Abstract
Aims:
Previous research on relative clause attachment has found that late Spanish–English bilinguals tend to parse both languages with a single strategy that is consistent with that of the language in which they are currently immersed. One recent investigation using an off-line measure has suggested that the same may not be true of early Spanish–English bilinguals, who did not show effects of greater exposure to English in the comprehension of ambiguous relative clauses (RCs). The present study sought to determine whether on-line RC attachment behavior in heritage Spanish was similarly unaffected by extensive exposure to English.
Design:
Forty-six monolinguals and 28 heritage speakers of Mexican Spanish participated in a self-paced reading experiment in which stimuli contained temporarily ambiguous relative clauses that were biased toward either high or low syntactic attachment.
Data and analysis:
Analysis of variance by subject and by item was performed on reading time data and on responses to post-stimulus comprehension questions.
Findings:
Statistical analyses revealed similar preferences for high attachment in Spanish among both monolinguals and heritage speakers.
Originality:
These results suggest that Spanish heritage speakers, as early bilinguals, may be less affected by ongoing exposure to English than the late bilinguals tested in prior studies.
Significance:
The outcome also suggests that, within Spanish as a heritage language, relative clause attachment may be unique compared to other aspects of sentence processing in that it shows little influence from English.
Keywords
Heritage speakers are known to show differences when compared to monolingually raised users of the minority language, but have also been shown to have a number of advantages over second language (L2) users (i.e. late bilinguals). On the one hand, they frequently exhibit language knowledge and behavior in the heritage language that diverges from that of other native speakers (i.e. monolinguals, especially those whose schooling was in the language) due to the factors of incomplete acquisition, attrition, and cross-linguistic influence (Benmamoun, Montrul, & Polinsky, 2010), as well as language contact-induced change that occurs at the level of the speech community. At the same time, there are also many advantages that heritage speakers can have over late bilinguals. For instance, phonology seems to be the area of greatest strength and heritage bilinguals tend to resemble fully competent native speakers very closely, particularly on measures of perception, performing more similarly to monolinguals than do L2 learners (e.g. Au, Knightly, Jun, & Oh, 2002; Knightly, Jun, Oh, & Au, 2003). Within morphosyntax, early exposure appears to be especially beneficial in the acquisition of gender agreement, for example, another area in which heritage speakers can outperform L2 learners (Montrul, Davidson, de la Fuente, & Foote, 2014). Heritage bilinguals are also more like native speakers educated in the minority language than are L2 learners when it comes to interface phenomena like the interpretation of null and overt pronouns (Keating, VanPatten, & Jegerski, 2011). Thus, while much of the research on heritage language has focused on describing areas of divergence, the identification of unique advantages is also a fundamental goal (see Montrul, 2013, for a more in-depth discussion of this point).
A fair amount of research has examined the linguistic competence of heritage bilingual users of Spanish, but thus far there has been relatively little focus on identifying and describing performance issues that might arise during on-line processing. 1 Thus far, some of the research that has begun to explore heritage language processing suggests that this may be another area of relative advantage for heritage speakers, who tend to resemble monolingual native speakers quite closely, more so than L2 learners. For instance, monolingual-like processing by heritage bilinguals was observed by Montrul (2006) in a study of the syntax and semantics of unaccusative and unergative verbs, in which heritage Spanish speakers and monolinguals showed a similar pattern of response times in a probe recognition task (although the heritage responses appeared to be slower overall). Montrul’s (2010) study of Spanish clitics and word order also found that heritage bilinguals were quicker than L2 learners to provide a correct response to a visual sentence interpretation task, in addition to showing higher accuracy in their responses. Both heritage bilinguals and L2 learners are dominant in the majority language and can therefore show signs of cross-linguistic influence in the minority language, so such evidence suggests that the developmental context of heritage language acquisition – including an earlier age of onset of bilingualism and more extensive exposure to input over time as compared to adult L2 acquisition (Montrul, 2013) – must render heritage speakers better able to suppress or otherwise overcome at least some types of cross-linguistic influence from English.
The present study targets a type of sentence processing that has previously been shown to be unusually susceptible to cross-linguistic influence from the dominant language among bilinguals. The interpretation of ambiguous relative clauses (RCs) like Miguel argued with the boss of the clerk who had a very long mustache has long been of interest in sentence processing research on monolinguals, is known to present a special challenge in the L2 in late bilingualism (Felser, Roberts, Gross, & Marinis, 2003; Papadopoulou & Clahsen, 2003; but cf. Frenck-Mestre, 2002), and appears to be a point of pronounced effects in L1 attrition in late bilingualism (Dussias, 2003, 2004; Dussias & Sagarra, 2007). In addition, monolingual Spanish typically shows one of the two possible processing preferences, high syntactic attachment of the RC to the boss, while English tends toward the opposing preference, low syntactic attachment to the clerk. Relative clause attachment is therefore an ideal experimental target in the study of Spanish–English bilingual populations like heritage speakers, because cross-linguistic influence might be observed. The low attachment seen in monolingual English is also believed to reflect a universal or default parsing strategy (Frazier & Fodor, 1978) on which bilingual processing might rely in order to maximize efficiency when faced with limited cognitive resources. Thus, either factor or both in conjunction might lead heritage bilinguals to employ a low attachment strategy in Spanish, so there is good reason to make such a prediction. Interestingly, the results of the present study will show just the opposite: a tendency toward high attachment in Spanish that did not differ from that of a monolingual comparison group.
