SUBJECT EXPRESSION IN A SOUTHEASTERN U.S. MEXICAN COMMUNITY

Through an analysis of immigrant Spanish in Georgia, potential contactinduced language change is investigated through the lens of subject pronoun expression. Pronoun variation among Mexican speakers is examined using sociolinguistic interview data. Tokens of overt and null subject pronouns (N = 4,649) were coded for linguistic variables previously shown to constrain subject expression (e.g. person/number, tensemood-aspect [TMA], polarity) as well as social variables (e.g. English proficiency, age), and then analysed using multivariate analyses in Rbrul. Results indicate an overall pronoun rate of 27%, which is slightly higher than what has been reported for monolingual Mexican Spanish. Several linguistic variables (e.g. person/number, switchreference, morphological ambiguity, polarity) and one social variable (age) played a significant role in pronoun variation. Moreover, differential effects were revealed when compared to monolingual Mexican Spanish for variables such as TMA. These findings point in the direction of dialect contact influences and the presence of a unique variety of Mexican Spanish in the U.S.


Introduction
The present study investigates the linguistic behaviour of newly-forming bilingual communities in the U.S., particularly at the initial stages of language contact between Spanish and English. Looking to the Southeastern U.S. as a relatively new contact situation, this analysis aims to understand the processes of language shift in situations of language contact and bilingualism. Latin American immigration to the Southeast has led to recent and significant demographic shift in this region, and thus substantial Spanish-speaking populations have emerged that historically were not part of the Southeast in the way that they were in the Southwest or Northeast. Due to the lack of an established presence, Southeast Spanish is understudied in sociolinguistic research compared to varieties in areas such as California, New Mexico, and New York (but see, e.g., Smith 2006;Carter 2014;Lamanna 2012;Wilson 2013;Montes-Alcalá & Sweetnich 2014;Limerick 2017). This study will focus on Spanish spoken in Georgia, observing the Latino community in Roswell, which is part of the Atlanta metropolitan area. Roswell represents the larger demographic shift in the Southeast, with recent emergence of Spanish-speaking communities, making it an excellent test site for studying emerging speech communities (Limerick 2017).
(1) Sí, ∅ estudié para abogada y… en la ciudad de México… [F39Mex] 1 'Yes, (I) studied to be a lawyer and…in Mexico City…' (2) Ah, no, yo lo estudié cuando recién llegué aquí [M34Mex] 'Uh, no, I studied it when I first arrived here' Preliminary research has explored SPE among ten Mexican speakers in Roswell (Limerick 2017), suggesting that bilingualism impacted their SP usage in relation to processes of grammatical simplification when compared to monolingual Spanish. Building on this research, the current study seeks to further explore potential early language contact effects in Roswell. I examine variables previously unexplored in this community and rarely studied in previous SP research (English proficiency, preferred media language, polarity, mood). Through a study of 20 first-generation Mexican immigrants, I investigate the frequency of and constraints on SPE to investigate contactinduced change.
The influence of linguistic and social factors on SP variation is analysed and compared with non-contact varieties of Spanish as well as other contact varieties by addressing the following questions: (a) What is the overall distribution of overt and null subject pronouns (SPs) in the Roswell data? How does the overt SP rate compare to that of non-contact varieties of Mexican Spanish? (b) How do language-internal factors condition SP variation and how do such patterns compare to previous research? (c) Do language-external factors (e.g. age, English proficiency) play a role in how speakers use SPs? (d) Does bilingualism play an influencing role in SPE? (e) What do the rates/constraints of SPE tell us about the emergent nature of Roswell Spanish?
The data employed for this analysis come from sociolinguistic interviews conducted in Roswell by the author. To answer the above questions, a quantitative analysis is carried out using variationist sociolinguistic methods (Labov 1972;Tagliamonte 2012). The article is structured as follows: Section 2 provides a background of Spanish in the Southeastern U.S. and Section 3 reviews previous studies of Spanish SPE. Section 4 outlines the current study's methodology and Section 5 presents the results. A discussion of the results is presented in Section 6, and Section 7 concludes the article.

Spanish in the Southeastern U.S.
Within the last 30 years, the Southeastern U.S. has seen immense immigration from Latin America, with many Southern cities experiencing growth rates in their Latino populations above 500% since 1990 (Winders 2005) and several states above 100%. For instance, between 2000 and 2011, Georgia, North Carolina, and South Carolina had growth rates of 102%, 119.6%, and 154.5%, respectively (Motel & Patten 2013). Such demographic shifts are not only seen in a few cities or states, but have been evidenced in the Southeast as a whole. Overall, the Southern U.S. had the highest growth rate (57.3%) of all regions of the country in the 2010 U.S. Census.
This influx of Latin American immigrants led to a widespread use of Spanish in Southeastern states which, during most of the twentieth century had been almost entirely English-speaking, such as in Georgia and North Carolina (Klee & Lynch 2009).
Some notable studies on Southeast Spanish include those on the imperfect/imperfect progressive distinction in North Carolina (Lamanna 2012) and the analysis of Carter (2014), which involved examining discourse concerning Spanish and U.S. Latinos in North Carolina. In Northeast Georgia, Smith (2006) studied codeswitching among Latin American immigrants. Moreover, Montes-Alcalá and Sweetnich (2014) studied language shift among Latinos in Atlanta. A few studies have examined Spanish in Roswell, Georgia, the city of focus for the present investigation (Wilson 2013;Limerick 2017). Wilson (2013), for example, examined narrative structure among Roswell speakers.
Regarding SPE in particular, Limerick (2017) examined pronoun use among ten Mexican immigrants utilizing Wilson's (2013) interview data. Limerick analysed SP patterns and considered possible English contact effects. He found that SPE was constrained by person/number, switch-reference, tense-mood-aspect (TMA), 2 lexical content, and clause type, the same factors (among many others) reported as significant in numerous other studies. Regarding potential English contact effects, little to no evidence was found as the overt pronoun rate was similar to that of monolingual Mexican Spanish reported in other studies, and, crucially, the social factors analysed (length of residency [LOR]; age of arrival [AOA]), were not significant. Nevertheless, a more indirect influence of bilingualism was suggested due to a reduced effect for the switch-reference constraint. This reduced effect was interpreted as a process of simplification on part of bilinguals whereby they were potentially losing sensitivity to switch-reference as evidenced by increased overt SP usage in same-reference contexts, a context that more strongly favours nulls in monolingual varieties. (See Section 3.2 for a discussion of investigating English contact through subject expression). To expand on this research, the present study draws from a larger corpus of interviews and more closely examines language contact effects by analysing additional factors (e.g. English proficiency; preferred media language) and comparing SPE with non-contact varieties of Spanish.

