú cantabas ‘you sang’) favor tú omission. This tendency is weaker among NYRs.Footnote ¹⁹ Perhaps, then, as speakers move away from attending to ambiguity of tokens within a verb tense, they increase their focus on ambiguity of the verb tense as a whole. If this is correct, the complexification in the current study might represent a change from a strategy that targets specific verb tokens to one that applies at the verb paradigm level.

A third possibility can be derived from an observation about newcomers' lack of sensitivity to TMA as a predictor of él/ella. If the TMA effect represents an ambiguity avoidance strategy, then why is it absent from the speech of newcomers? One possible answer is that, for newcomers, imperfect verbs without a subject pronoun are assumed to refer to third-person referents by default. In other words, bailaba ‘danced’ is automatically interpreted as él/ella bailaba ‘he/she danced’. Then, if a newcomer deviates from the default, choosing a 1sg rather than a 3sg referent, he will signal this departure by including yo ‘I’ with the imperfect verb. Analyses of yo lend some support to this hypothesis, as newcomers produce significantly higher rates of yo with verbs in the imperfect (49% of 5060 verbs) than in the preterit (30% of 7129 verbs) (X ² (1,1) = 452.76, p < .0001). If newcomers' grammar includes a strategy of assigning 3sg reference as a default, then it is also possible that another intergenerational change in NYC is the loss of this default strategy. Seen this way, increasing sensitivity to TMA could be compensation for the loss of the default 3sg reference assignation.

A complete explanation for grammatical complexification in Spanish in NYC is likely to be multifaceted, potentially related to decreasing sensitivity to reference and –s deletion, and perhaps, too, to the loss of a default strategy. There are a number of other changes in Spanish in New York that are likely to occur simultaneously, some of which may hinder clear communication and, therefore, trigger other changes to increase communicative efficiency. Thus, the idea is that a decrease in attention to one factor coincides with an increase in attention to other factors. Put more generally, linguistic changes do not occur in a vacuum. Understanding the nature of how changes relate to each other is a worthy pursuit (see Erker, Reference Erker2012).

A final point worth mentioning is that it is not yet clear whether the changes detected in NYC are generalizable to other bilingual settings. Some additional evidence that sensitivity to TMA increases with bilingualism comes from Torres Cacoullos and Travis (Reference Torres Cacoullos and Travis2011:254) in their study of 1sg pronoun yo ‘I’ in New Mexico, where TMA is significant among bilinguals who use both Spanish and English regularly, but not among Spanish-dominant speakers who rarely use English. On the other hand, Silva-Corvalán (Reference Silva-Corvalán1994:156) found that first-generation and second-generation, but not third-generation, Spanish speakers in Los Angeles are sensitive to the impact of verb form ambiguity on pronoun expression. Thus, while the ongoing changes in NYC appear to be the result of increasing bilingualism, it is also possible that different bilingual settings yield different linguistic changes.