Relative clause attachment in Spanish
RC attachment is an interesting and well-documented area of sentence processing that is known to vary cross-linguistically. In sentences like (1), taken from Cuetos and Mitchell (1988), the RC is ambiguous in that it could refer to either of the noun phrases (NPs) in the first clause of the sentence, the servant or the actress. When asked to interpret such stimuli, however, groups of native speakers of some languages show a tendency toward high attachment or NP1 attachment (typically in the range of 60 to 70%), understanding that it was the servant who was on the balcony, while speakers of other languages show a preference toward low or NP2 attachment, indicating instead that it was the actress that was on the balcony. Such strategies also appear to guide on-line processing, typically observed when RC sentences like (1) are disambiguated (e.g. by adding the phrase con su marido “with her husband” at the end) and participants show longer reading times or P600 effects while processing the dispreferred type of attachment.
(1) Alguien disparó contra el criado de la actriz / que estaba en el balcón / con su marido.
“Somebody shot the servantM of the actress / who was on the balcony / with her husband.”
Spanish and English differ with regard to the RC attachment preferences of native speakers. After initial research focusing on English showed a consistent preference for low attachment (although a number of studies have also shown no reliable preference in English, particularly with on-line measures; Carreiras & Clifton, 1993), it was proposed that this tendency reflected a universal principle known as late closure (Frazier & Fodor, 1978), but a study of Spanish brought into question the universality claim with evidence of a high attachment preference in Spanish. Cuetos and Mitchell (1988) found the high attachment strategy both off-line, as measured by an experimental questionnaire, and in an on-line clause-by-clause self-paced reading task, using stimuli such as that illustrated in (1). Subsequent work has served to classify a number of languages and varieties as preferring either high or low attachment, in addition to corroborating the initial proposal that Spanish prefers high attachment with on-line evidence from additional self-paced reading (Gilboy, Sopena, Clifton, & Frazier, 1995), as well as eye-tracking (Carreiras & Clifton, 1999) and event-related potentials (ERPs) (Carreiras, Salillas, & Barber, 2004). To our knowledge, all published work on Spanish RC attachment to date has targeted Peninsular Spanish, so it is not known whether variation might exist among different varieties of Spanish with regard to RC attachment preferences.
Theoretical accounts as to why RC attachment preferences might vary cross-linguistically are numerous, and widespread agreement has not yet been reached as to the specific linguistic and cognitive mechanisms underlying the behavior. For instance, linguistic tuning suggests that processing is experience-based, so the parser will tend to resolve a given ambiguity one way or the other based on which parse has been encountered more frequently by the individual in the recent past (Brysbaert & Mitchell, 1996). Construal (Frazier & Clifton, 1996) proposes that RCs are non-primary phrases exempt from late closure and that the default strategy is actually high attachment, because NP1 is the one in focus with regard to the discourse, but that this tendency is trumped by the syntactic availability of another form to express the possessive relationship between two NPs in languages that also have the unambiguous Saxon genitive in addition to the potentially ambiguous Norman one (e.g. the actress’ servant and the servant of the actress, respectively). The implicit prosody hypothesis claims that implicit prosody is represented during silent reading and this can, much like the overt prosody that occurs in spoken language, influence the placement of syntactic boundaries and thereby influence sentence processing (Fodor, 1998). Attachment-binding suggests that RC attachment involves a competition between two simultaneous and conflicting processes – the syntactic tendency of late closure and the discourse tendency toward attaching to the discourse-prominent NP1 – and that the syntactic principle is more likely to prevail in languages without overt relative pronouns, like English (Hemforth, Konieczny, & Scheepers, 2000). Finally, the recent pseudo-relative-first hypothesis takes into account the nearly perfect correlation of high attachment with the occurrence of pseudo-relatives across about 20 different languages and proposes that it is a default preference for an RC to be read as a pseudo-relative that motivates high attachment where possible (Grillo, 2012; Grillo & Costa, 2012). Thus, although it is not yet entirely clear what precise mechanisms motivate RC attachment, there is consensus that it involves syntax to some degree and that it may be linked to other aspects of cross-linguistic variation, including aspects of language that go beyond the syntax. For the purposes of the present study, the most important points are that RC attachment is syntactically complex, that it exhibits cross-linguistic variation – and in particular that Spanish and English differ in this regard – and that it has been shown to be subject to interesting cross-linguistic influence among bilinguals, so it stands to reason that the phenomenon may prove interesting in the study of heritage languages.