Pronoun Rates and Factors
Conditioning SPE It has been demonstrated that, in general, varieties from Mexico and Spain tend to have the lowest overt pronoun rates overall (16-25% overt), South American varieties such as Argentina, Colombia, and Chile exhibit mid-range rates (30-40% overt), and Caribbean speakers (e.g. Puerto Rico, Dominican Republic) have the highest pronoun rates out of all Spanish dialects (45-55%). 3 Regarding the factors that systematically guide SP variation, those shown to play a role across most dialects are person/number (Silva-Corvalán 1994a), same vs. switch-reference (Bayley & Pease-Alvarez 1996), clause type (Morales 1997), TMA (Cameron 1994), lexical content of the verb (Travis 2007), specificity of referent (Cameron 1992), reflexivity of the verb (Otheguy, Zentella, & Livert 2007), and priming (Travis 2007), as well as social factors including age and gender (Carvalho & Child 2011) (See Sections 3.3 and 3.4 for an in-depth discussion of these factors).

Pronoun Use and English contact
Contact varieties sometimes show higher overt pronoun rates than monolingual varieties, which is commonly attributed to English contact, as overt SPs are nearly obligatory in English but not in Spanish. Therefore, increased exposure to English has been argued to engender higher usage of overt pronouns in Spanish, with many studies finding higher pronoun rates among more established immigrants, the U.S. born, younger arrivals to the U.S., and those with higher English proficiency (e.g. Lapidus & Otheguy 2005;Otheguy, Zentella, & Livert 2007;Otheguy & Zentella 2012;Abreu 2012;Shin & Montes-Alcalá 2014). However, other researchers have not observed increased overt pronouns in U.S. contact varieties (e.g. Silva-Corvalán 1994a; Flores-Ferrán 2004;Torres Cacoullos & Travis 2010;Limerick 2017). The reason for this discrepancy has been explained in terms of differential social classes and social networks (Shin & Otheguy 2013;Shin & Van Buren 2016). Specifically, Shin and Otheguy's (2013) analysis of different nationality groups in NYC Spanish revealed that more affluent speakers, who tend to have looser social networks (Cubans and Colombians), had higher SP rates than less affluent speakers with more dense social networks (Mexicans, Dominicans, and Puerto Ricans). The latter group tends to have more tight-knit social networks and consequently maintain their linguistic patterns. Shin and Van Buren's (2016) study of US-born bilingual Mexican children in Washington/Montana corroborates this argument, finding little to no change in SPE when compared to monolingual adults. These authors posit that since the children were part of a more tight-knit community and were mostly in contact with monolingual Spanish-speakers, they retained their rates and usage patterns of SPs.
Furthermore, what constitutes evidence for language change (contact-induced or not) is debated among scholars. Poplack and Levey (2010) posit that changes in SP frequencies do not constitute evidence for change. They affirm that in order to consider contact-induced change, analysts must consider the ranking of factors that are probabilistically shown to favour or disfavour SP occurrence, that is, constraint hierarchies. Torres Cacoullos and Travis (2010:189) agree by stating that 'reliance on overall rates of use to determine contact-induced change is problematic because it is well known that regional dialects vary enormously' and suggest that divergences in grammatical patterning across dialects are necessary to substantiate contact-induced change.
Other scholars attribute changes in SP usage to English more indirectly, through a process of simplification on part of the bilingual, whereby they lighten the cognitive load of managing two linguistic systems by simplifying certain morphosyntactic or discourse constraints (e.g. Silva-Corvalán 1994b;Sorace 2004). That is, speakers may exhibit a weakening of certain constraints due to bilingualism in general, which indirectly is an influence from English in the U.S. context. Specifically, speakers have been shown to simplify pragmatic constraints on SP usage. For instance, various scholars have argued that Spanish-English bilinguals show a weakened sensitivity to switch-reference whereby they exhibit an increased use of overt SPs in same-reference contexts, an environment where nulls are more strongly favoured in monolingual varieties (e.g. Silva-Corvalán 1994a; Shin & Otheguy 2009).
I present below the reported overall overt pronoun rates for monolingual varieties (Table 1) as well as Spanish-English contact varieties (Table 2). Notice the crossdialectal variability in rates among monolingual Spanish as well as between monolingual and contact varieties.  Regarding contact Spanish, one can see from a comparison of Tables 1 and 2 that SP rates are sometimes higher, but sometimes equal or lower than rates for monolingual varieties, as highlighted above. For instance, higher rates were observed for New York Dominicans (48%) than were for Dominicans in Santiago (39%). Likewise, New York born/raised Spanish-speakers showed an increase (38%) when compared to New York Newcomers (30%). In contrast, we see in other cases that pronoun rates are virtually the same in contact varieties when compared to monolingual varieties, for instance in the case of North Carolina (17%) and Georgia Mexicans (21%) when compared to Yucatan (16%) and Mexico City (22%). Next, I discuss in detail the linguistic and social factors that are relevant to the present study.

Linguistic Variables
This section will outline the main findings according to previous SPE research, particularly for the variables employed in the current study. It is important to note that several additional factors not discussed here such as reflexivity, priming, and lexical frequency also greatly influence SPE (see, e.g., Travis 2005; Carvalho & Child 2011; Erker & Guy 2012 for further discussion).

Person/number
Person/number of the verb highly influences SP variation and has been shown to be the strongest predictor of variable SPE cross-dialectally (Orozco 2015). Specifically, first-person singular and third-person singular verbs tend to favour overt pronouns (e.g. Silva-Corvalán 1994a; Flores-Ferrán 2002;Shin 2012;Lastra & Martín Butragueño 2015). In fact, most studies have found that all singular forms in general are more likely to appear with overt SPs compared to plural forms (Orozco 2015).

