2.1 The contact site: Spanish and English in New Mexico

Northern New Mexico is home to “arguably the oldest continually spoken variety of Spanish anywhere in the Americas that has not been updated by more recent immigration” (Lipski, Reference Lipski2008, p. 193). Following settlement in 1598 from New Spain (what is, today, Mexico), Spanish speakers in the northern section of the state had minimal contact with speakers of other varieties of Spanish (Gonzales-Berry & Maciel, Reference Gonzales-Berry and Maciel2000, p. 4; Lipski, Reference Lipski2008, pp. 195, 202), developing over time their own distinct variety which we refer to here as Traditional New Mexican Spanish, following Bills and Vigil (Reference Bills and Vigil2008, p. 7). Traditional New Mexican Spanish is generally Mexican grammatically but has some “independently developed” words and phonetic features (Bills & Vigil, Reference Bills and Vigil2008, p. 15).

New Mexico became a Territory of the United States in 1850; however, English speakers were in the minority for longer than in the surrounding region: in 1890, 70% of the population of New Mexico could not speak English, a figure which had dropped to 33% by 1910 (Fernández-Gibert, Reference Fernández-Gibert, Rivera-Mills and Villa2010, p. 48). New Mexico was awarded statehood in 1912, and English increasingly displaced Spanish in the educational system (see Gonzales-Berry, Reference Gonzales-Berry and Maciel2000). Children were punished for speaking Spanish in schools, as recounted in (2), by one of the speakers in the NMSEB Corpus.

1. (2)

   

From the 1900s (with the Mexican Revolution, 1910–1920), immigration from Mexico has led to increasing contact with contemporary varieties of Mexican Spanish. While this augments the presence of Spanish overall, Traditional New Mexican Spanish is stigmatized in comparison with monolingual varieties, and thus the northward spread of Mexican Spanish also threatens the maintenance of Traditional New Mexican Spanish (Bills & Vigil, Reference Bills and Vigil1999, p. 56), as does the teaching of educated standard Spanish in schools as a foreign or second language (Gonzales-Berry, Reference Gonzales-Berry and Maciel2000). This disparagement of the local variety is displayed in example (3), about the speaker's granddaughter getting her homework – which the speaker had helped her with – marked as wrong. In this way, contact with English, with Mexican Spanish and with standard “school” Spanish are all said to endanger Traditional New Mexican Spanish (Bills & Vigil, Reference Bills and Vigil2008, p. 313; Travis & Villa, Reference Travis, Villa, Norrby and Hajek2011).

1. (3)

   

For students of language contact – the locus of which, as stressed by Weinreich (Reference Weinreich1963), is the bilingual speaker – the remaining speakers of Traditional New Mexican Spanish and English provide a precious window into bilingual speech phenomena, as their speech allows us to examine long-term grammatical repercussions of contact.

2.2 The community-based corpus

Speakers comprising the NMSEB Corpus are minimally third generation Nuevomexicanos “New Mexican Hispanos”. They are bilingual in that they regularly use both languages with the same interlocutor in the same domain, “the appropriate code for the Hispano community” in New Mexico (Gonzales, Reference Gonzales, Galindo and Gonzales1999, p. 29). Rather than administering a battery of proficiency tests, we adopt this criterion of regular use of both languages, as observed by the fieldworkers and subsequently confirmed in the recordings (see Poplack, Reference Poplack and Preston1993, p. 254).

Speech samples were recorded with the goal of best approximating observation of informal, everyday vernacular speech. Eight minimally third generation Nuevomexicano students of the University of New Mexico, who, importantly, were all community in-group members (see Clyne, Eisikovits & Tollfree, Reference Clyne, Eisikovits, Tollfree, Blair and Collins2001, pp. 235–236; Poplack, Reference Poplack and Preston1993, p. 260), conducted interviews with one or more family members or acquaintances, who make up the participants in the corpus. The fieldworkers were instructed to speak in both English and Spanish, switching between the two as they naturally would. Thus, the language switching and linguistic structures that occur in the corpus did not arise in response to direct elicitation from the interviewers.