Relative clause attachment in bilinguals
As with any case of clear cross-linguistic variation, RC attachment provides fertile ground for examining the fundamental question in bilingualism as to whether there is cross-linguistic influence between a bilingual’s two languages. Based on investigations conducted so far, RC attachment preferences appear to be one aspect of sentence processing that is particularly susceptible to such influence. In fact, the similarity between bilinguals’ two languages has been generally so striking as to lead researchers to a linguistic tuning (Brysbaert & Mitchell, 1996) account of bilingual RC attachment, in which a single processor operates on both languages, according to the relative amount of recent exposure it has had to each.
Initial off-line evidence along these lines was found by Fernández (2002), who investigated the on-line and off-line RC attachment preferences of two different groups of Spanish–English bilinguals in New York City. The results of a pen-and-paper questionnaire revealed that an English-dominant group preferred high attachment 48% of the time in English and 55% with Spanish, while a Spanish-dominant group exhibited a 73% high attachment strategy in both languages. The author concluded that each bilingual group tended toward a single processing preference that was most similar to that of monolingual speakers of their dominant language (assuming that the neutral attachment tendency sometimes observed on-line in English could also occur occasionally with off-line measures). The question of how such strategy might play out during on-line sentence processing still remained, however, as the on-line results from self-paced reading for this study were difficult to interpret, showing low attachment with both English and Spanish monolinguals and no reliable attachment preferences among the bilingual participants. This led the author to conclude that the specific methodology was inadequate to measure the on-line behavior of interest.
Evidence that a single, language-independent processing strategy can also apply during on-line processing was found by Dussias (2003), who examined the on-line and off-line processing of ambiguous RCs by two groups of late bilinguals residing in the United States. The participants were L1 English–L2 Spanish and L1 Spanish–L2 English, all with a very high level of proficiency in their L2. On a written questionnaire, both groups preferred low attachment in both their languages. The same low attachment strategy was also evident during self-paced reading in Spanish among both groups, which shows that the native speakers of Spanish who had acquired English in adulthood had acquired a native-like strategy in their L2 and that this strategy seemed to have in turn affected the L1 to the extent that RCs in both languages were processed in a similar fashion. Dussias suggested that this influence of the L2 on the L1 is due to long-term and ongoing exposure to the L2, as with linguistic tuning, given that the native Spanish speakers had resided in the U.S. for an average of 7.5 years at the time of testing. A subsequent eye-tracking study (Dussias, 2004) showed that the same immersion effect could be present among a similar group of late Spanish–English bilinguals who had resided in an English-speaking country for an average of only 3.7 years. Another eye-tracking study in this line of research (Dussias & Sagarra, 2007) corroborated the exposure-based account with yet another long-term L2 exposure group (mean 7.1 years L2 English immersion) that showed an L2-like low attachment strategy in both of their languages, as well as a short-term exposure group (mean 8.5 months L2 English immersion) that showed an L1-like high attachment strategy in both of their languages. Also potentially relevant to the exposure account was that the long-term exposure group was still immersed in the L2 at the time of testing, while the short-term exposure group resided in Spain.
To summarize thus far, there is evidence that bilinguals tend to employ a single processing strategy with RC attachment and that this strategy reflects the language history of the individual, in particular bilingual language dominance and language exposure, two variables that are perhaps inevitably entwined. The bilingual participants in the abovementioned research studies were primarily late bilinguals, meaning they had acquired the second language in adulthood, or, in the case of the participants in the Fernández (2002) study, near the cusp between early and late bilingualism, having begun to acquire the L2 in late childhood or early adolescence. Relatively less is known about the RC attachment preferences of early bilinguals such as heritage speakers, who also have a clear asymmetry between their languages with regard to dominance and exposure, but typically acquire both of their languages earlier in life.
Although evidence to date is scarce, there has been one off-line study that examined the comprehension of RCs among heritage bilinguals. Jegerski, VanPatten, and Keating (2014) looked at the attachment preferences of two types of Spanish–English bilinguals, a group of heritage bilinguals who had acquired both languages in early childhood and a group of late bilinguals who had all acquired Spanish from birth and English as adults. The late bilinguals patterned with participants from previous research, showing similar behavior in English and Spanish that was in this case a neutral preference for attachment (i.e. English-like, based on previous research). This finding was also interpreted as consistent with Dussias’ (2003, 2004) research on bilinguals of the same profile – native Spanish speakers who acquired English in adulthood and were living in the U.S. at the time of testing – in which there was both evidence that the low attachment preference had been acquired in the L2, English, and evidence that this preference had replaced the typical high attachment preference in the L1, Spanish. The heritage bilinguals in Jegerski et al. (2014), on the other hand, employed distinct attachment strategies in both of their languages, with a clear high attachment preference in Spanish and neutral attachment in English. Thus, there is reason to believe that the advantages of early bilingualism might be at play with RC attachment, if early bilinguals are less likely than late bilinguals to show effects of ongoing exposure to English when processing RCs in Spanish. It remains to be seen whether this result can also be seen during on-line sentence processing in Spanish.