Switch-reference
Switch-reference has also shown a strong influence on SP variation (e.g. Bentivoglio 1987;Cameron 1994;Silva-Corvalán 1994a;Bayley & Pease-Alvarez 1997;Travis 2005;Prada Pérez 2009;Torres Cacoullos & Travis 2010;Carvalho & Child 2011;Otheguy & Zentella 2012;Michnowicz 2015;Orozco 2015). Specifically, where there is a switch in subject referent, the SP is often overt; when there is no switch, null SPs are preferred. This pattern is generally thought to have a functional influence that has to do with referential tracking (Shin & Otheguy 2009) and facilitating interpretation of the antecedent for the listener (Cameron 1994). Certain contact varieties have shown a weakening of this constraint (Flores-Ferrán 2002;Shin & Otheguy 2009;Michnowicz 2015). For example, Flores-Ferrán (2002) found that NYC-born Spanish-speakers exhibit an increased use of overt SPs in same-reference contexts, demonstrating their loss of sensitivity to null pronoun usage in such contexts (e.g yo vivía aquí, yo me iba a Alpharetta, caminando, yo me iba a Sandy Springs 'I lived here, I would go to Alpharetta, walking, I would go to Sandy Springs' [M32Mex]). Furthermore, some researchers include an additional category, partial-switch, which analyses cases of switch in subject where the subject is coreferential with the immediately preceding object, as in (3)

Tense-mood-aspect (TMA) 5
Certain TMAs favour overt SPs while others favour nulls. Several studies have found that imperfects and conditionals favour overt SPs while presents and preterits are more likely to appear with nulls (Silva Corvalán 1982;Cameron 1994;Travis 2007;Carvalho & Bessett 2015). It has been proposed that imperfects and conditionals favour overt SPs due to their potential ambiguity. Since their first-and third-person singular verb forms are morphologically indistinct, the use of overt SPs would serve to disambiguate the referents of such forms (see Hochberg 1986). However, other studies have found no such correlation (e.g. Enríquez 1984;Bentivoglio 1987;Ranson 1991).
An alternative explanation has been proposed by Silva-Corvalán (2001), who discusses imperfects and preterits in relation to discourse functions and suggests that more overt SPs are used with imperfects due to the backgrounded nature of imperfect aspect, and that fewer overt SPs are used with preterits since they tend to foreground events. That is, for events that are more backgrounded the focus is on the subject (hence the use of an overt SP) whereas the focus is more on the action with more foregrounded events (hence the lack of an overt SP).

Morphological ambiguity
Similar to TMA, many researchers have shown that, when examining verb forms that are morphologically ambiguous, these forms favour overt SPs while unambiguous forms prefer nulls (e.g. Hochberg 1986;Travis 2007;Prada Pérez 2009;Erker & Guy 2012;Lastra & Martín Butragueño 2015;Michnowicz 2015). Ambiguous verb forms are those that have indistinct person inflections in the first and third-person singular, specifically in the imperfect, subjunctive (both present and past), conditional, and pluperfect. The idea is that speakers use overt SPs to clarify the referent for the listener. However, some scholars have found the opposite effect and argue that true ambiguity is rare and that it is the context that clarifies the antecedent. For instance, Ranson (1991) found fewer overt SPs with ambiguous verb forms relative to unambiguous forms and posits that contextual markers such as previous mention of the referent or background knowledge help to clarify the speaker's intended referent when the SP is null.

Verb class
Several researchers have found that verb class can also determine how an SP is manifested (Bentivoglio 1987;Silva-Corvalán 1994a;Travis 2007;Otheguy & Zentella 2012;Orozco 2015). It has been noted that verbs of psychological activity (e.g. creer 'believe'), verbs of communication, and copulas tend to be expressed with overt pronouns, with psychological verbs showing the highest probability. On the contrary, motion verbs tend to disfavour overts (Bentivoglio 1987;Silva-Corvalán 1994a;Travis 2007). Regarding psychological verbs, it has been hypothesized that overts are frequently used because these verbs tend to express the point of view of the speaker and because of the implied contrastive function that is often carried out in such contexts (Silva-Corvalán 1994a). Additional categories have also been employed for verb class, such as stative and activity verbs (e.g. Orozco & Guy 2008;Erker & Guy 2012;Otheguy & Zentella 2012;Orozco 2015). These studies have generally found that stative verbs favoured overt SPs while activity verbs favoured nulls.

Verbal mood
The influence of mood on SPE has not been extensively studied in previous research. The subjunctive mood has been examined in relation to specific tenses and aspects, but not so much in direct relation to the indicative and imperative moods in general. Lastra and Martín Butragueño (2015) have begun studying this variable in Mexico City Spanish, finding that non-indicative moods (subjunctive and imperative), disfavoured overt SPs. Regarding the indicative mood, while showing a higher frequency of overt SPs compared to imperatives, it showed neither a statistically significant favouring nor disfavouring effect on SPE.

Specificity
The use of a specific referent as opposed to a more general referent (e.g. impersonal tú), also tends to condition whether a pronoun is overt or null. Previous research has shown that the effect of specificity on SPE can depend on the dialect. Cameron's (1992Cameron's ( , 1993) work on San Juan and Madrid Spanish demonstrates that non-specific subject referents favoured overt SPs in San Juan while specific referents favoured overt SPs in Madrid. For Mexican Spanish, both Solomon (1999) and Michnowicz (2015) found that, like Madrid Spanish, overt SPs were favoured by specific referents.

Polarity
Polarity considers affirmative vs. non-affirmative clauses and has occasionally been analysed in previous research, with some researchers finding a significant effect (Lastra & Martín Butragueño 2015;Geeslin & Gudmestad 2016) and others reporting a lack of effect (Travis & Torres Cacoullos 2012;Torres Cacoullos & Travis 2015). 6 For example, Lastra and Martín Butragueño (2015) found that non-affirmatives (negatives and interrogatives) disfavoured overt SPs and that affirmatives showed a neutral effect. These authors give a possible explanation for this pattern in terms of the frequent clustering together of negative clauses in their data, which tend to be coreferential. Likewise, Geeslin and Gudmestad (2016) found that negation disfavoured overt SPs while affirmative clauses favoured them.

Style
Speech style has also been found to constrain SPE (e.g. Ávila-Jiménez 1996; Lastra and Martín Butragueño 2015). Ávila-Jiménez (1996) reported that more casual speaking favoured overt SPs in Puerto Rican Spanish. Similarly, Lastra and Martín Butragueño (2015) found that Mexico City speakers tended to use more overt SPs for more conversational styles, particularly when they had longer speech turns in the latter part of their interviews, as opposed to shorter speech turns at the beginning of the interview (less conversational), for which they used fewer overt SPs.