We employed the technique of the sociolinguistic interview to elicit “narratives of personal experience” during which monitoring of speech is minimized and for which the participant, not the interviewer-researcher, is the indisputable expert (Labov, Reference Labov, Baugh and Sherzer1984, pp. 32–42). Covered were what can be considered general interest topics, such as childhood, family and work, as well as community-particular, or insider, topics, such as hunting bears, making adobe bricks and pole vaulting in the acequia (irrigation ditch). Often arising naturally in the recordings was information about the speaker's linguistic history and their language attitudes (such as experiences with use of Spanish and English in school, as illustrated in the excerpts in (2) and (3) above), providing insights into the social context within which the bilingual phenomena of interest arise.

The 23 NMSEB Corpus participants in the present report are all from northern New Mexico, were born between 1922 and 1964, comprise 14 women and 9 men and include miners, ranchers and schoolteachers; see Appendix B for particpants’ background. They can be considered “early” bilinguals, with the caveats that such ways of classifying bilinguals remain disputable (see Grosjean, Reference Grosjean1998, pp. 133–144), and, particularly relevant in the case at hand, “age of acquisition” estimates are subject to the vagaries of self-reporting for people who have grown up in bilingual communities. Participants report that their first language was Spanish (18/23), they learned English upon starting school (18/23) or (also) at home (5/23), they prefer to speak English (13/23), Spanish (6/23) or both (4/23), and they use both languages not just in the home, but also with friends and at work, corroborating the bilingual nature of the community in which these speakers live. (For an overview of the NMSEB Corpus and a sociolinguistic profile of participants, see Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2013.)

A corpus that can be used for accountable quantitative analysis requires exhaustive transcription of the audio data. Transcriptions were done in ELAN, a software program that aligns the audio file to the transcription (Lausberg & Sloetjes, Reference Lausberg and Sloetjes2009). We use standard orthography for phonetic variants but represent lexical and morphological variants of the community, for example muchita “girl”, seen in example (10) (vs. muchachita), and the first person plural suffix -nos, seen in (18) (vs. -mos). The transcription adopted for the NMSEB Corpus follows the method for discourse transcription outlined in Du Bois, Schuetze-Coburn, Cummings and Paolino (Reference Du Bois, Schuetze-Coburn, Cumming, Paolino, Edwards and Lampert1993) and is prosodically based, relying on the notion of the Intonation Unit (IU), “a stretch of speech uttered under a single coherent intonation contour” (Du Bois et al., Reference Du Bois, Schuetze-Coburn, Cumming, Paolino, Edwards and Lampert1993, p. 47). Each IU is represented on a distinct line in the transcription, followed by punctuation indicating its prosodic contour. The relevance of the IU, which has been described as the “most plausible basic unit of the grammar of spoken language” (Croft, Reference Croft1995, p. 875), will become apparent in Section 6.1. The present data constitute approximately 18 hours of speech (60,000 IUs and 202,000 words), of which approximately 15 hours (45,500 IUs and 196,000 words) were produced by the participants.Footnote ²

Having met the extralinguistic requirements for testing convergence impelled by code-switching – an appropriate contact setting and speaker sample – in the following section we delimit a linguistic variable that is an appropriate candidate for grammatical convergence, namely one which involves points of both overlap and discrepancy between the languages in contact.

3.1 Conditioning of variation to ascertain (change in) grammatical structure

Any pronouncement of convergence that is based on departures from an idealized monolingual norm as perceived by the analyst rests shakily not only on unverifiable characterizations of the “monolingual norm”, but also on the precarious equation of variation with change. But we know that, while all change involves variability, “not all variability and heterogeneity in language structure involves change” (Weinreich, Labov & Herzog, Reference Weinreich, Labov, Herzog, Lehmann and Malkiel1968, p. 188). How, then, can grammatical alteration be observed?