The present study
The present study addressed the following research questions:
Do monolingual speakers of Mexican Spanish show a high attachment preference during on-line processing of temporarily ambiguous RCs in Spanish?
Do Spanish heritage speakers of Mexican descent who have extensive immersion exposure to English also show a high attachment preference during on-line processing of temporarily ambiguous RCs in Spanish?
Participants
The monolingual Spanish speakers included 46 degree-seeking undergraduate students (24 females; age range 18–26, M = 20.7) enrolled at the University of Guanajuato located in the city and state of Guanajuato, Mexico. They were all born and raised in central Mexico, had lived in Mexico their entire lives, and pursued majors in fields unrelated to the language sciences. One participant from an original pool of 47 was removed for having childhood exposure to English.
The Spanish heritage speakers consisted of 28 undergraduate students (23 females; age range 18–26, M = 20.2), including 27 degree-seeking students at San Diego State University and one non-degree-seeking student from San Diego Mesa College. Of the 28 participants, 25 were born in the U.S., three were born in Mexico but moved to the U.S. by age five, and all were simultaneous bilinguals, having been exposed to both Spanish and English by age five. The participants were all of Mexican descent and had at least one parent that was a first-generation immigrant from Mexico. Twenty-three participants from an original group of 51 heritage speakers were removed from analyses. To control for dialect, one bilingual of Guatemalan descent and another of mixed Mexican–Salvadoran descent were removed. Nine sequential bilingual participants were removed for having first immersion exposure to English after age five (ages 6–12). Next, to avoid proficiency level in Spanish as a confound and maximize comparability with previous research, the two participant groups were roughly matched in terms of global proficiency. They all took a 20-item cloze passage taken from the DELE (Diploma of Spanish as a Foreign Language) exam, the same passage used in studies conducted by Montrul and colleagues (e.g. Montrul, 2005; Montrul, Foote, & Perpiñán, 2008), and 12 heritage participants were removed for scoring below the monolingual range of 15 to 20. Finally, one participant was removed for not performing the tasks according to instructions. Univariate ANOVAs revealed no significant differences between the monolingual and bilingual participant groups with respect to age, F(1,72) = 1.109, p = .296, or global proficiency, F(1, 72) = 2.130, p = .149 (monolingual M = 17.4/20; heritage M = 16.8/20).
As is typical with heritage bilinguals, who by early adulthood have had more exposure to the majority language, English was the dominant language for the group. Table 1 displays the heritage speakers’ self-rated abilities in reading, comprehending, and speaking English and Spanish on a scale from 1 to 10. Self-ratings for each skill and language ranged from a low of 7 to a high of 10, indicating strong abilities in both languages. Between-language comparisons for each of the three skills suggest that the heritage speakers’ reading and speaking abilities were stronger in English relative to Spanish, whereas comprehension abilities in the two languages were similar.
Heritage bilingual participants’ mean self-rated proficiencies in Spanish and English (10 = highly proficient, 1 = minimal ability).
The difference refers to the mean self-rating in English minus the mean self-rating in Spanish.
Materials
The materials for the self-paced reading experiment consisted of complex sentences comprised of two clauses: a main clause that introduced two NPs of different genders, followed by an RC that modified one of the two. The RC could refer to either of the two NPs in the main clause, but was manipulated so that each RC was strongly gender-biased toward either the first or the second NP, as shown in (1). This manipulation thus yielded two stimulus conditions that varied with regard to forced attachment of the RC, which was either high or low (i.e. NP1 or NP2). 2 Each of 24 critical sentences was used to create both stimulus types, appearing in each of the two conditions exemplified in (2). Two counterbalanced presentation lists were created with each sentence appearing only once per list in either its (a) or (b) version.
(2) a. Miguel discutió con el jefe de la vendedora que tenía un bigote muy largo. High
b. Miguel discutió con la jefa del vendedor que tenía un bigote muy largo. Low
“Miguel argued with the bossM/F of the clerkF/M that had a very long mustache.”
Procedure
The native speakers residing in Mexico completed the study in a multiuse room at the Centro de Auto-Aprendizaje de Idiomas (CAADI), a self-learning language laboratory housed in the Language Department at the University of Guanajuato. The Spanish heritage speakers were tested in a language processing laboratory at San Diego State University. Participants first completed a background questionnaire followed by the self-paced reading task. The DELE cloze task was administered next. Afterward, participants completed two additional tests for a separate experiment. Most participants completed the tasks in 60–75 minutes.