Age
Age is one of the main social variables that has been shown to influence SP usage. Carvalho and Child (2011) found a favouring of null pronouns among younger speakers in Uruguay (ages 16-29) while older speakers favoured overt SPs, a trend they hypothesise is influenced by the Portuguese substratum (p. 23). Similarly, Orozco and Guy (2008) observed that older Colombians favoured overt SPs while adolescents favoured nulls, which they explain as a possible change in progress attributed to greater access to education by young people compared to prior generations. They also hypothesise that these results are due to influence of the Colombian highlands dialect. Lastra and Martín Butragueño (2015) found the same effect whereby older speakers favoured overt SPs and younger speakers were more likely to produce nulls. These researchers raise the issue of a possible change in progress toward less overt pronoun use in Mexico City. Contrastively, Flores-Ferrán (2002) found that older speakers (50+) of Puerto Rican Spanish in NYC exhibited lower rates of overt SPs than younger speakers and attributes this to the tendency of older speakers to prefer a more conservative use of language following prescriptive grammar norms.

Gender
Numerous authors have found that females and males differ in their SP frequencies such that women use more overt SPs than men (e.g. Bayley & Pease-Alvarez 1996;Solomon 1999;Carvalho & Child 2011;Otheguy & Zentella 2012;Shin & Otheguy 2013;Alfaraz 2015), while others found gender to be nonsignificant (e.g. Orozco & Guy 2008;Lastra & Martín Butragueño 2015;Michnowicz 2015). For instance, in their study of Uruguayan Spanish, Carvalho and Child (2011) found that females favoured overt SPs whereas males favoured nulls. Similarly, Shin and Otheguy (2013) observed a higher rate of overt pronouns among women in NYC, which they attribute to women's extensive contact with U.S. born bilinguals. Bayley and Pease-Alvarez (1996) hypothesized that women use narrative discourse more often than men and thus more SPs. Moreover, Alfaraz (2015) suggests that this gender pattern, particularly in the Dominican Republic, is linked to prestige.

Educational level
A relatively understudied variable, educational level has occasionally been found to condition SPE. For example, Ávila-Jiménez's (1996) study of Puerto Rican Spanish found that overt SPs were disfavoured among the college educated and favoured among those with lower education levels. The few additional studies that have examined this variable (Otheguy & Zentella 2012;Lastra & Martín Butragueño 2015), while not finding a statistically significant effect, have reported trends consistent with Ávila-Jiménez's (1996) analysis. Lastra and Martín Butragueño (2015) suggest that this pattern could indicate that null SP use has more prestige.

Socioeconomic status (SES)
Socioeconomic status (SES) is another understudied variable in relation to pronoun use (exceptions include Otheguy and Zentella 2012, Alfaraz 2015, and Orozco 2015. While neither Orozco (2015) nor Alfaraz (2015) have found an effect for SES, Otheguy and Zentella (2012) found that speakers with lower SESs favoured overt SPs while higher SES speakers favoured null SPs among the Mainland population in NYC.

Length of residency (LOR), age of arrival (AOA), and English proficiency
Important factors to consider regarding potential influence of English on Spanish are LOR in the U.S., suggesting greater exposure to English, and AOA to the U.S., indicating speakers' susceptibility to change. Drawing from Veltman (2000), Otheguy, Zentella, and Livert (2007:779) state that older arrivals who speak Spanish more frequently and fluently should show greater resistance to English influence. In addition to finding that longer LORs in NYC correlated with higher frequencies of overt SPs, the authors found that younger arrivals, being more susceptible to English influence, exhibited higher pronoun rates than older arrivals, who tend to use less English and whose Spanish is more proficient, thereby making them less susceptible to the influences of language contact.
Additionally, English proficiency has been examined in studies of U.S. Spanish, with some researchers finding a positive correlation between English proficiency and overt pronoun use (e.g. Otheguy & Zentella 2012) and others finding no such correlation (e.g. Silva-Corvalán 1994a). Another factor to explore for examining contact effects, which has not been addressed in previous studies to my knowledge, is media language. It could be the case that those who prefer English for media exhibit a higher use of overt SPs than those who prefer Spanish. This factor is explored in the current study.

Data Collection
In 2015 I conducted sociolinguistic interviews with Spanish-speakers who were either living or working in Roswell, GA. The sample for the present investigation consists of 20 first-generation Mexican immigrants. Regarding the recruitment of participants, the majority took place at a local plaza containing several Latino businesses, including a grocery store, clothing stores, barber shops, restaurants, and many others. I also recruited participants at a local church and at an ESL school. The interviews lasted between 30 minutes and one hour. They were informal, conversational, and addressed topics of personal history, local community life, differences between Mexico and the U.S., and experiences adapting to life in Roswell, among others.

The Speakers
The speakers' sociodemographic backgrounds can be summarized in the following way (see Tables 3 and 4 (1), and the state of San Luis Potosí (1). They consist of 12 females and 8 males, and their ages range from 25 to 60. Additionally, their LORs in the U.S. range from 2 to 25 years (average = 12 years), and their AOAs range from 11 to 56 (average = 27). In terms of education levels, they range from primary school to university. The speakers have a variety of occupations, nearly half of them being small business owners. Finally, their English proficiency ranges from very poor to good. English proficiency was measured using speakers' self-ratings on a scale from 1 (very poor) to 5 (very good), with an average rating of 2.75. 7

The Variable Context
In order to analyse SPE, the audio data were first transcribed. I then extracted all finite verbs from the interviews in order to locate each instance of SP usage. In order to isolate only cases in which variation between an overt and null SP can occur in Spanish, I have excluded the following types of tokens that fell outside the variable context: verbs within subject headed relative clauses (e.g. mi mami este es una gran mujer ah que ∅ también nos ha inculcado…'my mom umm is a great woman uh that has also instilled in us…' [F39Mex]); verbs appearing with full noun phrases (e.g. Mi esposa es de, nacida en Virginia 'My wife is from, born in Virginia' [M27Mex]); existential structures (e.g. haber, ser); hacer + time expressions (e.g. ∅ hace diez años vine 'ten years ago I came' [M34Mex]); verbs with inanimate referents (e.g. ∅ son buenos 'they're good' [M30Mex, referring to tacos]); impersonal se expressions (e.g. en la casa ∅ se habla español 'at home Spanish is spoken' [F43Mex]); imperatives (e.g. siempre me dijo, "∅ estudia, ∅ agarra un libro…" 'He always told me, "study, get a book…"' [F52Mex]); set phrases where an overt or null SP was categorical (e.g. ¿Qué sé yo? 'I don't know'; ∅ digamos 'let's say'). Speakers did not alternate between an overt and null SP in any of the above cases; thus, these structures were excluded.
For all tokens within the variable context (N = 4,705), I first coded whether each verb appeared with a null or overt SP. Subsequently, I coded for the independent variables using the categorizations discussed in the following sections.