Linguistic variability is conditioned by contextual features, which contribute to speakers’ choices among the set of variants constituting a linguistic variable (Labov, Reference Labov1969) – here, speakers’ choice of Spanish pronoun yo vs. an unexpressed 1sg subject. The prediction of the convergence-via-code-switching hypothesis (e.g. Gumperz & Wilson, Reference Gumperz, Wilson and Hymes1971) has been that code-switching to English results in higher rates of expressed subjects in Spanish. Here, however, we do not rely on overall rates of linguistic forms alone, but look to the configuration of conditioning factors (see Poplack, Zentz & Dion, Reference Poplack, Zentz and Dion2012b, p. 250), which provides “a more penetrating characterization of grammatical structure” (Erker & Guy, Reference Erker and Guy2012, p. 546). Thus, following the variationist comparative method (Poplack & Meechan, Reference Poplack and Meechan1998), we determine the linguistic conditioning of yo expression in the NMSEB Corpus, and then compare this conditioning with the structure of variability in monolingual benchmarks for Spanish and English.

3.2 Data: Circumscribing the variable context

The first step in determining the linguistic conditioning of yo involves defining the variable context, the sum of contexts in which speakers have a choice between variants (Labov, Reference Labov, Ammon, Dittmar, Mattheier and Trudgill2005). It is only by accounting for all non-occurrences as well as occurrences of the phenomenon of interest that we can establish the factors influencing speaker choices (Labov, Reference Labov1972, p. 69).

We began by extracting all tokens of finite Spanish verbs with (expressed and unexpressed) 1sg subjects produced by our participants (N = 2,324).Footnote ³ This included tokens where the pronoun and verb were separated by intervening material (N = 32), namely an adverb such as también “also”, nunca “never”, quizás “maybe”, nomás “only, just”. We also included the one token in the data of apparently emphatic yo sí “yes”, as yo expression with sí is variable in these data. Cases of unmarked contrast have been excluded in past studies as presumed obligatory contexts (e.g. Silva-Corvalán, Reference Silva-Corvalán, Fernández, Menéndez, Samper, Gutiérrez Araua, Vaquero and Hernández2003, p. 850), but it has also been noted that there is variability in such environments (Amaral & Schwenter, Reference Amaral, Schwenter and Eddington2005; Otheguy, Zentella & Livert, Reference Otheguy, Zentella and Livert2007, pp. 775–776). Further, without a clear operationalization of contrast, there is a great risk of circularity, whereby tokens are interpreted as contrastive because of the presence of yo, and then yo is described as a marker of that contrast (for operationalizations and tests of contrast, see Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2012, pp. 738–741; Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2014, pp. 367–371).

We then implemented the following exclusions. We excluded the notably small number of tokens (N = 9) that were contextually ambiguous, of which only two can be considered genuinely ambiguous. In four cases the identity of the subject makes no difference to the interpretation of the event (what Ranson, Reference Ranson1991, p. 145, refers to as “person irrelevant” and Ono & Thompson, Reference Ono and Thompson1997, p. 488, describe as the “referent” being left “open”). One such example is seen in (4), where the subject of tenía que meter el dedo could mean “I had to stick in my finger”, but it is equally possible that it means “one had to”. There are two tokens where the ambiguity arises because the speaker doesn't complete the utterance, and another because the surrounding speech is unclear.

1. (4)

   

We excluded the substantial number of postverbal tokens of yo (16% of expressed subjects in the data, 108/664), as postverbal subject pronouns are unlikely to follow the same constraints as preverbal ones (see Silva-Corvalán, Reference Silva-Corvalán2001, p. 165). (On the patterning of first singular subject position in this corpus, see Benevento & Dietrich, published online, January 14, Reference Benevento and Dietrich2014.) Also excluded were cases where the verb appeared in a distinct IU from the pronoun (N = 30) as it is not always possible to tell in such cases whether this pronoun should be considered the subject of the verb (illustrated in (5), where yo precedes a subordinating conjunction and there is also truncation) (see Otheguy & Zentella, Reference Otheguy and Zentella2012, p. 236). Other cases of expressed subjects in rare configurations (N = 5) were excluded (for example yo quizá como que tengo suna m- -- “I maybe kind of like have a --” [06 El túnico 0:35:08]).