The self-paced reading was administered in a phrase-by-phrase, non-cumulative moving window display using SuperLab 4.5. The first press of the button on a response pad displayed the main clause, the second displayed the RC, and the third a comprehension question. Each question was a binary choice item that tested the reader’s ultimate interpretation of the RC in the stimulus (e.g. ¿Quién tenía el bigote muy largo? a. el jefe b. la vendedora “Who had a very long mustache? a. the bossM/F b. the clerkF/M”). Each presentation list included 104 distractors in addition to the 24 critical items. Distractors tested null and overt subject pronouns and third-person direct object pronouns. Of the 104 distractors, 40 were followed by a who-type comprehension probe and 64 were followed by a question that ascertained where events in the sentence likely took place, also with two counterbalanced choices as answers. Critical items and distractors were divided into two presentation lists and each list was pseudorandomized so that items from the same experiment and condition never appeared consecutively. Participants read the items in one block and were required to take a short break after reading half of the items.
Results
For the self-paced reading task, all items were analyzed regardless of whether the post-stimulus comprehension question was answered correctly. Data points less than 200 ms were removed from the reading times (RTs) of each clause given the unlikelihood that a clause comprised of six to seven words could be read in a fifth of a second. When this procedure resulted in the removal of data from the first clause, the corresponding data for the second clause were also removed to ensure that RTs in the critical region were not affected by premature advancement of the first clause before it was read entirely. Likewise, data pertaining to comprehension questions that followed such items were also removed. In addition, RTs in critical main clauses that were 10,000 ms or greater were discarded given that such slow RTs were likely due to variables other than those under investigation, such as distraction or fatigue. Combined, these data trimming procedures resulted in the removal of 2.7% of the RTs. The descriptive statistics for the self-paced reading data are given in Table 2.
Mean reading times (in ms) for clause 2 by group and stimulus condition.
NP: noun phrase
Given that reaction time data are positively skewed, the RT data were logarithmically transformed prior to statistical analysis. The transformed values were all within 3 SDs of the mean for each participant and item in each condition and therefore were not trimmed further. Data starting at the second clause are of greatest interest to the present study given that it was this region that contained the RC and the disambiguating segment. Data from the second clause averaged by subject and by item were submitted to mixed two-way ANOVAs (group × attachment bias), which revealed a significant main effect of attachment bias, F1(1, 72) = 39.229, p < .001; F2(1, 46) = 16.165, p < .001. This reflects that RCs biased in favor of high/NP1 attachment were read significantly faster than those biased in favor of low/NP2 attachment. Bias did not interact with group, F1(1, 72) = 1.759, p = .189; F2(1, 46) = .345, p = .560, nor was there a main effect of group, F1(1, 72) = .646, p = .424; F2(1, 46) = 1.407, p = .242. In light of the unequal group sizes, we conducted Levene’s test to ensure homogeneity of variances in the two samples. The test was met in analysis conducted by participants (high/NP1: F(1, 72) = .503, p = .480; low/NP2: F(1, 72) = .749, p = .390) and by items (high/NP1: F(1, 46) = .058, p = .811; low/NP2: F(1, 46) = 2.256, p = .140).
Post-stimulus comprehension questions that were skipped and those that were answered in greater than 10,000 ms were excluded from analysis, as were questions affected by incomplete reading of the preceding stimulus. Comprehension questions answered incorrectly were retained in analysis of response times. Total data loss amounted to 5.3%. The descriptive statistics for the comprehension question data are given in Table 3.
Mean % target responses and RTs (in ms) for comprehension questions by group and stimulus condition.
Note. The “target response” was consistent with the bias of the stimulus.
RT: reading time; NP: noun phrase.
The analyses of the responses to the post-stimulus questions yielded a significant main effect of bias, F1(1, 72) = 19.583, p < .001; F2(1, 46) = 29.685, p < .001, which reflects the fact that responses were more often consistent with the gender bias of stimulus items that were biased in favor of high attachment. There was no main effect of group, F1(1, 72) = 3.277, p = .074; F2(1, 46) = 1.859, p = .179, and no interaction between bias and group, F1(1, 72) = .672, p = .415; F2(1, 46) = .634, p = .430. Log-transformed response times greater than 3 SDs from participant or item means in each condition were removed from analysis, accounting for .2% of the data. The analyses of the log-transformed response times yielded no effect of bias, F1(1, 72) = 2.779, p = .100; F2(1, 46) = 1.999, p = .164, no effect of group, F1(1, 72) = .454, p = .503; F2(1, 46) = .617, p = . 436, and no bias and group interaction, F1(1, 72) = .004, p = .952; F2(1, 46) = .010, p = .920. Levene’s test for homogeneity of variances was met in the high/NP1 condition, F1(1, 72) = .248, p = .620; F2(1,46) = .536, p = .468, but was violated in the participants analysis of the low/NP2 condition, F1(1, 72) = 9.511, p = .003; F2(1,46) = .073, p = .789. Visual inspection of box and whisker plots revealed that the smaller sample (i.e. the heritage speaker group) was associated with the population with the larger variance. When the smaller sample is associated with the population with larger variances, F statistics tend to be greater, thus increasing the chance of finding significant differences in means that do not exist. However, our analyses did not generate large F values or significant main effects, which suggests that a Type I error has not obtained.