Linguistic Variables
Given that one of the primary objectives of this analysis is to investigate issues of contact-induced language change, I have chosen to include independent variables that would facilitate comparability between my results and studies of monolingual Mexican Spanish. Therefore, I have included the same linguistic variables that were observed by Lastra and Martín Butragueño (2015) in their study of Mexico City Spanish: 8 Person/number, Switch-reference, TMA, Morphological ambiguity, Verb class, Verbal mood, Specificity, Polarity, and Style (see Table 5). First part (shorter speech turns, more dialogic pairs, more interviewlike) 8 Two variables not included in the present study that were included by Lastra and Martín Butragueño (2015) are textual genre (argumentation, description, dialogue, narrative) and pronoun position (preverbal, postverbal). Textual genre was not included due to the scarcity of narrative tokens in the current data. Pronoun position, while presented as an independent variable in Lastra and Martín Butragueño (2015), is not in fact an independent variable and does not feasibly function as one. In an earlier version of their study, these researchers clarify that they analyse how other independent variables constrain pronoun position rather than employing pronoun position as an independent variable itself. This factor was not included in the current study.  Lastra and Martín Butragueño (2015). In the current data, first part tokens were those that occurred within the first 10-15 minutes of each interview when speakers were asked about their demographic background and when the questions were less open-ended. This is when the speakers' responses were generally much shorter. After the first 10-15 minutes, the questions were more open-ended and the speakers produced much longer responses. The interview flowed more like a conversation at this point, and tokens were coded as Second part.
Second part: longer speech turns by interviewee (more conversationlike

Social Variables
I have included the following social variables: Age, Gender, AOA, LOR, English proficiency, Preferred media language, Educational level, Socioeconomic status (SES) (see Table 6). The selection of these variables was primarily based on their inclusion in previous studies of SPE in Spanish, especially those in situations of language contact (Otheguy & Zentella 2012;Michnowicz 2015). 14 Certain variables, such as preferred media language, educational level, and SES, were also chosen due to there being relatively little research with regard to their influence on SPE. Regarding SES, I have adopted the operationalization of Otheguy and Zentella (2012:271). These researchers used a combined measure based on both education and occupation of the speakers. Specifically, speakers were given points ranging from 1 to 4 in the following way: Education: elementary (1), secondary (2), college (3), graduate (4) Occupation: unskilled (1), skilled blue collar, clerical (2), store owner, manager, white collar (3), professional, business owner (4). . 15 An anonymous reviewer asserted that including SES and education as separate variables is problematic in this particular study, especially since SES comprises only education and occupation (and not additional factors such as income, place of residence, etc.) as part of its operationalization. Additionally, s/he pointed out the redundancy of studying both education and occupation as part of SES. I thank the reviewer for these insights and will consider these issues in future research. For the current study, however, the methods were maintained as to facilitate comparability with previous research. 1-2 points 3-4 points 5-6 points 7-8 points

Statistical Methods
To determine the statistical significance of the linguistic and social variables and the relative weight of each factor regarding its conditioning on SP use, mixed-effects multivariate analyses (logistic regression) were carried out using Rbrul (Johnson 2009) with the inclusion of the speaker as a random effect. A multivariate analysis allows us to determine the relative effect of multiple factors at once. Specifically, it will facilitate the production of ranking for the statistically significant factors based on the amount of variation explained by each factor. Moreover, the quantitative results assist in creating a ranking of values within a single factor group or variable (e.g. person/number)-i.e. values within a factor group that favour overt SPs (e.g. first-person singular) vs. those factors that disfavour SPs (e.g. third-person plural). In addition, the inclusion of the speaker as a random effect in the multivariate analysis is a way of controlling for the individual speaker and ensuring that the results obtained are generalizable to the data set as a whole, and that the patterns are not due to particular speakers skewing the results

Pronoun Rate
Out of 4,649 total verbs, 1,239 appeared with overt SPs and 3,410 appeared with null SPs, indicating an overall overt pronoun rate of 27% for Mexicans in Roswell. When we compare this rate to other Mexican dialects (see Table 7), we see that it is marginally higher than monolingual Mexican varieties, but not far outside the range (16%-25%) of these varieties. Regarding the comparison with the benchmark variety of Mexico City, we see a five percent rate increase in Roswell, a finding that will be further discussed in Section 6.

Linguistic Predictors
In the multivariate analysis, six variables were selected as significant: Person/Number, Switch-reference, Morphological Ambiguity, Verb Class, Polarity, and Style. TMA, Specificity, and Verbal Mood were not significant. Table 8 shows the variable hierarchy of the significant factors from strongest to weakest effect. The topranked factor, Person/Number, has the most predictive power for variable SP usage while the bottom factor, Style, is the least powerful. The variable hierarchy was determined based on factor weight ranges, which are an indication of the relative strength of each variable (Tagliamonte 2006(Tagliamonte , 2012. If we compare this hierarchy to that of Mexico City, we see both similarities and differences in ranking. The similarities include the following: Person/Number and Switch-reference are ranked first and second, respectively, for both varieties. Additionally, Style is ranked last for both varieties. Finally, Specificity was not selected as significant in either study. The differences between varieties involve the following: First, Morphological Ambiguity is stronger in Roswell, and Polarity is weaker (falling closer to the bottom of the hierarchy). Secondly, some constraints were lost in Roswell, namely TMA and Verbal Mood. Moreover, other constraints emerge as significant in Roswell, namely Verb Class. 16 To facilitate comparison, only common factors employed between the studies appear here.