1. (5)

   

Several non-variable contexts were also excluded. In identifying such contexts, we do not ask whether variability is theoretically possible with any given verb, but rather “we formulate . . . broad definitions of clausal and lexical types where variability is low enough to disqualify them from the study” (Otheguy et al., Reference Otheguy, Zentella and Livert2007, p. 776). The non-variable contexts we identified for these data include wh-questions, an environment where preverbal subjects do not occur in this variety (N = 27, e.g. qué te puedo contar? “what can (I) tell you?” [13 La acequia, 0:01:27]) (see Silva-Corvalán, Reference Silva-Corvalán, Amastae and Elías Olivares1982, p. 103); subject relatives (N = 2, e.g. yo era la única que Ø no sabía arrear “I was the only one who didn't know how to drive” [06 El túnico, 0:52:02]) (see Otheguy & Zentella, Reference Otheguy and Zentella2012, p. 246; Silva-Corvalán, Reference Silva-Corvalán, Amastae and Elías Olivares1982, p. 103); and fixed expressions (N = 6, e.g. ahí voy “coming” [13 La acequia, 0:00:19]).

These protocols leave us with 2,220 tokens for analysis.

4.2 Coreferential 1sg subject priming

Structural priming, the tendency to repeat the same structure or form, has been observed in both community-based studies of natural speech and in psycholinguistic experiments. The priming effect found here takes into account the realization of the speaker's most recent reference to themselves as subject, what we call coreferential 1sg subject priming. This effect is illustrated in the following two examples: in (8) the previous coreferential 1sg subject and the target token are both realized pronominally, and in (9) both are unexpressed.

1. (8)

   

2. (9)

   

Coreferential 1sg subject priming in a Human Switched Reference context is illustrated in (10), where there are two intervening human subjects (Nancy and they) between coreferential mentions.

1. (10)

   

It has been found that priming is stronger “production-to-production” than “comprehension-to-production” (Gries, Reference Gries2005, p. 374, and references therein), in other words, speakers are primed more by themselves than by their interlocutor. Therefore, we consider here only coreferential 1sg mentions. Coreferential mentions across quotative contexts, as in line 2 in (11), were coded separately as such, since it is not known how quotation may affect priming (N = 245).Footnote ⁵

1. (11)

   

How far back from the target can the prime be? Measures of distance between prime and target for other variables have included elapsed time and number of words or parsing units (modeled logarithmically in large-scale corpus studies, see Gries, Reference Gries2005, p. 120; Szmrecsanyi, Reference Szmrecsanyi2005, p. 371). Here we used clauses as a measure (following Weiner & Labov, Reference Weiner and Labov1983), reasoning that for a syntactic variable, such as subject expression, finite verbs “interfere” more than arbitrary material in the strength of a previous subject prime.

We operationalized the clause as any finite verb, in a main or subordinate clause, with a referential or non-referential subject, produced by any speaker. Set aside were fixed expressions that are not fully clausal, which had to be determined for both Spanish and English in this bilingual corpus.

In Spanish, not counted as clauses were fixed impersonal expressions (such as es que “it's that”, quién sabe “who knows”, hay veces que “sometimes”) as well as discourse markers (such as non-literal uses of anda/e(n) “go ahead”, ¿sabes qué? “you know what?”, mira “look”, as in (12)) We did, however, count as clauses instances of (yo) no sé “I don't know” and yo creo “I think” (as do Otheguy & Zentella, Reference Otheguy and Zentella2012, pp. 234–235) because, although on frequency measures they qualify as particular constructions, they are not entirely autonomous of other instances of yo + (cognition) verb by the measure of shared constraints, and do still behave as clauses (Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2012, pp. 738–741) (see Torres Cacoullos & Walker, Reference Torres Cacoullos and Walker2009).