Discussion
The first research question guiding the present study sought to establish whether monolingual native speakers of Mexican Spanish followed a high attachment strategy during on-line comprehension of disambiguated RCs, as this tendency had previously been observed fairly consistently with Spanish, in several studies and using different measures, but only with speakers of Peninsular varieties (Carreiras & Clifton, 1999; Carreiras et al., 2004; Cuetos & Mitchell, 1988; Gilboy et al., 1995). The outcome of the present study suggests that the same RC attachment preference also occurs in Latin American varieties of Spanish, both on-line and off-line, at least among monolingual speakers in central Mexico. While this result is not surprising, it could not have been taken for granted a priori.
The second research question guiding the present study sought to determine whether heritage bilingual speakers of Mexican ancestry followed a monolingual-like high attachment strategy during on-line comprehension of disambiguated RCs in Spanish. The results of the present study suggest that they do, on-line as well as off-line, which is consistent with the results of the one prior study that had looked at the RC attachment preferences of heritage bilinguals but had employed only an off-line measure (Jegerski et al., 2014). This outcome from the current investigation, taken in conjunction with the off-line results of the previous work, is different from what might be predicted based on previous research, both from the perspective of heritage speakers and, more broadly speaking, from the perspective of bilingual sentence processing in general.
To begin with, the outcome of the present study is somewhat unexpected from the perspective of heritage bilingualism because the heritage participants exhibited entirely monolingual-like behavior, both in the qualitative sense, as seen in the bias toward high attachment in the RTs for the stimuli and in the accuracy rates for responses to the post-stimulus comprehension questions, and in the quantitative sense, as seen in the statistical similarity of overall reading speed and response times. Previous research has found that Spanish heritage speakers tend to differ from monolingual speakers on a wide array of grammatical knowledge and behavior, including at least one other aspect of on-line sentence processing (Keating, VanPatten, & Jegerski, 2016), even when there is some degree of qualitative similarity in processing patterns (Montrul, 2006) and even when heritage bilinguals exhibit more monolingual-like processing than do L2 learners (Montrul, 2010). It therefore seems unusual for heritage speakers to appear so completely monolingual-like, particularly with a type of sentence comprehension that is as complex as RC attachment. This discontinuity with previous research suggests that some aspect of the present study was unique. More specifically, there may have been something different about the experimental method, the participants, or the linguistic form targeted that can explain the unexpected level of monolingual-like behavior exhibited by heritage bilinguals. Let us examine each of these factors in turn.
First, the experimental method, self-paced reading, has thus far been admittedly rare in research on heritage speakers. There is, however, ample empirical evidence that the method can reveal differences between participant groups in general, including between L2 learners and native speakers (see, e.g. Jegerski, 2014, for a selective review). On the other hand, a specific concern in the study of heritage bilinguals is that they tend to have underdeveloped literacy skills, which means that reading-based experimental measures should be employed with caution because they might underestimate heritage language ability, but the monolingual-like performance of the heritage participants in the present study (i.e. qualitatively similar RT differences as well as statistically similar RTs overall) suggests that the self-paced reading method can be used with success in the study of heritage languages, at least at higher levels of proficiency. Still, the phrase-by-phrase method employed in the present study is admittedly coarse-grained, so there remains the possibility that a more fine-grained measure like word-by-word self-paced reading or eye-tracking would reveal timecourse differences between the two participant groups. Monolinguals might show evidence of a high attachment preference earlier on in the sentence than heritage bilinguals, for instance, although this would represent a quantitative rather than qualitative difference. In other words, the fundamental outcome of the present study would still hold, so it seems unlikely that the use of the self-paced reading method can explain the nonconformity with previous research.
Second, it is possible that the early bilingual participants were somehow different from the ones in most other studies of heritage speakers. It is a fairly frequent finding that there can be variation among heritage bilinguals, both according to age of onset of bilingualism (e.g. Yoon, Kim, Kim, & Montrul, 2014) and to global proficiency level (e.g. Montrul & Bowles, 2009). Age of onset of bilingualism was on average 3.1 years (SD 2.1) for the heritage speakers in the present study, so it would be difficult to argue that their unexpected monolingual-like behavior in Spanish is due to having been exposed to English later on in childhood than is typical among participants in other research. As concerns proficiency, the bilingual participants in the present study did show a monolingual-like level of proficiency, albeit on an imprecise measure with only 20 items, so perhaps they had a higher than typical level of proficiency in Spanish. As mentioned in the introduction of this paper, there was one prior investigation of bilingual RC attachment preferences that had a similar outcome to that of the present study, with only an off-line measure, but both that study and the present one were conducted in Southern California on roughly the same participant population, so the consistency of outcomes between the two studies does little in the way of clarifying this point regarding participant proficiency. It could be the case that college-aged heritage Spanish speakers in that general area tend to have higher proficiency in Spanish than in other regions of the country, perhaps because of additional ongoing exposure afforded by close proximity to the Mexican border, and that they would closely resemble monolinguals in other aspects of sentence processing as well.