Roswell, GA (present study) Range
Mexico City ( An additional difference is revealed for Person/Number and its relation to Switchreference upon comparing the ranges. Particularly, we see that Person/Number, although ranking first in both varieties, appears to be much stronger in Roswell (54) than in Mexico City (32). Further, the predictive power compared to Switch-reference more than doubles in Roswell (Person/Number = 54 vs. Switch-reference = 26) and these two variables barely differ in relative importance for Mexico City (32 vs. 31). Table 9 below presents the constraint hierarchies for each factor group. The first column shows each factor group along with their particular levels, and the second column presents the factor weights (FW) for each constraint from highest to lowest probability of appearing with an overt SP. When a FW is closer to 1, this indicates a relative favouring of overt SPs. When it is closer to 0, it generally indicates a disfavouring of overt SPs (see Tagliamonte 2006:145, 156).

Morphological ambiguity
Also consonant with previous research (e.g. Hochberg 1986;Travis 2007;Prada Pérez 2009;Erker & Guy 2012;Lastra & Martín Butragueño 2015;Michnowicz 2015), ambiguous verb forms favour overt SPs with a FW of .60 while unambiguous forms promote null SPs with a FW of .40. This finding lends support to the Functional Hypothesis (Hochberg 1986), which predicts an increased use of overt SPs with ambiguous verbal morphology for first-person singular and third-person singular forms in order to compensate for the absence of explicit person marking.

Verb class
Regarding Verb Class, mental process verbs slightly favour overt SPs (FW = .57) and activity verbs favour null SPs (FW = .43). Stative/communication verbs show neither a favouring nor disfavouring, with a FW of .50. The pattern for mental verbs follows the trend in the variationist literature (e.g. Silva-Corvalán 1994a;Travis 2007;Torres Cacoullos & Travis 2010Otheguy & Zentella 2012;Orozco 2015) and is perhaps attributable to the implied contrast expressed by the speaker (Silva-Corvalán 1994a). The finding for activity verbs also aligns with previous studies (e.g. Orozco & Guy 2008;Erker & Guy 2012;Otheguy & Zentella 2012;Orozco 2015). However, the effect of stative and communication verbs is inconsistent with previous research since these verb classes showed a neutral effect in the present study and a favouring of overt SPs in other studies (e.g. Travis 2007;Orozco & Guy 2008;Erker & Guy 2012;Otheguy & Zentella 2012;Orozco 2015).

Polarity
Despite its overall modest effect, the results for Polarity show that affirmative clauses slightly favour overt SPs and that non-affirmative forms slightly favour null SPs. This finding agrees with the results of Lastra and Martín Butragueño (2015) and Geeslin and Gudmestad (2016). It is possible that subject coreference across contiguous negated clauses explains the favouring of null SPs (Lastra & Martín Butragueño 2015).

Style
Albeit the weakest effect, Style also significantly constrains pronoun use, with second-part style (more conversational) slightly favouring overt SPs and first-part style (shorter speech turns) slightly favouring nulls. This pattern is consistent with the findings of Lastra and Martín Butragueño (2015) and Ávila-Jiménez's (1996) findings showing a favouring of overt SPs with more casual speech.
It is worth noting that the Style effect could be influenced by the data distribution. There could simply be a higher occurrence of contexts that favour overt SPs in the second part of the interview compared to the first part (e.g. switch-reference), thereby producing the result of more overt SPs in more conversational speech. After examining such distribution, I indeed found this to be true, particularly for switch-reference and mental verbs, whose occurrences more than doubled for second-part tokens (e.g. 1485 switch-reference tokens in second-part of interview vs. 707 tokens in first part).

The unexpected TMA finding
Aside from the significant variables, one of the most striking findings emerges when we consider one of the non-significant variables, namely TMA. As discussed in Section 3, TMA has been shown to be significant in an overwhelming number of studies in the variationist literature, and with a very clear distinction: imperfect forms, which typically favour overt SPs, and preterit forms, which typically disfavour such forms. There is generally a wide percentage difference of overt SPs in these two contexts in the literature. The result in the current data, however, shows a drastic narrowing of this distinction with a virtually identical percentage of overt SPs in both contexts: 30% with imperfect and 29% with preterit (see Table 10). This divergence from previous data evidences a loss of the TMA constraint in the Roswell data, particularly with regard to the imperfect and preterit environments, and also contradicts Silva-Corvalán's (2001) hypothesis concerning the discourse functions of foregrounding and backgrounding. These functions are not observed in the present data since the speakers did not exhibit differential pronoun behaviour according to whether they used imperfect or preterit aspect. The alternative explanation discussed previously for the TMA effect has to do with the potential ambiguity of the subject referent. Despite the lack of statistical significance for the TMA variable, the current results do not necessarily contradict such explanation since there was indeed a significant effect for the Morphological Ambiguity variable.

Social Predictors
For the multivariate analysis of social variables one factor, age, was found to be a significant predictor of SP variation. None of the other social variables included-Gender, LOR, AOA, English Proficiency, Preferred Media Language, Socioeconomic Status, and Education-showed a significant effect. Table 11 shows the constraint hierarchy for Age. These results indicate that older speakers (50+) display the highest rates of SP usage at a 34% rate while younger speakers (35-49 and 20-34) slightly disfavour overt SPs at a rate of 24%. This distinction between older and younger speakers is generally in line with previous studies (e.g. Orozco & Guy 2008;Carvalho & Child 2011;Alfaraz 2015;Lastra & Martín Butragueño 2015;Orozco 2015Orozco , 2016, including those of other Mexican Spanish varieties (Lastra & Martín Butragueño 2015;Orozco 2016).

5.4
Length of Residency (LOR), Age of Arrival (AOA), and English Proficiency Although the factors of LOR, AOA, and English Proficiency were not selected as statistically significant, it can be informative to explore their general trends, particularly for shedding light on issues of language contact. Table 12 below shows the SP rate for each factor. Overall, the trends contradict a contact hypothesis based on English transfer. First, regarding LOR, the highest pronoun users are those that have been in the U.S. for the shortest amount of time (1-7 years) while more established immigrants have lower pronoun rates. Secondly, regarding AOA, the highest pronoun users are those with the oldest ages of arrival (30-56) whereas younger arrivals have lower rates. English Proficiency is difficult to interpret as there is not a linear trend. Nonetheless, we see no evidence in favour of English transfer (e.g. speakers who rated their English as 1-2 and 4 have virtually identical pronoun rates). These trends differ from what researchers have found for U.S. Spanish varieties under contact-induced change (e.g. Lapidus & Otheguy 2005;Otheguy, Zentella, & Livert 2007;Otheguy & Zentella 2012).