1. (12)

   

In English, non-clauses include several discourse formulae. Classified as formulaic were 1sg subject–present tense cognition verb collocations I mean, I guess, I think, I remember, I'm sure, I know, I don't know when they occur as parentheticals in the clause (I guess in (13)), or on their own in an IU, that is, as prosodically independent (I don't know in (14)), as opposed to when they occur introducing clausal material in the same IU (Travis & Torres Cacoullos, Reference Torres Cacoullos and Travis2014, pp. 363–366). Also skipped over in the clause count as formulaic were expressions such as you know (except where it occurs with a complement clause), you know what (on its own in an IU), and go figure.

1. (13)

   

2. (14)

   

Potential primes were counted within four clauses from the previous coreferential mention. Coreferential 1sg subject priming with this cutoff applies to approximately two-thirds of the data.

Overall, as shown in Table 1, previous realization as a pronoun (yo or I) most strongly favors occurrence of yo, which is expressed almost three times as often in this context than with a previous unexpressed 1sg subject (31% vs. 11%). When we examine this more closely, we find that the priming effect dissipates with increasing distance (see Travis, Torres Cacoullos & Kidd, to appear). At a distance of zero intervening clauses (that is, in coreferential contexts), the rate of expression in the context of a previous yo (seen in (8) above) vs. a previous unexpressed 1sg subject (seen in (11) above) is more than four times greater (35%, N = 148 vs. 8%, N = 551); at a distance of one or two clauses it is three times greater (40%, N = 68 vs. 13%, N = 226); and at a distance of three or four clauses, it is less than twice as great (45%, N = 38 vs. 24%, N = 86).

Note that in Table 1, previous realization as a pronoun includes both yo (N = 254) and I (N = 208). We will explore the cross-language priming effects in relation to code-switching in Section 7.3.

4.3 Cognition verbs and particular constructions

As to the remaining constraints in the multivariate analysis, for Semantic class of verb, the strong favoring effect of cognition verbs observed here replicates that reported in numerous previous studies across different varieties of Spanish (e.g. Bentivoglio, Reference Bentivoglio1987, p. 60; Enríquez, Reference Enríquez1984, p. 240; Silva-Corvalán, Reference Silva-Corvalán1994, p. 162; Travis, Reference Travis2007, pp. 116–117). Frequent among cognition verbs are the 1sg subject–present tense collocations yo creo “I think”, with a notably high yo rate of 87% (94/108) (also reported to favor yo in other varieties (Erker & Guy, Reference Erker and Guy2012, p. 536, 539; Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2012, pp. 739–741), and (yo) no sé “I don't know” with a yo rate of 41% (56/138).

Second, yo expression is disfavored by reflexive me-marked verbs, most frequent among which is the cognition verb acordarse “remember” (tokens of which account for 40%, 125/314, of the reflexive verb data). Though the overall rate of expression for acordarse (12%, 15/125) corresponds to the average for reflexive verbs, the yo rate may tend to be lower in the negative polarity collocation no me acuerdo (at 8%, 4/49) than in affirmative me acuerdo (at 14%, 10/69) pointing again to lexically-particular constructions with distinctive tendencies in subject expression rates.

Finally, considering the Tense-aspect-mood constraint, yo is disfavored with Preterit (past perfective) verb forms. This is as predicted, and though it has been interpreted as an effect of ambiguity of person–number morphology (Preterit suffixes distinguish grammatical persons), this interpretation is not applicable for these data as the Present (which also distinguishes grammatical persons) patterns identically to the Imperfect (which is morphologically ambiguous), with relatively high yo rates of 32% (287/908) and 30% (152/515), respectively.

Having established the patterning for the NMSEB Corpus, we turn now to the comparisons, first with Spanish, then English, benchmarks.

5.1 The equivocality of overall rates

Spanish subject expression has been enlisted as a candidate for grammatical convergence under the assumption that Spanish and English subject pronouns are similar enough that bilinguals may come to use Spanish subject pronouns at a higher rate in tandem with the overwhelming rate of subject pronoun use in English. But a fundamental problem with confining tests of convergence to comparisons of overall rates of variable occurrence is the question of the threshold for a “high(er)” rate, given the susceptibility of rates to genre, topic or other extragrammatical situational considerations, as well as disparities in overall rates between dialects.