Fortunately, the heritage bilingual participants from the present study were taken from the same original group as those that participated in another study that was conducted concurrently with the present one (Keating et al., 2016) so the results of both can be compared to determine whether this group is unusual among heritage speakers in that they would perform as monolinguals on any measure. Keating et al. (2016) looked at the on-line comprehension of potentially ambiguous subject pronouns by heritage Spanish speakers using the self-paced reading method. Specifically, they assessed the tendency of heritage bilingual and monolingual speakers of Mexican Spanish to link a Spanish null or overt subject pronoun in the matrix clause of a sentence with either the subject or object from a fronted subordinate clause in stimuli such as After the journalistM had lunch with the athleteM, he/Ø began writing the article (translation), as per a hypothesis of intrasentential pronominal reference known as the position of antecedent strategy (Carminati, 2002). The monolingual group showed the predicted biases for a subject antecedent with null pronouns and an object antecedent with overt pronouns, but the heritage bilinguals showed a similar strategy only with null pronouns and showed no reliable preference of antecedent for overt pronouns. Thus, there was an important difference between the heritage bilingual group and the monolingual group with regard to the processing of pronominal reference, which shows that the heritage bilingual participants that participated in the pronominal reference study and in the present one are not monolingual-like in every regard. This supports the notion that different areas of language can be differentially affected by the contextual factors that shape heritage language acquisition. In addition, it suggests that unusually high proficiency in the heritage language cannot solely account for the outcome of the present study.
Third, if the experimental method and participant population cannot explain the difference in outcomes between the present study and most previous research, then the last possibility that remains is the linguistic target, on-line RC attachment preferences. It would appear that there is something unique about this type of sentence processing that makes heritage bilinguals more likely to exhibit monolingual-like behavior in the minority language – perhaps while also employing a different monolingual-like attachment strategy in the majority language (Jegerski et al., 2014), although the off-line result for English has yet to be replicated with an on-line measure. This proposal is consistent with previous research with late bilinguals that had found RC attachment to be uniquely malleable (Dussias, 2003, 2004; Dussias & Sagarra, 2007), presumably more so than some other aspects of sentence processing.
As for what, linguistically speaking, might make RC attachment so special, further research is needed to test plausible hypotheses. In terms of competence versus performance, either or both could potentially be at play here. For instance, if, as discussed in further detail below, early bilinguals have greater inhibitory control, then maybe inhibition is more of a factor with RC attachment than with other aspects of on-line processing, including the type of pronominal reference that yielded a different result with the same participant group from the present study (Keating et al., 2016). With regard to competence, it is difficult to speculate precisely what type of linguistic knowledge heritage bilinguals might have acquired early on in Spanish that would lead them to prefer high attachment well into adulthood, given the wide range of linguistic underpinnings that have been proposed to explain RC attachment preferences, the diverse factors that can affect them, and the apparent difficulty that late bilinguals have in developing high attachment in an L2. Nevertheless, one factor that makes this type of sentence processing stand out from many others is the potentially critical role of phonology, in the form of implicit prosody (Fodor, 1998). If high attachment in Spanish can occur primarily on the basis of the phonological principles that guide prosody, then it makes sense that RC attachment might be more likely to look monolingual-like than other types of sentence processing in heritage bilingual because phonology appears to be relatively less susceptible to the incomplete acquisition, attrition, cross-linguistic influence, and language contact phenomena that can affect heritage language (Au et al., 2002; Knightly et al., 2003).
In addition to the outcome of the present study being somewhat unexpected from the perspective of heritage language research, the results are even more remarkable from the broader perspective of sentence processing among bilinguals in general. Most previous research on bilingual processing of ambiguous and disambiguated RCs had shown that a single attachment preference is used for both languages and that this strategy reflects the typical monolingual strategy of either the bilingual’s dominant language or the language to which the bilingual is most frequently exposed. Most college-aged heritage speakers of Spanish in the U.S. are unequivocally dominant in English and also receive more exposure overall to English than to Spanish, particularly in the formal and academic registers that are more print-based and more likely to contain syntactically complex language of the type in the stimuli for the current experiment. It therefore logically follows that Spanish–English heritage bilinguals would be predicted to show a single RC attachment preference in both of their languages and that the preference would reflect the low attachment strategy that is typically seen in English. One important difference between the present study and previous on-line research is that the participants were early bilinguals as opposed to late bilinguals, so age of onset of bilingualism appears to make a difference with this aspect of bilingual sentence processing.