SP Rate Comparisons
Roswell speakers demonstrated an SP rate of 27%, which slightly diverges from most monolingual Mexican varieties. For instance, the rate found for Mexico City was 22% (Lastra & Martin Butragueno 2015) and the rate for NYC newcomers was 19% (Shin & Otheguy 2013). As stated above, however, 27% is still within the general dialectal range for Mexican Spanish; it is a relatively low pronoun rate when compared to other regional varieties, which is characteristic of Mexican Spanish. In comparing Roswell Mexicans to other contact varieties, Table 13 shows that Roswell Spanish falls approximately in the middle of the New Jersey (24%) and Los Angeles (29%) Mexican varieties. 18

Language-internal and Language-external Variables
Overt SPs were found to be favoured in the following environments: first-person singular and third-person singular verbs; switch in subject referent; morphologically ambiguous verbs; mental process verbs; affirmative polarity environments; and conversation-like speech style. In comparing these results with previous sociolinguistic studies, they are consistent with what other researchers have found, and also parallel the Mexico City data analyzed by Lastra and Martín Butragueño (2015). Three additional linguistic variables that were included in the present analysis but that were not found to be significant were Specificity, TMA, and Verbal Mood. One of these nonsignificant factors, TMA, was particularly striking as this was an unexpected finding given its widely-replicated significance in previous research. It appears that Roswell speakers are experiencing a loss of sensitivity to TMA; however, further research is warranted in order to explore the motivations for this finding.
The language-external factor that played a role in SP usage for Roswell speakers was their age. Younger and middle-aged speakers slightly disfavoured the use of overt SPs (24%), while older speakers showed a favouring of overt SPs (34%). This finding corroborates previous research that has reported a distinction between younger and older speakers (Orozco & Guy 2008;Carvalho & Child 2011;Alfaraz 2015;Lastra & Martín Butragueño 2015;Orozco 2015Orozco , 2016. This observation in the current data could be an incidental reflection of the differing regional origins of the speakers according to age. That is, it could be the case that the older speakers simply hail from regions in Mexico in which overall overt SPE is higher and that the younger speakers come from lower overt SPE regions. Further research of this issue is warranted. 19 This finding deserves further discussion, specifically as it relates to language continuity and dialect contact in Roswell. As similar age effects have been observed in monolingual Mexican varieties (Lastra & Martín Butragueño 2015;Orozco 2016), it could be the case that this pattern in the Roswell data simply reflects a continuation of trends that began in Mexico. Contact with other Mexican Spanish-speakers may be reinforcing the age effect, resulting in the maintenance of this trend for Roswell speakers and an inhibiting of the type of pronoun increases among younger speakers that might be expected in more established U.S. Spanish varieties. After all, there are several Mexican Spanish dialects in contact in Roswell, which are represented in the 18 It should also be noted that the current rate of 27% differs from the rate found in Limerick (2017), which also examines Mexican Spanish in Roswell (21% overt SPE). This rate discrepancy could be attributed to sample size and coding differences between the two studies. For instance, 20 speakers were employed in the current study while only 10 were included in Limerick (2017). Furthermore, there are differences in circumscribing the variable context between the studies (See Otheguy and Zentella 2007 for a similar perspective on discrepancies in findings). 19 I thank an anonymous reviewer for suggesting this potential explanation of the age effect. current sample. Further, the Roswell speakers reported having more interaction with Spanish-speakers than English-speakers overall in the sociolinguistic interviews. Thus, the speakers' SP patterns for age are likely being reinforced.