Figure 1 depicts rates of 1sg Spanish subject expression in the NMSEB Corpus and as reported in eight other studies. Note the wide range here, where the highest, at 50%, doubles the lowest, at 25%. Also note that the NMSEB Corpus, with a rate of 25%, is among the lowest.

Figure 1.

Rates of expression of Spanish 1sg subject pronoun yo across studies. Sources: Mexico City, Mexico (Lastra & Butragueño, Reference Lastra and Butragueño2015); Madrid, Spain (Enríquez, Reference Enríquez1984) and Santiago, Chile (Cifuentes, 1980–1) as reported in Silva Corvalán (1994, p. 153); Castañer, Puerto Rico (Holmquist, Reference Holmquist2012, p. 211); Caracas, Venezuela (Bentivoglio, Reference Bentivoglio1987, p. 36); Cali, Colombia (Travis, Reference Travis2007, p. 113); San Juan, Puerto Rico (Cameron, Reference Cameron1994, p. 31); Puente Genil, Andalusia (Ranson, Reference Ranson1991, p. 135, 138).

On the one hand the lower rate of subject expression in the NMSEB Corpus is dissonant with convergence in terms of the prediction that the overall rate of yo would be pulled up by virtue of bilinguals associating it with English I. On the other hand, it could be consonant with convergence, if convergence takes the form of bilinguals’ loss of constraints favoring expressed subjects (see Silva-Corvalán, Reference Silva-Corvalán1994, pp. 153–162). The comparison of overall rates of use across dialects and studies is, as anticipated, unrevealing as to change, contact-induced or otherwise.

Crucially, no study has identified differences in the linguistic conditioning of subject expression in Spanish. Regardless of overall rate differences, the same probabilistic constraints on subject expression in Spanish are consistently reported, including in comparisons both across dialects (e.g. Cameron, Reference Cameron1993, Reference Cameron1994) and genres (Travis, Reference Travis2007). We therefore turn to the linguistic conditioning, and compare the constraints on subject expression in bilingual New Mexican Spanish and non-contact varieties, in order to corroborate grammatical similarities or differences.

5.2 Shared probabilistic constraints across varieties of Spanish

Table 2 depicts the constraints on variable Spanish subject expression in the NMSEB Corpus and as reported in four other studies. This shows that the linguistic conditioning of variable yo subject expression for New Mexican Spanish–English bilingual speakers (detailed above in Table 1) replicates that found in numerous studies across Spanish varieties. As indicated by the arrows, Spanish 1sg pronominal subjects in both the NMSEB Corpus and non-contact varieties are favored in non-coreferential contexts, by preceding pronouns, and with cognition verbs; they are disfavored with reflexive-marked and Preterit verb forms.

Table 2.

Linguistic constraints on 1sg Spanish subject expression across studies.

Sources: Cali (Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2012, p. 726); Caracas (Bentivoglio, Reference Bentivoglio1987, pp. 36, 60); San Juan & Madrid (Cameron, Reference Cameron1994, pp. 32, 38, 40); New York Newcomers (Otheguy & Zentella, Reference Otheguy and Zentella2012, pp. 163–165).

* Not tested or not reported.

^a For all singular grammatical persons, Cameron (Reference Cameron1994, pp. 39–40) finds priming in coreferential contexts and a weaker effect under switch reference.

In short, the constraints on subject expression in the bilingual data of the NMSEB Corpus do not display differences with non-contact varieties of Spanish.

6.1 Unexpressed subjects in English vs. Spanish

The overall rate of subject pronoun I in conversational English is considerably higher than that of yo in any Spanish variety, with unexpressed 1sg subjects in the vicinity of just 2% (151/~9,000) (Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2014). But given the uninformativeness of rates discussed above, a more telling difference between the two languages lies in the variable context. In a sample of (expressed and unexpressed) 1sg subjects from the Santa Barbara Corpus of Spoken American English (SBCSAE, Du Bois, Chafe, Myer, Thompson, Englebretson & Martey, Reference Du Bois, Chafe, Myer, Thompson, Englebretson and Martey2000–5), Torres Cacoullos and Travis (Reference Torres Cacoullos and Travis2014, pp. 22--23) find no cases of unexpressed I in interrogatives, relative clauses or subordinate (complement, adverbial, or if) clauses (other than ones involving coordination, as in if I go out and Ø ask for it (SBCSAE 17, Jim, line 7)). Thus, for I expression, variability is found only in declarative main clauses. Such a restriction to declarative main clauses is absent from Spanish.