This observation raises the question of the underlying cause for any difference due to age of onset of bilingualism. Why might the early Spanish–English bilinguals in the present study show more monolingual-like behavior than the late Spanish–English bilinguals in prior investigations, particularly in what is a minority language for the early bilinguals and presumably the stronger language for the late bilinguals? In terms of linguistic competence versus performance, it seems quite unlikely that heritage bilinguals would have stronger linguistic competence in Spanish than late Spanish–English bilinguals. The difference can therefore most likely be traced to some aspect of performance, or language processing as it plays out in real time. For instance, if both groups of bilinguals have equivalent levels of the relevant grammatical competence in both of their languages, then the realization of one RC attachment strategy or the other at a given moment could be a matter of inhibiting the opposing strategy. Such an interpretation would not necessarily be at odds with the linguistic tuning explanation favored by other investigators of bilingual processing of RCs (Dussias, 2003, 2004; Dussias & Sagarra, 2007), in which RC attachment was assumed to be most influenced by exposure among late bilinguals, if increased exposure to a language strengthens the level of activation of the corresponding attachment strategy, thereby rendering it harder to suppress. Of course, some type of stipulation would have to be appended to linguistic tuning to account for the apparent lack of exposure effects among early bilinguals, such as a tentative proposal that the basic mechanisms behind tuning would still occur, but that sufficient inhibitory control could override their effects on processing. In other words, both the early bilinguals in the present study and the late bilinguals in previous research would have a high level of activation of low RC attachment due to ongoing exposure to English, but only the early bilinguals would have sufficient inhibitory control to suppress the tendency while processing Spanish. Such an account is based on evidence that early bilingualism is associated with cognitive advantages and inhibitory control in particular (Luk, DeSa, & Bialystok, 2011).
Notwithstanding, this proposal is speculative and there is one alternative explanation for the difference between the early Spanish–English bilinguals in the present study and the late bilinguals from previous research that was suggested by an anonymous reviewer. In addition to the clear difference in age of acquisition of English between the two participant populations, it is possible that amount of ongoing exposure to Spanish also differed between the two groups and was a confounding factor. For instance, if the late Spanish–English bilinguals in the Dussias studies – mostly professionals who were pursuing graduate degrees in the U.S. – were less integrated into established Spanish-speaking communities in the U.S. than were the early Spanish-bilingual participants in the present study, then the late bilinguals might also have had less ongoing exposure to Spanish at the time of testing than the early bilinguals. On the other hand, it seems quite unlikely that late bilinguals raised and formally educated abroad in Spanish would have had less overall, long-term exposure to Spanish than early heritage bilinguals in the U.S. Still, if current exposure is more important than total lifetime exposure to RC attachment preferences, it may have been greater ongoing exposure to Spanish rather than early onset of bilingualism that minimized attrition among the bilingual participants in the present study. This question can only be resolved by further research that takes into consideration the amount and timing of exposure to Spanish experienced by individual participants, so the conclusions of the present study should be taken as tentative until such evidence is obtained through future study.
Another avenue for further investigation is to examine on-line relative clause attachment preferences among heritage Spanish speakers at lower levels of proficiency. The heritage bilingual participants in the present study were similar to the monolingual participants with regard to their preference for high attachment of RCs, but they were also similar in terms of global proficiency and overall reading speed, so the same high attachment preferences may very well not occur among participants with less advanced ability in the heritage language. If this is indeed the case, it would be interesting to see whether there is instead a clear preference for low RC attachment.
In conclusion, the somewhat nebulous phenomenon of RC attachment has again revealed some interesting tendencies in bilingual sentence processing. Rather unexpectedly, the comprehension of RCs appears to be an area of sentence processing in which heritage bilinguals can be indistinguishable from monolingual native speakers who were formally educated in the heritage language. Within bilingualism more generally, RC attachment may also prove to be an area of language comprehension in which an early onset of bilingualism confers a special advantage. Thus, the outcome of the present study underscores the notion that the early bilingualism and extensive exposure that characterize heritage language acquisition can have long-term advantages.
Footnotes
Acknowledgements
We owe special thanks to Douglas Goodwin, Lisa Conaway, and Luis González Martínez of the University of Guanajuato for their warm hospitality and generous assistance during our data collection in Mexico. Versions of this work were presented at the 2013 joint conferences on Spanish in the U.S. and Spanish in Contact with Other Languages (McAllen, TX), the 2013 Annual Conference of the American Association of Teachers of Spanish and Portuguese (San Antonio, TX), and the 2014 Symposium on Spanish as a Heritage Language (Lubbock, TX).
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by a Critical Thinking Grant in the Humanities and Social Sciences and a Microgrant for research travel awarded by the College of Arts and Letters at San Diego State University, for which we are grateful.