The Role of English/bilingualism
In terms of overall pronoun rates, we observed a marginally higher rate for Roswell Mexicans (27%) than what was reported for Mexico City Spanish (22%, Lastra & Martín Butragueño 2015). Is this sufficient evidence to conclude that Mexicans in Roswell have been influenced by English in terms of pronoun rate? From the perspective of this investigation, an increase in overall pronoun rate alone is not sufficient to make a determination regarding contact-induced change. As previously noted, using overall frequencies to assess language contact has been called into question (e.g. Poplack & Levey 2010;Torres Cacoullos & Travis 2010), and a similar perspective is adopted here. Specifically, it has been widely observed that frequencies of overt SPs differ substantially across dialects (e.g. Poplack & Levey 2010;Torres Cacoullos & Travis 2010;Otheguy & Zentella 2012;Alfaraz 2015;Orozco 2015), and it cannot be ruled out that the increased usage observed in the current data is not simply a reflection of regional variation. 20 In fact, it was shown earlier that within Mexico SP rates vary. Could it be then that the differing regional origins of the current speakers explain this increase when compared to Mexico City? Subsequent calculations of the pronoun rate for exclusively Mexico City speakers (N=8) vs. speakers of other Mexican dialects (N=12) in the current data revealed the following: Mexico City speakers alone show a rate of 24% (closer to the 22% benchmark ), and the remaining speakers, 28%, falling closer to the overall average (27%). Thus, the data suggest that dialect contact is playing a role. Since MC speakers are in contact with other Mexican speakers, these trends for frequency that have a source in Mexico appear to be reinforced due to the dialect contact situation. For example, the source dialects of the Roswell speakers may also exhibit SP rates at the level of 27%. The evidence from the current data suggests, then, that the higher rate observed for the non-Mexico City dialects is what is driving up the overall pronoun rate. The different regional Mexican varieties in contact with each other in Roswell would be reflected in the overall rate of 27%.
Furthermore, upon observing trends in the current data for pronoun rates in relation to social factors such as LOR, AOA, and English proficiency, we do not see direct correlations. Longer established immigrants or those with higher English proficiency do not have higher pronoun rates. In fact, the trends observed pointed in the opposite direction of what an English transfer hypothesis would predict (Silva-Corvalán 1994a; Otheguy & Zentella 2012). The highest pronoun users were the least established immigrants and the oldest arrivals to the U.S. In addition, speakers with both the lowest and highest self-reported English levels had virtually the same overt frequencies.
Turning now to the linguistic constraints on SP variation, does bilingualism seem to be playing a role here? When comparing variable hierarchies between Mexican Spanish in Roswell and Mexico City Spanish, speakers in Roswell demonstrated some distinctions in relative position in the hierarchy and range values. For instance, the variables of morphological ambiguity and person/number, were stronger in Roswell, the range for morphological ambiguity was 20 (ranking third overall) compared to 13 (ranking sixth overall) in the results reported by Lastra and Martín Butragueño (2015). For person/number, a range comparison relative to switch-reference in both varieties revealed that the gap was much wider in Roswell (range difference of 28) than in Mexico City (range difference of 1), showing that person/number is a much more important factor than switch-reference for Roswell speakers whereas the predictive power of the two variables is about the same for speakers in Mexico City. In comparing this finding with other monolingual dialects, however, we find that the Mexico City data is an exception. Other studies of monolingual Mexican Spanish mirror the current findings in that person/number is undoubtedly the strongest and the gap in strength between person/number and switch-reference is notably wide (Solomon 1999:250;Michnowicz (2015:112); Orozco (2016:5,7). 21 Contrastively, the gap in the Mexico City data is almost non-existent, indicating a rather atypical pattern for Mexico City speakers. Thus, Roswell speakers are actually showing patterns more like most other monolingual dialects. Another constraint divergence was for polarity. This was shown by the range value of 10 in the current results, which is more than twice as weak as the strength observed in Mexico City (23). Finally, TMA was not significant in the Roswell data, but was significant for Mexico City monolinguals.
What do such divergences suggest in terms of the potential linguistic processes at work for speakers in the current data? As discussed in previous literature, processes such as simplification--the weakening of discourse or morphosyntactic constraints to lighten the cognitive load (Silva-Corvalán 1994b; Sorace 2004)--as well as complexification (the emergence of constraints) (Shin 2014) can sometimes play a role as an influence of bilingualism in other contact varieties of Spanish. However, for the present analysis, there is little evidence of bilingualism having such effects. Rather, as in the case for SP frequencies, the evidence here suggests that any differences in constraints are simply due to dialectal variation in the data. Crucially, since English proficiency (level of bilingualism) was not shown to be a significant factor, there is no evidence that bilingualism is playing a role. Despite the divergences in the variable hierarchies and the apparent weakened sensitivity to certain constraints for Roswell speakers, we would only be able to point in the direction of a bilingualism effect if we had seen correlations between speakers with the highest English proficiency and higher pronoun rates, a piece of evidence absent from the current investigation. 22 .
In addition, we see ample evidence of linguistic continuity in the variable pronoun system between Mexico City and Roswell speakers. It was found that the first two variables in the hierarchy were the same and in the same order according to importance: Person/number ranked 1 st and switch-reference ranked 2 nd in both studies; style was ranked last, showing the least amount of importance for both varieties; and the two varieties shared the lack of influence of specificity on SP variation. More importantly, upon comparing the constraint hierarchies (ordering of constraints within each variable), it was demonstrated that Roswell and Mexico City speakers share the same 21 Michnowicz (2015): 57 vs. 31; Solomon (1999): 43 vs. 11;Orozco (2016): 50 vs. 28 for verb class [switch-reference was not included in Orozco's study; verb class was the second strongest predictor]. 22 It may also be the case that differences in data distribution and configuration between the current analysis and that of Lastra and Martín Butragueño (2015) are motivating these constraint divergences in terms of strength of effect. For example, regarding person/number, more than half of the tokens in the Roswell data are 1sg (strongly favouring overt SPs); 1pl tokens (the second most frequent), which strongly disfavour overt SPs, account for 20% of the data. In Lastra and Martín Butragueño's (2015:43) study, however, these two extremes for person/number make up smaller proportions of their data (46% vs. 9%), which may make the Person/Number effect appear weaker when compared to the current study. patterns for each significant variable in terms of which contexts favour the use of overt SPs, and regarding the hierarchical ordering of these contexts. Therefore, considering the above comparisons, we have primarily a case of linguistic continuity in the Mexican Spanish of Roswell, Georgia. Since Roswell speakers use Spanish most of the time and are primarily in contact with Spanish-speakers who also have limited exposure to English, their SP rates and usage patterns are generally maintained. This is consistent with what has been found for other U.S. varieties with tight-knit social networks and limited exposure to English (Shin & Otheguy 2013;Shin & Van Buren 2016). Moreover, the differences in SP usage observed in the current data, rather than having their root in bilingualism and English contact, point toward a reflection of dialectal variation and dialect contact. Further research with more robust dialect groups is needed, however, to confirm this (see, e.g., Otheguy & Zentella 2012 for a substantive treatment of dialect contact effects).

Conclusion
Since no evidence of English contact or bilingualism was found, it is more likely the situation of dialect contact emerging in Georgia that promotes the formation of this unique variety. That is, it is not bilingualism that is driving this emergence, but rather the heterogeneous interaction and combination of different Mexican varieties. Being only in the initial stages of language contact (first generation immigrants) and not yet being a well-established variety (i.e. second or third generation immigrant Spanish) with extensive contact with English, we are not (yet) observing evidence of English contact influence. Rather, we are seeing structural continuity on one hand, while simultaneously seeing a reflection of signs of dialects in contact in Georgia regarding SPE, especially since the speakers overall interact more with other Spanish-speakers than English speakers. The present findings reveal insights into the linguistic mechanisms at work in the initial stages of contact (i.e. Spanish in Georgia as a site of recent emergence of Spanish-speaking populations), which is something obscured in the sociolinguistic SP research on Spanish in the U.S. that primarily examines more established varieties.
Future studies should investigate several issues. It would be useful to incorporate a wider range of linguistic variables and to also consider interaction effects between variables. Additionally, it is crucial to explore issues that will shed more light on the role of language/dialect contact. For instance, cross-generational data would be helpful in revealing how the Spanish of second-and third-generation immigrants within this community compares with that of first-generation speakers. Concerning dialect contact, additional data of other dialect groups are essential to fully explore this issue. Aside from the opportunity to study contact between dialects, employing speakers of diverse regional and national origins would contribute to a more representative analysis of Spanish-speakers in the Southeastern U.S.