Within declarative main clauses, variation between expressed and unexpressed I occurs with both non-coordinated and coordinated verbs. We see in Table 3 that in English, coordinated verbs, defined here as verbs with coreferential subjects conjoined with and, present a notably lower rate than non-coordinated verbs. A similar pattern is seen in the NMSEB Corpus for Spanish, where the subject expression rate is also lowest in coordinated verbs with y “and”, at just 5%. However, a shared pattern cannot be construed as a point of convergence in and of itself because it does not speak to whether change has taken place. This is determined through comparisons with non-contact Spanish benchmarks. A similar effect has been reported for “and” coordination across varieties of Spanish, including in Colombia (the figures for which are given in Table 3, based on the Corpus of Conversational Colombian Spanish), Mexico City (Lastra & Butragueño, Reference Lastra and Butragueño2015) and Puerto Rico (Cameron, Reference Cameron1992, p. 206)).

Table 3.

Rate of 1sg subject expression by “and” coordination and according to position in prosodic unit (Intonation Unit, IU), in the NMSEB Corpus, compared with non-contact English and Spanish.

^a English (Santa Barbara Corpus of Spoken American English) and non-contact Spanish (Corpus of Conversational Colombian Spanish) from Torres Cacoullos and Travis; in English sample I rate is an artificial 34%, with two tokens of expressed I extracted for each unexpressed.

^b Excluded from total are tokens of an initial adverb with an unexpressed subject (e.g. ya no me acuerdo “I don't remember anymore” [03 La Marina, 0:37:00]) (N = 95) and tokens of o “or” (N = 7).

Most striking in English is the prosodic constraint that applies to non-coordinated verbs. As seen in Table 3, with non-coordinated verbs, unexpressed 1sg subjects occur virtually only in absolute initial position of the IU, as in (15), a constraint that has been intuited by a number of scholars (e.g. Napoli, Reference Napoli1982) as an application of a more general phonological process of (variable) “left-edge deletion” (Weir, Reference Weir2012). Thus, outside “and-” coordination, the variable context must be restricted to prosodic-initial position in English.

1. (15)

   

We find no such initial-position prosodic constraint for Spanish subject expression. Prosodic position thus qualifies as a “conflict site” (Poplack & Meechan, Reference Poplack and Meechan1998, p. 132), according to which the structure of variability differs for Spanish and English 1sg subject expression, beyond patent overall rate differences.

If bilingual Spanish 1sg subject expression has been influenced by the English prosodic constraint, with non-coordinated verbs we should see higher rates of unexpressed subjects where unexpressed subjects occur in English, namely in IU-initial position (as in line 1 in (6) above). That is, we should see lower rates of expressed subjects in IU-initial position than in non-IU-initial position. (Non-IU-initial position includes verbs preceded by conjunctions such as cuando “when”, pero “but” (as in (9) above, line 3), adverbs, fillers, or other (more substantial) material (as in (12) above).

Instead, returning to Table 3 we see that, whereas in English (SBCSAE) the highest rate of I expression is found in non-IU-initial position, in the NMSEB Corpus yo expression is highest in IU-initial position, as is also the case in the non-contact variety of Spanish.

Where we do find English parallels with Spanish is in terms of cross-linguistic tendencies for priming and subject continuity. Coreferential 1sg subject priming has an effect: though rare, when unexpressed I does occur, it tends to do so in clusters. The subject continuity effect in English – a higher rate of expression in Switched Human Reference contexts – is bound to unexpressed-to-unexpressed priming, which tends to occur in a coreferential context, thus raising the rate of unexpressed I in this context (Torres Cacoullos & Travis, Reference Torres Cacoullos and Travis2015b).

Thus, while the NMSEB Corpus and non-contact varieties of Spanish and English all share candidate cross-linguistic patterns, on the more restricted variable context in English and on the clear “conflict site” with Spanish (that of prosodic initial position), the NMSEB Corpus data demonstrate no convergence with English.

6.2 Expressed yo vs. stressed I

It could be countered, nevertheless, that, given the supremely lopsided rates of the variants, I expression is not a pertinent English variable enticing bilinguals to convergence. Alternatively, English subject pronoun stress has been interpreted as equivalent to Spanish expressed pronouns. For example, Payne (Reference Payne1997, p. 43) claims that “Spanish pronouns correspond [functionally] to English stressed pronouns (roughly speaking)”. Givón's (Reference Givón1983b, p. 17) topic accessibility continuum also suggests equivalency, placing unstressed pronouns in a language such as English at the same level as person-number agreement with unexpressed subjects in a language such as Spanish (both coding more continuous or accessible participants), and placing stressed and expressed (independent) pronouns at the same level (both coding less continuous participants).

Looking again at data from the SBCSAE, Travis and Torres Cacoullos (Reference Travis and Torres Cacoullos2014, p.373) found a rate of stressed I of 14% (163/1,133), setting aside formulaic units (discourse marker and quotative uses of I think and other collocations, as noted above) which, in general, lack stress on I. Multivariate analysis reveals similarities and differences in the linguistic conditioning of stressed I and expressed yo. The hypothesized cross-linguistic constraint of coreferential subject priming is shared: stressed I to stressed I priming is operative, such that stressed I is favored when the preceding clause subject was a coreferential stressed I, just as we observed for yo-to-yo and unexpressed-to-unexpressed priming. However, the subject continuity constraint, though congruent with referent accessibility, is configured differently, with stressed I being subject to a distance effect as opposed to the local coreferentiality effect that is relevant for expressed yo. Stress on I is not favored when there is a switch in reference from the subject of the preceding clause (even in the case of Human Switched Reference) but when the previous mention is at a distance of two or more clauses. Yo expression in the NMSEB Corpus adheres to the Spanish, not the English, pattern, with the sharpest difference in yo rate being found between contexts with none vs. at least one intervening human subject.

The strong cognition verb class effect on yo is absent for stressed I, such that lexical types that might be classified as cognition verbs show disparate behavior. I think (the putative translation counterpart of yo creo) displays a stressed I rate close to the average (15%, 23/157), but I don't know (+ clause) shows a significantly higher rate of stress (20%, 12/59). Other cognition verbs (I guess and I mean) are virtually never stressed (0/36 and 1/188 respectively). In Spanish, on the other hand, cognition verbs tend to behave as a class, which favors yo expression (Travis & Torres Cacoullos, Reference Travis and Torres Cacoullos2012, pp. 734–742). The NMSEB Corpus is again aligned with monolingual Spanish, with cognition verbs favoring expressed yo. Besides this general verb class effect, we find among the cognition verbs a yo rate that is twice as high for yo creo “I think” (87%, 94/108) as for (yo) no sé “I don't know” (41%, 56/138), also conforming to the monolingual Spanish pattern of high rates of yo creo while differing from the English pattern of higher rates of stressed I in I don't know, thus demonstrating distinct lexically particular constructions in the two languages as well.

In sum, as with expression of I, we find no evidence for convergence with patterns of stress on I. The bilinguals in the NMSEB Corpus evince no alteration of Spanish constraints on 1sg subject expression and no adaptation to English language particular constraints. As characterized by the linguistic conditioning of variant choice, yo in the NMSEB Corpus is grammatically similar to expressed yo in other varieties of Spanish, and different from candidate English counterparts expressed I and stressed I.

Further, recall that, despite copious code-switching in the corpus, we found no significant effect for recent use of English in the multivariate analysis (Table 1). We now turn to consider the role of code-switching.